Escolar Documentos
Profissional Documentos
Cultura Documentos
to Climate Change
by
Predrag Prodanović
c
Predrag Prodanović 2008
The University of Western Ontario
Faculty of Graduate Studies
Certificate of Examination
Supervisory Committee
The thesis by
Predrag Prodanović
entitled:
Date
Chair of the Thesis Examination Board
ii
Acknowledgements
The person that made all this research possible is Professor Slobodan P. Simonovic, to
whom I wish to express my deepest appreciation. He has provided an exceptional level
of guidance and supervision throughout the years, and for that I am most grateful.
Thank you very much Professor.
I would also like to thank Dr. Juraj Cunderlik, a former co-worker and a good
friend, for providing assistance and guidance on hydrologic modeling and a variety
of other tasks. I owe much appreciation to Shawn Gettler, who agreed to spend a
considerable chunk of his time on various programming tasks. To the staff of the
Upper Thames Conservation Authority (Mark Helsten and Rick Goldt in particular)
I offer my thanks for numerous meetings, phone calls and data requests.
Financial support from the Natural Sciences and Engineering Research Council,
Ontario Graduate Scholarship in Science and Technology, and the Canadian Founda-
tion for Climate and Atmospheric Sciences is acknowledged.
iii
Abstract
The response of water resources systems is obtained by using meaningful and locally
applicable climate change scenarios. In this work, the scenarios are derived through
the use of a non parametric weather generator algorithm (able to produce arbitrary
long time series records of climate data using locally observed data, as well as Global
Circulation Model outputs used to drive the direction of climate change). A method
developed for post-processing the weather generator outputs (i.e., the non-paramteric
disaggregation algorithm) is able to produce rainfall time series records at arbitrary
fine time scales, thus being useful in studies looking at changes to urban water re-
sources management practice resulting from climate change. On a basin wide scale,
the inverse approach (where weather generator model outputs are linked with hydro-
logic models) is used to study the basin response, risk and vulnerability to changing
climatic conditions. Flood and low flow frequency analyses are used to quantify
changes in magnitude and recurrence intervals of critical hydrologic exposures, there-
fore providing support for future basin wide changes to water resources management
practice.
In order to test ways of responding to changing climatic conditions, an illustra-
tive integrated water resources managment model is developed in this work. The
development, water use, etc.). The concept of feedback is used in the integrated model
to couple physical and socio-economic characteristics, thereby allowing the model to
test water resources managment policy options in combination with climate change
iv
impact scenarios.
The methodology developed is applied to the study area of the Upper Thames
River basin, located in Southwestern Ontario, Canada. The main results from the
modeling work reveal that changes within the next century will bring more frequent
(and more severe) flooding conditions, with a shift to more summer flooding. Ex-
treme low flow conditions are expected to remain at their current levels, where severe
droughts periods will still be likely. Results of the integrated model simulations
revealed that availability of fresh water supplies and water use play the the most
bour weather generator modeling, intensity duration frequency curves, extreme rain-
fall analysis.
v
Contents
Acknowledgements iii
Abstract iv
List of Tables x
1 Introduction 1
1.1 Climate change impacts, adaptation and mitigation . . . . . . . . . . 1
1.2 Water resources impacts and adaption to climate change . . . . . . . 10
2 Methodology 33
2.1 Generating climatic input for impact assessment studies . . . . . . . . 35
2.1.1 Spacio-temporal storm generation . . . . . . . . . . . . . . . . 39
vi
2.3.1 Causal loop diagrams . . . . . . . . . . . . . . . . . . . . . . . 99
2.3.2 System dynamics simulation method . . . . . . . . . . . . . . 100
2.3.3 Criticism of the system dynamics method . . . . . . . . . . . 128
vii
4.1.3 Intensity-duration-frequency analysis . . . . . . . . . . . . . . 228
4.2 Inverse analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
4.2.1 Impact of changing climate on flood flows . . . . . . . . . . . 241
4.2.2 Impact of changing climate on low flows . . . . . . . . . . . . 247
5 Conclusions 293
References 314
Appendix 328
viii
B Weather generator model 331
ix
List of Tables
14 IDF for London: use of stations within a 100 km radius of London . . 236
15 IDF for London: % difference between 100 and 200 km simulations . 237
16 Critical flood exposures at the Ealing stream gauge . . . . . . . . . . 242
17 Critical flood exposures at the St. Marys stream gauge . . . . . . . . 243
x
List of Figures
1 Methodology overview . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2 Graphical representation of the synthetic storm model . . . . . . . . . 43
3 Clark’s unit hydrograph method . . . . . . . . . . . . . . . . . . . . . 74
4 Exponential baseflow recession method . . . . . . . . . . . . . . . . . 77
xi
25 Auxiliary variables of the urban business activity sector . . . . . . . . 175
26 Table functions for flood and drought . . . . . . . . . . . . . . . . . . 176
27 Level and rate diagram of the urban business activity sector . . . . . 180
28 Table functions of the business activity sector . . . . . . . . . . . . . 184
xii
51 IDF curves for London: undisturbed historic, historic and MSC . . . 233
52 IDF curves for London: historic, dry and wet WG output . . . . . . . 234
53 Flood frequency analysis under different climates . . . . . . . . . . . 244
54 Low flow frequency analysis under different climates . . . . . . . . . . 248
xiii
1
1 Introduction
Changes in the global climate will seriously impact human and natural environments.
For water resources management, changes in climatic variables are expected to signif-
icantly impact spacial and temporal distribution of water. Such changes can create
problems in the availability of water for life, and impact the region’s ability to cope
streamflow, etc.). Changes in the frequency of hydrologic extremes are expected for
many locations, possibly resulting in a combination of increased risk of spring time
flooding and reductions in summer streamflow. Such projected changes in hydrologic
extremes will have serious implications for planning, operation and design of water
resources systems.
This study focuses on the response of water resources systems to changes in cli-
mate conditions. A variety of computer simulation models are developed that (i)
generate climatic input needed for water resources impact assessment, (ii) quantify
changes in hydrologic extremes and assess risk and vulnerability of river basins un-
der climate change, and (iii) operationalize the concept of integrated water resources
management, by dynamically coupling physical and socio-economic aspects of water
management.
The climate is Earth’s most complex system, and one that is (and has been) stud-
ied extensively by many over the years. The Intergovernmental Panel on Climate
2
Statements like this are drawn from decades of scientific research by a consortium of
the world’s top climate (and other) scientists.
One effect of warming is the intensification of the world’s hydrologic system. Ob-
served trends from 1900 to 2005 reveal that precipitation has increased significantly
in eastern parts of North and South America, northern Europe, and northern and
central Asia, but declined in the Sahel, Mediterranean, South Africa and in southern
Asia. Increased intensification of the hydrologic system has caused extreme events
to change in frequency and intensity over the last century. The IPCC (2007) states
that:
• It is very likely that cold days, cold nights and frosts have become less
frequent over most areas, while hot days and hot nights have become more
frequent.
• It is likely that heat waves have become more frequent over most land
areas.
• It is likely that the frequency of heavy precipitation events (or proportion
of total rainfall from heavy falls) has increased over most areas (emphasis
in original, p. 3, topic 1).
Vincent and Mekis (2006) examine changes in various temperature and precipita-
tion indices for Canada, revealing that the period between 1950-2003 has seen more
days with precipitation, a decrease in daily precipitation intensity, and a decrease in
the maximum number of dry days. In terms of temperature, their analysis reveals
occurrence of fewer cold nights, cold days and frost days, and conversely more warm
3
nights, warm days and summer days across the country. The authors also report
that annual snowfall significantly decreased in southern Canada, and increased in
the north and north-east during the past century. Fernandes and Korolevych (2007)
conduct assessments of annual trends in actual evapotranspiration at 101 locations
in Canada, and find that evapotranspiration increased up to 0.73% per year for 81
locations, and decreased by 0.25% per year for the remaining 20 stations. Statistically
significant, increasing trends are detected for 35% of the stations, mostly in Atlantic
and Pacific coastal regions (resulting from increasing temperature, surface radiation
in the east, and increasing temperature, surface radiation and precipitation trends in
the west). For regions with year-round snow cover, warming is shifting the timing and
magnitude of hydrologic events. Whitfield and Cannon (2000) indicate earlier spring
runoff across Canada, while Schindler and Donahue (2006) show that summer flows
(May to August) in the Athabasca River have declined 20% since 1958. Vulnerability
to extended periods of drought is increasing across North America, as population
growth and economic development create more demands from agricultural, municipal
and industrial uses (Postel and Richter, 2003).
ing infrastructure design standards. Cheng et al. (2007) also perform historic trend
analyses for both rainfall and streamflow for years 1958-2002, and find that both are
increasing. Seasonal rainfall totals for each of the four river basins (Rideau, Humber,
Grand and Thames) have been increasing by about 20 mm per decade. Furthermore,
mean daily streamflow volume has been increasing over the period 1958-2002. Future
estimates of climatic conditions are made by use of statically downscaling global cir-
culation model outputs from global to local scales (outputs from five different models
are used). In terms of rainfall, Cheng et al. (2007) projects that (i) number of days
with measurable rainfall could increase by 20%, (ii) frequency of future heavy rainfall
events could increase anywhere from 25-50% during this century, (iii) seasonal rain-
fall totals (Apr-Nov) could increase by about 20-35%. For streamflow, the authors
project that (i) frequency of low flow events (< 5th percentile of the historic obser-
vations) could increase by 45-120%, (ii) frequency of future high flow events (> 95th
percentile of the historic observations) could increase by 45-90%, (iii) seasonal mean
daily streamflow volume (May-Nov) could increase by 15-35% during this century.
IPCC (2007) has also demonstrated that the warming of the climate system is
having discernible impacts on physical and biological systems. Researchers are con-
fident that hydrologic systems are already experiencing increased runoff and earlier
spring melt, and that lakes and rivers are gradually warming and experiencing lower
water quality. Researchers are also confident that warming is strongly affecting ter-
restrial biological systems for a range of species. Changes include, but are not limited
to, earlier timing of spring events (leaf-unfolding, bird migration and egg-laying), as
well as poleward migration of a variety of plant and animal species. Based on new
evidence, confidence exists that observed changes in marine and freshwater biological
systems are linked to rising water temperatures, as well as to changing ice cover,
salinity, oxygen levels and circulation in the oceans. Shifts in algal, plankton and
5
fish species in high-altitude oceans and lakes are already observed, as are a range of
changes to fish migration patterns.
Observed regional climatic changes of managed natural and human ecosystems
are more difficult to distinguish, because adaptation measures and non-climatic in-
fluences mask changes imposed by the altered climate. Therefore, observed effects of
increased temperature in these ecosystems are documented with medium confidence
only; the effects include changes to agriculture and forest management in the North-
ern Hemisphere (earlier spring planting of crops, increased disturbances of forests to
fires and pests), increased human mortality from excess heat in Europe, earlier onset
of production of pollen in the Northern Hemisphere, and shorter hunting seasons in
the Arctic.
Changes in global climate are attributed to four causes: emissions of anthro-
pogenic greenhouse gases into the atmosphere (resulting from burning of fossil fuels),
large-scale land use changes (converting forests into agricultural lands, increased ur-
banization, etc.), solar radiation (energy reaching the Earth from the Sun), and to a
lesser extent natural variations (volcanic activity for example). All three causes have
an ability (alone and in combination) to alter the energy balance in the climate system
and to change global temperature and precipitation patterns. Greenhouse gases trap
excess heat in the atmosphere (that would otherwise escape into space), thus causing
warming. Land use changes take away Earth’s ability to absorb carbon dioxide from
the atmosphere, and correspondingly influence the patterns of warming. Even small
changes in solar radiation and/or volcanic activity can significantly disrupt the energy
balance in the climate system, and consequently induce large temperature changes.
Of all the above causes, the consensus in the scientific community is that warming
experienced in the second half of the 20th century is “very likely due to the observed
use changes (IPCC, 2007, p. 6, topic 2). Measurements in solar radiation patterns
indicate that only minor changes were observed: warming can therefore be attributed
to human activities. Current concentrations of greenhouse gases are higher than those
observed in the historic record spanning many thousand years (obtained through
sions will continue to grow over the next few decades” (p. 1, topic 3). The growth of
greenhouse gas concentrations in the atmosphere will ensure warming in the future,
with the possibility of greater changes to the global climate system than have been
observed during the 20th century. The projected changes for the 21st century include
patterns of warming for most places of the world, together with changes in wind
patterns, precipitation, as well as climate extremes (rain, snow, wind and ice storms,
hurricanes, heat waves, etc.). Based on a number of global circulation models, it is
likely that extreme events will become more frequent (IPCC, 2007, p. 4, topic 3).
Ecosystems
• Many ecosystems are likely to experience reductions in resiliency due to the un-
precedented combination of climate change, associated disturbances (i.e., flood-
ing, drought, wildfire, insects, ocean acidification), and other influences (land
use change, pollution, fragmentation of natural systems, over-exploitation of
resources).
7
• About 20-30% of plant and animal species assessed are at increased risk of
extinction, if increases in global average termperature exceed 1.5-2.5◦C.
Food
Coasts
Industry
• The most vulnerable industries, settlements and societies are those in coastal
zones and river flood plains; they are expected to face increased incidence of
Health
Water
of glacial mass and reductions in snow cover are projected to increase, thereby
reducing water availability, hydro power potential, and changing flow seasonality
for regions dependent on glacial melt-water (Asia, South America).
• The potential benefits of increased runoff will likely be outweighed by the nega-
tive effects of increasing variability in precipitation, by seasonal shifts in water
supply and water quality, and by the increased flood risk.
• Future increases of heavy rainfall events are projected for many regions, even
when annual precipitation is expected to decrease. These rainfall events will
increase exposure to floods, water quality problems, etc.
of natural and social environments. Examples of adaptation are many, of which only
few are summarized below.
In the water resources sector expanding rainwater harvesting, storage and conser-
vation techniques, water re-use, desalination and improved efficiency include a set of
adaptation options. A unifying framework including national, regional, and local in-
tegrated water resources policies is needed, as are water related hazards management
strategies (to cope with the increased risk of floods, droughts, and other extreme
events). In the agricultural sector, adjustments to planting dates and crop variety
offset the projected growth of global emissions or reduce emissions below current
levels” (p. 3, topic 4). Mitigation strategies for various sectors are offered by IPCC
(2007). Key mitigation strategies for the energy supply include: (i) improving the
efficiency of the supply and distribution of power, (ii) switching of fuel from coal
to natural gas, (iii) using nuclear power and renewable energy (hydro power, solar,
wind, geothermal), etc. Policies that have shown to be effective in this area include
the following: a reduction of fossil fuel subsidies, taxes or carbon charges on fossil
fuels, renewable energy incentives, etc. It is noted that there is strong evidence that
improvement to human health from reduced pollution (from greenhouse gas emission
reductions) can be substantial, and may offset a large fraction of mitigation costs.
10
IPCC (2007) also finds high agreement and much evidence that a wide range of
national policies and instruments are available to governments to include incentives
for such mitigation action. Some of the policies and instruments may include the fol-
lowing (i) climate policies integrated into broader development policies, (ii) emissions
regulations and standards, (iii) taxes and charges for carbon emissions, (iv) financial
incentives (subsidies and tax credits) for development of non-carbon technologies,
(v) voluntary agreements for reduction of emissions, (vi) information instruments in
promoting and raising awareness, and (vii) research and development of new tech-
nologies.
Adaptation and mitigation measures need to be developed with a risk management
perspective over the long term. Furthermore, it is noted that neither adaptation nor
mitigation alone can avoid all climate change impacts, although many can be avoided,
reduced or delayed.
This section examines water resources impacts and adaptation measures resulting
from changed climatic and physical conditions. The section focuses on water resources
impacts on global, regional and local scales, and ends with an explanation of prime
importance to Canada.
Major impacts on water resources from changes in climate are related to increased
temperature and precipitation variability. Anthropogenic climate change is one of
many stresses on water resources systems. Others include changes in population,
economy, lifestyles, technology, environmental practices, etc. The current state of the
knowledge of water resources impacts resulting from changes in climate are summa-
rized below (after Kundzewicz et al., 2007):
11
• The effects of streamflow and ground water recharge will vary regionally and
• Peak streamflow will likely move from spring to winter for rivers depending on
snowmelt, with significantly lower flows during the summer season;
• Flood frequency and magnitude is likely to increase in most regions, while vol-
umes of low flows are likely to decrease;
latitudes and the tropics, and decreases for subtropical areas). Even though temper-
atures are expected to rise in the future, precipitation may increase in one season
12
fluences. Such influences consist of land use changes, reservoir construction, pollution
emissions, water use and irrigation, waste water treatment, etc. Changes in water
use are strongly linked to changes in population, food consumption, economic policy,
technology, lifestyle, and society’s value of water resources (Kundzewicz et al., 2007,
p. 181).
The future vulnerability of surface water resources is strongly linked to warming
and the seasonality of river flows. A robust finding often cited is that snow cover will
reduce in extent over many regions, thus shifting the peak spring runoff to earlier
in the year. This change in runoff may increase winter flows, but may also reduce
summer flows, and therefore reduce water supplies for many areas. For regions with
little or no snowfall, changes in runoff are more dependent on precipitation than on
temperature. Seasonal flow characteristics will change in these regions, with higher
flows during the peak season, and either lower flows or drier periods during the low
flow season (Kundzewicz et al., 2007, p. 184).
Demands for ground water will increase in the future. This is in part due to
increasing population and socio-economic activities, and to decreased availability of
surface water supplies in snow-dominated basins. Spring ground water recharge will
shift to winter, while summer recharge will decline, thus profoundly effecting veg-
etation and ecosystems. Sea-level rise in vulnerable areas will increase salt water
intrusion into aquifers, therefore degrading ground water quality.
A warming climate with its increased precipitation variability will increase risks
from floods and droughts. A robust finding by Kundzewicz et al. (2007) states that
13
higher precipitation extremes in warmer climates are very likely to occur, with pre-
cipitation intensity increasing almost everywhere in the world (p. 187). This change
increases the risk of flash flooding and the possible failure of urban drainage systems
(Waters et al., 2003). Storm drainage systems will have to be designed to accom-
modate the increased rainfall intensity resulting from climate change. Droughts and
drought like conditions are expected to be exacerbated for many regions. With more
than a billion people relying of melt waters from glaciers and seasonal snowpacks for
water supply, “the consequences of projected changes for future water availability,
predicted with high confidence and already diagnosed in some regions, will be ad-
verse and severe” (Kundzewicz et al., 2007, p. 187). The impact of extreme flood and
drought events on human populations will occur disproportionately, leaving the least
adaptable countries most vulnerable.
Variations in surface runoff, coupled with higher water temperatures, are likely go-
ing to adversely affect water quality for many regions of the world. The unfavourable
impacts will undoubtedly affect human health, ecosystems, and patterns of water
use. Lowering of water levels in lakes, streams and rivers will lead to re-suspension of
Costs associated with expected impacts of climate change are difficult to quantify,
as the performance of water resources systems depends on climatic and non-climatic
(i.e., human) factors. The impact of floods and droughts could be lessened by in-
creased investment in infrastructure, and by changes in land use management, but
these options carry considerable costs. Increased annual runoff may entail benefits,
14
as more water resources are made available, but the extra runoff may also increase
the likelihood of flood damage. The increases in runoff may prove harmful, especially
if it arrives during the high flow seasons (when it is needed the least).
In Canada, there are a number of agencies committed to producing high qual-
ity research on the impact of climate change, as well as possible adaptation and
mitigation measures. The report by Bruce et al. (2000) presents the first compre-
hensive top-down treatment of climate change impacts for Canada. Research of
Lemmen and Warren (2004) closely follows previous work and outlines the Canadian
ture to such adverse impacts of climate change, the report by Infrastructure Canada
(2006) is available (even if only as a literature review report).
Lemmen and Warren (2004) present detailed impacts of climate change on the
Canadian water resources sector. Northern communities that rely on glacial melt
water will see earlier occurrence of spring floods, with a possible reduction of summer
streamflow. This will seriously impact hydro power generation potential, fisheries,
tourism and recreation. For the Arctic, thinner ice cover and an increases in ice-free
seasons (from increased temperatures) will create severe ecological problems, and im-
pact the traditional way of life for northern communities (i.e., hunting, fishing, etc.).
Prairie provinces will see changes in annual streamflow, with possible declines in sum-
15
frastructure. Atlantic provinces may see a decreased amount and duration of snow
cover, saline intrusion into coastal aquifers, as well as possible large reductions in
streamflow, increased spring flooding, coastal erosion, loss of potable water, etc.
Changes in timing, magnitude and frequency of extreme weather events presents
veying the increased loads. As a result, extreme events may override the capacity of
storm sewer containment infrastructure and may infiltrate into drinking water sup-
plies. Increased runoff from contaminated surface water also poses risks to drinking
water safety.
the goal of water resources management. The additional challenge now is to take into
consideration the impacts of climate change, and to implement management plans
and strategies that will reduce its adverse effects.
Climate change adaptation strategies are typically a combination of structural
construction of infrastructure, and include the following (i) reservoirs (often simul-
taneously used for flood peak reduction and low flow augmentation), (ii) retention
basins (used to collect surface runoff before reaching a stream), (iii) levees and flood
walls (employed to control the level of flood inundation), (iv) channel modification (in-
creasing hydraulic capacity and stability of channels), and (v) municipal water man-
agement infrastructure (sanitary and storm sewers, flood control facilities), among
others (after Simonovic, 2006).
Non-structural measures on the other hand, concentrate on altering the damage
potential of water with minimal physical intervention, and focus on modifying current
management practices. Examples of non-structural measures include the following
(after Simonovic, 2006) (i) zoning (limiting development in vulnerable areas such as
floodplains), (ii) insurance (providing compensation when damage from floods and/or
droughts are not avoidable at acceptable cost), (iii) protection of individual proper-
ties (waterproofing of the lower floors of existing buildings for example), (iv) flood
and drought warning systems (to inform about possible damages), and (v) disaster
preparedness (to prepare the community for effective responses to an emergency).
Increased water stress due to population growth and economic activity is driving
additional infrastructure investment and institutional changes. This is referred to
as autonomous adaptation—adopting measures not deliberately designed to address
climate change, but which lessen already observed consequences and build resilience
in water resources systems. Planned adaptations on the other hand take climate
change explicitly into account, and build flexible management protocols in order to
deal with risks and uncertainties associated with climate change. Integrated water re-
sources management is an instrument recommended for exploring adaptation options
to changed climatic conditions (Kundzewicz et al., 2007; GWP, 2008). The integrated
public and private agencies across all scales (community, regional, national, global).
The integrated approach aims to take into account society’s views and to change the
traditional water resources planning process (where structural measures and strate-
gies of increasing supply have dominated the practice). An integrated framework can
also help bring about consensus among opposing sides facilitating communication and
raising awareness.
General ideas relating to the concept of integrated water resources management have
been developed some time ago as a result of increasing pressure on water and other
natural resources by the growing world population and its associated socio-economic
development. Problems related to changed climatic conditions have further exacer-
bated already heavily stressed water resources systems in many places, and therefore
offer support to adopting integrated management strategies. The challenges in water
resources management include the following (after GWP, 2000) (i) securing adequate
water for human society (sustenance and food production), (ii) protecting ecosystems
(plant and animal species), (iii) managing the variability of water and its associated
risks (too much or too little water), (iv) raising public awareness and creating the
political will to act, and (v) ensuring collaboration across all sectors and boundaries
(involvement of stakeholders in policy making, planning and implementation).
Principles relevant to the process of integrated water resources management have
been developed from an international consultative process at the International Con-
ference on Water and the Environment in Dublin, as well as the United Nations Con-
ference on Environment and Development in Rio de Janeiro. They are sometimes
referred to as Dublin-Rio principles (GWP, 2000) and include the following:
18
The above principles embody the ideas that holistic approaches to water resources
management are necessary, and must therefore include interactions between the hy-
drologic cycle and natural, social and environmental systems. Such interaction in-
evitably recognizes that all natural resources (including water) have physical limits,
and that human societies affect their productivity and sustainability.
The principles postulate that holistic institutional approaches are needed to cre-
ate co-ordination between policy makers at all levels (national, regional, local) and
at all sectors (ecologic, economic, social and political). The challenge arises in imple-
menting a participatory approach across such a wide range of scales, especially since
everyone is a stakeholder (from a national minister, to a community group). The
participation advocated is more than consultation, and requires that all stakeholders
Many past failures of water resources management can be traced to water being
viewed as a free good, where its full value was not properly appreciated. The full value
of water includes two parts (GWP, 2000, p. 19): (i) economic (value to users, benefits
19
from return flows, indirect uses, societal objectives), and full (including economic and
intrinsic). Treating water as a economic good that can not be substituted (there is
no replacement for water) will help balance future demand with supply, and therefore
sustain the human society and the environment in the long term.
Papers by Mitchell (2005, 2006) outline the Canadian experience with integrated
water resources management. Most notable examples include a linked management
system for the Fraser River Estuary in British Columbia, and a multi-barrier approach
to drinking water safety in Ontario. Although there are a number of difficulties
(especially in implementation), rewards of integrated water resources management
are far greater than those of non-integrated approaches. Mitchell (2006) concludes
that if integrated water resources management is to be useful and effective, emphasis
should be placed on the following (i) stakeholders’ shared vision for the future, (ii)
the need for systemic planning and implementation, (iii) the importance of different
approaches are needed. These bottom-up approaches include interactions among var-
ious bio-physical, socio-economic and institutional sectors, for the purpose of build-
ing the management capacity needed for effective administration of water resources
(Loucks and van Beek, 2005, p. 26). This latter approach is gaining acceptance, as it
Typically, projects start by identifying physical features of the water resources sys-
tems under consideration, such as meteorological and climatic characteristics, land
cover and soil types, dominant land and water use patterns, runoff in rivers and
streams, as well as regional ground water and aquifer properties. Regional socio-
economic features are also studied, such as population and economic growth. All this
information is needed, because the amount of runoff (and ground water recharge)
depends both on physical (climate, land cover, soil types, vegetation, etc.) and socio-
economic (population, economic activity, management practices, etc.) processes op-
erating in the basin.
such sector individually. The interaction between the socio-economic and physical
forces drive the overall functioning of the basin. For example, availability of clean
water can place limits on the economic development and growth of a region. Growth
and increasing land development, if excessive, can increase the timing and magnitude
of flood peaks, while reducing ground water recharge rates (through excessive conver-
sion of agricultural and forested lands to residential and business uses). Such growth
and development can eventually limit the availability of water. Socio-economic forces
are therefore inseparable from their physical counterparts and must be included in
(integration and communication across government levels and public and private sec-
tor groups, micro and macro economic effects of water developments, etc.). Overriding
criteria for such an integrated policy (recognized as challenging to implement) include
economically efficient water use, equity (all people should have adequate water qual-
agreed upon, where the administrative functions of all levels of stakeholders are es-
tablished. Lastly, management instruments for operation, regulation, monitoring and
enforcement must be developed.
The only way to practically achieve an enabling environment is to move away from
The role of governments in this process should be that of enablers, regulators, and con-
flict resolvers. One way of achieving such an environment is through the involvement
of the private sector in providing water services, but with proper and transparent
regulation and monitoring.
Management instruments are tools that enable the water resources decision maker
to make rational and informed choices between alternatives. A wide range of methods
(quantitative and qualitative) offered by the systems approach is available to provide
support for these tasks. These methods should be complemented by knowledge in
strategies. Some of the tools include guidelines for how to perform water resources
assessments, how to synthesize current and acquired knowledge base (data, methods,
monitoring approaches), how to perform and quantify environmental impact and risk
assessments, as well as how to use risk management in defining acceptable levels of
risk. All management instruments promote partnerships and encourage stakehold-
ers to work together, share knowledge and experiences with the aim of searching
constructively for possible solutions. Even stakeholders with opposing interests can
“enhance each others capacity for mutual benefit and achieve a common vision by
sharing risks, responsibilities, and rewards” (Mitchell, 2006).
The examples in the Global Water Partnership Toolbox (GWP, 2003) demonstrate
23
Canada and the European Union in their December 2006 special issue.
Three sets of regulatory instruments (GWP, 2000) exist for setting appropriate
management structures and procedures for use by water resources authorities. They
include direct controls, economic instruments, and encouraged self-regulation. Direct
controls are means with which to guide the formulation of management regulations,
and to examine the legal ramifications of water ownership, land use planning, and
public and private utility regulation (water servicing). Economic instruments are tools
that provide incentives to changing behaviour, raise revenue to finance investments,
der climate change. The systems approach can capture the principles of horizontal
(across multiple disciplines) and vertical (across multiple hierarchies) management
styles, and thus provide support for integration and interdisciplinary communication
among stakeholders involved in water resources projects. One way to achieve this
processes of various scales, and represent both physical and social elements. Such
models have the potential to reveal behavioural implications of physical and socio-
economic systems in response to policy and management options, and offer guidance
for management action. Such computer models can help identify policy levers and
provide stakeholders with the means to enhance management capacity. The basics of
the systems approach is presented next, as are introductory concepts in mathematical
modeling.
Development of the general systems theory in the 1950’s and 1960’s introduced a par-
ticular way of thinking, where no problem can be studied in isolation, but must be
considered alongside the entire context it is embedded in. Such an approach means
that problems encountered in water resources can not be studied individually and
separately, but should be considered from a much broader context. For example,
before future municipal plans and strategies (floodplain development, or storm water
drainage standards) are formulated, potential changes in physical, social and environ-
mental conditions (flood and drought frequency, water availability, urbanization and
development, etc.) should be taken into account. For the historical development and
evolution of the systems approach, from its early days to the present, the reader is
referred to the work of Richardson (1991); Jackson (2000, 2003), on the other hand,
provides an outline of different systems approaches and how they can be applied to
management in general.
The first applications of the systems approach to water resources management in-
clude the research of Maass et al. (1962) and Hufschmidt and Fiering (1966); in these
studies, explicit methodologies for system design and simulation are formulated. This
25
methodology includes four steps: (i) identifying design objectives, (ii) translating ob-
jectives into design criteria, (iii) using the criteria to devise plans for water resources
development, and (iv) evaluating the consequences of plans. The above work encour-
ages new ways of performing engineering and economics for the purpose of developing
better system designs, particularly through the use and development of mathematical
models.
A model is defined as a simplified representation of reality. Meadows et al. (1982)
postulate that a model is not just any simplified representation, but rather one that
main types of mathematical modeling approaches are often used in systems theory:
optimization and simulation. The former approaches are referred to as prescriptive
modeling approaches (as they recommend an optimal course of action, a design, or a
policy), while the latter approaches are regarded as descriptive (“because they offer,
for a given set of inputs and initial conditions, a description of the outputs through
time of the phenomena under study”) (ReVelle et al., 2004, p. 2). The focus in this
study is on the development and use of mathematical simulation models that assist
in the management of complex water resources systems.
Optimization models are developed by describing the problem in terms of objec-
tives, design (or decision) variables, and constraints. Design variables are a set of
26
controllable or partially controllable set of inputs, whose optimal values are sought
after. Objectives (or objective functions) are used to measure the effectiveness of a
particular policy, and are always expressed in terms of design variables. Constraints
are sets of physical, economic, social, environmental, and other restrictions applied
to the model. The outcome of an optimization study typical in water resources in-
cludes, for example, optimal reservoir release that maximizes a reservoir’s ability to
reduce downstream flooding, while meeting all the specified constraints. Examples
of reservoir optimization in water resources management include those of Yeh (1985)
ture policies. Examples of hydrologic simulation models include those of Soong et al.
(2005), Cunderlik and Simonovic (2005), among others.
Another set of model classification of interest to water resources engineering and
management includes deterministic and probabilistic/stochastic models. Determinis-
tic models assume that each variable and parameter in the problem can be represented
27
as a fixed number, or series of fixed numbers for any given set of conditions. Alter-
nately, probabilistic/stochastic models attach a degree of uncertainty or randomness
to variables and parameter values. Models specifically looking at time dependent
properties of systems are further classified as dynamic, while those not explicitly con-
sidering time are referred to as static. Ways in which models consider spacial and
temporal variations in processes and parameters form another relevant model classifi-
cation scheme. Two types are commonly used: distributed and lumped. The former
category refers to models which take into account detailed variation in behaviour and
parameters throughout the system; the latter category ignores detailed variations and
considers them as homogeneous through the system, or its components (Simonovic,
2003).
The problem addressed in this study includes the dynamic response of water re-
sources systems to changing climatic conditions by using various statistical and dy-
namic simulation models. Observed warming and the changing climate eventually
intensifying the hydrologic cycle, are going to have a significant influence on water re-
sources systems and their management. Issues related to timing, variability, quantity
and quality of streamflow will influence the management practice, as will planning,
design and maintenance of future flood protection infrastructure. These factors will
force governments to alter current practices in which design standards and manage-
ment guidelines are set. Such alterations are not always easily achieved, as conflicting
viewpoints from different stakeholders, decision makers, and water users will almost
certainly emerge and will need examination. The system approach coupled with ideas
of integrated water resources management together comprise a promising set of tools
for assisting in such a process.
28
models, algorithms and procedures developed for the purposes of supporting climate
change impact assessment of water resources at various spacial scales. The contribu-
tion is summarized in the following categories:
basin scale;
erator algorithm;
(c) Flood and low flow frequency analysis under the changing climate;
Three different models are developed to generate climatic input for water resources
integrated assessment models. The first model includes a synthetic storm algorithm
able to produce spacial and temporal distribution of rainfall storms moving across a
basin. The novelty of this model consists in its being able to spacially and temporally
distribute statistical rainfall information (i.e., take a single value of an n year return
period rainfall, and distribute it temporally and spacially across an area of interest).
This model has been previously published as a conference paper (Prodanovic et al.,
2005), and is an extension of the work originally developed in 2004, put published in
hourly intervals, later used to develop rainfall Intensity Duration Frequency (IDF)
curves under different climatic signals (used for designing storm water and flood con-
trol infrastructure). An original procedure is developed that synthesizes (and comple-
ments) weather generator and disaggregation algorithms to generate synthetic local
data under climate change that is statistically similar to the observed data. Published
work in this area includes a report to the City of London (Prodanovic and Simonovic,
2007a), as well as a journal paper currently under review (Prodanovic and Simonovic,
2007b).
30
low flows. As a result of this research, recommendations for revising existing water
resources management standards and guidelines are developed and provided to the
local authorities. The novelty of this particular research contribution includes linking
weather generator outputs to event and continuous hydrologic models for the purpose
of testing local response of water resources systems under changed climatic conditions.
Publications originating from this work include a conference paper (Simonovic et al.,
2007) as well as a published journal paper (Prodanovic and Simonovic, 2007c). The
published work part of this research contributions draws heavily from calibrated con-
tinuous and event hydrologic models developed by Cunderlik and Simonovic (2005)
and Cunderlik and Simonovic (2007), respectively.
Lastly, an integrated water resources management model is developed that links
detailed physical (climatic, hydrologic) and socio-economic (population, housing,
business, land use) processes operating on the basin scale. The argument is that
neither the physical nor the socio-economic character of an area can be studied in
isolation, as each influences and is influenced by the other. An original simulation
model is developed, where detailed physical processes (represented by hydrologic mod-
els) are linked with regional socio-economic characteristics (captured with system dy-
namics models) via feedback. This clarifies the fact that physical processes change
dynamically with socio-economic changes (i.e., surface runoff increases with urban-
ization); the model also shows the reverse, where the socio-economic conditions are
dynamically altered with physical changes (i.e., limited water availability impacts
regional growth and development). The integrated water resources model is capa-
31
ble of testing a wide range of physical, social, economic, environmental and other
policies, thus assisting in operationalizing the process of integrated water resources
management. The publications stemming from this work include two conference pa-
pers (Prodanovic and Simonovic, 2006a, 2007d) and one journal paper under review
The methodology is presented in Chapter 2 where the theoretical framework for all
methods, components and models is developed. This chapter includes the mathemat-
ics necessary to reproduce all components of: (i) climatic input generation (spacial
as is the application of the inverse approach to the study of regional flood and low
flows. Hydrologic outputs are quantified with statistical analysis methods, therefore
facilitating estimation of impacts (of high and low flows) under the influence of cli-
32
matic change. The results of the integrated water resources management model are
illustrated next, and are used to show possible impacts of the interconnected climatic
and socio-economic scenarios within the study area, thus offering suggestions that
current policy makers should examine in the future.
2 Methodology
In order to assess the impact of climate change on water resources systems and to
develop adaptation strategies, means for constructing meaningful climate change sce-
narios are required. Section 2.1 covers a number of different ways to generate climate
change scenarios for use in evaluating the response of water resources systems to cli-
mate change. The developed climate change scenarios are employed, using the inverse
approach, to evaluate local impacts of the hydrologic flow regime (Section 2.2), and to
assess climate change impacts on both physical and socio-economic aspects of water
resources in an integrated manner (Section 2.3).
generator are synthetic data sets of daily duration that are statistically similar to the
observed data. The generated daily data includes precipitation, and maximum and
minimum temperature. One way of disaggregating daily precipitation is accomplished
by using the spacio-temporal storm model (that generates, spacially and temporally,
that are statistically similar to the observed data. The hourly generated data (for a
number of different climate change scenarios) are used as inputs in an event based
hydrologic model; this model quantifies the change in flooding characteristics under
34
Socio-Economic
Observed Local Time Series Output
Daily Data Policy Options
K-NN
GCM Data Weather
Generator
Socio-Economic
Model
Long term
Generated Local
Hydrograph
Daily Data
Spacio-Temporal Continuous
Storm Model Hydrologic
Model
Observed Local
Spacio-Temporal Hourly Data
Storm Hourly Data
Generated Local
Hourly Data
Event Rainfall
Hydrologic Low Flow
Frequency
Model Management
Analysis
Guidelines
Flood Flow
Flood
Frequency
Hydrograph
Analysis
Storm Water
Management
Guidelines
Flood
Management
Guidelines
are used as inputs for a continuous hydrologic model, with the purpose of assessing
effects of climate change on regional low flow and drought characteristics. Frequency
analyses are similarly used to quantify regional low flow characteristics, based on
which recommendations for future low flow management guidelines are also proposed.
In order to be able to assess the impacts of climate change on water resources systems,
methods to generate locally applicable climate scenarios are needed. Currently GCMs
are the most comprehensive and the best available tools in estimating future climatic
changes. The GCMs include outputs for a variety of variables (temperature, pre-
cipitation, humidity, wind speed, solar radiation, etc.) needed for impact assessment
studies. GCM outputs are subject to numerous modeling uncertainties (Mearns et al.,
2001), meaning that if different GCMs employ different methods for representing the
36
same physical phenomena, they may produce different behaviours. Such characteris-
tics increase uncertainties associated with GCM outputs, as it is possible for different
GCMs to produce different outputs even through they are conditioned on identical
inputs (emission scenarios). Comparison studies of different GCMs and their outputs
found that simulated temperature patterns agreed with observed trends significantly
better than estimates of precipitation (Kundzewicz et al., 2007). Therefore, more
confidence is placed in simulated temperature changes than future estimates of pre-
cipitation. This situation creates a problem for analyzing water resources systems,
are sometimes incompatible with temporal scales of river basins. For example, the
rainfall-runoff modeling requires data with relatively short time steps (daily and/or
hourly); the global models, however, are only able to produce monthly outputs with
a higher degree of accuracy (Cunderlik and Simonovic, 2007). This incompatibility
have resolutions between 3-5◦ in latitude and 5-10◦ in longitude (equivalent to a grid
of 100 km by 100 km), and are thus significantly larger than most river basins. Such
coarse resolution is inadequate for the representation of many relevant smaller scale
phenomena, and is therefore inappropriate for determining climate change impacts
Uncertainties in climate change impact analysis of water resources systems are due
to uncertainties in the following: (i) GCM precipitation inputs (and to a lesser extent
in temperature), (ii) climate sensitivities (by how much the climate will change in
response to a specified forcing), and (iii) hydrologic models (Kundzewicz et al., 2007,
p. 181).
In order to overcome difficulties associated with GCMs, techniques have been
developed that downscale GCM outputs to finer spacial and temporal resolutions.
These techniques include the following (Kundzewicz et al., 2007, p. 181): (i) dy-
p. 590), thereby widening their general applicability in climate change impact assess-
ments. The main limitations of RCMs include systematic errors in the driving fields
determined from GCMs. Depending on spacial extent and grid resolution, RCM
simulations require specification of regional boundary conditions, thus compound-
ing uncertainty. Lastly, RCMs are often computationally expensive, and have been
cal model that relates large scale (“predictors”) to regional scale variables (“predic-
tands”), and then feeding GCM predictors to the statistical model to estimate local
and regional climate information (Giorgi et al., 2001, p. 591).
An advantage of statistical downscaling is that analyses employed using this
data may not be applicable under conditions of future changed climate). Another dis-
advantage of these approaches is that they can not explicitly account for systematic
changes in regional forcing conditions or feedback mechanisms (Giorgi et al., 2001,
p. 591). More information on statistical downscaling methods is provided in Section
2.1.2.
The rest of this section develops the means for producing the input data needed for
climate change impact assessment studies of water resources systems. Two different
means are used in this research to synthetically generate input sequences for use in
water resources impact assessment. The first is a simple method that generates spacio-
temporal rainfall over an area of a basin using pre-defined temporal distributions and
geometric properties of storms. This method should be used in preliminary assessment
studies that do not warrant use of more sophisticated means. The second method is
an advanced statistical downscaling technique used to synthetically generate weather
scenarios conditioned on both locally observed data and GCM outputs (used to drive
the direction of climate change). Both approaches represent ways of synthesizing
potential weather sequences that river basins may experience.
39
All hydrologic studies require the use of precipitation in order to drive the rainfall
runoff process. For situations where flood peaks are of interest (design of water
management infrastructure, flood plain delineation, etc.) design storm hyetographs
are needed (hyetographs are plots showing rainfall intensity versus time). Design
storm hyetographs act as inputs to hydrological models, with discharge produced as
output.
A variety of methods able to generate design storm hyetographs exist within the
current research literature. Veneziano and Villani (1999) suggest that most of the
methods can be classified into one of the following categories: (i) specification of sim-
ple geometrical shapes anchored to a single point of the Intensity-Duration-Frequency
(IDF) curve; (ii) use of the entire IDF curve; (iii) use of standardized profiles obtained
directly from rainfall records, and (iv) simulation from stochastic models.
Methods that generate rainfall based on a single point of the IDF curve re-
quire specification of the return period and storm duration. Oftentimes rectangular
hyetographs are used, where the average intensity is used throughout the storm du-
ration. This procedure is used in combination with the rational method for design of
storm water management infrastructure. However, this method is criticized because of
its tendency to underestimate the total rainfall volume (Veneziano and Villani, 1999).
Use of alternate (other than rectangular) geometric distributions of rainfall are there-
fore required (examples include triangular hyetographs of Yen and Chow (1980) and
linear/exponential hyetographs of Watt et al. (1986)). Veneziano and Villani (1999)
claim that such methods are simple, intuitive and easy to construct, but “do not have
a strong conceptual basis and may produce biased flow estimates” (p. 2726).
proposed that use the entire set of duration-intensity values for a particular frequency.
Methods of this kind are those of Keifer and Chu (1957), known as the Chicago
method. This method for temporal distribution of rainfall is often used in the design
of urban drainage systems in Canada (MTO, 1997).
tually constitutes a rainfall event and because of the randomness of the rainfall phe-
nomena itself. Because of these limitations, standardized profiles must use some sort
of temporal smoothing or ensemble averaging (Veneziano and Villani, 1999, p. 2726).
Lastly, stochastic generation of rainfall are covered in detail in the next section.
All methods above specify temporal distribution of rainfall, while implicitly as-
suming a uniform spacial distribution (i.e., that a same hyetograph is applicable over
the entire study area). This assumption is justified when considering the response of
relatively small catchments but not when basin-wide response is sought after. In the
text that follows, a synthetic storm model is presented; this model is applicable for
testing different storm scenarios and their hydrologic response on the basin scale.
The storm model presented here is based on models developed by Prodanovic et al.
(2005) and Cunderlik and Simonovic (2007). The complete set of model equations is
presented in this section, together with a description and meaning of each process and
parameter used. It is noted that some parameters are physically (or geometrically)
based, some are statistical, and some are simply modeling parameters that allow a user
to adjust and fine-tune the model and are neither physical nor statistical. The storm
41
can be deemed equally valid). On the other hand, the calculation of PMP is quite
lengthy and requires a considerable amount of time and effort.
This model does not provide the means for incorporating travel speed of storms;
in other words, it considers storms as stationary during the simulated time period.
Storms usually travel in the “same direction as mean cloud layer winds”, which are
best determined from observed data (Cunderlik and Simonovic, 2007, p. 569). In this
model, storm movement is captured by a single parameter that controls its shape.
The basic structure of the synthetic storm model is a series of randomized concen-
tric ellipses, used to represent a rainfield of a particular region. Both Huff (1967) and
NWS (1978, 1982) use the concept of concentric ellipses to represent storms in their
work; Huff (1967) even claims that elliptical shapes dominate most often, regardless
of the the depth of rainfall or storm duration (p. 1016). The storm model therefore
assumes elliptical isohyetal patterns; it also assumes that rainfall intensity decreases
exponentially with distance from the center of the storm.
The storm model requires a coordinate location of the centroid of each spacial unit
(or subcatchment) in the study area, or (xi , yi), to be specified. The units of xi and yi
can be in any units of length [L], or alternately, in degrees of longitude and latitude,
is the total depth of rainfall for the time period in question. In most cases total
42
rainfall depth, D [L], is known; total depth can also be found by multiplying an
average intensity (obtained from an IDF curve) to storm duration. Other necessary
parameters include the following:
• The orientation of the storm (denoted by the rotation of the ellipses), or α [deg];
• The parameter am [L], identifying the the spacial extent of the storm;
The parameters described above are graphically depicted in Figure 2, where the
basic structure of the model is shown. After the storm depth and shape parameters
are specified, the algorithm begins. The location of each subcatchment is described
in polar coordinates with a radial distance di [L]:
p
di = (yi − y0 )2 + (x0 − xi )2 (1)
" #
(yi − y0 )
θiI = Tan−1 ∀ di ∈ I (2)
(xi − x0 )
θiII = 180 − θiI ∀ di ∈ II (3)
where an equation is given for each quadrant in the xy plane (quadrant I is top right,
II is top left, III is bottom left, and IV is bottom right). Now that the radial distance
di is known, the semi-major component ai can be determined for each concentric
43
(xi , yi )
di θi
bi
α x
(x0 , y0 ) ai
am
a2i b2i
d2i = (6)
sin2 γi + b2i cos2 γi
r
1
ai = ρdi sin2 γi + 2 cos2 γi (7)
ρ
To specify the spacial extent of the storm (i.e., the location beyond which no
rainfall occurs), the following statement is used: if the calculated ai is greater than
specified am , assign zero to rainfall for that particular subcatchment; otherwise, pro-
44
ceed with calculations as usual. In the case where ai is smaller than am , we use
the following fourth order polynomial approximation as an area reduction to point
rainfall, defined as ci :
4
X
ci = ck Aki (8)
k=0
in current research literature that distribute rainfall temporally (Wood and Goldt,
2004). Only two such methods are selected for use in this study—the SCS (1986)
method and that of Huff (1967). Both of these methods provide a means of dis-
tributing total rainfall for a given storm duration via mass distribution functions.
For example, if a total daily rainfall is specified, the aforementioned methods can
where u(0, 1) is a uniformly distributed random number, and σ [−] is a coefficient used
m being the total storm duration. The exponential term in equation (10) portrays
a decay of rainfall intensity as a function of distance from the storm center. The
coefficient cf [−] is calculated in such a way that the exponential term is unity at the
storm’s center, with ǫ (an arbitrary small value) at distance am away from the center
(i.e., its spacial extent). The intensity decay coefficient cf is computed as:
ln(ǫ)
cf = (11)
−am
As the coefficient cf becomes large, the intensity decays much more quickly from
the storm’s center, thus producing a storm with a smaller area; on the other hand, if
a small cf is used, more uniform rain field will result.
The array SCS2(j), or alternately HUFF3(j) in equation (10) represents the tem-
poral mass curve used to distribute the daily rainfall total into m time steps, one
for each time step j. It is noted that Wood and Goldt (2004) recommend that, for
the Upper Thames River basin type II distributions be used for the method of SCS
(1986) and either quartile III or IV storms for the method of Huff (1967); for this
study, only quartile III were used. Finally, the two dimensional array Sij is one that
gives temporal (subscript j) and spacial (subscript i) rainfall for the entire basin.
The above spacio-temporal storm method represents simple means of generating lo-
cal scale precipitation estimates for use in impact assessment studies. If additional
climatic variables are needed (temperature, snowfall, solar radiation, wind speeds,
etc.), more sophisticated means must be used. Statistical downscaling techniques (of
which weather generators are subsets) can be used to extend the observed climate
record, as well as to generate scenarios of possible climatic changes.
46
(i.e., local climate variable), L the predictor (i.e., large scale climate), and F a de-
terministic/stochastic function conditioned by L and determined from observed data.
Three key assumptions are made by applying statistical downscaling techniques: (i)
the predictor variables are realistically modelled by GCMs and are applicable to the
region of interest, (ii) the predictors adequately represent the changed climate sig-
nal, and (iii) the relationship F is valid under conditions of altered climate (which
is untestable) (Dibike and Coulibaly, 2005, p. 146). Two main classes of statisti-
cal downscaling methods include regression downscaling techniques and stochastic
weather generators.
Regression downscaling methods rely on statistical functions to relate predic-
tor and predictant variables (i.e., function F ). Methods in this category differ in
their choice of mathematical transfer functions, predictor variables, or statistical fit-
ting procedures. Different types of regression downscaling methods use linear and
non-linear regression, artificial neural networks, canonical correlation and principal
component analysis to derive predictor-predictant relationships (Conway et al., 1996;
Schubert and Henderson-Sellers, 1997). The advantages of regression methods are
their low computational requirements and ease of use. However, they are only able
to explain climatic variability within the range of historically observed data; there-
fore these methods are not ideal for the study of extreme events lying outside of the
observed range (Dibike and Coulibaly, 2005, p. 147).
Stochastic weather generators are sophisticated algorithms that produce long time
series climatic records based on locally observed climate patterns. Weather generators
47
are complex random number generators, whose mechanisms are conditioned through
various means (global circulation models, resampling of specified years of record, El-
Niño and El-Niña cycles, wet, dry, warm or cold seasons, among others). Weather
generators are usually classified into two categories (Sharif and Burn, 2006): paramet-
ric and non-parametric. The former are stochastic tools that generate weather data
by assuming a probability distribution function and a large number of parameters
(often site specific) for the variable of interest. The latter do not make distribu-
tion assumptions or have site specific parameters, but rely on various shuffling and
sampling algorithms.
Parametric weather generators focus on independent generation of precipitation,
with other variables conditioned on the occurrence of precipitation. Daily precip-
itation values are generated using a two-state first-order Markov model, where an
assumed probability distribution is used to fit the observed values (Sharif and Burn,
2007, p. 42). The most famous weather generator models are those originally proposed
by Richardson (1981), referred to as WGEN models (see also Richardson and Wright,
1984). Modifications and further refinements of the Richardson (1981) weather gen-
erator can take into account non-normal distribution of wind speeds and relative
humidity. However, a major drawback associated with the above weather generators
is that persistent events (drought and prolonged rainfall) are not easily reproduced
(Sharif and Burn, 2007). In order to overcome this problem, a number of differ-
ent alternatives have been suggested. An example of one such alternative is the
weather generator LARS-WG, for which sequences of dry and wet days are mod-
elled first, while precipitation (and other variables) are conditioned on the wet/dry
states. Nevertheless, both WGEN and LARS-WG have difficulties in reproducing
annual variability in monthly means and simulating weather data at multiple sites
Sharif and Burn (2007) offer the following deficiencies of the parametric weather
generators: (i) they do not adequately reproduce spacial and temporal dependence
between variables, (ii) assumptions have to be made regarding the probability dis-
tribution of variables, (iii) non-Gaussian features in the data can not be adequately
captured, and (iv) a large number of parameters have to be fitted for each time
period, therefore increasing computation time (p. 42). Alternatives to parametric
weather generators are non-parametric generators. These generators alleviate the
aforementioned drawbacks, and have recently shown great promise in being state of
the art in statistical weather generation. The most popular non-parametric weather
generation techniques are the K-Nearest Neighbour (K-NN) resampling approaches.
Work by Lall and Sharma (1996), Sharma et al. (1997), Buishand and Brandsma
(2001), Yates et al. (2003), Sharif and Burn (2006, 2007) describes applications of
ments, as they help generate possible critical extremes that have not yet occurred.
Weather generators with perturbation mechanisms, conditioned by outputs of
GCMs, can therefore produce adequate synthetic data of high spacio-temporal res-
olution for a number of different scenarios needed to assess local climate change
impacts.
49
The weather generator developed in this study is a K-Nearest Neighbour model. The
model draws heavily from the previous work by Sharif and Burn (2006, 2007), which
is in turn based on the work of Yates et al. (2003).
Nearest neighbour algorithms are popular for the following reasons: (i) they are
capable of modeling non-linear dynamics of geophysical processes, (ii) they do not
data set. In the study of weather sequences, minimum and maximum temperature,
as well as precipitation, are typically used.
The procedure starts by assembling a historic data set free of missing values for a
number of stations in and around the study area. To produce weather for a new day,
all days with similar characteristics are extracted from the historic record, from which
a single nearest neighbour is selected according to a defined set of rules (explained in
detail later).
The nearest neighbour algorithms found in the current research that are applied
to generate weather sequences recommend use of daily data. Daily data is recom-
mended because most long term simulation models used in environmental impact
assessment (such as hydrologic models) require such data; also, daily data is typi-
cally more widely available than data of shorter durations. Furthermore, the work of
Wojcik and Buishand (2003) shows that second-order statistics are better preserved
when the weather generator operates on a daily time step, followed by disaggrega-
50
tion into shorter intervals, than when a weather generator that operates on intervals
shorter than a day. For this reason, a daily weather generator is selected for use in
this research.
In the K-Nearest Neighbour algorithm used in this work, p variables are selected
days in the observed historic record. A feature vector is defined in an expanded form
as:
Xtj = (xj1,t , xj2,t , . . . , xjp,t) (12)
where xji,t stands for a value of the weather variable i = 1, 2, . . . , p for station j for
day t.
For the sake of simplicity, assume that the simulation starts on January 01, and
continues for as long as synthetic data is desired. The algorithm steps are the follow-
ing:
1. Regional means of p variables are computed, across all q stations for each day
in the historic record:
X̄t = (x̄1,t , x̄2,t , . . . , x̄p,t ) (13)
where,
q
1X j
x̄i,t = x (14)
q j=1 i,t
2. Data block of size L is computed next, as it is the one that includes all po-
tential neighbours to the current feature vector Xtj . Temporal window of size
w is then selected, and all days within that window are selected as potential
51
neighbours to the current feature vector. In the work of Yates et al. (2003)
and Sharif and Burn (2007), w is selected as 14 days. This implies that if the
current day in the simulation is September 17, then all days between September
10 and September 24 are selected from all N years of record, but excluding the
September 17 for the given year (this prevents the possibility of generating the
same value as that of the current day). Therefore, the data block consisting of
all potential neighbours is L = (w + 1) × N − 1 days long for each variable p.
3. Regional means are also computed for all potential neighbours across all q sta-
tions for each day in the data block L using equation (14).
4. The covariance matrix, Ct , for day t is computed using the data block of size
L × p. For the special case when p = 1, the covariance matrix is simply the
variance of the nearest neighbour vector (L × 1 in this case).
5. If the start of the historic data record is January 01, then for the first time step
a value is randomly selected for each variable p from all January 01 values in
the record of N years. This value is used for the first simulated time step, that
is January 01.
6. Next, the Mahalanobis distance is computed between the mean vector of the
current days weather X̄t (step 1) and the mean vector X̄k of all nearest neighbour
values (step 3), where k = 1, 2, . . . , L. The distance is computed as follows:
q
dk = (X̄t − X̄k )Ct−1 (X̄t − X̄k )T (15)
where T represents the transpose matrix operations, while C −1 stands for its
inverse.
52
8. The Mahalanobis distance dk is sorted from smallest to largest, and the first
K neighbours in the sorted list are retained (they are referred to as the K
nearest neighbours). Furthermore, a discrete probability distribution giving
higher weights to closest neighbours is used for resampling out the set of K
neighbours. The weights are calculated for each k neighbour according to:
1/k
wk = Pk (16)
i=1 wi
j
X
pk = wi (17)
i=1
Through this procedure the neighbour with the smallest distance gets the high-
est weight, while the one with the largest distance gets the least weight.
9. In order to determine which one of the K neighbours will be selected as the one
used for the current weather, a uniformly distributed random number u(0, 1) is
generated first. The next step in the algorithm is to compare u to p, calculated
previously; note that p exists for each one of the K neighbours. If p1 < u < pK ,
then day k for which u is closest to pk is selected, where k = 1, 2, . . . , K. On
the other hand, if u ≪ p1 , then the day corresponding to d1 is selected; and if
used as the weather for all stations in the region. This means that if weather
of September 19 from year n is selected, then all stations in the area will use
the weather of September 19 from year n, from its respective data sets, as the
simulated weather for day t.
10. The above steps simply resample (or reshuffle) the historic record of data, which
may not be enough in a study where unprecedented values of weather variables
are of interest. Sharif and Burn (2007) present an approach that is able to
perturb the historic resampled data, and therefore generate data not observed
historically. For each station, for each variable, the authors fit a non-parametric
distribution to K neighbours of step 8, and estimate a conditional standard
deviation, σ, and bandwidth, λ. The conditional standard deviation is estimated
from the K neighbours, while λ is calculated based on the work of Sharma et al.
(1997) using the following equation:
(a) Let σij be the conditional standard deviation of variable i and station j
computed from K neighbours. Assume that zt is a normally distributed
random variable with zero mean and unit variance, for day t. The new
(perturbed) value of the weather variable i, for station j, for day t, is
computed as:
j
yi,t = xji,t + λσij zt (19)
where xji,t is the value of the weather variable obtained from the basic
54
j
K-Nearest Neighbour algorithm (up to and including step 9); yi,t is the
weather variable value from the perturbed algorithm; λ the bandwidth
(depending in the number of samples); and σ the standard deviation of
the K nearest neighbours.
(b) Since some weather variables are bounded (i.e., precipitation), there exists
a possibility that equation (19) may generate values that are impossible
equation (18) is larger than λa , then λa is used instead (Sharif and Burn,
2007).
(c) If the value of the bounded weather variable (i.e., precipitation) computed
previously is still negative, then a new value of zt is generated, and the
step 10(b) repeated.
(d) Step 10(c) must be repeated until the generated value for the bounded
weather variable becomes non-negative.
The steps of this version of the weather generator model are repeated as long as
the synthetic data are required in the simulation.
55
The weather generator algorithm described above provides daily climatic information
for a number of different variables, for different climate change scenarios. However,
this data is not sufficient for small urbanized basins with quick response times, as
floods may form within hours. Studies interested in quantifying local flooding impacts
of climate change require meteorologic data with finer temporal resolution than a
approaches exist that distribute daily precipitation into hourly intervals: paramet-
ric, and non-parametric. Parametric disaggregation procedures rely on determining
statistical factors from the observed historic record, and then estimate rainfall of
shorter durations. Non-parametric methods do not rely on statistical characteristics
of observed data, but rather include sophisticated shuffling and perturbation mecha-
nisms to achieve disaggregation. Applications and examples of both types are briefly
presented next.
In the work of Connolly et al. (1998), statistical model parameters are derived
from relatively short records and are used to simulate starting time, duration, rainfall
amount, and time to peak intensity for each event. A double exponential probabil-
ity distribution is used to generate the internal intensity of each event. Gyasi-Agyei
(2005) describes a parametric model where repetition techniques and proportional
adjusting procedures are used to disaggregate daily into sub-daily scales, while suffi-
ciently preserving statistical properties of observed data. This model prevents the
56
to accurately estimate hourly rainfall from daily totals. Sivakumar et al. (2001) point
out the difficultly in using statistic and stochastic processes in transforming rainfall
data from one scale to another, and suggest using a chaotic approach to disaggrega-
tion.
The only examples in the literature that include non-parametric examples of rain-
fall disaggregation include the work of Wojcik and Buishand (2003), and Wey (2006).
Both are variants of the method of fragments: a spacial case of K-Nearest Neighbour
resampling. The method of fragments uses observed historic data to select a set of
fragments (fractions of daily rainfall), which are used to distribute daily rainfall to-
tals. The selection of fragments closely resembles the resampling structure of the
K-Nearest Neighbour approaches.
In the present study a non-parametric approach is used for rainfall disaggregation.
The main justification for using the non-parametric approach lies in using existing
historic data as is, and not having to make assumptions regarding rainfall probability
distributions. Another advantage of non-parametric approaches is that site specific
parameters are not needed. This makes non-parametric models directly transferable
to other regions. The disaggregation method developed in this approach is a variant of
the K-Nearest Neighbour approach, and distributes daily data using resampled locally
observed data of fine temporal scales. Since disaggregation is achieved based on a
resampling algorithm from observed data, temporal characteristics of disaggregated
rainfall have a high likelihood of being preserved.
The disaggregation algorithm starts by taking the observed record of hourly (or
57
other finer resolution) data, and extracting all rainfall events from the record. Rainfall
events are defined as periods of non-zero rainfall, separated by an arbitrarily defined
period free of rain, ns . Thus, if a rainfall event starts on any given day, the algorithm
will continue recording it until ns hours are found free of rainfall. In other words, this
procedure is analogous to a filter that is used to separate the historic record (which
contains mostly zeros) into a record consisting of only rainfall events (with few zeros).
An important assumption regarding the rainfall disaggregation procedure must be
revealed at this point. The assumption consists of neglecting all rainfall events lasting
longer than 24 hours. This is because the weather generator algorithm presented in
the previous subsection operates by shuffling the historic record at a daily time step,
and thus does not explicitly consider patterns of multi day events. Nevertheless,
multi day rainfall events are indeed generated with the weather generator algorithm.
Furthermore, events of longer durations have typically smaller average intensity, and
are thus not critical in the study of annual extremes required for this project. Of
more interest to this project are short-term high-intensity events normally produced
by convective type storms.
After hourly rainfall events are extracted from the historic record (for each sta-
tion with available data), they are disaggregated based on the K-Nearest Neighbour
approach. This part of the algorithm starts by reading daily rainfall produced by
the weather generator for day t, for each station considered. Then, a set of poten-
tial neighbouring events is selected from the observed record, and one such event is
chosen. The daily weather generator value is then disaggregated based on this event.
The rule for selecting neighbouring events from the observed record are based on
the following two criteria:
1. Only events within a moving window of wh days are selected for further consider-
58
the historic record of events for all stations, as a potential set of neighbours.
2. In order for the disaggregation procedure to work properly, daily totals (ob-
tained from the weather generator model) are compared to the set of neighbour-
ing event totals; this step is performed in order to make sure disaggregation of
like events takes place. Only those events are selected whose totals fall between
a lower (lpd ) and an upper bound (upd ) of the daily amount to be disaggregated
(l and u are the lower and upper bound fraction, and pd the daily total rainfall
for day t). This selection process ensures that if a daily amount of 100 mm is
to be disaggregated, with upper and lower fractions of 1.2 and 0.8, all events
within the prescribed window with totals between 120 and 80 mm are selected
as a potential set of events from which to choose.
After a smaller set of events are selected, one of the events is chosen at random and
used in the disaggregation for that day. It is possible for the generated daily rainfall
to be much greater than all event totals within the prescribed moving window; in
such a case, the event with the highest total within the moving window is selected for
disaggregation purposes. This procedure is applied for all days and stations in the
study area.
cate the inter-station character to the observed (hourly) data. (Note that using the
daily weather generator does ensure replication of the inter-station characteristics of
the observed daily data.) For studies looking at rainfall intensity-duration-frequency
of single sites, using the above disaggregation method is entirely appropriate as the
In order to test the output of the weather generator model, a number of different
statistical tools are traditionally employed. Definition of tools employed in this report
are given in this section.
One way of presenting a summary of statistical data graphically is via box and
whisker plots. These plots show the lower quartile (25th percentile), median (or
50th percentile), and upper quartiles (75th percentile) of data with boxes, while its
whiskers are typically constructed with minimum and maximum data extents from
the boxes. Box and whisker plots are used in this report to show the comparison
between generated and observed data.
Three types of box and whisker plots are used in this report: monthly totals, auto
and cross-correlations. The first box and whisker plot is one that shows monthly
totals, where the horizontal axis shows the month of year, while the vertical axis
displays the monthly total. The second type of box and whisker plot is one where
auto-correlation of data is presented. Auto-correlation is a statistical tool used to
test the correlation of a time series with a lagged version of itself, for the purpose of
identifying patterns of randomness. It is also used to identify a temporal character
60
n−k
1 X
autocorr(k) = (xi − x̄)(xi+k − x̄) (20)
(n − k)σ 2 i=1
where k is the time lag, xi is the data value, x̄ the mean and σ 2 the variance of
where xi and yi are two time series for which comparison is to be made. Similarly,
the range of cross-correlation is between [-1,+1], with zero indicating no correlation,
while its bounds show maximum correlation.
The extent of the generated data displayed with box and whisker plots is shown
on a monthly time scale. This means that the weather generator output is aggregated
to a monthly time step, in cases when generated monthly totals are contrasted with
totals historically observed. Auto and cross-correlations are computed on a daily time
61
step and then averaged on a monthly basis before being shown as outputs in box and
whisker plots.
IDF analysis is typically used to capture the essential characteristics of point rain-
fall for shorter durations. Such analysis provides a convenient tool for statistically
summarizing regional rainfall information, and is often used in municipal storm water
management and other engineering design application. The IDF analysis starts by
gathering time series records of different durations. After time series data is gath-
ered, annual extremes are extracted from the record for each duration. The annual
tributions to fit a time series of rainfall data. The most widely accepted probability
distribution used in analysis of extreme rainfall statistics is the Gumbel Extreme
Value I distribution (WMO, 1981). Although other distributions are sometimes used,
the Gumbel distribution is used in Canadian engineering practice (Watt et al., 1989;
MTO, 1997; Vasiljevic, 2007).
Gumbel’s type I extreme value distribution is used in this study to fit the annual
extremes of rainfall. The Gumbel probability distribution has the following form
(Watt et al., 1989):
xT = µx + KT σx (22)
where xT represents the magnitude of the T -year event, µx and σx are the mean
and standard deviation of the annual maximum series, and KT a frequency factor
62
depending on the return period, T . The frequency factor KT is obtained using the
relationship: √ " #
− 6 T
KT = 0.5772 + ln ln (23)
π T −1
The Meteorological Service of Canada (MSC) uses this method to estimate rainfall
frequency for durations of 5, 10, 15 and 30 minutes, as well as for 1, 2, 6, 12 and 24
hours. However, most stations do not have time series records for durations shorter
than 1 hour, and therefore character of shorter rainfall durations must somehow be
estimated. The World Meteorological Organization (WMO, 1994), provides a method
to estimate short-duration rainfall, where:
Average ratios of rainfall amounts for five-, 10-, 15- and 30-minutes to one-hour
amounts, computed from hundreds of station-years of records, are often used
for estimating rainfall-frequency data for these short-durations. These ratios,
which have an average error of less than 10 per cent, are:
Duration (min) 5 10 15 30
Ratio (n-min to 60-min) 0.29 0.45 0.57 0.79
Use of the above data implies that if a 10-year one-hour rainfall is 70 mm, then
the 10-year 15-minute rainfall is 0.57 × 70 ≈ 40 mm. For additional information on
this and other statistical distributions as they pertain to rainfall, see Chow (1964),
Benjamin and Cornell (1970) and WMO (1994).
The IDF data derived using above methods is typically fitted to a continuous
function, in order to make the process of IDF data interpolation faster. For example,
a designer may require a 10-year intensity for a duration of 45 min, which is not readily
available in the published IDF data. Ontario’s Drainage Management Manual (MTO,
1997) recommends fitting the IDF data to the following, three-parameter function:
A
i= (24)
(td + B)C
63
where i is the rainfall intensity (mm/hr), td the rainfall duration (min), and A, B,
and C are coefficients. After selecting a reasonable value of parameter B, the method
of least squares is used to estimate the values of A and C. The calculation is repeated
for a number of different values of B, in order to achieve the closest possible fit of the
data. Details of this procedure are provided in MTO (1997), Chapter 8. After IDF
data is fitted to the above function, plots of rainfall intensity vs. duration (for each
return period) are produced.
The classical (or traditional) approach to climate change impact assessment of water
resources systems includes the use of GCM emission scenarios. As discussed in Sec-
tion 2.1, GCM outputs (temperature and precipitation) are scaled down from global
to regional scales (either with dynamical or statistical downscaling techniques), and
are used as inputs into hydrologic models. Hydrologic models simulate the movement
of water and moisture over land and water bodies. These models use climatic vari-
ables as input (temperature, precipitation), as well as catchment characteristics (size,
shape, slope, land use, storage capacity, etc.) to simulate the flow of water between
various components in the hydrologic cycle (canopy interception, surface runoff, soil
infiltration, ground water percolation, etc.). Changes in the climate signal results in
changes to the regional hydrologic flow regimes, and have the potential to adversely
affect a variety of human and natural systems (see Section 1.2).
The study by Southam et al. (1999) evaluate impacts of climate change for On-
tario’s Grand River Basin, under 21 scenarios of future surface water supply, reservoir
operation, population growth and water use. The conclusion of their research points
out that climate change may bring serious impacts on the capability of the Grand
64
River to assimilate waste water and provide a reliable source of water for municipal
purposes, while maintaining existing water quality standards. The hydrologic water
balance model used by Southam et al. (1999) operates on a monthly time step, and
uses 14 nodes to represent regional hydrologic characteristics of a basin 6,800 km2 in
areal extent. Their model is not able to consider ground water (which may be used
as additional water supply in the future), nor capture possible changes in flooding
characteristics under future climate scenarios (its time step is too large).
Research by Palmer et al. (2004) looks at the potential impacts of climate change
on a river basin in northwest Oregon, with a focus on reservoir operation for the pur-
poses of water supply. Outputs of six GCMs are downscaled and used as inputs into
a continuous hydrologic model. The study findings point to significant influences of
annual streamflow patterns (decreased summer flows between 10 and 20%, increased
winter precipitation and temperatures, higher risks of droughts). Such changes will
reduce the ability of local reservoirs in providing water supply.
In the study by Coulibaly and Dibike (2004), the authors compare a number of dif-
ferent downscaling techniques and hydrologic models to assess the impacts of climate
change on flood flow frequency for the Saguenay River system in Quebec. They find
significant differences between downscaling methods themselves (i.e., some downscal-
ing techniques produce an increase, while others a decrease in mean annual flow for
the same global input). The choice of downscaling techniques can drive the outcome
of the analysis, and thus can mask the overall system behaviour under the conditions
of altered climate. The choice of climate models themselves (and their associated data
sets) may also significantly impact the overall results. The work by Simonovic and Li
(2003) shows that input data (obtained from three different climate models) is sub-
stantially different, thus explaining large variations in the final results. Consequently,
the end-users (water management authorities, government policy makers and other
65
stakeholders) have became sceptical of results of such analyses, and can rarely form
new guidelines and management strategies as a result of them.
An alternative to classical approaches (of downscaling GCM outputs for use in
local hydrologic impact assessment) is the inverse approach. The inverse approach
constitutes a set of steps used for evaluating hydrologic risk and vulnerability to
changing climatic conditions at a local scale. At its core is a weather generator
algorithm that produces arbitrary long, time series records of climatic data, eventually
used as inputs for hydrologic models. The weather generator algorithms employ
various shuffling and perturbation mechanisms to local climate data, thus generating
a statistically similar, locally applicable, climate record. Weather generators can be
conditioned on GCM outputs, and thus can take into account directions of change
consistent with more detailed global models. Avoiding downscaling of GCM outputs
for use in local climate change impact assessments is the greatest advantage of the
inverse approach.
The strength of the inverse approach is that it addresses some of the deficiencies
of the classical approach, and it provides for the active participation of end users
in the assessment of climate change impacts. Such involvement builds confidence
that the obtained results may be used as a basis for further action. The inverse
approach as implemented in this study includes the following steps (taken from
Prodanovic and Simonovic, 2007c):
of water resources systems are first identified, together with their risk/hazard
levels. The end-users are involved at this stage, as they are most familiar
66
with the particulars of the water resource system under consideration. This
stage may simply involve consultation with end users and gathering information
on extreme events of interest to them. The user may specify a particular set
of threshold values that, when exceeded, lead to extensive damage from high
precipitation).
GCMs.
4. Weather generator output is used as input into a hydrologic model for trans-
forming meteorologic conditions into their corresponding hydrologic exposures.
As a large number of meteorologic conditions is used in the simulation, an
67
5. After a frequency analysis is completed for each climate scenario, the critical
hydrologic exposures from step 1 are used to look up their recurrence intervals
for present and changing climatic conditions. This way the user of the water
resources system may estimate local shifts in the frequency of occurrence of
critical hydrologic exposures resulting from changed climatic conditions.
The inverse approach offers an alternative method for the analyses of climate
change impacts and addresses many of the limitations of the classical approach. Its
efficiency is strongly dependent on the tools used for the simulation of meteorological
conditions (weather generator) and on the transformation of meteorologic conditions
into corresponding hydrologic situations (hydrologic model). All of the limitations of
weather generator and hydrologic models selected for the implementation of the in-
verse approach affect the quality of the overall results. However, the inverse approach
is quite flexible for use of new, better, tools that may result from future research
efforts.
gain insight into natural phenomena resulting from extreme meteorological events (ex-
cess or deficit of precipitation). This insight can be used as a basis for future decision
making and for formulating management strategies and plans.
Depending on the detail required in modeling projects, hydrologic models are di-
1988, p. 313). They typically have no provision for moisture balance between storms,
and are therefore not appropriate for analysis and simulation of dry weather flows
(i.e., seasonal or long term flows). Event models are typically used to ascertain re-
gional flooding characteristics, to determine the extent of floodplains, to evaluate
input rainfall (Linsley and Franzini, 1979, p. 67). This means that a T-year return
period precipitation corresponds exactly to a T-year return period flow, which may
not be true. Suppose, for example, that a region receives a large amount of precipi-
tation in a short time interval; also, assume that the region was previously very dry,
and hence its soils depleted of moisture. Even though rainfall is of a large magnitude,
the produced flow may not correspond exactly to the return period of the rainfall. A
100-yr return period rainfall may produce only a 25-yr flood flow. The implication
of this assumption is that proper flood flow frequency analysis can not be performed
with event-based hydrologic models.
One way around this limitation is to use a continuous-based hydrologic model, and
69
thus take into account detailed soil moisture balance of both dry and wet weather.
Then, and only then, will a true flood flow frequency analysis be performed, as hydro-
graphs are produced year round (during both wet and dry periods). Continuous-based
hydrologic models simulate the long term rainfall-runoff process, over large amounts
of pervious land (Bedient and Huber, 1988, p. 313). Unlike event-based models, con-
tinuous models take into consideration long term soil moisture balance; they are
therefore applicable for simulations of daily, monthly and seasonal flows. Continuous
hydrologic models are usually not well suited for studies of extreme flood flow, because
biasing their calibration to extreme flood events lowers their performance in repro-
ducing long term low flow characteristics (Bennett, 1998; Cunderlik and Simonovic,
2005). Despite these limitations, event-based models are the best tools for simulating
flood flows, while continuous-based models are best for longer term flow analysis (i.e.,
models. Because the goal of event-based models is to capture basin response to short-
duration high-intensity events, they need not consider long term detailed movement
of moisture nor their associated evaporation or evapotranspiration losses. This is
warranted because evaporation and evapotranspiration losses are negligible during
the course of an event. The event-based models are usually much easier to set up and
70
(USACE, 2006); Storm Water Management Model, SWMM (Huber and Dickinson,
1988; Rossman, 2004); Visual OTTHYMO (Greenland International Consulting Inc.,,
2001), as well as others. The most common continuous models are WATFLOOD
(Kouwen, 2001), Hydrologic Simulation Program FORTRAN, HSP-F (Bicknell et al.,
• Do model outputs estimate flow peaks (stage, discharge), volumes, and hydro-
graphs at special interest points such as junctions, reservoirs, weirs and other
hydraulic structures?
• Can main hydrologic processes be captured by the model and estimate the
• Can the inputs required by the model be provided within a reasonable time with
minimal costs? This includes stage-discharge relationships for river reaches
and reservoirs, precipitation and temperature data, potential evapotranspira-
tion data, land use, soil types, and other physiographic data, etc.
71
All HEC-HMS models consists of three parts: (i) a meteorologic module (which
includes methods describing precipitation input and/or calculation of evapotranspi-
ration), (ii) a basin module (consisting of methods describing the physical properties
of a catchment), and (iii) a control module (where start and end times of a simulation
are specified). The meteorologic and basin modules consist of a collection of methods
allowing the user to specify and describe climatic and physical properties of the basin.
For example, different loss methods (i.e., representing evaporation and/or evapotran-
spiration) are available depending on whether the user wishes to study short or long
term hydrologic characteristics of the basin. Event and continuous hydrologic models
are used in the inverse approach to evaluate the response of climate change to water
resources systems.
72
The HEC-HMS event based hydrologic models uses the following set of methods to
Losses
cepted by vegetation (or surface depressions), or infiltrated directly into the soil. In
this study, Initial and Constant loss rate method is used (USACE, 2000, p. 46). This
method satisfies a specified initial loss at the start of the event (water intercepted
by vegetation and/or stored in surface depressions), followed by a constant loss rate
(mimicking the process of soil infiltration) for the time remaining in the precipitation
event.
The Initial and Constant loss rate method works by taking a discrete value of
precipitation, pt as input, and calculating precipitation excess, Et , as output using
the following:
P
0 if pi < Ia
Et = pt − fc (25)
P
if pi > Ia and pt > fc
P
0
if pi > Ia and pt < fc
P
where pi represents the sum of precipitation from i = 0, 1, 2, . . . , t, while Ia the
initial, and fe the constant rate of precipitation loss. The parameter Ia captures the
initial losses due to interception and depression storage, while parameter fe represents
the portion of precipitation that infiltrates into the soil.
Equation (25) shows that initial loss is satisfied first (interception and surface
73
etc.), losses are assumed as zero. Surface excess is used as input to a transform
method that converts it into direct runoff by convoluting a hydrograph.
Transform
In order to transform surface excess to direct runoff, unit hydrograph theory is typi-
cally employed. Clark’s unit hydrograph is used in this study (Hoggan, 1996; USACE,
where AI is the cumulative fraction of a basin area and t/Tc is a fraction of the time
of concentration. Adoption of such a relationship assumes that velocity of overland
flow is uniformly distributed over the catchment area, and that the time required for
runoff to reach the outlet is directly proportional to distance travelled (Hoggan, 1996,
Flow, Q
4
2
0 1 2 3 4 5 Time, t
1
Tc
(a) Time Area Map of Basin (b) Translational Hydrograph
Flow, Q
Storage
R
Q
1 Outflow
Time, t
(d) Instantaneous Unit Hydrograph (c) Linear Reservoir Routing
size, and time of concentration (the time needed for a drop of water falling on the
farthest point in the catchment to reach its outlet, denoted by Tc ). The translational
unit hydrograph describes the flow passing through the catchment outlet if no at-
tenuation takes place. A translational hydrograph can be computed by knowing the
time-area relationship, along with a basin’s total area. The volume of the transla-
tional hydrograph corresponds to the uniform rainfall of one unit (either mm or in.)
falling over the basin for the duration equivalent to the time of concentration.
Since hydrograph attenuation does occur, the translational hydrograph is thus
routed through a linear reservoir [Figure 3(c)]. At this step the slope of the storage-
outflow function is specified (represented by R), as it is needed to route the trans-
lational hydrograph through a linear reservoir so as to obtain an instantaneous unit
hydrograph. The linear reservoir routing is represented by a discrete approximation
Ot−1 + Ot St − St−1
I¯t − = (28)
2 ∆t
where I¯t represents an average inflow at time t, or the ordinate of the translational
hydrograph, and Ot and St represent the outflow and storage during ∆t. Storage in
the linear reservoir is approximated by the following relationship:
St = ROt (29)
Substituting
2∆t
c= (31)
2R + ∆t
where Ut represents the ordinate of the unit hydrograph. Finally, to obtain direct
runoff of the catchment Qdt , the unit hydrograph is transformed (via the convolution
equation in discrete form):
t
X
Qdt = Ei Ut−i+1 (34)
i=1
where Ei is the rainfall excess, computed previously. Direct runoff is then added to
Baseflow
The recession method computes baseflow differently at the start and mid way
77
Peak flow, Qp
Flow, Q
Initial
Baseflow, B0 Threshold flow, Q∗
Baseflow, Bt
Threshold time, t∗
Time, t
through the event (USACE, 2000). From the start of the event until a threshold
value on the recession side of the hydrograph, baseflow is calculated using equation
(35). After that point, total flow is computed first using:
Qt = Q∗ k (t−t∗ ) (36)
where Q∗ is the threshold flow, t∗ the threshold time, and Q the total flow. Threshold
flow Q∗ is computed by:
Q∗ = aQp (37)
where a is a peak-to-threshold ratio, and Qp the peak of the total flow hydrograph.
Baseflow for the period after t∗ is then calculated as total flow minus direct runoff.
Bt = Qt − Qdt (38)
The reservoir and river routing component of the hydrologic model describes the
movement of a flood wave (total runoff from multiple catchments) as it passes through
a series of river reaches and/or a reservoir. The mechanism employed here is a simple
hydrologic routing method (Chow, 1956; Hoggan, 1996; USACE, 2000), referred to as
the Modified Puls method. Its computational procedure is analogous to one described
for the linear reservoir method above, with the only exception being that non-linear
storage-outflow functions are employed here. The method takes a hydrograph as
input [Figure 5(a)], sends it through a nonlinear reservoir [Figure 5(b)], and produces
a modified flood hydrograph [Figure 5(c)] as output. The assumption here is that
a channel (or a reservoir) provides enough storage and/or resistance to flatten and
attenuate the original hydrograph.
79
Flow, Q
Time, t
(a) Inflow Hydrograph
Inflow
Flow, Q
Outflow
Storage
Q
Outflow
Time, t
(c) Outflow Hydrograph (b) Reservoir/River Routing
The continuous hydrologic modeling used in this study is somewhat different than
the original HEC-HMS framework, because it is adapted for the needs of the cur-
rent problem (assessment of risk and vulnerability to changing climatic conditions).
The hydrologic model is divided into a number of different modules, each represent-
ing different processes captured by the model. The overall structure comprising the
continuous model is given next, followed by details of the individual components (or
modules).
The overall structure of the continuous hydrologic model is shown in Figure 6, where
each box represents a module that mathematically represents a physical processes
functioning in the basin. Precipitation and temperature (obtained from the weather
generator model) are used as inputs to a snow module, where adjustments are made
to account for both solid and liquid precipitation. The output of the snow module
is adjusted precipitation, used for computation of losses (used here in plural to em-
phasize existence of more than one kind of water loss). The losses module represents
returned to the stream from ground water), surface excess (the portion of the flow
that does not infiltrate into the soil), and ground water recharge (the flow that enters
deep aquifers and does not return to the stream). The surface excess is used by a
transform module, or one which converts surface excess to direct runoff by convolut-
81
Temperature
Precipitation
Snow
Module Evapotranspiration
Adjusted
Precipitation
Losses Transform
Module Module
Surface
Excess Surface
Runoff
Baseflow Direct
GW Recharge Runoff
Routing
Module
Streamflow
ing a unit hydrograph. The output of the transform module is surface runoff, which
is then combined with baseflow to produce direct runoff. Direct runoff, on the other
hand, is used as input of a flood routing module, which calculates the propagation of
a flood wave eventually producing channel streamflow.
The snow module (shown in Figure 7) takes time series data of temperature and pre-
cipitation (provided by a daily weather generator model) and uses it to compute snow
accumulation and melt, based on the degree-day method (Cunderlik and Simonovic,
2004b, p. 65). An interpolation method is used to spatially interpolate the precip-
itation and temperature, as the climatic data provided is usually available only for
selected stations within and around the basin. Such a method is needed because the
hydrologic models typically contain multiple subcatchments, which receive precipi-
tation as input. The interpolation of temperature and precipitation is accomplished
with the Inverse Distance Weighting Method (USACE, 2000, p. 23). After the cli-
matic data is interpolated, it is separated into solid and liquid forms of precipitation.
The solid form of precipitation (or snowfall) is then subject to an accumulation and
melt algorithm, eventually producing snowmelt. Snowmelt is then added to liquid
precipitation (or rainfall), producing a new variable defined as adjusted precipitation.
The adjusted precipitation is used next as input to the losses module and the rest of
the model.
The set of equations that represent snow accumulation and melt module are pre-
sented next. After the climatic data (temperature and precipitation) is interpolated,
the algorithm starts. If the average daily temperature Tt is less than the minimum
83
Weather
Generator
Data
Temperature Precipitation
Interpolate
Process
Temperature Precipitation
Separation Snowmelt
Process Snowfall Process
Rainfall
Snowmelt
Adjusted
Precipitation
St = P t (39)
Rt = 0.0 (40)
" #
T+ − Tt
St = Pt · (41)
T+ − T−
Rt = Pt − St (42)
Similarly, if the average daily temperature is greater than the maximum temperature
for snowmelt, all precipitation is assumed to fall as rainfall, represented with:
St = 0.0 (43)
Rt = Pt (44)
where MR is a parameter representing melt rate [mm/◦ C/day], and is set to 4.0; Tcr
85
is a critical temperature for melt, and is set at zero. Snowmelt previously obtained
is accumulated with the following equation:
St = St + St−1 (46)
St = St − Mt (47)
Pa = Rt + Mt (48)
where Pa represents adjusted precipitation in [mm/day]. If, on the other hand, all
accumulated snow melts, the adjusted precipitation becomes:
Pa = Rt + St (49)
Lastly, if no snowmelt occurs, the adjusted precipitation takes on simply the vale of
rainfall.
Pa = Rt (50)
The most complicated component of the hydrologic model is the losses module. This is
because it takes into consideration a large number of processes that are simultaneously
functioning, which must be properly accounted for. The conceptual diagram of this
86
module is shown in Figure 8 and is based on the work of Bennett (1998), where details
of the Soil Moisture Accounting algorithm are presented. Furthermore, Bennett’s
algorithm is based on the work of Leavesley et al. (1983), and their Precipitation-
Runoff modeling System.
The losses module uses a series of conceptual reservoirs to represent the storage
and movement of water in each subcatchment of the basin. The storage reservoirs are
the following: (i) canopy interception, (ii) surface interception, (iii) soil profile, and
(iv) a number of ground water layers (only two shown here). The inflow and outflow
rates between the reservoirs regulate the amount of water stored in each conceptual
reservoir. These rates include evapotranspiration, infiltration, percolation, surface
runoff and ground water flow.
The canopy storage layer includes the precipitation captured by vegetation (such
as trees, shrubs, bushes, grasses, etc); furthermore, the precipitation is the only inflow
that can fill this storage volume. When precipitation occurs, it fills this storage
layer first, provided it is not already at capacity. The only process that can remove
the moisture out of this layer is the process of evapotranspiration. After filling the
canopy layer, the precipitation becomes available to fill the surface storage, and/or to
infiltrate into the soil. The surface storage layer represents the volume of water held
by shallow depressions and cracks on the ground surface. The water stored in this
layer (in addition to precipitation excess from the canopy) is available to infiltrate
into the soil, provided that the soil is not fully saturated. The inflow to the surface
storage layer is water that does not infiltrate the soil layer; in other words, this water
is a combination of the precipitation excess from the canopy layer, and its own volume
left over after infiltration took place. The outflow from this layer is evaporation and
surface runoff. The surface runoff is the flow produced when the surface storage layer
is at capacity, and thus can not absorb water that previously has not infiltrated.
87
Precipitation Evapotranspiration
Canopy
Interception
Storage
Surface
Interception
Storage
Surface Excess
Soil Storage
Tension Upper
Infiltration Zone Zone
Percolation
GW Flow 1
Ground Water
Layer Storage 1
Percolation
Deep Percolation
During large precipitation events, the canopy and surface storage layers become full
quickly, thus producing high amounts of surface excess—as infiltration alone is usually
never able to absorb all surplus precipitation.
The soil profile storage layer corresponds to the top layer of the soil. Water that
infiltrates is the only inflow to this layer, while the outflows include percolation to
the lower ground water layer and evapotranspiration. Soil storage is further divided
into two zones: the upper zone and the tension zone. The former zone is that portion
of the soil layer that can lose water to both percolation and evapotranspiration,
while the latter zone is one that looses water only through evaporation, but not
percolation (Bennett, 1998, p. 5). This is because the upper zone represents water
held in the pores of the soil (which can freely percolate and/or evaporate), while the
tension zone constitutes water held by capillary tension. It should be mentioned that
evapotranspiration rates from the soil vary, as it is more difficult to remove water
held by capillary tension than water held between the pores of the soil.
Evapotranspiration is the process that removes moisture from canopy, surface, and
soil profile storage. In the Soil Moisture Accounting algorithm, evapotranspiration
The water that percolates from the soil profile storage is used as an input to the
ground water layer immediately beneath. The outflows from this layer represent the
ground water flow (one that is returned to the stream channel as baseflow), and further
percolation either to another ground water layer or to deep percolation—representing
Since all equations for the Soil Moisture Algorithm can not be reproduced here
(Bennett (1998) takes 90 pages to do it), only few key relationship are presented. The
following set of differential equations represent the dynamics of canopy A, surface B,
soil C, and top D and bottom E ground water layers, all in [mm]:
ETtB and soil ETtC storage layers), PtB precipitation going beyond the canopy, StE
surface excess, It infiltration, Rt percolation (from/to soil RtC , ground water 1 RtD
and 2 RtE layers), and GW Ft lateral ground water flow (from layers 1 GW FtD and 2
GW FtE ), all in [mm/hr].
The process of soil infiltration in the Soil Moisture Algorithm is captured by first
as the absolute maximum), and depends linearly on the current volume of water held
in the soil. If the soil contains little or no water, the potential infiltration can be
90
as high as the infiltration capacity; however, if the soil is at, or near saturation,
the potential infiltration will be small. The actual soil infiltration It at time t is
computed as the minimum value between the water available for infiltration W At
and the potential soil infiltration Itp . Therefore,
reflecting that the amount that actually infiltrates can never be more than is currently
available, despite how high the potential infiltration is.
Similar to the process of infiltration is percolation, or the water that is transferred
from the soil storage to a ground water layer immediately beneath. The potential soil
RtC,p = Rm
C
(Ct /Cm ) · (1 − Dt /Dm ) (58)
C
where Rm stands for maximum soil percolation in [mm/hr]; Ct for water volume in
soil storage in [mm]; Cm for soil storage capacity in [mm]; Dt for current ground water
layer storage, and Dm for maximum ground water storage both in [mm]. Equation
(58) indicates that potential soil percolation depends on amounts of water in soil and
ground water. For example, if the ground water storage is at or near saturation, large
amounts of water will not be able to percolate to the ground water aquifer. Potential
percolation also depends on the amount of water currently in the soil, demonstrating
that if soil is void of water, no percolation can occur. Actual soil percolation (RtC )
is computed in the same manner as actual soil infiltration, and thus is the mini-
mum between the potential soil percolation and the amount of water available for
percolation.
91
The outflows from the ground water layers represent the lateral ground water flow
(one that is eventually returned to the stream channel as baseflow). The Soil Moisture
Algorithm represents this lateral flow as:
where GW Ft stands for ground water flow in [mm/hr]; P GW Pt for potential ground
water percolation in [mm/hr], and k for a ground water storage coefficient [hr]. The
derivation of this equation is presented in (Bennett, 1998, p. 57). Such lateral flow
from both ground water layers is averaged, before being routed with a series of linear
reservoirs to produce baseflow.
Clark’s method is used to convert surface excess (obtained from the Soil Moisture Ac-
counting algorithm) into direct runoff. An identical formulation to the one presented
for the event model is used. Similarly, the Modified Puls method is used for rout-
ing flow in river reaches and in reservoirs, and a series of linear reservoirs method is
used to transform lateral ground water flow (obtained as outputs from Soil Moisture
Accounting algorithm) into baseflow.
The procedure applied to calibrate and verify hydrologic models starts after the flow
and precipitation regime of the study area are assessed. Knowledge of spacial and
temporal aspects of streamflow and precipitation must be acquired, as well as past
density of the observed hourly streamflow and rainfall record, (ii) a number of differ-
ent rainfall events, (iii) adequate spacial coverage of rainfall-runoff events, preferably
extending over the entire basin, and (iv) durations of rainfall events exceeding basin’s
the time of concentration.
Similarly, criteria for selection of data for calibration and verification for contin-
uous hydrologic modeling include (Cunderlik and Simonovic, 2003, p. 21): (i) long
sequences of daily streamflow, precipitation and temperature data; (ii) representa-
tive hydrologic variability (mean as well as extremes) in the selected sequences; (iii)
followed by initial parameter estimates. Initial parameter sets are estimated from
known physiographic properties of the basin (land use characteristics, soil types,
etc.), as well as from modeling experience. The hydrologic model is then simulated,
and model output is compared to observed flow hydrographs. A number of indices
used in comparing simulated to observed hydrologic characteristics are presented in
ulating a sequence of flows not previously used (i.e., not used in calibration), and
comparing the results with observations for the same time period. The verification
procedure acts as a test of how good the calibrated model parameter set is. Sensi-
tivity on the other hand, simply alters parameter values incrementally with the aim
of determining which has the largest influence on computed model behaviour. The
limitations of models are therefore often revealed through sensitivity analysis.
Flow frequency analysis is often employed to synthesize extreme flows (flood peaks
and/or low flows) to a theoretical probability distribution, thus quantifying and stan-
ical details of these and other methods applicable to hydrologic statistical analysis,
the reader is referred to Benjamin and Cornell (1970), Linsley and Franzini (1979),
Bedient and Huber (1988), Watt et al. (1989), and Hoggan (1996).
Outputs of frequency analysis are typically shown as plots, with frequency (or
return period) shown on the horizontal, and the quantity of interest (peak or low flow,
rainfall) on the vertical axis. Such plots are useful in comparing impacts resulting
from changed climatic or policy conditions that can eventually lead to formulations
94
This section aims to introduce a new modeling framework that dynamically inte-
grates and links various components used in the modeling of water resources. This
approach extends the traditional way of water resources modeling, as it not only
includes models of physical processes, but also models of socio-economic processes.
A number of models have been developed that provide support to the integrated
nature of water resources management (Cai et al., 2002, 2003; Ward et al., 2006;
Mainuddin et al., 2007). All cited studies combine physical based management of
water resources (through hydrologic water balance models or exogenous inputs) with
mathematical optimization techniques that quantify regional economic characteris-
tics. Furthermore, they explore interconnections between hydrologic basin response
and socio-economic policy options, and thus have the potential to be useful in various
integrated water resources management projects. The goal of the above studies is
to determine optimal water allocation policies among conflicting water users, while
minimizing costs and maximizing profits.
The methodology developed in this work focuses on developing a modeling ap-
proach that assists the process of integrated water resources management through
system simulation (as opposed to using optimization). The emphasis is placed on
modeling and simulation of key characteristics of complex water resources systems
including: a feedback based system structure, representations of physical and socio-
economic processes (and their linkages), complexities of spacial and temporal scales,
multiple stakeholder participation and involvement, and long term planning and man-
agement. In other words, the proposed methodology provides for an integrated way to
95
simulate complex water resources systems and to capture, in a single model, detailed
physical and socio-economic processes.
The system dynamics simulation method can be used effectively for coupling phys-
ically based hydrologic processes to larger scale socio-economic activities. In current
systems model, developed using system dynamics (p. 30). Their integrated model
simulates the storage and movement of water throughout the basin for current and
future operational policies, and computes flows under current and future safe yields.
However, feedbacks between the hydrologic and operational components are entirely
missing from their model, as it is assumed that operational policies and management
practices do not significantly alter regional hydrologic characteristics.
Water resources management studies consider a large number of socio-economic
processes, which include population growth, urbanization, water use rates, or man-
mate population growth rates, industrial production, jobs, urbanization and develop-
96
ment, among others. These models (upon which most urban master plans are based)
normally use social, political and administrative constructs of current structures of
government as their basis for decision-making regarding development and land use,
social, economic and environmental prosperity, etc. The time horizon of such models
is typically a few decades at most. It should be said that these socio-economic models
also use the latest methods and techniques of their field, and represent the state of
the art.
Many in the current water resources management practice do not emphasize dy-
namic feedback links between physical and socio-economic components. Such a de-
coupled approach misses some of the key elements that are a part of today’s complex
systems, as it assumes that (from Simonovic and Davies, 2006):
1. We can predict the character of all interactions between the two systems,
despite their [individual] nonlinear nature.
2. The interconnections between the systems do not fundamentally deter-
mine the future state of the [region]. In other words, the interactions
between these systems are largely irrelevant to the behaviour of each.
3. The biophysical and socio-economic systems are essentially separate, so
that feedbacks between the systems are external to both.
4. We can separate human effects ... from natural effects (p. 432).
Classical water resources management modeling can not help and bring about
better understand regarding the functioning of complex water resources systems, de-
spite exchanges of information that take place when scenarios are compared. The
fundamental building block of systems, the feedback loop, is not evident in the clas-
sical approaches when physical and socio-economic links are considered; the systemic
framework of problem solving is also not considered. Both of these are absolutely
necessary to better understand the functioning of socio-economic processes and the
physical world in which they are embedded in.
97
a result, knowledge and information is rarely generated that challenges the ways in
which the organizational structure of current water resources management is set and
operates. Oftentimes, defensive routines exist that protect the current status quo; in
classical studies examples of such defensiveness are never brought to the surface, where
of climate change on water resources systems and their management. Integrated wa-
ter resources modeling, implemented using system dynamics, has an ability to capture
the complexity of both physical (hydrologic, climatic) and socio-economic (popula-
tion, land and water use) systems, as well as their interactions. The flexibility of using
scenarios, together with an ability to examine the sensitivity of different management
strategies, give the methodology an added practical benefit (Simonovic and Li, 2003,
p. 365). The modeling framework can also help uncover key leverage points, where
small changes (often found in unexpected places) can bring substantive positive and
encouraging effects.
Progress by practitioners in this direction has indeed been slow. One encouraging
98
Despite this realization, the report by the Policy Planning Section (which is some
320 pages long, including the appendices) provides no concrete means by which these
feedback links could be formulated, tested, or otherwise examined. In the above re-
port, the systemic structure of the socio-economic systems is never explicitly defined—
not with feedback loops or by other means. The report does offer ample verbal
descriptions and associated statistical information which provide the basis for a de-
scription of a systemic structure, but such a structure is never explicitly formulated
or proposed. This means that the systemic structure can never be questioned, and its
defencive routines (responsible for maintaining the status quo) can never be exposed.
The report does, however, provide ample socio-economic data containing “snapshots
capturing information on demographics, local economic clusters and niche markets,
land uses and the environment” (MMAH, 2002, p. 1). This quotation simply illus-
trates the trend in classical studies of valuing increasing data resolution of individual
components, while altogether neglecting dynamic feedback interrelationships between
be brought to surface, and thus properly examined. Evidence shows that people who
use the systemic way of thinking, coupled with the concept of feedback and computer
simulation models, experience genuine learning incapable of being produced otherwise
(Senge, 1990; Senge et al., 1994; Sterman, 2000).
eling lies in its ability to capture, in a single unifying model, systems and subsystems
that would traditionally be studied separately (such as hydrologic and socio-economic
characteristics). Understanding the behaviour of complex dynamic systems demands
the use of such an integrated framework. As a first step in building integrated models
of this kind, causal loop diagrams are used as precursors to building detailed system
dynamics models. They are presented next.
Before system dynamics models are built, causal loop diagrams are used in prelimi-
nary modeling stages to formulate dynamic hypotheses regarding causes of behaviour
under study. These diagrams are maps that show how variables within a problem
might be related to one another. Causal loops quickly capture the feedback struc-
ture of systems, elicit mental models of individuals, and communicate important
feedbacks that are thought to be responsible for overall system behaviour (Sterman,
100
2000, p. 137).
Causal loops consist of variables linked with arrows (indicating causality). Signs
shown adjacent to each arrow represent link polarity, and are defined by Sterman
(2000) in the following way:
A positive link means that if the cause increases, the effect increases above
what it would otherwise have been, and if the cause decreases, the effect de-
creases below what it would otherwise have been.
A negative link means that if the cause increases, the effect decreases below
what it would otherwise have been, and if the cause decreases, the effect in-
creases above what it would otherwise have been (p.139, emphasis in original).
Causal diagrams therefore show the direction of causal influences between vari-
ables; they also show the polarity of feedback loops (reinforcing and/or balancing)
that make up the system structure. A word of caution is offered at this point, as some
assume that system behaviour can be inferred by looking at the number of feedback
loops in a causal diagram, as well as their polarities. In fact, this is not true. It is im-
possible to infer a system’s behaviour simply by examining (or counting) the feedback
loops in a causal diagram. Such inference is impossible because behaviour of com-
plex systems exhibits dynamic shifts of feedback loop dominance, where behaviour is
governed by strength of each feedback loop. Since strength of feedback loops can not
of sound system dynamics simulation models. Originating from general systems the-
ory (and its paradigm of systems thinking), system dynamics is a school of thought
that uses feedback control theory to build computer simulation models. The funda-
mental building block of this approach is the feedback loop. The system dynamics
approach has been used to address a wide range of problems, including physically
based hydrologic processes (Li and Simonovic, 2002; Sehlke and Jacobson, 2005), or-
ganizational behaviour and management (Forrester, 1961; Senge, 1990; Richardson,
1991; Sterman, 2000), global change modeling (Meadows et al., 1982; Simonovic,
(1999) applied it to modeling of long term planning and policy analysis in the Nile
River Basin in Egypt; Ahmad and Simonovic (2000) used it in simulations of reservoir
operation for flood control, while Simonovic and Li (2003) have used it to evaluate
the impact of climate change on a large scale flood protection system for the city of
Winnipeg, Canada. Winz (2005) applies system dynamics to the study of availability,
quality and distribution of water in urban areas of Auckland, New Zealand.
The research by Li and Simonovic (2002) demonstrates one of the first applications
of system dynamics simulation and systems thinking to studying strictly hydrologic
processes in North American prairie watersheds. The work of Saysel et al. (2002)
develops a system dynamics model for the purpose of studying agricultural develop-
ment in Turkey. The model consists of sectors representing water resources, land use,
demography and agricultural pollution. Stave (2003), on the other hand, uses a very
simple model to bring about public understanding of water management options in
Las Vegas, Nevada. Problem definition similar to Saysel et al. (2002) is adopted by
102
Fernández and Selma (2004), who studied water scarcity of irrigated landscapes in
Spain. Sehlke and Jacobson (2005) proposed a model to investigate the utility of the
system dynamics approach in modeling large complex hydrologic systems and their
interaction between surface and ground water in the western United States.
System dynamics modeling is different than other modeling constructs; its central
point of view states that its model’s structure (how feedback loops are arranged and
linked together) is the source of its overall behaviour. This implies that different
model structures cause different system behaviour. The behaviour referred to here
is dynamic; in other words, it shows how systems (or their components) evolve and
change over time.
System dynamics is different than other computer simulation methodologies as it
is equally capable of describing physical and social processes. For example, hydro-
logic models only study land-based rainfall-runoff processes operating within a basin;
however, they do not take into account social, economic, political or other equally
relevant processes that affect the hydrologic cycle. Socio-economic processes are ab-
solutely critical in the management of water and other resources, because important
management decisions are rarely based on purely physical (or technical) considera-
tions alone. System dynamics has the power to integrate such components into a
single unifying framework, therefore being useful to water resources management.
The system dynamics methodology was the first computer simulation framework
to emphasize the feedback structure of social systems, and the powerful links be-
tween social and physical systems. Prior to the development of system dynamics,
the explicit linkages between social and physical domains (although acknowledged as
existent, significant, and sometimes even necessary) were not considered, because an
appropriate theory for such interaction was simply not available. Even today, some
half a century after its initial introduction, system dynamics studies receive much
103
unfounded criticism, mostly from people stuck in the traditional world view, where
compartmentalization, divide and conquer attitudes, and top-down organizational
management strategies still prevail.
System dynamics modeling, because it equally well applies to the study of both
physical and social systems, is showing promise in being able to discover and generate
knowledge that will help map ways the physical world interacts with its social realm.
In the field of water resources systems management such an approach is extremely im-
portant, as the professional practice has much to gain from applying system dynamics
to its complex problems. Even though various modeling efforts are nowadays used in
water management, for the purpose of discovering better management strategies, very
few such efforts consider how water resources systems affect and are affected by the
social and environmental systems in which they are embedded. For example, those
studying flood, drought, water quality and sedimentation rarely build models where
the impact of the water resources systems are felt in social and environmental sys-
tems, which can eventually influence the functioning of the water resources systems
themselves.
Perhaps such modeling is not done because many current water resources man-
agement models are based on representations of very detailed physical processes, and
are therefore very data intensive. Such data intensity means that large amount of
observed data be gathered to adequately calibrate and verify the models in order to
build confidence in the produced outputs. However, for modeling of social systems
such detailed data is in many cases simply not available, thereby leading many (es-
pecially those used to physical modeling) to conclude that social systems modeling
is not possible. The approach adopted in this study is that all available data should
be used, and when critical data is not available, models should still be built with as-
sumptions that are clearly stated and fully explained. Allowing critical assumptions
104
it adopts a position that embraces finding and mapping relationships that link com-
ponents of complex dynamic systems, utilizing all available data. The approach also
emphasizes long term dynamic continuous behaviour. Such an approach is entirely
different from the static world view traditionally accepted, which approximates dy-
mental models that shape decisions, and that, through action, bring about changes
in future states of the world, thus leading to new decisions (Forrester, 1961).
Definition of feedback
its input, in order to further regulate its output. The state of the system provides
information cues about an action, which then continuously acts and changes the
system’s state. As an example of a feedback consider the process of soil infiltration
common in hydrologic analysis. The moisture content of the soil (i.e., its system state)
governs how much water can infiltrate (i.e., its system action)—highly saturated soils
will allow little or no infiltration, while dry soils can absorb much more water. The
process of infiltration increases the amount of water in the soil, thus changing the
105
system state, and further limiting more infiltration. It is important to note that
“action” in this context represents a process that operates continuously; this definition
is contrasted with the every day meaning of the word, emphasizing an event-oriented
nature of the term.
There exist two types of feedback loops: positive and negative. Positive loops are
self reinforcing, meaning that an initial change is amplified in the same direction as
all of the components are traced through the loop. For example, more people in an
area implies that more development will take place, as people need adequate housing
Negative feedback loops are referred to as balancing loops, because they counteract
and oppose initial change as components are traced through the loop. This means
that initial change in a variable eventually produces a result of the opposite sense
(in the same variable) after all are traced through the loop. For example, as a
less attractive than initially thought (Sterman, 2000, p. 12), thus causing a decline of
in migration. This hypothetical situation is an example of a self-regulating feature of
social systems.
106
Definition of delays
Another important characteristic of system dynamics are delays. Delays are critical
to the dynamics of systems, because they can “breed danger by creating instability
and oscillation” (Sterman, 2000, p. 409). Furthermore, delays can help filter noise
from variability and thus provide clearer signals upon which better management de-
cisions can be made. Delays are found everywhere, and are inseparable from feedback
loops—ultimately because feedback loops could not exist without delays. According
to the above definition of a feedback loop, a system state provides information cues
to a decision (or action), which then continuously alters the system state, eventually
leading to adjustments of original decisions. The process by which information is
generated includes delays, as it always takes time to measure, transfer and report
referred to as delays.
To further understand delays, consider examples illustrating the differences be-
tween short and long delay times. Short delays imply that a consequence of an action
is observed quickly, because the system responds promptly to the initial change. In
the case of soil infiltration, the information about the state of the soil (its moisture
content) is communicated quickly to the process of infiltration, which then quickly
adjusts the inflow rate to the soil further changing its moisture content. Long delays,
however, are more difficult to grasp. A commonly cited example is that of global
warming—where continual anthropogenic emissions of carbon dioxide do not alter
the global average temperature immediately, but only decades and centuries after the
107
fact. The inability to understand the function of long delay times explains why so
many people today still refuse to accept the reality of global warning.
In the system dynamics literature, there exist two types of delays: material and
information. The former refer to processes that capture delays in physical or material
flows. In this case, the inflow to a process (whether it be the flow of water to a
reservoir, or the influx of applications seeking water taking permits, for example) is
temporarily stored (in a reservoir, or on a desk of a government official) and remains
in transit for some time. After a specified period (or a time delay), the quantities
The latter type of delays do not include any physical flows, but rather represent
the “gradual adjustment of perceptions or beliefs”, and thus correspond to mental
(or social) processes (Sterman, 2000, p. 412). An example of an information delay is
a change between, a normal inflow rate to a reservoir, and an operator’s belief about
the likely future inflow rate. Suppose that a large flood creates an inflow rate to a
reservoir greater than the normal rate, and that such floods increase in magnitude
and frequency as a result of climatic change. In this case, the operators are unlikely
to immediately adjust their practice and instantly accept the higher inflows that
may be the consequence of changing climate. If higher magnitudes floods persist, the
expectations and beliefs of operators will also adjust: operators will expect more such
floods, but only after some time. The delay between the initial signal of change, and
operators’ realization that climate change is bringing higher flood flows, represents
information delays. Furthermore, as Sterman (2000) explains:
108
Material and information delays constitute an important set of tools for systems
analysis, as they provide better ways of representing both the physical and the mental
processes that form the basis of all human decision making.
All system dynamics simulation models consist of only four basic elements, which
are introduced in this section. The elements are depicted as graphical objects in
current system dynamics simulation software (Stella, Vensim, Powersim). Commer-
cially available software is helpful because as graphical objects are manipulated, the
software automatically writes the computer code associated with the manipulated
elements. The user is relieved of the burden of writing computer code from scratch,
and is free to simply populate the model with custom data.
Figure 9 shows the graphical representation of the four elements together with
their meaning, while Figure 10 presents a simple system dynamics model to facilitate
definitions. Accumulations in system dynamics models are represented with boxed
variables, and are referred to as levels, stocks and/or state variables. Level variables
are used to hold any quantity that accumulates—storage of water in a reservoir may
be represented by such a variable, as can the amount of knowledge in a person’s
head (despite the fact that it can’t easily be measured). An advantage of the system
dynamics method is that level variables can represent material (such as volumes of
109
Symbol Meaning
Storage
[m3 ]
Inflow Outflow
[m3 /s] [m3 /s]
Water
Storage Elevation
[m] Gate
Elevation Operation
Function [−]
[m3 vs. m]
and learning can increase the knowledge content in a person’s mind, while a rate of
forgetting can decrease it. Rates are depicted by arrows with valves in the centre, to
denote their responsibility in regulating level variables. Clouds are also part of the
rate variables and refer to inputs coming from outside of the defined model boundary.
The units of rate variables are those of its corresponding level variables divided by
time. Therefore reservoir volume is expressed in units of volume [m3 ], the inflow and
outflow rates must be represented in units of volume over time [m3 /s].
Levels and rates are the only variables necessary for building system dynamics sim-
ulation models. However, it often proves helpful to add additional variables (called
auxiliary variables) as intermediate steps that keep track of other relevant informa-
tion. For example, the operation of a reservoir may depend on how fast water level
rises in the reservoir. In a simulation model water level in a reservoir is determined
the above example, the information from the reservoir level is carried (by the arrow)
to the auxiliary variable, representing reservoir water elevation, which, together with
gate operation procedures determine the outflow.
Each elements in the system dynamics method demands a precise and unambigu-
ous mathematical meaning. Therefore, any set of stocks, flows, auxiliaries and arrows
111
dSt
= It − Ot ; S(t0 ) = S0 (60)
dt
where S represents reservoir storage, and I and O its inflow and outflow, respectively.
The initial condition of the model is defined as the amount of storage in the reservoir
at t0 . Equations for water elevation (e) and reservoir outflow (O) are further specified
as:
et = f (St ) (61)
Ot = g(et ) (62)
dSt
= It − g(f (St)); S(t0 ) = S0 (63)
dt
which can be approximated using any numerical scheme for solving ordinary differ-
ential equations. If Euler’s method is used, equation (63) becomes:
h i
Si = Si−1 + Ii − g(f (Si)) ∆t (64)
where i represents discrete time intervals with an increment of ∆t. Other numerical
schemes can be used, thereby leading to different discrete representations and ulti-
mately different solutions. The reader is reminded that system dynamics emphasizes
112
logical implications of assumptions embedded in its simulation models, not precise nu-
merical predictions. Therefore, absolute values of outputs are not nearly as important
as resulting behaviour modes.
Converting a level and rate diagram to an ordinary differential equation becomes
apparent by looking at Figure 10 together with equation (60). The state variable in
the resulting differential equation is always the level variable, which is regulated by
its rates (inflows and outflows). If system dynamics model structure is depicted, a
differential equation can always be specified by stating that the rate of change of the
state variable is equal to the sum of its inflows, minus the outflows. This concept
is equally applicable for modeling material (a flow of water from a reservoir) and
immaterial (the belief held by a person or group) processes and states.
Different numerical techniques are often used in solving ordinary differential equa-
tions; most common techniques are Euler, Runge-Kutta, and difference equations.
Euler’s method is the fastest and simplest, as it relies on linear Taylor series expan-
sion within the time step. Runge-Kutta methods achieve similar things, but evaluate
integral values at four points within the time step. Runge-Kutta methods are a mod-
p. 906). Faster rates of change (inflows minus outflows in equation (64)) and/or
larger time steps normally lead to greater integration errors. The magnitude of the
integration errors therefore depends on the stability of the dynamic system.
Selecting an appropriate time step is critical for simulation of dynamic systems.
Magnitude of the time step must be appropriately selected, so that it does not lead to
113
large integration errors and system instability. When numerically solving differential
equations, time steps are selected that are divisible by 2, and therefore take values such
as 1, 0.5, 0.25, 0.125, and so on. A useful guideline in selecting time step magnitudes
suggests that a time step should be approximately one-quarter to one-tenth the size
identical behaviour, the original time step is deemed acceptable. If the simulation
outputs are not identical, the time step should be appropriately reduced, simulation
repeated, and comparisons made. This procedure should be repeated until no further
changes in behaviour between the simulations are observed (Sterman, 2000, p. 907).
The general acceptance of physical processes explains why the structure of phys-
ical models is generally undisputed. The feedback links that describe such processes
are typically known, and are represented with a high degree of confidence. All hy-
drologic models, for example, start by using precipitation as input, calculate losses to
evaporation and infiltration, and estimate surface runoff and ground water recharge
as output. Essentially, this is a description of the phenomenon of mass balance—a
universally applicable phenomenon. Mass balance explains why the same hydrologic
114
parameter set for describing how much evaporation takes place, how much water in-
filtrates, or ends up as streamflow, when compared to another region. However, the
same system structure may be applied.
When it comes to understanding the behaviour of social systems, things are en-
tirely different. The social processes that operate in one part of the world (or even
a region) can be completely different than those that operate in another part or re-
gion. These differences in social processes are a result of differences that exist in
population, cultural and personal values, belief systems, regional economies, growth
prospects, wealth, structure of governance and institutions, etc. This variety means
that there can be no universally accepted social system structure. A model structure
has to be built that will describe the social processes operating in that region. More
importantly, the model will have to describe how social processes interact with the
more known physical ones, as the two are inextricably linked. System dynamics mod-
eling offers guidance in this task, as it has the ability to formulate and test custom
social structures required in modeling social systems.
The system dynamics modeling process is outlined next. In nearly all cases, the
following set of steps are applied when formulating simulation models (after Sterman,
2000, p. 86).
1. Problem Articulation. This is the first, and probably the most important step in
the modeling process. With this step model boundary is defined, a time horizon
is set, and a clear purpose specified for the model. The general rule of thumb
115
is to model the problem at hand, not the entire system. This task is difficult,
as the premise of system approach claims that problems can not be studied in
isolation but must be considered alongside the entire context within which they
are embedded. A good model must contain enough detail to adequately address
the problem at hand, but must also avoid the clutter of unnecessary detail.
details.
with time. The system structure that is used to formulate the hypothesis is
endogenous, which produces behaviour from the interaction of variables within
the model. The interaction of variables and system components is sought after,
and includes conceptualizing the causal structure that describes major feedback
loops. Causal loop diagrams are one set of tools that can easily formulate a feed-
back structure by capturing the interaction of the structure’s many variables.
Dynamic hypotheses are causal maps showing dominant feedback loops, and
how these loops are arranged in the proposed model structure.
loop diagram that preceded it, as precise mathematical relationships and equa-
tions need to be specified for each variable. These equations try to capture
real world system structure, along with a set of decision rules. In this step all
model details are provided, with all equations, parameters, and initial condi-
(problematic) behaviour.
4. Testing. After the first iteration of the simulation model is completed, it needs
to be tested for errors and inconsistencies in structure, equations, and parame-
ters. One aspect of model testing involves comparing historically observed data
117
(if available) with data generated by the model. However, model testing involves
more than just ensuring historical data is replicated; testing also includes re-
viewing every variable and ensuring that its behaviour is consistent with the
real world. Sterman (2000) claims that model sensitivity should also be eval-
uated in light of assumptions made, both parametric and structural (p. 103).
Such evaluation is extremely important, as model assumptions reflect current
mental models (or world views) and determine the structure of the modelled
system, which in turn determines the overall model behaviour.
Extreme model tests likewise form an important set of tools for system dynam-
ics model testing. These tests push the model to the limit, and seek outputs
for cases that are highly unlikely to happen in the real world. These tests are
important, as they attempt to evaluate the model’s ability to meet basic laws
of physics. For example, a model output that shows limitless economic or pop-
ulation growth might also assume that land resources will always be available,
when in fact they are not. As trivial as this example sounds, many models fail
to meet this basic test (Sterman, 2000), even though they replicate historical
data rather well.
the model is established, the modeller (and the client) can start formulating
policies for improving the problematic behaviour being studied. Policies do not
simply involve changing model parameter values; policies may include revision
of system structures, or specification of new decision rules. High leverage poli-
cies (those that dramatically alter system behaviour by making small changes)
usually involve changing the dominant feedback loops, eliminating time delays,
and changing the amount and quality of information available to different agents
118
It is emphasized that system dynamics models are never built by following the
above steps sequentially. The modeling process is iterative, meaning that after com-
pletion of any one step, the modeller usually goes back to adjust and/or reformulate
the previous steps until satisfactory results are produced.
etc.), as is historical flow hydrograph data used in calibration. This procedure of cal-
ibration is contrasted with model evaluation techniques used in system dynamics,
where emphasis is placed on subsystem interaction and behaviour modes as opposed
to precise numerical predictions. Because the objective of system dynamics modeling
is never to reproduce the historic data in the same way that physical models do, the
use of calibration (and verification) is therefore not entirely appropriate. In system
119
dynamics modeling, model tests are used instead of calibration and verification.
In order to draw parallels between the model calibration and verification that
is used in physical modeling and model testing used in system dynamics, one must
consider the following factors. One of the factors guiding the level of detail in system
provided by different people. The system dynamics community sees this subjectivity
as a point of advantage, because viewpoints and deeply held beliefs of key decision
makers and stakeholders can be exposed, criticized and ultimately improved. Such
critical engagement has the potential to improve system management as a whole, in
• Boundary adequacy. These tests examine (i) if important processes are cap-
tured in the model endogenously (i.e., via its feedback structure), (ii) if model
behaviour varies significantly when boundary assumptions are relaxed, and (ii)
if policy recommendations change as boundaries are extended. Causal diagrams
120
and stock and flow maps assist in model boundary expansion, as do interviews,
workshops, or other means used to solicit expert opinion.
• Structure assessment. The purpose of these tests is to evaluate (i) if the model’s
• Surprise behaviour. This test assess the model’s ability to generate behaviour
previously unobserved or unrecognized, as well as to evaluate whether the model
models require testing and evaluation according to their accepted standards, keeping
in mind that one set of objective model evaluation standards is insufficient.
122
The advantages of using the system dynamics simulation method for the study of
integrated water resources management include the ability to link detailed models of
physical processes (such as hydrologic models) with models describing socio-economic
characteristics (population growth, land use, business activity, water use, etc.). Ap-
plying the systems approach to integrated water resources management motivates
the development of simulation models that are able to test a wider range of poli-
cies and management structures than either physical or socio-economic studies can
test by themselves. Endogenous (feedback-based) interaction among the components
may more accurately capture behaviour of real world systems, where socio-economic
processes influence and are influenced by physical basin characteristics. For exam-
ple, widely available water can encourage development; excessive development can
increase areas of paved surfaces, thus increasing timing and peaks of floods, degrad-
ing water quality, and ultimately limiting future development possibilities. Feedbacks
like these regulate the behaviour of water resources systems.
neers and municipal officials assigned with day-to-day tasks of management of water
123
resources. Without specific goals, targets, evaluation measures and other prescrip-
tive aspects, typical water resources managers may not be able to achieve the full
benefits that integrated water resources management promises. The latter approach,
on the other hand, does define specific targets, goals and objectives through its use
In order to illustrate the system dynamics modeling process, a relatively small sim-
ulation example is formulated. Suppose that a manager wants to explore the problem
of population growth in a region that depends heavily on water (for drinking, eco-
nomic and environmental well-being). For the sake of simplicity, an assumption is
made that water availability is the condition that has the potential to limit develop-
ment, growth and expansion. After all, if water resources are depleted quicker than
they can be naturally replenished, further growth of any kind will not be possible.
It is hypothesized that population growth in this area happens through increased
birth and in migration rates—meaning that as an area increases in population, more
couples are expected to marry and have children of their own; also, as more people
move into an area, the possibility of finding employment increases, thus attracting
more people. An increase in births (and in migration) increases the overall population
(shown as reinforcing loop (R1) in Figure 11). As the population grows and the
inhabited area expands, so will the demand for water; this means that governments
124
+ +
Desired
Births Population
(R1) + (B1) Water Use
+
Water
Actual
− Availability
Water Use
Legend:
R1 = Growth Loop
B1 = Limit/Constraint Loop
will have to provide additional sources of water by drilling new wells, and/or drawing
more water from existing lakes and rivers. However, the supply of water in any area
is limited—there is only so much rainfall each year that recharges the ground water
aquifers; surface lakes and rivers, however large, are finite supplies. The available
water is thus taken as a constant, assuming that water transfer from adjacent areas
is either not available, or too expensive for the local economy. Therefore actual water
use is regulated by availability and a growing population. Furthermore, it is assumed
that as the area uses increasing amounts of its available water, local governments will
increase the price of servicing water to deter further expansion and, ultimately, water
use. If this happens, new families and businesses are assumed to seek alternate areas
with lower living expenses, thus lowering growth rates of the present area (shown as
a balancing loop (B1) in Figure 11).
The system dynamics model schematic shown in Figure 12 has been built from
the causal loop diagram of Figure 11. Note that this diagram contains more variables
than the causal loop diagram, as precise mathematical equations are needed for all
variables (provided in Table 1). In this example, Euler’s method is used to represent
the state variables Population (see row 1 in Table 1). Other equations in the example
125
Population
Births (R1) Deaths
Births
Deaths
Normal (B1)
Water Use Normal
Water Availability
Multiplier Water Use Per
Person Per Year
Water Use To
Availability
Water Availability
Ratio Water
Multiplier Table
Availability
Figure 12: State and rate diagram for population growth example
further growth (see Figure 13). The logic behind choosing such a function assumes the
following: if area residents are using only a small fraction of available water (denoted
by a use to availability ratio that is much smaller than unity), then future regional
growth can be sustained and is indeed encouraged (multiplier of growth is larger than
unity). However, as the population increases and residents use a higher fraction of
available water, further growth is severely restricted. The multiplier effect in this
case is smaller than unity, implying a halt of further growth. The transition point
in this example occurs when the multiplier reaches unity, or the point at which the
growth due to water availability stops. In this example, that key decision point is set
when the water use to availability ratio is about 0.4. Of course, if the area residents
end up using all available water (i.e., use to availability ratio is unity), future growth
will reduce to zero, as it becomes impossible for the area to support more people
without adequate water. The key feature of system dynamics models is that when
such assumptions are criticized, others can easily be put in their place (i.e., another
table function can easily be specified).
Other assumptions used in the model relate to normal rates of births and deaths.
In this example, the regional birth rate normal is set as seven births per one hundred
people in any given year; similarly, the normal death rate is selected as five deaths per
one hundred people, per year. An assumption is also made that each person in the
area uses, on average, 200 Liters of water each day; annually this converts to a per
capita water use of 200 L/d × 365 d/yr = 73,000 L/yr. Regional water availability in
1.6
1.2
0.8
0.4
0
0 0.2 0.4 0.6 0.8 1
Water Use to Availability Ratio (−)
Water Availability Multiplier Table
The simulation output of our example is shown in Figure 14, with an assumption
that the area, at the time that the simulation starts, has 100 people. The simulation is
performed for 100 years, with a yearly time step. The top plot in Figure 14 shows the
state variable Population as it evolves with time. During the first four or so decades,
the area has plentiful water resources thus allowing growth of exponential proportions.
During this time, the reinforcing growth loop (R1) dominates the dynamic. As long
as the birth rates are larger than death rates, growth will be sustained. However, after
four or so decades (and a ten fold increase in population), water resources begin to
deplete. This information gets communicated through the multiplier effect according
to the rules of function shown in Figure 13, where increasing use to availability reduces
the potential for further growth. At this stage in the dynamic, the balancing loop
(B1) dominates the system behaviour, while the reinforcing loop lies dormant. Note
that in this example population never actually runs out of water. What happens
instead is that local government officials initiate polices that curb growth in light of
scarce resources.
Like any other methodology and school of thought, the system dynamics method
has received much criticism in its recent past. The majority of criticism offered is
traced to persons belonging to different paradigms, holding different ideas about how
systems in particular domains can (and should) be modelled. Any departure from
the familiar is deemed unsatisfactory. Further criticism is attributed to those with
“imperfect command of the available [system dynamics] literature” (Lane, 2000, p. 3).
The following list of criticisms is compiled from Cole et al. (1973) and Jackson (2003),
as well as from the personal experience of the author.
129
1600
Population (people)
1200
800
400
0
0 20 40 60 80 100
Time (yrs)
Population
100
Rate (people/year)
80
60
40
20
0
0 20 40 60 80 100
Time (yrs)
Water Availability Multiplier (−)
Births Deaths
1.6
1.4
1.2
1
0.8
0.6
0 20 40 60 80 100
Time (yrs)
Water Availability Multiplier
1. System dynamics models, especially those modeling social systems, are impre-
cise, lack rigour, and therefore can not give accurate predictions of the future.
Furthermore, models require extensive data for calibration and/or verification
purposes; when building models of social systems (or when linking them with
2. Those building system dynamics models often tend to use social factors that
dynamics models.
3. System dynamics often produces models with over simplification and a large
amount of aggregation. Many processes operate on finer spacial and tempo-
ral time scales, thus modeling at high levels of aggregation may actually miss
features that are supposed to be captured.
5. System dynamics models are difficult to understand, despite the relative sim-
plicity of their evolution rules.
world views lead to different paradigms that adopt different criteria regarding what
data should be included, how a model should be built, and how a model’s perfor-
mance should be evaluated. Using the criteria of one paradigm and applying it to
models produced by another may not be entirely appropriate. This realization helps
in addressing the first criticism in the above list. For example, those involved in mod-
eling approaches that emphasize precise, short term predictions of states (i.e., flood
volumes after a storm, performance of the economy after a recession) have difficulty
accepting the main focus of system dynamics—where interest resides in discovering
mechanisms responsible for the behaviour of complex systems. The system dynamics
method emphasizes uncovering relationships and assumptions (represented via feed-
back loops) that are responsible for observed patterns of behaviour. Revealing the
logical implications of such assumptions, and formulating alternate policy options
and scenarios, are seen as more valuable pursuits than precise, short term predic-
tions. Once exposed, feedback structures of systems can be changed, thereby leading
to different patterns of behaviour.
Data requirements are acknowledged as serious limitations when building models
that require the inclusion of social components. System dynamicists recognize this
problem, but still claim that it is better to build models based on assumptions (which
can be easily questioned and later changed once additional data becomes available),
than to abandon modeling altogether in such difficult situations. Meadows et al.
(1982) claim the following: (i) that omitting elements (because of lack of knowledge or
132
available data) implies they are not important, which is not true, (ii) that providing
correct signs to feedback loops is ultimately more important than including more
accuracy in any relationship or parameter value, because the goal of the analysis is
to obtain the dynamic character of the problem, not to produce precise predictions,
and (iii) that “if one believes an unmeasured psychological, environmental or political
factor is important in a system, one should state this belief explicitly and precisely, to
encourage observation or experimentation that will increase knowledge of this factor”
(Meadows et al., 1982), p. 116.
that inclusion of more detail (whether in the form of more variables or more spacial
regions) does not necessarily alter the tone of conclusions drawn. Selecting the level
of aggregation in the modeling work is therefore not a limitation of the methodology,
but rather the preference of the modellers.
are exposed and incorporated into the simulation model. By their definition mental
models are subjective, and their inclusion into system dynamics models represents
anything other than an objective representation of the world. The belief that system
dynamics models claim to discover objective social structures is incorrect; these mod-
els are simply trying to show the logical consequences of assumptions incorporated
133
system dynamics models. Meadows et al. (1973) recognize that without this training
many tend quickly to jump to conclusions about the inappropriateness of the method-
ology or a particular model. A more fruitful approach would be for critics to offer
alternatives ways (or models) of addressing problems that system dynamics is trying
to tackle. To simply say that system dynamics is inappropriate, and then offer no
viable alternative(s), is insufficient.
134
The purpose of this chapter is to show how tools developed in this research are
practically applied to the Upper Thames River basin, in order to asses the impact of
climate change on water resources systems. The Upper Thames River Conservation
Authority (UTRCA) administers all aspects of water and land management, and is
delegated by the Province of Ontario to provide hazard land management input to all
municipal development plans. Floods constitute the most severe hydrologic hazards
in the Upper Thames River basin (impacts from low flows and droughts constitute
significantly less severe hazards).
Flood plain limits are based on a 1937 flood, the highest recorded and documented
flood in the basin. Severe flooding in the Upper Thames River basin occurs either
as heavy rainfall events typical during the summer, or from rainfall on snowmelt
events representative of the spring season. Low flows and droughts are most common
during the summer months, during which some streams consist entirely of sewage
discharges. Growing development pressures from urban and rural communities, ageing
water management infrastructure, and changes in hydrologic extremes resulting from
climate change are forcing the UTRCA to review its existing water management
operating protocols and design standards.
The models developed in this study aim to assist the UTRCA in their managing
water resources systems. Different methods (for an overview see Figure 1 in Chapter
2) are applied to obtain local meteorological conditions under climate change, and
are developed by using locally observed data as well as outputs of global circulation
models. Meteorologic data under changed climatic conditions are used as inputs into
event and continuous hydrologic models, thus estimating future regional impacts of
floods and low flows. Generated meteorological data is also used to asses changes in lo-
135
within the river basin, offering water resources managers, stakeholders and govern-
ment and other officials a practical tool for evaluating climate change adaptation
policies and management strategies required in integrated water resources manage-
ment.
climate change scenarios, and the application of daily/hourly weather generator sim-
ulations), (iii) the application of the inverse approach (through event and continuous
hydrologic modeling and analysis), and (iv) a description of the integrated water
resources management model (with details of socio-economic model components, as
The Thames River formed after the retreat of the Wisconsinian Glacier from Ontario,
during our most recent ice age ending some 10,000 years ago. In fact, some of the
northern branches still flow through the ancient spillways, while the southern reaches,
with their lower gradients, emerged after a long existence as a large glacial lake
land. The renaming occurred after Simcoe proposed that a new capital of Upper
Canada be called London, to be located between Lakes Huron and Erie at the forks
of the Askunessippi River (in place of the original Algonquin village Kotequogong).
Interestingly, the proposal for the capital was rejected in favour of Toronto, but the
new names for the river and the city were kept.
A large part of the Thames river basin was originally part of deciduous Carolinian
forests, most of which are now cleared to permit farming, agriculture and urban
development. Despite this deforestation and rezoning, the Thames River is still one
of the country’s richest basins in natural and human heritage. The basin is home to
some 2,200 species of plants, including the Wood Poppy (found only in two locations
in Canada). Its biodiversity consists of a large and diverse populations of clams, fish,
and other wildlife. Most notably, the Eastern Spiny Soft Shell Turtle and the Queen
137
Snake (now registered as endangered species in Ontario) call the Thames River their
home (Wilcox et al., 1998).
The Thames river basin consists of two majors tributaries (see Figure 15) of the
river Thames: the North Branch, flowing southward through Mitchell, St. Marys, and
eventually into London, and the East Branch, flowing through Woodstock, Ingersoll,
and east London. The two branches meet in London near the centre of the city,
referred to as the Forks. The Thames then flows westwards and exits the basin near
Byron, eventually draining into Lake St. Clair. The length of the Thames River
(from Tavistock to its mouth at Lake St. Clair) is approximately 273 km, while its
annual discharge (measured at the Byron stream gauge) is about 35.9 m3 /s. The
Upper Thames River basin receives about 1,000 mm of annual precipitation, 60% of
which is lost through evaporation and/or evapotranspiration, is stored in ponds and
wetlands, or is recharged as ground water (after Wilcox et al., 1998, p. 15). The slope
of the Thames River is about 1.9 m/km for most of its upper reaches, while its lower
reaches are much flatter with a slope of less than 0.2 m/km.
The Upper Thames River basin includes three major water management reser-
River basin includes extensive dyking systems in the cities of St. Marys and London,
as well as a diversion channel in the town of Ingersoll.
Floods and droughts represent major hydrologic hazards in the Upper Thames
River basin. On average, 25% of all floods occur during March, while more than 50%
of all floods happen between February and April (Cunderlik and Simonovic, 2004a).
138
Perth
County
Mitchell
Whirl Cr.
Branch
Flat Cr.
North
Black Cr. Seberingville
Avon River
Shakespeare
Kirkton Stratford
Tavistock
St. Marys Trout Cr.
Fish Cr.
Oxford
Wildwood Dam
County
Gregory Cr.
Middlesex Mud Cr. Innerkip
County Wye Cr. Embro
Pittock Dam
Medway Cr. ch
Thorndale an
Br
h
nc
e
dl
ra
.
Fanshawe Dam id
Cr
B
Cr
Arva M
ow
Woodstock
ut
no
xb
So
bu
O
.
Beachville
r Cr
au
W
Thamesford
Cede
Reynolds Cr.
Killworth
Ingersoll
Komoka Dorchester
London Dingman Cr.
Delaware
Early springs therefore see basin-wide floods, resulting from sudden warming and
snowmelt on a yearly basis. The period between December and April may also see
severe floods, resulting from a combination effect of snowmelt and intensive precip-
itation. Summer frontal storms are also known to produce severe flooding in and
around the basin, but such storms are less frequent—even though they can exceed in
magnitude (sometimes significantly) spring time flooding. In fact, some of the highest
observed floods have occured during the summer months.
Drought conditions are possible at any time of the year, although they are most
frequent between the months of June and September. In the summer months the
mean monthly flows are to the order of 25 to 30% of long term annual mean flow;
as a reference, mean monthly flows in spring time are as large as 300% of the mean
annual value (Cunderlik and Simonovic, 2004a). During the summer months, a large
be problematic.
The human influences in the basin date back to 500 A.D., as the area was thought
to have been inhibited by Aboriginal peoples, who were one of the first to practice
agriculture in Canada. Subsequently, the area attracted a multitude of French fur
traders and European settlers, as abundant fish and game were present. Since then
however, agriculture and modern industry have overtaken once vast forested lands.
(The reports by Wilcox et al. (1998) and UTRCA (2001) outline these, and many
other interesting historical facts about the basin.)
140
Today, the majority of the river basin is covered with agricultural lands (80%),
with forest cover and urban areas comprising about 10% each (basin’s land area is
approximately 3,500 km2 ). The population of the basin is about 420,000, of which
350,000 are residents of the City of London, the largest urban center in the basin.
Other urban centers are those of Mitchell, St. Marys, Stratford, Ingersoll and Wood-
stock. The region is divided into three counties: Perth to the north, Oxford to the
east, and Middlesex to the west.
Land use characteristics of the basin are shown in Table 2, and are based on
information provided in UTRCA (2001). It is noted that data in the table represents
only that portion of each county that lies in the river basin; in other words, the
land area of each county located outside the basin’s boundary is not shown here or
otherwise considered.
The strength of the regional economy in and around the Upper Thames River basin
resides in its diversification (MMAH, 2002). The region’s economic strength exists
because of its close proximity to Highway 401 corridor (a major international land
based transportation link), thus giving it easy access to many Canadian and US cities.
The makeup of the regional economy consists of strong manufacturing, education,
health care and agricultural sectors. The regional economy depends on a number of
manufacturing plants (particularly around London, Ingersoll and Woodstock). The
141
educational sector is also very strong influential in the region’s economic activity, as
it contains first class, internationally recognized, educational, medical and research
institutions. The agricultural sector is classified as diverse, as it supports a variety
of activities from specialized crops to major livestock operations. The counties in
the study area provide some of the highest agricultural sales and productivity figures
among all counties in the province (MMAH, 2002, p. 2). Although not as strong as
other economic sectors, tourism forms a significant generator of economic activity,
particularly in Stratford and other smaller towns, where parks, conservation areas
and other cultural attractions draw visitors from within the region and abroad.
Table 3 provides additional information regarding socio-economic characteristics
of the basin, and includes population, dwellings/farms, labour force and jobs for
both urban and rural regions. The information in this table is based on data from
Statistics Canada and its 2001 Community Profile Census (STATCAN, 2001). Water
use information, on the other hand, is provided in Table 4 for both urban and rural
sectors, and is divided into uses of surface and ground water. This information is
taken from county ground water studies (Kell, 2004; Brown, 2001; Merry, 2003). As
with land use, the socio-economic and water use characteristics are shown only for
that part of each county that is within the basin. Both the 2001 census and county
ground water reports provide information on a county-wide basis, and thus represent
a different physical boundary than the study area. The county based information is
thus adjusted (based on land area, location of major urban centres, and population
density) in order to produce data that corresponds to the boundary of the basin.
It is interesting to note that detailed characteristics of water use data for Oxford
County (the largest consumer of water in the area) is only available county wide.
Details regarding consumption of water on a smaller scale is not available to the
Rural Sector
County Population Farms Labour Force Jobs
(people) (no.) (people) (people)
Middlesex 21,410 889 2,915 2,752
Oxford 21,676 1,163 3,060 2,915
Perth 16,718 1,371 2,282 2,218
Totals 59,804 3,423 8,257 7,885
143
Rural Sector †
County SW Res SW Ind/Comm GW Res GW Ind/Comm
(m3 /yr) (m3 /yr) (m3 /yr) (m3 /yr)
Middlesex 155,037 232,555 1,783,824 2,428,579
Oxford 740,921
Perth 783,393 148,584
Totals 155,037 232,555 3,308,138 2,577,163
This section presents details regarding the application of methods used to generate
climatic input for use in the Upper Thames River basin. Discussion of model param-
eters (spacio-temporal storm model), climate change scenarios, data selection for the
weather generator model, as well as for rainfall disaggregation and IDF analysis is
given.
relatively quick decay of rainfall intensity from the storm’s center. The coefficients
describing the areal reduction factor for equation (8) are specified as : c0 = 1.000,
c1 = −6.700 × 10−4 , c2 = 2.588 × 10−6 , c3 = −5.018 × 10−9 , and c4 = 3.721 × 10−12 . It
is noted that equation above coefficients are valid for subcatchment areas up to 500
145
km2 , for the Upper Thames River basin. These parameters are region specific and
may not apply to other areas. The ratio of major to minor ellipses is set to about
2.5, thus controlling the shape of the ellipses.
axes ten times larger as the minor axes) a distinctive rainfall field will be observed.
The parameter set corresponding to simulation 6 is similar to previous cases, with
only its orientation parameter adjusted. Simulation results of the above simulation
cases are shown in Section 4.1.1.
Climate change scenarios in this work are based on outputs of global circulation
model simulations, and are used to condition the input data used by the weather
generator. The scenarios do not represent future predictions or forecasts, but simply
146
offer possibilities of what might happen if humanity follows a certain course of action
(i.e., continual growth of population, increased carbon dioxide emissions, increased
urbanization, etc.). With regards to global circulation models, all scenarios have been
standardized by the report of Nakicenovic and Swart (2000).
In this work, the climate change scenario data is obtained from the website of
the Canadian Climate Impacts Scenarios group at the University of Victoria, Canada
(http://www.cics.uvic.ca). Time series data is obtained for the grid point containing
the site of interest, for a particular time slice. For this study the time slice of 2040-2069
is used for all climate change scenarios, thus representing average climatic conditions
for the year 2050. Historic global circulation data, also obtained from the University
of Victoria, consists of data for period between 1961-1990 and represents baseline
conditions against which climate change scenarios are compared.
Based on the global circulation data obtained above, change fields for each scenario
are computed based on percent difference from the baseline case for each variable
(monthly precipitation, and minimum and maximum temperature). Monthly values
of global data are then averaged for all years of output to obtain the change fields.
After the change fields are computed, the climate change scenarios are formulated as
follows: Locally observed climate data is obtained first for a number of stations in the
study area. Then this data is modified using the monthly percentage change values
previously obtained. This means that if the precipitation change field for the month
of January is +10%, then all January values in the historic record are multiplied by
1.10; similarly, if the change field is −15% for the same month, all historic data is
multiplied by 0.85. Temperature changes are applied to the observed data by adding
the change field values to to the observed record. If, for example, the maximum
temperature change field is +5◦ C, then this change field is added to the observed
value of temperature. The locally modified data set is then used by the weather
147
The criteria used for selecting scenarios is based on the ability of the scenarios to
produce a range of possible future climates. The scenario CCSRNIES B21 (wet) is
selected as the upper range of possible future precipitation generated by the global
circulation models. Similarly, the scenario CSIROM2kb B11 (dry) is regarded as
the lower range of possible future precipitation magnitude. The two scenarios are
therefore selected to show the widest possible range of climate change impact on
magnitude of precipitation.
The change fields for the dry and wet scenario are shown in Table 6, while the
summary of their meaning is provided in Table 7. Based on this information, locally
observed station data is modified (multiplied or added by the change fields for each
month in the historic record) in order to formulate climatic inputs appropriate for
the weather generator of the Upper Thames River basin. Development of climate
change scenarios in this way considers all available climatic data (local and global) to
look into the future, each emphasizing different climatic conditions. The wet climate
148
climate (examining cooler and drier conditions) can be used in assessment of future
low flows. In light of the above discussion, the wet climate therefore represents the
upper range of climate change impacts, while the dry climate scenario corresponds to
the lower range.
Note that both wet and dry scenarios are equally likely. However, when dealing
with questions regarding the potential increase in precipitation magnitude and fre-
quency, the wet climate scenario should be used (as it is more extreme and therefore
more critical for design purposes). Similarly, if the focus is on drought conditions,
the dry climate is more critical and should therefore be used in the analysis.
The weather generator model is used to simulate climatic input for a number distinct
purposes. They include generation of: (i) hourly data for event modeling; (ii) daily
data for continuous and integrated modeling, and (iii) hourly data for rainfall intensity
duration frequency analysis. In this study, items (i) and (ii) require data with a high
degree of spacial resolution, thus demanding use of many stations in and around the
basin. Item (iii) however, as it focuses on local short-duration high-intensity rainfall
events, does not depend on high spacial resolution of data.
Weather generator outputs obtained from Wey (2006) and Sharif and Burn (2007)
are used in hydrologic and integrated modeling. Details of their application are
described next.
150
Event modeling
The weather generator used in event modeling is developed by Sharif and Burn (2004,
2006, 2007). The weather generator operates on a daily time step, and is thus inade-
quate for producing short-duration high-intensity storm events that frequently cause
summer flooding in the basin. To cope with this problem, a modified version of the
weather generator is developed by Wey (2006). This version takes synthetic daily data
generated by the original model, together with historically observed hourly rainfall,
and disaggregates it to produce hourly rainfall values. Disaggregated hourly data are
used as inputs to the event hydrologic model.
The original version of the weather generator synthetically generates 100 years of
data with a daily time step for each climate scenario, for each station. Stations used in
the work of Sharif and Burn (2004, 2007) are shown in Table 8, and use historically
observed precipitation and temperature data for the period between 1964-2001 to
condition the weather generator (in addition to two global circulation model outputs).
Generated data at each station contains 100 years × 365 days/year = 36,500 days of
precipitation, for each of the three scenario (historic, dry and wet). To disaggregate
all precipitation data into hourly values is futile, as large number of days see little
or no precipitation. In the work of Wey (2006), only days which receive more than
25 mm of rainfall are disaggregated into hourly intervals, and are defined as events.
shortest duration) is selected for use in event based modeling. Since 100 years of
151
weather generator outputs are provided, a single critical event is selected from each
year as input for the event model, for each scenario. Peak flows resulting from the
100 events are used in flood flow frequency analysis to estimate the potential impact
of changed climatic conditions on the hydrologic flow regime.
Continuous modeling
Same weather generator climates are used in the simulation performed by the con-
tinuous hydrologic model. The synthetic data are generated for the same fifteen
stations as for the event model (shown in Table 8), for the same three scenarios.
Daily precipitation, maximum and minimum temperature are used as inputs to both
continuous hydrologic, and the integrated model (whose major component is the
continuous model). Output hydrographs of the continuous model are used in the
frequency analysis to study climate change impact on low flows and droughts.
Instead of focusing on weather generator outputs for use in event and continuous hy-
drologic modeling, this section focuses on changes in extreme rainfall at the local level.
Changes in extreme rainfall patterns could demand new regulations in the form of
152
storm water management strategies, guidelines and design practices, as well as altered
municipal infrastructure design standards. In some cases changing hydro-climatic
conditions may also require upgrading, retrofitting, rebuilding, or even constructing
additional water management infrastructure.
Intensity duration frequency analysis is used to take look into current engineer-
ing design standards for municipal infrastructure (such as roads, bridges, culverts,
drains, sewer systems, etc.), and provide data and information based on which they
can be changed. Since design standards for much of municipal water management
Service of Canada (MSC) for both hourly and daily durations for stations in South-
western Ontario with 30 or more years of data (see Table 9). Data of shorter durations
are available for a only a handful of stations with relatively short records, and were
thus not used further. This is because in order to capture properties of long term
many stations large portions of the record are unavailable; for data of daily duration
however, the amount of missing data is considerably less (see Table 9). It is further
noted that the number of stations used in the weather generator algorithm influence
the computation of regional means (thereby affecting the calculation of Mahalanobis
distance), which affects the choice of the nearest neighbour. In other words, the
153
weather generator outputs may be sensitive to the number of stations used. This
sensitivity is explored further in a later section.
It is pointed out that only rainfall data is used in this portion of the study (as
opposed to total precipitation in event and continuous modeling). This is because
MSC database for the hourly time step consists of mostly rainfall data, but not total
precipitation data (rainfall plus snowfall). The reason is that hourly gauges oftentimes
freeze during the winter season, and can not therefore accurately estimate values of
total precipitation. Daily amounts for total precipitation are available for all stations,
but were not used since only rainfall data is available for disaggregation purposes. For
example, it would be inconsistent to use daily total precipitation data for the weather
generator, and hourly rainfall data for disaggregation purposes.
Using only rainfall data in this part of the study is supported with current MSC
practice. In other words, MSC also uses rainfall data in estimates of intensity dura-
tion frequency that it provides for various stations within Canada; for such data for
London, Ontario, the reader is referred Ontario’s Drainage Manual (MTO, 1997). For
the sake of maintaining consistency with MSC estimates, only rainfall data is used in
this study. For estimates of rainfall on snow melt, the interested reader is referred to
Watt et al. (1989) and Wood and Goldt (2004).
mate missing rainfall values using the Inverse Distance Weighting method (USACE,
2000). The hourly and daily time series used consist of data between 1965-2000. At
the time of writing, the latest available hourly data provided by the MSC includes
155
data ending in year 2000; daily data however, is available up to present. The hourly
data has a significant portion of the record missing, sometimes up to 40-50% (Table
9). Key stations (London, Windsor and Toronto) had 10% or less of its hourly record
missing. Data for the daily time step was of significantly better quality, with key
• Historic input, meaning that historic data is not altered by the outputs of global
circulation model outputs. This is used as the base line for comparison with
other scenarios.
ing.
• Dry input, produced by altering the historic record with change fields provided
by CSIROM2kb B11, also given in Table 6. The dry input postulates a cli-
mate that is drier and cooler than normal, leading to prolonged dry spells and
droughts.
(N = 36 years of record, simulated three times, thus giving 108 years of simulated
output) for each scenario. Such a long record is sufficient for estimating events with
return periods as high as 100 years, currently demanded by the study.
156
days was used in the extraction of nearest neighbour events used in the disaggregation.
This value was estimated based on the work of Wey (2006).
Note that historical (unmodified) data is used in the disaggregation procedure,
because only non-dimensional rainfall mass curves were needed from this data set.
Modifying the historic record of hourly data according to the change fiends of Table 6
did not significantly affect the resulting temporal distribution. Therefore, historically
observed rainfall data is deemed sufficient for disaggregation purposes.
This section covers the application of the hydrologic models for use in the inverse
approach for the Upper Thames River basin. Both event and continuous models are
included in the discussion. Lastly, schematics showing the summary of application of
the inverse approach to the study of flood and low flows are shown.
Throughout this report, terms low flow and drought, are used synonymously.
This use is justified by the Ontario’s Low Water Response manual (OLWR, 2003)
which defines drought conditions based on below normal precipitation or below nor-
mal streamflow. This means that if either streamflow or precipitation is below a
defined norm, drought occurs. This terminology may not be in agreement with other
hydrologic literature. For additional definitions of low flow and its difference from
drought, please refer to Smakhtin (2001).
157
The event hydrologic model has been originally applied to the Upper Thames River
in the work by Cunderlik and Simonovic (2004b, 2007). The event hydrologic model
consists of thirty three spacial units (also called subbasins or subcatchments), twenty
one river reaches, and three reservoirs. The schematic of the model is shown is Figure
16. The methods used to describe losses, transform, routing and baseflow components
are outlined in detail in Section 2.2.2. All parameters used for the event model
are listed in Cunderlik and Simonovic (2004b), as well in Prodanovic and Simonovic
(2006b).
Cunderlik and Simonovic (2004a) analyze the hydro-climatic data for the Upper
Thames River basin, and conclude that events of July 2000 and November 2003
should be used for calibration, while events of September 2000 and November 2001
for verification purposes. The events for calibration and verification include both
convective and frontal rainfall driven floods, and thus form a representative set of
events that produce flooding in the study area. The calibration event of July 2000 is
of alternative methods for capturing reservoir releases is beyond the scope of this
158
sb1 Legend
sb2
Junction
jnMitchellSG
sb3 River reach
R560 sb4
jn640 SubBasin
sb5
R640 sb7
Reservoir
jnAvonSG
jn750
sb8
R750 sb9
jnDuttonSG
120 600
100 Mitchell 500 St. Marys
Flow (m3 /s)
600 300
400 200
200 100
0 0
0 50 100 150 200 250 0 50 100 150 200 250
Time (hrs) Time (hrs)
Computed Observed Computed Observed
project, and should be a subject of a separate study. The encouraging fact is that the
entire source code for the hydrologic model is available, which can easily be adopted
to custom modeling of the regional reservoir operating rules, and thus representing
hydrologic conditions more accurately.
Since local reservoir operating procedures have significant influences on flow char-
acteristics downstream, calibration of subcatchments downstream of reservoirs be-
comes problematic. In the case of the July 2000 event, Cunderlik and Simonovic
(2004b) remove the reservoirs from the event model and use the observed reservoir
Cunderlik and Simonovic (2004b) also discuss in length the sensitivity analysis of
the event hydrologic model. Their results show that, with respect to flood magnitude,
the model is most sensitive to Clark’s storage coefficient. In terms of peak volume
however, performance of the event model is most sensitive to Initial and Constant loss
rate parameters, baseflow recession, and Clark’s storage coefficient. The parameter
set selected for the event model is adequate for simulating flood peak magnitudes,
despite that recession parts of the hydrograph are slightly underestimated, therefore
leading to 5-10% underestimation of total flood volume (Cunderlik and Simonovic,
2004b).
The continuous based hydrologic model applied to the Upper Thames River basin
is described in Cunderlik and Simonovic (2004b, 2005). The continuous hydrologic
model similarly consists of thirty two spacial units, twenty one river reaches, and
161
three reservoirs. The schematic of the continuous model is identical to one for the
event based model, with only one less subcatchment. The methods used to describe
losses, transform, and river and reservoir routing are presented in Section 2.2.2.
The continuous hydrologic model used is seasonal in nature, with a different pa-
rameter set describing summer and winter seasons. Simulation of seasonal variations
with a hydrologic model requires change of parameters for each season. This amounts
to providing a parameter set for the start of the simulation, performing the simulation
for the season, then stopping the simulation and providing a different parameters set
for the next season. This process is repeated until the end of the simulation period.
Parameter set for summer and winter seasons is given in Cunderlik and Simonovic
(2004b), as well as in Prodanovic and Simonovic (2007f). Seasonal parameter sets
for the HEC-HMS continuous model are also recommended by Fleming and Neary
(2004).
Cunderlik and Simonovic (2004b) describe in detail the calibration and verifica-
tion procedures applied to the continuous model of the Upper Thames River basin.
The data used for model calibration consists of daily data for the period between
November 1979 to October 1988; the time period used for verification include Novem-
ber 1988 to October 1997. This data set represented highest available spacio-temporal
hydro-meteorologic data density for the Upper Thames River basin, and was there-
fore used for calibration and verification. Figure 18 shows the calibration output
for a set of stream gauges in the study area. The simulation results show that the
continuous model (coupled with the external snow accumulation and melt algorithm)
adequately captures regional long term hydrologic characteristics. The model shows
no systematic bias for peaks produced during spring or summer flood events, even
though it has the tendency to underestimate total streamflow volumes by 10-15%
180
150 Mitchell
Flow (m3 /s)
120
90
60
30
0
0 60 120 180 240 300 360 420 480 540 600 660 720
360
300 St. Marys
Flow (m3 /s)
240
180
120
60
0
0 60 120 180 240 300 360 420 480 540 600 660 720
420
360 Ealing
Flow (m3 /s)
300
240
180
120
60
0
0 60 120 180 240 300 360 420 480 540 600 660 720
1050
900 Byron
Flow (m3 /s)
750
600
450
300
150
0
0 60 120 180 240 300 360 420 480 540 600 660 720
Time (days)
Computed Observed
It is important to point out that calibration of both event and continuous model
is based on spacially and temporally interpolated precipitation. The interpolated
precipitation may not correspond to the actual temporal and spacial distribution of
precipitation, and is therefore reflected in the calibrated model parameters. Compari-
son of computed and observed hydrographs in such cases inevitably leads to differences
(resulting from different input signals).
Clark’s storage coefficient, together with parameters describing the physical prop-
erties of the soil (maximum soil infiltration rate, and depths of tension zone and soil
profile storage) represent the most sensitive components of the continuous hydrologic
model in terms of flood magnitude (Bennett, 1998; Cunderlik and Simonovic, 2004b).
In terms of hydrograph volume, the continuous model is most sensitive to the Soil
Moisture Accounting parameters describing ground water layers. These parameters
are most important for the simulation of low flow characteristics, and therefore affect
the overall goodness-of-fit the continuous model.
The schematic presented in Figure 19 is used to illustrate the application of the inverse
approach to the problem of flooding in the Upper Thames River basin. The analysis
starts by gathering global circulation model outputs together with locally observed
daily data (precipitation, temperature) for use in weather generator modeling. The
weather generator produces synthetic daily data which is disaggregated with the help
of locally observed hourly data. The synthetically generated hourly data is divided
into events which are then used as input to the event based hydrologic model. The
output of the event hydrologic modeling are flood hydrographs, whose peaks are syn-
thesized via frequency analysis. The outputs of frequency analysis are plots showing
return period versus flow peaks for each climate scenario considered. Local critical
164
Observed Local
Observed Local
Daily Data
Hourly Data
Synthetic
Daily
GCM Data Data
K-NN
Weather Disaggregation
Generator
Guidelines
Critical
Exposures
thresholds (i.e., floods over topping a dyke for example) are used to estimate the
changes in frequency of occurrence of such events resulting from changed climatic
conditions. Based on knowledge of frequency (and/or magnitude) shifts of local crit-
ical thresholds, suggestions for revision of existing water management guidelines and
similarly starts by gathering outputs from global circulation models together with
locally observed daily data. This data is used by the weather generator model to
produce an arbitrary long synthetic sequence of weather at a daily time step. The
daily data is used as an input into a continuous hydrologic model, whose outputs
are daily flow hydrographs. After accumulating and averaging the outputs to longer
durations (weekly and monthly), they are put through a frequency analysis where
appropriate methods of analysis are used. Likewise, critical local low flow hydrologic
exposures (i.e., minimum flow in a stream required for waste assimilation) are used
This section describes the application of the integrated water resources modeling
to the Upper Thames River basin. The integrated model couples the continuous
Observed Local
Daily Data
GCM Data
Synthetic
K-NN Daily
Weather Data
Generator
Flow vs.
Return Period
Consequences
Frequency from changed
Analysis climatic
conditions
Guidelines
Critical
Exposures
(like changes in demographics, housing, jobs, land and water use practices), as well as
able to produce detailed hydrologic output (like frequency of floods/drought, timing
and regularity of flows) typically used for impact analysis and/or engineering design.
The continuous hydrologic model component applied is identical to one presented
and particularly how they can be applied in studying water resources management
from an integrated perspective. It therefore has the potential to be a practical and
useful tool to managers, policy makers and stakeholders. The objective of this model
is to assess water resources risk and vulnerability to changing climatic conditions in
the study are, and postulate behaviour modes of alternate climatic and management
scenarios. The socio-economic model component is built using the system dynamics
framework.
is developed and applied to the Upper Thames River basin study area, although its
structure may be applicable to other regions as well with appropriate modification.
The rest of this section describes the structure of the modelled system, and provides
168
details on each sector considered. As the overall model contains a large number of
equations, not will all be listed. The most significant functional forms are presented
together with few selected equations, while the Appendix lists them all in their en-
tirety.
The Upper Thames River Basin socio-economic model component is divided into
three spacial units. These units are selected as regional economic activity is easier
described using a larger aggregated area; in our case, the spacial units are defined as
areas belonging to each county located in the Upper Thames River basin (see Section
3.1.1 for a map of the basin), and are thus referred as Middlesex, Oxford and Perth.
County wide spacial units are appropriate as selected socio-economic data for the
study area is only available on the county wide basis—see Tables 2-4 in Section 3.1.2
of this report.
Each spacial unit contains identical model structure, as it is assumed that socio-
economic characteristics in each region are similar (based on the number of people,
unemployment level, jobs, etc.) As the region’s economic output heavily depends
on industrial as well as agricultural activity, separate sectors are built to account
for such features. Each spacial unit therefore contains sectors describing urban and
rural population, urban and rural business activity, housing, as well as land use.
The socio-economic model component architecture is shown in Figure 21, where each
major sector is depicted by a rectangle and linked to other sectors via variables shown
adjacent to the solid lines.
follow. The socio-economic model component is based on the work of Forrester (1969),
169
Availability of Housing
Demand for Housing
Population Weather Housing
Availability of Generator
Residential Water Climate Land
Demand for
Labour Available
Availability Hydrologic Residential
Force for Housing
of Jobs Component Land
Mass (1974), Schroeder et al. (1975) and Alfeld and Graham (1976).
The representation of population is subdivided into urban and rural sectors. These
sectors contain information that generate population dynamics based on current pop-
ulation, birth and death rates, as well as descriptions of in and out migration. Within
the scope of the model, the population sectors depend on the availability of water,
jobs and houses, illustrating the fact that each plays a prominent role in dynamics
of growth. Each of the three variables places a limit on the possibility of future
growth and expansion. For example, the area can not experience population growth
if adequate water, jobs or housing are not available. The business activity sector is
likewise divided into urban and rural categories, representing dynamics of business
and farm units, respectively, through investment and depreciation of capital. The
business activity sector depends on a labour force provided by the population sector
(where only a fraction of the total population represents labour force), water avail-
urban settings, as farm units are defined to represent places where rural families both
live and work. The land use sector is the most detailed part of the socio-economic
component as it represents dynamics of change in terrestrial land cover—defined in
this study as forest, agricultural, business and residential land use types. The resi-
171
dential and business communities exert pressure on the local government to rezone
and allocate additional land to their respective uses if favourable conditions exist.
An assumption is made that as business and population growth continues, remaining
forest cover is converted to agricultural land, while agricultural land is developed into
systems behaviour. The major level (or state) variables in urban sectors are busi-
ness structures, housing and population. Similarly, for the rural sectors, the levels
are farm units and rural population. The major flows include rates that increase
the state variables (such as construction, investment, births and in migration), or
decrease it (depreciation and demolition, deaths and out migration). The urban and
rural model sectors are shown as level and rate diagrams in Figures 22 and 23.
The level and rate diagram of the land use sector is shown in Figure 24, and depicts
the representation of land cover in the study area—forest, agricultural, residential and
business land. This sector describes the political process used to rezone land cover
from one type to another. For example, if demand for additional agricultural land
is strong, the agricultural community will place pressure on the local government to
clear remaining forest cover and thus convert it to farm land. Likewise, the same
logic is applied if the urban communities demand additional land for residential and
Business
Business Structures Business
Construction Demolition
Urban
Urban Housing Houses Urban Housing
Construction Demolition
Urban
Population
Urban In Urban Out
Migration Migration
Urban Urban
Deaths Births
Farm
Units
Farm Unit Farm Unit
Investment Depreciation
Rural
Population
Rural In Rural Out
Migration Migration
Rural Rural
Deaths Births
Forest
Cover
Forest To
Agricultural
Rezoning
Agricultural
Land
Agricultural
To Residential
Rezoning
Residential Agricultural
Land To Business
Rezoning
Residential Business
Land Land
Rezoning Rezoning
Business
Land
Figure 24: Levels and rates schematic of the land use sector
174
and business categories. In cases when an area contains only residential and business
land (such as cities, or when all agricultural land disappears) a competition structure
exists that converts one land use type to another.
This section describes inputs and structural relationships used in the urban business
activity sector, as well as selected relationships that are used as links to other sec-
tors. This is needed because certain information is shared by more than one sector of
the model. The sector described here presents the underlying dynamics of urban eco-
nomic activity. The sector includes a detailed breakdown of surface and ground water
ables (bolded), which are used to compute regional drought conditions based on the
definition provided by Ontario’s Low Water Response (OLWR, 2003) regulations.
An information delay is applied to the drought level variable producing a perceived
drought level, as it is assumed that action in response to drought is taken only after
some period of time (i.e., it takes time to measure drought and convince people it
where P DLt stands for perceived drought level, and DLt for actual drought level, both
as dimensionless indices between 0-3 (see OLWR, 2003); DrT represents a drought
delay time in [yrs]. The function Smooth3 is defined in detail in (Sterman, 2000,
175
Monthly Monthly
Precipitation Flow
Average
Drought Damage
GW Recharge Damage
Level
GW Recharge Drought
Delay Time Delay Time
Flood Damage Flood Damage
Perceived GW Perceived Investment
Investment
Recharge Drought Multiplier
Fraction Total Multiplier
Level Table
GW Allocated To
Urban Business
Business GW Perceived Flood Damage
Water Use Damage Investment Delay Time
Availability Reduction Multiplier
Ratio
Water Use
Fraction Total
Business SW Reduction
SW Allocated To
Availability Ratio Table
Urban Business
Business SW Business GW
Use in 2001 Use in 2001
Initial
Max Allowable Business
SW Pumping Structures
SW Use Per GW Use Per
Business Business
Structure Structure
p. 433), and corresponds to a third order information delay. This function has the
responsibility of smoothing as well as delaying the input signal (actual drought level),
upon which future decisions are made.
Response to drought in the model is accomplished through a water use reduction
ratio (W URRt ), implying that area residents and businesses actually reduce their
water consumption based on perceived drought conditions. The right plot in Fig-
ure 26 shows the functional relationship between the perceived drought level and the
water use reduction ratio. The relationship is based on the measures recommended
by OLWR (2003). As drought conditions worsen every water user is assumed (opti-
where W URRT is the water use reduction ratio relationship that relates perceived
drought level to reduced water use. Relationships like this simply convert an input
total damage (originally calculated from the hydrologic model component operating
on a six hour time step and aggregating the results) is compared against a monthly
average damage. This ratio of current to average damage is related to investment ini-
tiatives undertaken to rebuild and/or repair the damage. The functional relationship
between current to average damage and its investment multiplier is shown on the left
plot in Figure 26, and represented mathematically as:
where F DIMT stands for the flood damage investment multiplier table. This table
relates a damage ratio (damage D in [$] to average monthly damage D in [$]) to a
flood damage investment multiplier F DIMt in [-]. The logic of such a relationship
is as follows: if the ratio of current to average damage increases (i.e., the area sees
more flood damage) the investment multiplier will correspondingly increase, reflect-
ing the undertaking of initiatives designed to rebuild, repair and/or replace assets
damaged from a flood. Note that the function for this multiplier never dips below
unity, and thus assumes that decisions to recover from flood damage are always initi-
ated. Modifying this function results in testing alternate assumptions related to flood
damage recovery. For example, a function that shows a decline of future investment
in response to increasing flood damage could easily be tested. This would imply that
flood investment multiplier curve would dip below unity for high levels of damage,
thus discouraging future business investment.
The flood damage investment multiplier is also put through an information delay
implying that it takes time to assess damage and come up with extra funds to rebuilt
178
and/or repair the damage. Such a formulation takes the following form:
where P F DIMt stands for the perceived flood damage investment multiplier in [-],
and DmDT for damage delay time in [yrs]. In the simulation model the flood damage
is shown in the form of spikes (i.e., high damage amounts during short time intervals),
tions are never known precisely, an information delay is applied to this variable as
well. (Information delays also filter out fluctuations, thus providing noise free signal
for the quantity of interest.) Perceived ground water recharge P GW Rt [106 m3 /yr]
is therefore represented by:
P GW Rt = Smooth3(GW Rt , GW DT ) (69)
[106 m3 /yr], defined as maximum surface water allowed to be pumped (from Lakes
179
Huron and Erie) MAP in [106 m3 /yr], multiplied by a fraction allocated to urban
business fSW UB in [-].
BSW At = fSW UB · MAP (71)
All water allocation fractions used in the model are calculated based on current
use practices (see Table 4 in Section 3.1.2). In order to estimate average ground and
surface water use per business structure, we take the total respective water use and
where GW UBS and SW UBS stand for ground and surface water user per business
structure, in [106 m3 /yr/unit]; BGW U2001 and BSW U2001 represents total ground
and surface water used in year 2001 in [106 m3 /yr], and BS0 for initial number of
business structures.
The rest of the urban business activity sector is shown in Figure 27, where itali-
cized variables represent information obtained from outside (either from other sectors,
or other figures). The main state variable representing economic activity of the re-
gion is a business structure. Its definition is identical to one provided in the work
Alfeld and Graham (1976), where the authors explore a number of different defini-
tions of economic activity, but in the end select the business structure as the best
one (p. 10). According to the authors, a number of businesses does not describe
regional economic activity well because a single large company can create more ac-
tivity than many smaller ones. A monetary measure is also not satisfactory, as this
makes it difficult to relate economic activity to employment, migration, land use, and
subsequently, taxes.
180
Attractivness Of
Urban Jobs
Attractivness Of
Multiplier
Urban Jobs
Table
Business Labour Multiplier
Force Multiplier Urban Population
Table
Urban Labour Urban Labour
Business Labour Force To Jobs
Force Multiplier Force
Ratio
Figure 27: Level and rate diagram of the urban business activity sector
181
how much depends on number of employees, and land area occupied). Growth of
business activity, as it is defined here, takes place when area creates a demand for
investment and construction of new business structures. Similarly, an increase in the
rate of demolition and depreciation of business structures creates an economic down
where BSt represents a number of business structures in [units], and BCt and BDt
stand for business construction and demolition, respectively in [units/yr].
The main variable regulating dynamics in the business activity sector is business
construction multiplier (BCMt ); it is defined as a product of business land (BLMt ),
business labour force (BLF Mt ), business water availability (BW AMt ), and the per-
182
The multiplier effect works on the premise that any one of its dependent variables
has the potential to significantly affect its outcome. Consider an example where
water resources are depleted in an area, but labour force, jobs and business land
are plentiful, and floods are not an issue. Despite the fact that other conditions are
favourable, depleted water resources imply that growth can not be sustained. Similar
logic is applied for other combinations of the above variables (i.e., businesses can’t
generate additional structures if either labour force, land or water is unavailable).
The urban business activity sector sector consists of a number of feedback loops,
which are discussed next. Water use in this sector (the same logic is used in all other
sectors) is divided into surface and ground water uses. Each business structure is
assumed to use a certain amount of water (both from surface and ground sources
in Middlesex County, or from ground water alone in Oxford and Perth Counties).
Combining water use per business structure (SW UBS, GW UBS) with the total
number of structures (BSt ), multiplied with a water use reduction ratio W URRt ,
results in variables named business surface (BSW UD) and ground (BGW UDt ) water
use desired (see Figure 27). By dividing desired water use (BSW UD, BGW UD)
by water availability (BSW At , BGW At ) for each use type, desired business use
to availability ratios are computed. Under simulation conditions emphasizing high
growth it was discovered that desired water use to availability ratios can actually
become larger than unity, a physical impossibility. In order to deal with situations
such as this, a fuzzy minimum function is used to gradually convert desired use to
availability ratios to (actual) use to availability ratios that never exceed unity. The
183
where BSW UEt (BGW UEt ) stands for business surface (ground) water use effect;
W UET for water use effect table.
The logic of this relationship implies that as water use to availability ratios increase
(and more available water gets used), further industrial growth can not be sustained
because local governments are assumed to raise costs of servicing and delivering water
to discourage growth that can potentially make the area run out of water. Use effects
from surface and ground water are combined into a variable called business water
availability multiplier (BW AM) by a weighted average, which is then used as one of
Next, the model obtains a total area of land zoned for business purposes (BLt in
[km2 ]) from a land use sector (described later), from which a fraction of total business
184
1.5 1.5
Water Use Effect (−)
2 2
Attr of Jobs Mult (−)
1.5 1.5
1 1
0.5 0.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Labour Force to Jobs (−) Labour Force To Jobs (−)
Labour Force Mult Table Attr of Jobs Mult Table
This ratio is used next to compute a business land multiplier, and thus influence
construction of new business structures. Its relationship is shown in Figure 28 on
top right and its rationale is the following: Small fractions of occupied land imply its
wide availability; in such cases further increasing land occupancy leads to increased
business activity. In other words, it is attractive for businesses to move to such area as
servicing (water, electricity, sewer, etc.) is assumed to have been already completed,
with the added bonus of having plenty of available land for its operations. However,
A total urban labour force (ULFt [people]) is calculated by taking urban popula-
186
tion (UPt [people]) and multiplying it by an urban labour force participation factor
ULF P F in [-].
ULFt = UPt · ULF P F (80)
After this, by dividing total labour force (ULFt ) to available jobs (UJt ), a variable
labour force to jobs ratio is obtained (ULF JRt ).
the functional form given on bottom left in Figure 28. In this relationship, low labour
force to jobs ratio implies that an area has less available labour than jobs, thus having
the potential to hamper business investment (i.e., people just can not be found to fill
available positions). On the other hand, high ratios of labour force to jobs in an area
mean that there exists a labour force which exceeds available jobs. In such conditions
businesses are assumed to react positively, as they can pick and choose among a large
pool of available skilled workers to fill the needed positions.
Labour force to jobs ratio (ULF JRt ) is also used to determine an attractiveness of
jobs (AUJMt ), used later by the population sector to regulate the number of people
moving to an area. The relationship between labour force to jobs ratio and a jobs
attractiveness multiplier is shown on bottom right in Figure 28, where small labour
force to jobs ratio imply that labour force is smaller than available jobs—in other
words the area has more jobs than people to fill them. Although not attractive for
(where labour force is greater than available jobs), the area stops being attractive for
new families because the prospect of finding jobs becomes grim.
namics (without available housing people can not live or work in an area). Housing
also influences regional economic health, where lack of housing means lack of people,
reducing a potential labour force of skilled workers, thus halting potential growth of
the economy. The housing sector used in this work is taken directly from the work
of Alfeld and Graham (1976). Systemic structure of the sector consists of a stock
variable representing urban houses (UHt [houses]), modulated by rates representing
housing construction (HCt [houses/yr]) and demolition (HDt [houses/yr]), respec-
tively. The level and rate diagram of the housing sector is shown in Figure 29, and is
represented by:
UHt = UHt−1 + (HCt − HDt )∆t (82)
The feedback structure of the sector includes loops representing housing and land
availability. The principle for determining the housing land multiplier is identical to
one discussed in the business activity sector with the only exception that now houses
are used instead of business structures. Therefore, the details will not be repeated
here. The second feedback loop regulating dynamics is one that involves housing
availability, and is described next. Knowing the number of houses in an area (UHt
[houses]), an average household size (HS [persons/household]), and the population
(P [people]), a households to houses ratio (HHRt [-]) is computed (it is simply a ratio
188
Attractiveness Of
Urban Housing
Urban Housing Multiplier Table
Availability
Multiplier Table Attractiveness Of
Urban Housing
Multiplier
Urban Housing
Availability
Multiplier
Urban Housing
Land Multiplier
Urban
Table
Population
Urban Housing
Land Multiplier
Household To
Urban Housing Residential Land Houses Ratio Household
Construction Fraction Occupied Size
Multiplier
Urban
Urban Housing Houses Urban Housing
Construction Demolition
Figure 29: Level and rate diagram of the urban housing sector
189
1.5 1.2
1
1
0.8
0.5 0.6
0 0.4
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Household to Houses Ratio (−) Household to Houses Ratio (−)
Housing Avail Mult Table Attr Housing Mult Table
According to Alfeld and Graham (1976), “the households to houses ratio repre-
sents an aggregate of more tangible variables that respond to housing market con-
ditions, including price, vacancy rates, maintenance and service, finder’s fees, and
flexibility in choice of location (p. 153).” Similarly to the labour force to jobs ra-
tio, households to houses ratio is used to provide a signal to the housing industry
thus modulating the number of houses within an area by directly influencing housing
construction.
The housing availability multiplier is a variable that receives the signal provided
from the households to houses ratio and uses it to compute a multiplier effect directly
impacting housing construction. Its functional form is shown on the left in Figure 30.
Low households to houses ratios imply that an area has more available houses than
households, thus indicating plentiful housing conditions. Such cases therefore provide
signals to the housing construction industry that more houses are not needed, and
no additional funds are invested in construction. High households to houses ratios
190
on the other hand, imply that the region has more households (i.e., families) than
available homes, thus signalling the need for additional housing construction. Of
course, an assumption is made that housing construction multiplier has a saturation
effect, meaning that after a certain point no matter how high the demand for new
an area consists of more available houses than it needs, thus making it attractive for
those wishing to move there. On the other hand, high households to houses ratios
stand for conditions where households (i.e., families) outnumber existing houses, thus
making it unattractive for newcomers. The same saturation effect is assumed in this
relationship as well.
Urban population dynamics are described in this sector, which contain a number of
variables previously discussed. The input variables obtained from elsewhere in the
model (shown in italics) are depicted in Figure 31, and represent an identical formu-
lation to one already discussed in the urban business activity sector. The level and
rate diagram of the urban population sector is shown in Figure 32, and is described
next.
The structure of the population sector consists of a state variable denoting a cur-
rent level of urban population (UPt [people]), with its rates being in (UIMt ) and
out (UOMt ) migration, as well as birth (UBt ) and death (UBt ) rates, all in [peo-
ple/yr]. Normal values of all variables are chosen from the 2001 Canadian census
(STATCAN, 2001). The births and in migration have the ability to increase the pop-
191
Max Allowable
SW Pumping
Attractivness Of
Urban Housing
Multiplier
Attractivness Of
Urban Jobs
Multiplier
Urban Attractivness
Multiplier
Urban
Population
Urban In Urban Out Urban Out
Migration Migration Migration
Urban In Normal
Water Use
Migration
Reduction
Normal Urban Residential Ratio
SW Use Desired
Urban Residential Urban Residential
GW Availability GW Use Desired
SW Use Per
Urban Person
Urban Residential GW Use Per
GW Use Desired Urban Person Urban Residential
To Availability SW Availability
Ratio
FuzzyMin Urban Residential
SW Use Desired
Urban Residential GW To Availability
Use To Availabilty Ratio
Urban Residential
GW Use Urban Residential SW
Use To Availabilty
Water Use
Urban Residential Effect Table Urban Residential
SW Use Effect SW Use
Urban Residential
GW Use Effect
Urban Residential
Water Availability
Attractivness
Figure 32: Level and rate diagram of the urban population sector
193
ulation, while out migration and deaths can reduce the population. This is depicted
with the following equation:
The urban attractiveness multiplier is the only variable that has the ability to reg-
ulate the behaviour of the population sector. It consists of attractiveness multipliers
representing availability of housing, jobs, and water. The structure used to determine
the water availability multiplier is identical to one described above in the business
activity sector, and thus will not be repeated. The only difference is that variables
names are changed (where the word business is replaced by residential). The attrac-
tiveness of urban housing and urban jobs multipliers have been also described above,
in the housing and business activity sectors, respectively. The urban attractiveness
The rural business activity sector parallels its urban counterpart, and uses the same
principles in its formulation. The auxiliary variables of this sector are shown in
Figure 33, and the level and rate diagram is presented in Figure 34. Some differences,
however, are discussed next.
The state variable describing the rural business activity sector is selected to be
Perceived GW
Recharge
Fraction Total
GW Allocated To
Farm Units
Farm GW
Availability
Fraction Total
SW Allocated To Farm SW
Farm Units Availability
Farm SW Farm GW
Use in 2001 Use in 2001
Initial
Max Allowable
Farm
SW Pumping
Units
SW Use Per GW Use Per
Farm Unit Farm Unit
Attractivness Of
Rural Jobs
Attractivness Of
Multiplier
Rural Jobs
Table
Rural Labour Multiplier
Force Multiplier Rural Population
Table
Rural Labour Rural Labour
Rural Labour Force To Jobs
Force Multiplier Force
Ratio
Figure 34: Level and rate diagram of the rural business activity sector
196
stock, equipment, etc). A farm unit thus consists of an aggregate unit responsible
for agricultural operations. The total number of farm units can be modulated by
investment (i.e., construction of more farms) and reduced by depreciation (where old
farm structures are demolished).
The farm unit investment multiplier is defined in the same manner as its coun-
terpart in the urban business activity sector. As before, it has been formulated to
depend on availability of water, land, labour force, as well as the perceived flood dam-
age investment multiplier (see Figure 34). The water use reduction ratio (previously
calculated from inputs of the hydrologic model component) is applied to the desired
water use, and has the effect of reducing water usage rates in response to regional
drought conditions. The attractiveness of rural jobs is defined in the same manner as
its urban equivalent, and thus provides a signal to the rural population sector.
The rural population sector is identically defined as its urban counterpart; it is shown
in Figures 35 and 36 for completeness. Since a rural housing sector is not explicitly
defined (it has been aggregated into a definition of farm units), a separate attractive-
ness multiplier is therefore not present in the formulation of this sector. The rural
population therefore responds to two attractiveness multipliers, namely water avail-
ability attractiveness (RW AAt ) and attractiveness of rural jobs (ARJMt ), defined in
the same way as its urban equivalents. All feedback relationships in this sector are
indistinguishable in structure from ones used earlier, with the only difference that
variable names have been changed, and populated using different initial conditions.
197
Max Allowable
SW Pumping
Attractivness Of
Rural Jobs
Multiplier
Rural Attractivness
Multiplier
Rural
Population
Rural In Rural Out Rural Out
Migration Migration Migration
Rural In Normal
Water Use
Migration
Reduction
Normal Rural Residential Ratio
SW Use Desired
Rural Residential Rural Residential
GW Availability GW Use Desired
SW Use Per
Rural Person
Rural Residential GW Use Per
GW Use Desired Rural Person Rural Residential
To Availability SW Availability
Ratio
FuzzyMin Rural Residential
SW Use Desired
Rural Residential GW To Availability
Use To Availabilty Ratio
Rural Residential
GW Use Rural Residential SW
Use To Availabilty
Water Use
Rural Residential Effect Table Rural Residential
SW Use Effect SW Use
Rural Residential
GW Use Effect
Rural Residential
Water Availability
Attractivness
Figure 36: Level and rate diagram of the rural population sector
199
The land use sector consists of the most detailed component of the socio-economic
land to suit its respective needs. For example, if the business community finds itself
in need of additional land for development, it will place pressures on the local govern-
ment and its land use planners to allow such zoning. In the structure of the land use
sector, a conversion of land use types can come from a number of different sources.
The land use sector is divided into four categories (see Figure 24): forest cover (F Ct ),
agricultural (ALt ), residential (RLt ) and business land (BLt ) all in [km2 ].
The auxiliary variables of the land use sector are shown in Figure 37, which
describe the competition structure between residential and business land uses. This
is needed in cases if an area develops majority of it surrounding agricultural lands,
to residential land each year. Residential land fraction rezoned depends on a normal
rate of such conversion (RLRN [1/km2 ]), and multiplier effects named residential land
availability multiplier (RLAMt [-]) and residential land rezoning multiplier (RLRMt
[-]).
RLF Rt = RLRN · RLAMt · RLRMt (85)
Business
Residential Land Business
Land Occupied Land Occupied
Located near the top of Figure 37 are the variables that describe community
pressure that eventually leads land use planners to change the land zoning. The resi-
dential community pressure multiplier (RCP Mt [-]) is a variable that generates such
pressure for the conversion of residential land to business uses. It responds to the
households to houses ratio in the following manner (see top left plot in Figure 38):
for increasing values of the ratio (meaning the area has more families than houses)
the pressure to rezone residential land to business uses decreases. The business com-
munity pressure multiplier (BCP Mt [-]) is defined in the same way. Top right plot in
Figure 38 shows the relationship between an urban labour force to jobs ratio and its
corresponding business community pressure multiplier. For increasing labour force to
jobs ratios (implying a shortage of jobs generating the need for additional business
land on which new business structures can be built), the pressure to rezone residential
Since it takes considerable time for the community pressure to actually convert land
from one type of use to another, an information delay is thus applied to the residential
land rezoning pressure.
where RLRP At stands for residential land rezoning pressure average [-], and RLRP AT
202
2 2
1.5 1.5
1 1
0.5 0.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Households To Houses (−) Labour Force To Jobs (−)
Res Comm Press Mult Table Bus Comm Press Mult Table
Bus Land Rez Mult (−)
Res Land Rez Mult (−)
25 12
20 10
8
15
6
10
4
5 2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Res Land Rez Press (−) Res Land Rez Press (−)
Res Land Rez Mult Table Bus Land Rez Mult Table
Bus Land Avail Mult (−)
Res Land Avail Mult (−)
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Res Land Frac Occupied (−) Bus Land Frac Occupied (−)
Res Land Avail Mult Table Bus Land Avail Mult Table
for residential land rezoning pressure averaging time [yrs]. The delayed variable is
one that helps determine the residential (and business) land rezoning multipliers. The
relationship between the delayed pressure, and residential land rezoning multiplier is
shown in the middle left plot in Figure 38. As the average pressure for the residential
land to get rezoned increases, more residential land actually gets rezoned. Similar
principle is used for the business land rezoning multiplier, shown in the middle right
plot of Figure 38, where increased pressure to convert more residential land reduces
the demand to rezone business land to residential use.
The variable residential land fraction rezoned also depends on the land that is
actually available for rezoning. If land is occupied, the process of rezoning that land
can not take place until the structure occupying it (whether it be a business, house
or a farm unit) is demolished. Therefore, the residential (and business) land fraction
RLOt = UH · LP H (88)
The amount of residential land occupied (RLOt ) is divided by the total land zoned
for residential purposes (RLt ), to obtain a variable called residential land fraction
occupied (RLF Ot ):
RLF Ot = RLOt /RLt (89)
The residential land availability multiplier (RLAMt ) is obtained from inputs of res-
idential land occupied. The relationship for this multiplier is shown in the bottom
left plot in Figure 38, and uses the following logic: as more residential land gets oc-
cupied, less is available for rezoning to business use types. An identical procedure is
204
employed when specifying the relationship for the business land availability multiplier
(BLAMt ), shown on the bottom right plot in Figure 38, where business land fraction
occupied is obtained using the same way as its residential counterpart.
Residential and business land fraction rezoned (RLF Rt and BLF Rt ) are multi-
plied by the their corresponding state variables (residential (RLt ) and business (BLt )
land) to obtain rates of residential and business land rezoning (RLRt and BLRt ),
shown in Figure 39. The rate variables are the final outcome of the rezoning process
between residential and business land, and represents the actual land rezoned each
year. The rate of rezoning forest cover to agricultural land similarly depends on the
fraction occupied by farm units, and is presented next. The functional relationship is
shown in the left plot in Figure 40, and it assumes that after 30% of agricultural land
is occupied by farm units, growing pressures develop for additional agricultural land.
The actual rate variable forest to agricultural rezoning (F ARt ) is a product between
the current level of forest cover (F Ct ), a normal rate of conversion F ARN, and its
rezoning multiplier F ARM. Forest cover in this work is assumed to be able to only
be converted to agricultural land.
will be a growing trend to rezone the remaining agricultural land for urban uses. The
relationship is assumed to saturate at high urban occupancy values, implying that a
limit exists regarding how much (and how fast) such conversion can occur.
205
Forest To Agricultural
Fraction Occupied Rezoning Multiplier
By Farm Units Table
Figure 39: Level and rate diagram of the land use sector
Ag to Res/Bus Mult (−)
For To Ag Rez Mult (−)
2.5 2
2 1.5
1.5
1
1
0.5 0.5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Frac Occ Farm Units (−) Res/Bus Frac Occupied (−)
For To Ag Rez Mult Table Ag to Res/Bus Mult Table
This section describes the main contribution of the work presented in this report,
Even though previous modeling efforts have been prepared for different purposes
(see Chapter 2), all share one common feature: they study water resources manage-
ment problems using the framework of systems thinking and feedback. Nearly all
employ some sort of interactions between physical and socio-economic realms; this is
indeed their largest contribution. However, none consider feedback type interaction of
a modeling framework where physical and hydrologic processes influence, and are
influenced by socio-economic processes has not been adequately treated in the cur-
rent system simulation literature. The main contribution of this work develops a
framework where detailed hydrologic processes are linked (via feedback) to larger
The socio-economic model component presented above can not be simulated without
hydrologic inputs—it is postulated that available water, flood and drought conditions
are necessary determinants to regional social and economic well being of the river
basin. The hydrologic model component on the other hand, could be run (this is how
they are traditionally used) without inputs from the socio-economic model compo-
nent. Simulating the model in this way implies that change in regional socio-economic
character does not affect hydrologic processes. The Upper Thames Systems Model,
referred to the model that joins the physical and socio-economic components via feed-
back, captures processes and produces conditions with more realism than either one
of the model components by themselves.
The links between hydrologic and socio-economic model components used in this
work are shown in Figure 41 via a causal loop diagram. The left part of the diagram
shows feedback structure of the continuous hydrologic model component (and in par-
ticular the HEC-HMS Soil Moisture Accounting algorithm), while the right portion
points to a select few variables of the socio-economic model component. The model
consists of three major links between its components. Each is discussed next.
The hydrologic component of the model provides precipitation and temperature
(originally obtained from the weather generator model), alongside with ground water
recharge and direct runoff to the socio-economic model component. The links are
shown as heavy lines in Figure 41. As mentioned previously, the hydrologic component
operates on a six hours time step, while the socio-economic component uses a monthly
time step. The two model components interact in the following way (the interested
reader is encouraged to look at the code listing provided in the Appendix, where all
+
Potential
Evapotranspiration Vegetation
+ −
Temperature
Precipitation
Maximum − Urban
+
Surface Land
+ − + − Storage
Canopy and
Surface Storage Excess
− + + + Population and
+ Direct + Business Sectors
+ Infiltration Runoff
+
Infiltration Capacity Flood +
− + Damage
+ Soil Storage
− Water Use
+ + −
Percolation
−
+ −
Baseflow
Groundwater
Storage + −
Drought
Level
+
−
GW Recharge
then executed.
The socio-economic model component then simulates dynamics of the social con-
ditions of the study area, and thus updates its major state (population, land cover,
business structures, farms, etc.) and rate variables (in migration, business and farm
investment, rezoning rates, etc.). The socio-economic conditions resulting from the
execution of the socio-economic component update the physical conditions of the
hydrologic component through a number of parameters, embedding the assumption
that regional hydrologic conditions dynamically change in light of changing socio-
economic conditions. The updating of the parameters is done before the execution
of the hydrologic model (at the start of the second month) and includes the fol-
lowing (see Figure 41): potential evapotranspiration, maximum surface storage, and
soil infiltration capacity. The input used to determine the degree of change to the
hydrologic parameters come from the land use sector of the socio-economic model.
Assumed functions that relate land use variables (such as urban and vegetative land)
210
area occupied by business structures and houses, divided by the total basin area. The
relationship assumes that as a fraction of paved land increases, the surface storage
capacity decreases and excess runoff increases. Higher runoff causes an increasing
amount of rainfall (and snowfall) to end up in channels and ponds immediately after
the storm, thus reducing the amount of water that can potentially infiltrate into the
soil.
The middle plot in Figure 42 similarly shows the relationship between a fraction
of vegetation cover (defined as a sum of forest cover and agricultural land divided
by the total basin area) and maximum soil infiltration. The infiltration capacity of
the soil depends on many factors, some of which are the soil type, moisture content,
organic matter, vegetative cover, and season (Linsley et al., 1958). The soil type is by
far the most important regulator of infiltration capacity, as soil porosity modulates
the resistance of flow. More specifically, increasing soil porosity increases infiltration
capacity (for example, infiltration is greater in sandy soils than ones containing large
amounts of clay and silt). Vegetation also increases infiltration capacity of the soil,
as it tends to increase soil porosity. Furthermore, Linsley et al. (1958) claim that:
1.6
1.2
0.8
0.4
0
0 0.2 0.4 0.6 0.8 1
Fraction of Vegitation Land (−)
Max Soil Infil Effect Table
1.25
PET Effect (−)
1
0.75
0.5
0.25
0
0 0.2 0.4 0.6 0.8 1
Fraction of Vegitation Land (−)
PET Effect Table
The assumed relationship shows that as vegetative cover increases, so does the
maximum soil infiltration. However, the reverse is also true: as the vegetative cover
decreases, the infiltration capacity also decreases, thus increasing surface runoff and
even further lowering ground water recharge.
this study, it will suffice to say that evapotranspiration heavily depends on vegeta-
tion cover. Furthermore, as vegetation cover increases and an area sees more trees,
brush, grasses and other plants, the total amount of water lost to the atmosphere
from vegetation (i.e., evapotranspiration) also increases. The relationship in bottom
plot of Figure 42 is formulated to reflect this fact. The interested reader is referred
to Snyder et al. (2004), where more details are provided regarding vegetation cover
and its impact on climate.
All relationship shown in Figure 42 have been formulated to capture accurately
conditions of the study area. For example, the fraction of vegetative cover in the
Upper Thames River basin at the start of the simulation is somewhere close to 0.9;
therefore, the hydrologic model component receives a multiplier effect of unity at the
start. However, as simulated time goes on and vegetation cover is reduced (through
urbanization), soil infiltration capacity and potential evapotranspiration are lowered
(by having a multiplier effect of less than unity). The same is true for the effect of
maximum surface storage. At current low fractions of urban land cover, the surface
storage multiplier effect is close to unity implying no alteration of parameters of the
hydrologic model component. However, as the amount of urban land increases, the
multiplier effect reduces the surface storage capacity, therefore producing more runoff.
213
Section 2.3.2 outlines details regarding development, testing and sensitivity analyses
mate change impacts on water resources systems. All model parameters are estimated
using the latest available data from Statistics Canada community profiles (STATCAN,
2001), as well local sources (UTRCA, 2001).
Extreme conditions and surprise behaviour tests are used to measure the ability
of the model to respond to policies, external shocks and noise outside of what has
been historically observed. For example, tests that assume an infinite supply of water
214
1.75
for the Upper Thames River basin are simulated, as are management policies that
encourage a complete switch from ground to surface water use (see Section 4.3.3).
tionships in the integrated model. This formulation assumes that available water
(captured through water use to availability ratios) places a limit on future regional
growth and development. For low ratios of water use to availability (meaning plen-
tiful supplies) growth and expansion is encouraged. However, for high water use to
availability ratios (implying scarce supplies) growth is severely restricted as nearly all
available water is used, and therefore can not support additional development. Two
different formulations of this function are presented in Figure 43 for the purposes of
a sensitivity analysis.
Simulation output for a number of different variables is shown in Figures 44 and 45.
The results of the simulations indicate that sensitivity to changes in the water use to
215
600
600
400
400
200
200 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base Run 1 Run 2 Base Run 1 Run 2
GW Recharge (106 m3 /yr)
100 60
GW Use (106 m3 /yr)
80
40
60
20
40
20 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base Run 1 Run 2 Base Run 1 Run 2
SW Pumping (106 m3 /yr)
106 100
SW Use (106 m3 /yr)
105 80
104 60
103 40
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base Run 1 Run 2 Base Run 1 Run 2
30 1000
5 200
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base Run 1 Run 2 Base Run 1 Run 2
1.2 70
Rural Pop. (1000’s)
Farms (1000’s)
1 60
0.8 50
0.6 40
0.4 30
0.2 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base Run 1 Run 2 Base Run 1 Run 2
A key feedback relationship that links the hydrologic and socio-economic model
components relates a level of urbanization (measured as the fraction of paved land)
to soil’s ability to store water (measured by hydrologic variable maximum surface
storage). The original relationship is displayed on the top plot in Figure 42, where
increasing amounts of paved surfaces are reducing soil’s ability to retain water dur-
Climatic scenarios used to drive the hydrologic model component are presented in
Section 3.2.2, and are thus not further mentioned here. The socio-economic scenarios
(also referred as policy options) are described below.
Five different socio-economic scenarios are considered in this work, and are briefly
218
described next. The first scenario considered is referred to as the base case scenario.
This scenario shows the system behaviour under normal (or no change) conditions.
The usefulness of having a base case scenario is mainly for comparison of outcomes
with other scenarios. In computer simulation modeling of physical and social systems
is therefore derived from testing its different assumptions, policies and management
strategies, and comparing such outcomes to a reference (or base case) scenario.
The second scenario tested in the model is one which considers area’s water re-
sources to be infinite. Even though this is a rather unrealistic scenario, it is worth
testing as many area residents believe that water supply in the region can never be
exhausted. This belief stems from the fact that our study area is surrounded by the
Great Lakes, and the view that Great Lakes are an infinite source of water. This
view is further exacerbated by low cost of lake water, and willingness of governments
to build new pipelines to supply water to areas that had previously relied on ground
water (as in the case of city of Strathroy, located west of London). The structure
of this scenario is identical to the base case scenario, with the only exception that
water availability feedback loops in the socio-economic component are disabled (i.e.,
they are artificially set to unity for the entire simulation). This means that regional
population and economic activity can grow regardless of the condition of its water
resources.
The third scenario tested is one where area residents and businesses (in other
words all users of ground and surface water) voluntary agree to reduce their water
use by 30% when compared to the base case scenario. This scenario is aimed to show
219
the behavioural implication of a strict water conservation policy. Within the socio-
economic model component, this scenario is set by initially reducing per capita usage
rates of urban and rural population, as well as reducing water usage per business
structure and farm unit by 30%.
The fourth scenario is similar to the third scenario, as the same one time wa-
ter use reduction of 30% is employed, together with land rezoning policies empha-
sizing conservation of forest and agricultural lands. This scenario aims to reduce
rates of urbanization resulting from favourable socio-economic conditions (abundant
land, water, jobs, housing, etc); it also aims to reduce total water use in the basin.
This scenario achieves this by severely restricting the conversion of agricultural land
to business and residential uses, as well as implementing strict water conservation
policies. In the socio-economic model component, this scenario is implemented by
scenario, with the exception that no ground water is used in the basin. In other words,
this scenario looks to address the implications of performing a switch from ground
to surface water supply basin-wide. This means that counties that have historically
been using either combination of ground and surface water (like Middlesex), or only
ground water (like Perth and Oxford), are assumed to implement policies and initiate
construction projects that will provide water from the Great Lakes. This scenario
is also unrealistic, but it shows implications of extending the view point that Great
Lakes will always be there to provide the much needed water.
220
Summary of Scenarios
The simulation process undertaken in this study is performed in such a way to consider
all possible combinations of climate and socio-economic scenarios. This means that
for each climate scenario, five socio-economic scenarios are tested to determine the
overall system behaviour. Figure 47 shows the schematic of all scenarios (fifteen in
total), while Table 10 provides their brief summaries.
Climate Socio-economic
Scenarios Scenarios
Base
InfWat
Hist RedWat
RedWatLimLand
SwitchToSW
Base
InfWat
Scenario Dry RedWat
RedWatLimLand
SwitchToSW
Base
InfWat
Wet RedWat
RedWatLimLand
SwitchToSW
RedWatLimitLand SwitchToSW
Management strategies A scenario where a
that assume a switch from ground
reduction of per water to surface water
capita water use by is made, as
30%, together with governments allow
limiting conversion of further extraction of
agricultural (and Great Lakes water and
forest) land to abandon ground water
residential and sources completely.
business uses.
222
This chapter shows simulation results for all models, methods and algorithms devel-
oped in Chapters 2 and 3 for the Upper Thames River basin case study. Simulation
results are divided into the following categories: (i) generation of climatic input (spa-
cial storms, weather generator climates, intensity duration frequency analysis), (ii)
inverse analysis (showing impacts on flood and low flows), and (iii) integrated analysis
(where a continuous hydrologic component is dynamically coupled to a socio-economic
component via feedback). Each category is presented in sequence.
This section reveals the simulation results used for generation of climatic inputs for
impact assessment modeling. Spacio-temporal storm model simulations for the Up-
per Thames River basin are presented first, followed by detailed weather generator
outputs (daily and hourly) for the same study area. Results of rainfall Intensity-
Duration-Frequency (IDF) analysis for the City of London are also presented.
The results of the spacio-temporal storm model application to the Upper Thames
River basin, with parameters shown in Table 5, are shown in Figure 48. Outputs
are displayed as isohyetal maps showing the spacial distribution of rainfall across the
basin. The isohyetal maps are generated by using a spacial contouring algorithm that
takes daily rainfall totals at subcatchment centroids (locations for which time series
data is generated), and estimates rainfall totals for all other locations within the basin.
The result of the contouring algorithm is a map showing the spacial distribution of a
particular rainfall storm corresponding to parameters given in Table 5.
223
Lattitude (degrees)
Lattitude (degrees)
43.4 43.4
40
40
80
43.2 43.2
20
40
60
43.0 43.0
42.8 42.8
-81.6 -81.4 -81.2 -81.0 -80.8 -80.6 -81.6 -81.4 -81.2 -81.0 -80.8 -80.6
Longitude (degrees) Longitude (degrees)
(c) (d)
43.6 43.6
Lattitude (degrees)
Lattitude (degrees)
43.4 43.4
20
80
40
60
43.2
40
43.2 20
20
40
43.0 43.0
42.8 42.8
-81.6 -81.4 -81.2 -81.0 -80.8 -80.6 -81.6 -81.4 -81.2 -81.0 -80.8 -80.6
Longitude (degrees) Longitude (degrees)
Figure 48(a) shows the result of the spacio-temporal model for simulation 1 (see
Table 5); results for simulation 2 (where only the temporal mass curves are changed)
are not shown, as the outputs have a nearly identical spacial distribution. Simulation
3 is shown in Figure 48(b), displaying a storm covering the entire basin. Similarly,
the result for simulation 4 (where the temporal mass curve is altered) is not shown
due to its close resemblance to outputs of simulation 3. The results of simulation 4
are displayed in Figure 48(c), which shows a storm with a limited spacial extent in
one axis, while extending the entire length of the basin in the other axis. Storms such
as these are useful for studies in the Upper Thames River basin, as the region often
sees storms with similar patterns originating from Lakes Huron or Erie. Lastly, the
result of simulation 6 is shown in Figure 48(d), where a typical storm with an altered
orientation parameter is displayed, indicating the ability of the model to produce
alternatively be used.
Following the simulation of the daily weather generator model for all stations and
scenarios, the statistical tools of auto and cross-correlations (described in Section
225
2.1.2) are applied to test whether the methodology is able to replicate the temporal
and spacial character of regional rainfall. A similar analysis of weather generator
results, showing spacial and temporal character of total precipitation, maximum and
minimum temperature are presented by Sharif and Burn (2006, 2007) and are thus
not repeated here. Figure 49 shows sample output of rainfall analysis, using the above
mentioned statistical tools. The processing of simulation results for other stations are
similar, and is therefore not shown.
The statistics shown in Figure 49 depict monthly rainfall and lag-1 auto-correlation
for the London station, as well as lag-0 cross-correlation between London and Toronto,
and London and Windsor (the stations with the smallest number of missing hourly
data). The presentation of statistics is shown in box and whisker plots, where the
bottom and top of the box show the 25th and 75th percentile of data, with the median
in between. For all cases, historically observed averages are also plotted to identify
the ability of the model to represent the structure of temporal and spacial character
of rainfall in the area. In all cases, the model adequately replicates the historically
observed patterns.
A number of observations are drawn from the plots shown in Figure 49. First, the
winter rainfall is quite minor and not variable (with the smallest interquantile range
values from all months). This observation is explained by the freezing of gauges in
the winter when only a small amount of rainfall is recorded (i.e., most of the precipi-
tation falls as snow). Second, the summer months have the largest monthly rainfall,
and with the most variability. Third, the lag-1 auto-correlation structure is slightly
underestimated, when compared to the observed record. Since the observed values
are only slightly above zero, this underestimation does not pose a significant prob-
lem. Fourth, the cross-correlations between London and Toronto, and London and
Windsor are adequately replicated. These comparisons show that winter rainfall has
226
0.8
250
0.6
200
0.4
150
100 0.2
50 0
0
-0.2
-50
01 02 03 04 05 06 07 08 09 10 11 12 01 02 03 04 05 06 07 08 09 10 11 12
Month of year Month of year
Lag-0 Crosscorrelation London-Toronto Lag-0 Crosscorrelation London-Windsor
1.2 1.2
Lag-0 Crosscorrelation (−)
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
01 02 03 04 05 06 07 08 09 10 11 12 01 02 03 04 05 06 07 08 09 10 11 12
Month of year Month of year
result from convective patterns in the atmosphere. In such storms, intensity rapidly
changes over relatively small areas, thereby producing localized flooding conditions.
Convective type storms are most important to the design of storm water man-
agement infrastructure that take surface runoff away from the streets and safely
transports it to storm water management ponds and/or area rivers. The statisti-
cal properties of storms of this kind are captured by means of IDF curves, discussed
in Section 4.1.3.
Daily weather generator simulations, for use in the inverse approach (see stations of
Table 8), produce a daily record of total precipitation, as well as maximum and min-
imum temperatures. The generated daily data are used as inputs into the continuous
hydrologic model. Daily precipitation data are further disaggregated by Wey (2006),
to obtain hourly inputs for use in the event model and the study of flooding. A
different set of stations (shown in Table 9) is used to study local short-duration high-
intensity events for the City of London. The disaggregation mechanism developed in
Section 2.1.3 is used to produce hourly time series data for each station, to which a
frequency analysis is performed and IDF curves generated for each simulated climate
scenario. The method of Wey (2006) was not used in the disaggregation employed for
local IDF analysis, as it was not necessary to reproduce regional inter-station char-
acteristics at the hourly time scale. The daily weather generator on the other hand,
is able to reproduce the inter-station spacio-temporal character of observed data.
228
of the historical rainfall events for relatively short-durations, an extra weather gen-
erator simulation is considered (in addition to one representing historic, wet and
dry climates). Note that the weather generator model presented in Section 2.1.2
has a perturbation mechanism that is able to generated values not previously ob-
served. Historic, wet and dry inputs are considered. In order to attempt to replicate
the IDF curves currently used in Ontario as its design standard (MTO, 1997), an-
other simulation is considered. This output is named historic undisturbed, indicating
the removal of the perturbation component responsible for generating unprecedented
weather sequences (i.e., removal of Step 10 in the weather generator algorithm). Ta-
ble 11 summarizes the differences between the input climates and weather generator
outputs, and offers a brief description of each simulation output produced.
Table 12 shows the IDF data generated by the undisturbed historic scenario,
together with similar data produced by the Meteorological Service of Canada in 2001
(available from MTO, 1997). The IDF data acquired from processing the output of
three weather generator scenarios (historic, wet and dry) is shown in Table 13, from
which detailed comparisons are made. The data presented in Tables 12 and 13 is
graphically represented in standard plots for all scenarios, in Figure 50.
In order to examine the accuracy of reproduced IDF data for London, two addi-
tional figures are used. The plots of Figure 51 show the comparison between the undis-
turbed historic, historic and the MSC intensity data for various durations. Through
visual inspection, the following is concluded: (i) for rainfall of short-duration (1 hr
and less), the historic weather generator scenario most closely reproduces the MSC
intensity data, for all return periods, (ii) for intermediate rainfall durations (2, 6 and
229
Intensity (mm/hr)
100 100
10 10
1 1
0.1 1 10 0.1 1 10
Duration (hrs) Duration (hrs)
Dry WG Scenario MSC from 2001
Intensity (mm/hr)
Intensity (mm/hr)
100 100
10 10
1 1
0.1 1 10 0.1 1 10
Duration (hrs) Duration (hrs)
Wet WG Scenario
Intensity (mm/hr)
100 Legend:
The lower most curve represents the
return period of 2yr, followed by 5, 10
25, 50 and 100 yrs.
10
1
0.1 1 10
Duration (hrs)
Table 12: IDF for London: undisturbed historic and MSC IDF
Undisturbed Historic Return period, T
Duration 2 yr 5 yr 10 yr 25 yr 50 yr 100 yr
5 min 6.2 8.7 10.3 12.4 13.9 15.5
10 min 9.6 13.5 16.0 19.2 21.6 24.0
15 min 12.2 17.0 20.3 24.4 27.4 30.4
30 min 16.9 23.6 28.1 33.8 38.0 42.1
1 hr 21.3 29.9 35.6 42.7 48.0 53.3
2 hr 26.8 39.9 48.5 59.5 67.6 75.6
6 hr 34.4 49.6 59.7 72.5 82.0 91.4
12 hr 39.2 54.1 64.0 76.5 85.7 94.9
24 hr 42.8 57.8 67.7 80.3 89.6 98.8
Table 13: IDF for London: historic, dry and wet WG output scenarios
12 hrs), the undisturbed historic (i.e., weather generator up to step 9) most closely
replicates the MSC data, and (iii) for rainfall of longer duration (24 hrs), both the
undisturbed historic and the historic weather generator scenarios overestimate the
MSC data for all return periods.
Daily weather generator output, after being disaggregated into hourly rainfall,
is used to produce IDF data for three different climate scenarios. The comparison
of this data is shown in Figure 52. An important observation is made by visually
inspecting Figure 52: for all return periods and durations, the wet climate scenario
shows a much greater magnitude (and intensity) than either the historic or the dry
climate scenario. Since the wet scenario is designed specifically to study high-intensity,
short-duration flooding conditions resulting from climatic change, this observation
has major implications for the design, operation and maintenance of storm water
with staff from the MSC revealed that data prior to 1960 does exist in paper form,
but is not available to the general public. Daily data on the other hand is available
from 1943 to present for the London station, but this data is of limited use when es-
timating intensities and frequencies of durations shorter than 24 hrs. Using different
data sets (the present work uses 1965-2000, while the MSC uses 1943-2001) explains
differences observed in results presented in this work when compared to MSC esti-
mates. Therefore, direct comparisons between the the simulation outputs developed
in this study and MSC curves are not entirely appropriate.
233
60
40
Rainfall (mm)
Rainfall (mm)
50
30
40
20
30
10 20
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
2 hr Rainfall 6 hr Rainfall
100 120
100
80
Rainfall (mm)
Rainfall (mm)
80
60
60
40
40
20 20
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
12 hr Rainfall 24 hr Rainfall
140 140
120 120
Rainfall (mm)
Rainfall (mm)
100
100
80
80
60
40 60
20 40
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
Undisturbed Historic Historic MSC
Figure 51: IDF curves for London: undisturbed historic, historic and MSC
234
70
50
Rainfall (mm)
Rainfall (mm)
60
40 50
40
30
30
20 20
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
2 hr Rainfall 6 hr Rainfall
120 160
100 140
Rainfall (mm)
Rainfall (mm)
120
80
100
60
80
40 60
20 40
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
12 hr Rainfall 24 hr Rainfall
160 160
140 140
Rainfall (mm)
Rainfall (mm)
120
120
100
100
80
60 80
40 60
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
Historic Dry Wet
Figure 52: IDF curves for London: historic, dry and wet WG output
235
It has been noted that the number of stations used in the weather generator algorithm
can influence the generation of synthetic rainfall. Using stations surrounding the site
of interest is beneficial, because the structure of spacial and temporal correlations in
the region is maintained. This produces weather in the entire region that is consis-
tent with statistical properties of the observed record (auto and cross-correlations).
Furthermore, in cases where only limited rainfall data is available (for example some
stations may have up to 50% of its hourly record missing), using surrounding stations
is entirely appropriate.
In the study of short-duration, high-intensity local rainfall, use of too many sur-
rounding stations may be questioned, because convective storms are highly localized
weather patterns, operating on relatively small spacial scales. Using stations too far
from the site of interest may influence the weather generator results (i.e., with possi-
ble increases or decreases in rainfall intensity). In order to study this phenomenon,
a smaller study area is considered that uses only stations within a 100 km radius
of London. The stations included in this analysis (with distances from London in
brackets) include: Brantford (76 km), Delhi (51 km), London (0 km), Sarnia (94
km), Stratford (41 km), St. Thomas (26 km), Waterloo (79 km) and Woodstock (34
km). The output of this analysis is shown in Table 14 and Table 15 as percent differ-
ence between simulations using stations within a 100 and 200 km radius of London,
respectively. The percent difference is computed as:
i100 − i200
%Diff = × 100 (90)
i200
where i100 and i200 are rainfall intensities obtained from analysis considering stations
within a 100 and 200 km radius of London, respectively.
236
Table 14: IDF for London: use of stations within a 100 km radius of London
Historic WG Scenario Return period, T
Duration 2 yr 5 yr 10 yr 25 yr 50 yr 100 yr
5 min 7.5 10.5 12.5 15.0 16.8 18.6
10 min 11.7 16.3 19.4 23.2 26.1 28.9
15 min 14.8 20.7 24.5 29.4 33.0 36.7
30 min 20.5 28.6 34.0 40.8 45.8 50.8
1 hr 26.0 36.2 43.0 51.6 58.0 64.3
2 hr 33.4 52.0 64.4 79.9 91.5 102.9
6 hr 44.0 65.7 80.0 98.2 111.6 125.0
12 hr 52.1 73.6 87.9 105.9 119.3 132.5
24 hr 58.2 80.1 94.6 113.0 126.6 140.1
Table 15: IDF for London: % difference between 100 and 200 km simulations
Results shown in Table 15 indicate that using data from within a 100 km radius
of London results in insufficiently different simulations from those obtained using the
200 km data. Since most of the differences fall within 5%, the original simulation
results of Table 13 (for simulations using stations within 200 km) are kept without
further modification.
The City of London currently uses two different IDF curves as standards on which
water management infrastructure is designed, operated and maintained. Conveyance
systems are designed based on a curve provided by MacLaren (1962), while most other
storm water management facilities are designed based on criteria provided by the
City of London (2003). The IDF curve currently used to design conveyance systems
has been adopted from a 1962 study and is based on data from the 1950’s for the
Toronto area. To quote:
Since adequate rainfall data is not available for the direct determination of
intensity-frequency curves for London, the rainfall intensity-frequency curves
available for the Toronto area were compared with the corresponding data de-
rived from the regional maps, since the regional maps indicate no variation in
the rainfall intensity-frequency to be anticipated in either Toronto or London.
This comparison indicated that the intensities derived from the regional maps
were approximately 10 per cent greater than those obtained from the Toronto
curves. Accordingly, the two-year frequency curve [for London] ... was derived
by increasing the intensities obtained for a two-year frequency storm in the
Toronto area by 10 per cent (MacLaren, 1962, p. 100).
Despite being outdated and originally derived for a different geographic location,
the IDF curves used for designing conveyance systems are quite conservative. There-
fore, one of the recommendations of this report is for the City to undertake a study
239
to ascertain if using this curve is appropriate for either present (or future) conditions,
especially in light of information presented in this report.
The design of other storm water management infrastructure in the City of London
is based on the 1986 IDF curve provided by the Atmospheric Environment Service.
lower estimates for design rainfall than those from more recent sources. Markus et al.
(2007) also show that larger design rainfall can result in even greater design discharges
(due to the non-linear nature of the rainfall-runoff process); they too therefore provide
support for the recommendation that current design standards should be reviewed
and updated.
To illustrate the implications of climate change on municipal storm water manage-
ment for the City of London, IDF values from historic and wet simulation scenarios
are considered. Using current design standards (City of London, 2003) results in the
following:
• Design of small storm drains, combined sewer overflow trunks, and others with
a time of concentration less than 90 minutes (requiring use of a 5-yr storm),
will be subject to approximately an 8 to 15% increases in design peak flows.
• Design of larger storm drains and combined sewer overflow trunks (with times
240
• Design of storage volumes (in storm water management ponds, detention basins
and other storm water facilities) will also need to be increased. For designs
needing to meet water quality and erosion control standards (5 yr storm, with
plications on urban drainage practice may be worth considering. Watt et al. (2003)
divide the implications into the following categories:
be deeper (which can cause more erosion); pond volumes will have to increase,
thus requiring deeper ponds, larger land areas, or both; these volumes will also
impact the frequency of combined sewer outflows (where rain water is combined
with raw sewage and discharged into a receiving water body).
fall) are expected to prevail. For both quantity and quality control structures,
this maintenance may translate into policies demanding more frequent inspec-
tions, repairs and/or upgrades.
241
• Externalities. Factors that are not directly related to climate change but rele-
vant to design, operation and maintenance of urban drainage systems, also need
to be taken into account. Examples of such factors include long-term deteriora-
tion of existing structures, population growth, new models for financing urban
infrastructure, licencing storm water loadings, water pricing, etc.
Cities and communities need adaptation strategies that consider both short and
long term impacts of climate change on municipal infrastructure to adequately prepare
for the future.
The inverse analysis to climate change impact assessment includes the application
the inverse analysis, and shows the impact of changing climatic conditions on the
local hydrologic flow regime (flood and low flows).
The first step in the application of the inverse approach involves identifying the critical
flood exposures within the area under consideration. In consultation with the staff
242
1200 1400
Peak Flow (m3 /s)
from the Upper Thames River Conservation Authority, the information in Tables 16-
18 has been assembled. The table entries have been taken from a local flood time
operations manual, and show either exposures, or required courses of action that
government (and other) officials need to take. One of the final goals of this analysis
is to show how the frequency of such exposures might shift with a changed climatic
signal.
Results of weather generator simulations are processed by the hydrologic models
for historic, wet and dry climate scenarios. Annual maximum peak flows are selected
from the output hydrographs and used in flood flow frequency analysis. The Gumbel
and Log Pearson III statistical distributions are employed to fit the hydrologic output
data, thus producing frequency plots (flood magnitude vs. return period). Figure 53
shows the results of the statistical analyses for three different climates (historic, wet
and dry), and three different stations (St. Marys, Ealing and Byron). Note should be
taken regarding the magnitudes of flood flows for return periods greater than 100 yrs.
This is because large flood magnitudes (i.e., larger than 100 yrs) are extrapolated
from fitting 100 years of simulated data to a probability distribution (Gumbel or Log
Pearson III). Therefore, these values should be considered with caution. The obtained
results show similar outcomes for both distributions up to the 100 yr return period;
Log Pearson III distribution shows higher magnitudes for larger recurrence intervals
resulting from more conservative estimates in the extrapolation of extreme values.
The Log Pearson III distribution is selected as the basis upon which the current
study’s conclusions are made, as local regulators use this distribution to set regula-
tory standards in the basin (floodplain mapping and extent, infrastructure designs,
development possibilities, etc.). In order to apply the inverse approach, suppose that
the hydrologic exposure of interest (step 1) is the event that causes the dykes in Byron
(west London) to over-top. In Table 18, it is observed that this occurs at flow of 990
246
m3 /s. After completion of steps 2, 3, and 4 (calibrating the event hydrologic model,
obtaining weather generator inputs, and performing the simulation/frequency anal-
ysis), step 5 of the approach involves looking up recurrence intervals of the critical
hydrologic exposure for all climate scenarios. Using the bottom right plot of Figure
53, it is observed that a flow of 990 cms has a return period of 33 yrs under the
historic climate scenario.
Under the wet climate scenario, where precipitation magnitude is expected to
increase, the return period for the flow overtoping the dyke is only 17 yrs. This
therefore suggests that the level of safety offered by the dyke under the wet scenario
will be about two times lower than the level under present conditions. Under the dry
climate scenario, where dry spells dominate the regional hydrology, the return period
of the flow overtoping the dyke is expected to increase to 65 yrs. Such a climate
scenario indicates that if dry spells and low flow conditions dominate, flooding will
be less frequent. The dry climate is only shown here for comparison purposes, as it
is designed specifically to assist the study of drought effects and low flows. For the
purpose of understanding climate change impacts on flooding, the focus rests with
the wet climate scenario simulation (as this is the most critical scenario for flooding).
The main finding obtained by implementing the inverse approach to the assess-
ment of climatic change impacts is that flooding in the Upper Thames River basin will
occur more frequently, if the future climate resembles the wet scenario. Therefore,
property flooding will occur more frequently on average than it did in the past. This
impact is of serious concern especially for smaller communities that are already ex-
periencing regular flooding resulting in property damage (St. Marys, Mitchell). The
changing climatic conditions are expected to increase flood damage in the future.
247
extreme low flow characteristics (Durrans, 1996). Annual minimum seven day and
monthly flow durations are used in this study, as these are defined by Ontario’s
Ministry of Environment as flows providing a limiting condition for sewage treatment
and waste water disposal into a receiving water body (Pryce, 2004). Figure 54 shows
the statistical analyses of annual minimum seven day (left column) and monthly
(right column) flows for three different stations (St. Marys, Ingersoll and Byron),
for three different scenarios (historic, dry and wet). As an aside, unlike the case of
flood analysis, low flow frequency analysis uses return period flows of non-exceedence:
meaning the probability that flows are equal to (or less than) some specified value.
In order to illustrate the use the of inverse approach for the analysis of droughts
and low flows in the Upper Thames River basin, critical drought exposures are first
identified. Suppose that of interest is the occurrence of a level II drought (see Table
19) in the vicinity of the Greenway Pollution Control Plant in Byron. Greenway
is the largest sewage treatment plant in the basin, having the capacity of 150,000
cubic meters per day of waste water, and is always running at or near capacity. By
looking at the time series record of simulated flow for the historic climate scenario,
and performing the necessary analysis, the lowest average summer month flow for
the Byron stream gauge is 9.5 m3 /s. (For the Ingersoll and St. Marys stream gauges
the lowest average summer flow is 1.0, and 3.8 m3 /s, respectively.) This number is
obtained by selecting from each simulated year the lowest average summer month
flow; all such flows are then averaged over a period of record (100 years in our case).
In order for a level II drought to occur in the basin, the lowest average summer
monthly flow must drop to 50% of its average (see Table 19), which works out to
248
10 10
0.1
1 10 100 1 10 100
Weibull Return Period (yrs) Weibull Return Period (yrs)
Hist Dry Wet Hist Dry Wet
10 Monthly Flow (m3 /s) 10
7 Day Flow (m3 /s)
1 Ingersoll Ingersoll
1
0.1
0.01 0.1
0.001
0.01
1e-04
1e-05 0.001
1 10 100 1 10 100
Weibull Return Period (yrs) Weibull Return Period (yrs)
Hist Dry Wet Hist Dry Wet
Monthly Flow (m3 /s)
7 Day Flow (m3 /s)
10 Byron Byron
10
0.1 1
1 10 100 1 10 100
Weibull Return Period (yrs) Weibull Return Period (yrs)
Hist Dry Wet Hist Dry Wet
Table 19: Critical drought exposures in the Upper Thames River basin
Monthly Streamflow Level† Drought Exposure
Less than 70 % of I Potential water supply problems;
lowest average Request volontary conservation; Ask
summer month for 10 % reduction in water use;
Inform media of drought conditions.
about 5 m3 /s for the Byron station. Using the 5 m3 /s value of the monthly flow,
flow return periods for all climate scenarios are obtained. According to the historic
climate, a level II drought has a return period of 6.3 years (i.e., flow of 5 m3 /s or less is
expected to occur with a probability of 1/6.3 = 0.16 in any given year). Surprisingly,
simulation showed that under the dry climate, the return period for a level II drought
did not change, implying that future drought conditions are expected to be the same
as those currently observed. Simulation under the wet climate scenario, however,
showed the occurrence of a level II drought with a return period of 10 years (meaning
that a flow of 5 m3 /s or less is expected with a probability of 0.1 any given year, or
less often).
The main finding obtained by the implementation of the inverse approach to the
assessment of climatic change impacts is that low flow conditions in the Upper Thames
River basin will remain approximately the same as those currently observed (Figure
54).
The last step of the inverse approach involves formulating water resources manage-
ment recommendations. These recommendations need to address the possible adverse
effects of changing hydro-climatic conditions on flooding and on the occurrence of low
questions regarding what could/should be done about climate change, and how its
adverse effects could influence future policy and decision making. Some of the issues
identified by the stakeholders are discussed in this section.
Flooding
For the Upper Thames River basin specifically, altered hydrologic conditions, as a
result of changes in climate imply that the frequency of flooding will be altered as
well, especially for the case where higher precipitation magnitude is considered (wet
climate). Therefore, some of the current guidelines and management practices may
require revision. For example, the UTRCA, may need to revise and update their
concern, especially for smaller communities that are already experiencing regular
flooding resulting in property damage (St. Marys, Mitchell).
The City of London has a number of flood management works in place, namely
the Fanshawe Reservoir and an elaborate dyking system for the Thames River. These
flood management works were originally designed to ensure safety of the city from
damaging floods. As a result of climatic change, this infrastructure may in the future
provide a lower level of safety than it had been originally designed for. In the case of
252
Fanshawe dam, the maximum release of 840 cms represents a regulatory return period
of 100 yrs (historic case, obtained from the top left plot in Figure 53) which will, as
a result of the changing climate be reduced to only 50 yrs (wet climate). Therefore,
appropriate government officials may need to consider retrofitting or upgrading the
dam and other existing flood management infrastructure in the basin. Such retrofits
and upgrades will result in increased budgets for flood management operations.
A further consequence of the potentially changed climate is that current design
standards for bridges, roads, sewers, drains, storm water facilities and other municipal
infrastructure should be reviewed, in order to assess their ability to cope with the
changing hydro-climatic conditions.
The following presents a summary of recommendations for possible changes of
flood management guidelines, as a result of climatic change in the Upper Thames
River basin. Some of the issues included in these recommendations are taken from
comments and suggestions provided by the basin’s stakeholders during the aforemen-
tioned stakeholders meeting (held on November 22, 2005). The changes are catego-
rized in the following three categories:
lations currently employed by the UTRCA may require revision. This recom-
mendation is supported by the analysis showing that floods of all magnitudes
will increase in frequency (i.e., they will occur more often). Possible revisions
to existing regulation include those pertaining to: (i) the use of the river and its
adjacent park land for recreation (including boating, camping, swimming, fish-
ing, jogging, etc.), (ii) the removal of pumping stations near area rivers during
the course of a flood (see Tables 16-18), (iii) the patrolling of river banks during
periods of high water levels and the monitoring of the performance of critical
253
infrastructure (such as roads, bridges, culverts, drains, sewer systems), (iv) the
issuing of permits for floodplain development (including construction of roads,
buildings, bridges, culverts, drains, sewer systems, etc), and (v) the revision of
reservoir operating procedures (including smaller dams, weirs, outflows, etc.).
this happens, cracks, crevices and other problems will intensify, thus requiring
increased maintenance levels leading to higher costs. An evaluation of current
structural and non-structural measures used to reduce flood damage (such as
reservoirs, dykes, floodwalls or implementation of land use zoning practices,
flood warning systems, waterproofing, etc.) will need to take place. As flood
frequency increases basin-wide, so will budgets allocated to the maintenance of
flood protection measures. Next to changed maintenance budgets for existing
infrastructure, a need will arise for investment in additional infrastructure to
bring the flood protection to regulatory levels and meet the needs of the pop-
Although the province of Ontario has a low water response management framework
that includes officials across all sectors of the government and incorporates various
long and short term measures, local and/or regional drought management plans are
still missing. This omission is further reinforced by the fact that adverse drought
related impacts in the Upper Thames River basin are currently quite minor. Studies
documenting damage resulting from low flows and droughts in the basin are altogether
missing, despite the fact that potentially severe conditions, as the present analysis
shows, can occur with a chance of 1 in 10 any given year.
Local drought management guidelines are needed that address differences in re-
gional characteristics such as predominant land use patterns (urban, rural, agricul-
tural, forest), water use (surface or ground water), population and industrial growth,
as well as other important factors. For example, different regions within the basin
have different thresholds for drought triggers, which need to be carefully assessed and
documented. Currently, the thresholds used are those suggested by the OLWR (2003)
document, and are the same the entire province of Ontario (approximately 1,000,000
km2 of land). Guidelines of this kind can potentially mask local (or smaller scale)
drought conditions. Further exacerbating the problem is the fact that drought triggers
can change with time, as in the cases where infrastructure (such as reservoirs, canals,
255
and agricultural tile drainage) is upgraded or when water demands change. Urban
and rural drought triggers may also need to be set at different levels, as urban infras-
tructure can usually absorb the impact of shorter droughts (Loucks and van Beek,
2005).
It should be noted that water quality in the Thames River (both upstream and
downstream of London) is below the provincial water quality standard that the Min-
istry of Environment sets for Ontario’s watercourses. Like many other watercourses
in Ontario, the Thames River is classified as a Policy 2 watercourse, indicating that
its total phosphorous concentration is above the targets set by the provincial gov-
ernment. Phosphorous is a nutrient necessary for plant and animal life, but elevated
levels can cause excessive plant growth and algal concentrations, as well as eutroph-
ication of surface water bodies. Furthermore, SNC-Lavalin and RV-Anderson (2005)
found that in the Upper Thames River basin, elevated phosphorous concentration are
a result of the following factors:
SNC-Lavalin and RV-Anderson (2005) also found that the mass balance of total
phosphorous sources for London indicates that pollution control plants contribute
about 17% of the total annual phosphorous loading; tributary loads and stormwater
runoff represent 14%; 30% and 38% of total annual phosphorous loading originates
256
from the North and South Branches of the Thames upstream from the city, respec-
tively. It is also noted that total phosphorous loading in the Thames River down-
stream of London has either remained the same or has decreased since the start of
record keeping in 1978, despite an increase in the population of about 100,000 people
between 1978 and 2005. This stability is attributed to the successful removal of phos-
phorous and other pollutants by the area’s pollution control plants, as well as other
water quality improvement initiatives (sewer separation, stormwater management,
sewer use bylaws, etc.) in the City of London.
Despite the excellent work of London’s pollution control plants, the Thames
river consistently experiences water quality problems, particularly during dry sea-
sons. Therefore, it is our recommendation that increased emphasis be placed on
management practices (particularly in rural areas that drain most of the basin) so as
to limit total phosphorous loading. This emphasis is necessary because drought like
conditions can exacerbate already poor water quality in area rivers.
Another issue of concern when considering droughts is that of water use. Indi-
viduals living in smaller communities that are users of ground water may experience
drying of wells; thus the drought conditions may require the drilling of costly new and
deeper wells. This is further exacerbated by the fact that water use during times of
drought increases (due to warmer temperatures), while ground water recharge (due to
decreases in precipitation) usually decreases. This means that conditions are possible
where water demand exceeds the supply. Careful management guidelines requiring
strict conservation and possible restrictions (of water uses in commercial, industrial,
residential and agricultural sectors) are required to lower the impact of these poten-
tially troubling situations. Furthermore, modifying reservoir operations and enforcing
compliance of existing water taking permits may also need to take place.
A large amount of surface water is extracted from Lakes Erie and Huron and
257
is currently used primarily by residents of London and its surrounding areas. City
and other public officials often acknowledge the widespread belief of those living in
London that there is a virtually unlimited supply of water for the city. Such a belief
is troubling, since water levels in Lake Huron have recently been much lower than in
the past. Furthermore, users may not fully appreciate the impact of increasing water
withdrawals from the Great Lakes, especially in light of the knowledge that only about
2% of the area’s water is renewable. More aggressive drought education initiatives
are needed (in addition to current information bulletins provided by the UTRCA);
rigorous education initiatives are needed to make the residents more aware of the
potential for drought, potential impacts, and possible long term damage reduction
mechanisms.
Based on the results of our analyses, four categories of recommendations are made
regarding drought conditions in the Upper Thames River basin. Some of the issues in-
cluded in these recommendations are taken from comments and suggestions provided
by the basin’s stakeholders. The recommendations include the following:
etc.), (ii) tourism and recreation (including the impact of stagnant, foul smelling
water on trail and park visits for jogging, camping, fishing, swimming, etc.), (iii)
wetlands (including the loss of habitat for fish, reptiles, birds and mammals,
and changes in vegetation cover), (iv) reservoir operations (including conflicting
258
use of water for low flow augmentation and flood control), (v) ground water
withdrawal (including increased cost during peak drought times, the drilling of
deeper wells), and (vi) streamflow water quality (especially when flow consists
of treated sewage).
gers over time are possible. A flexible regulatory framework should be set in
place to allow for changes in definitions, should they be deemed necessary.
rous loading); new regulations should be set in place to monitor water quality
in both short and long term. A review of all existing water quality guidelines
should also take place. Some of these revisions should include a review of: (i)
agricultural fertilizer use, (ii) storm water management techniques for current
and future land development, (iii) combined sewer discharges, (iv) pollution
control plant effluent strategies, and (v) landfill use practices, including solid
waste management.
tating impact on people, plant and animal life is seen as paramount to sustaining
and preserving the natural environment. Novel educational programs (targeting
water resources managers and practitioners, municipal and other government
officials, farmers, and the general public) are urgently needed. The following
modifying long held beliefs that water is a limitless resource), and (iii) enhanc-
ing interactions between socio-economic and physical domains (emphasizing a
systemic approach with a realization that structure of socio-economic systems
has just as much to do with drought as current climatic and/or physical condi-
tions).
Securing funds for such studies is recognized to be a challenge, since the age-
ing water management infrastructure (reservoirs, dykes, sewer systems, etc.) in the
basin demands immediate attention. A fine balance between short and long term
goals needs to be achieved; after all, the sustainability of water resources is at stake.
changing climate will affect everyone living in the basin, from possibly higher taxes
to restricted floodplain development possibilities. Interested groups and individuals
should be given an opportunity to participate in the planning process, and should be
included in the decision-making.
260
The results presented in this section show the application of the integrated model to
the Upper Thames River basin. The simulation output is divided into the following
categories: First, a number of figures are presented that show the current state of
the basin, using the historically identical climate combined with the base case socio-
economic scenario. This output shows the implications of following current policies
scenarios reveals impacts that a changed climate can have on regional hydrologic char-
acteristics. After this, a number of socio-economic scenarios are tested, while keeping
the climate signal fixed. With this set of scenarios, alternate polices and management
strategies are tested in order to show socio-economic impacts of the Upper Thames
River basin. Lastly, scenario pairs are tested which couple all combinations of cli-
matic and socio-economic scenarios and reveal behavioural implications. The climate
change scenarios used are described in Section 3.2.2, while the socio-economic scenar-
ios are outlined in Section 3.4.4. Figure 47 and Table 10 show the summaries of all
The state of the physical (i.e., hydrologic) conditions are discussed first, and include
flood and drought conditions in the basin (shown in Table 20 and Figures 55 and
56). The output is presented in terms of probability frequency curves showing the
return period versus a flow of specified duration, together with timing and regularity of
hydrologic extremes (after Cunderlik and Burn, 2006). Frequency curves are obtained
application of the integrated water resources management model. These results are
contrasted with similar results presented in Sections 4.2.1 and 4.2.2, depicting the
results of the inverse analysis. In these sections climate change impacts on flood and
low flows is characterized using outputs of event and continuous hydrologic models
alone. The flood and low flow characteristics under climate change presented in this
section are derived from outputs of the integrated model, where hydrologic and socio-
economic components are coupled.
Indicators of hydrologic extremes employed by this study synthesize large amounts
of hydrologic data and provide a single value as an index upon which change (as one
induced by the altered climate) can be measured. The selected indicators measure the
changes in magnitude, timing, and regularity of hydrologic extremes. The magnitude
indicator selects typical return periods (100 year for high flows, and 20 year for low
flows) used in practice with duration of (either one, seven or thirty day), and examines
5 800
4
3 600
2
400
1
0 200
-1
-2 0
0 200 400 600 800 1 10 100
Daily Peak Flow (m3 /s) Return Period (yrs)
Data Gumbel Data Gumbel LP3
Log Pearson III Factor, k
3 1000
2
800
1
0 600
-1
400
-2
-3 200
100 1000 1 10 100 1000
Daily Peak Flow (m3 /s) Return Period (yrs)
Data LP3 Gumbel LP3
10 20
8
15
6
10
4
2 5
0 0
1 10 100 1 10 100
Return Period (yrs) Return Period (yrs)
Data Weibull Data Weibull
5 10
Monthly Flow (m3 /s)
7 Day Flow (m3 /s)
4 8
3 6
2 4
1 2
0 0
1 10 100 1000 1 10 100 1000
Return Period (yrs) Return Period (yrs)
Weibull Weibull
measured with an indicator taking values between 1 and 365, thus pointing out the
average day of the year on which extreme flow occurs. If mean day of flood turns out
to be n, this means that, on average, highest annual flow is expected to occur on the
nth day of the year. The regularity indicator, on the other hand is a measure used to
indicate the variability of timing, on a scale between 0 and 1. Values close to unity
indicate that extreme events occur roughly at the same time each year, while values
close to zero suggest greater variability in terms of timing. For formal definitions of
these indicators, the interested reader is referred to a paper by Cunderlik and Burn
(2006). It should be noted that magnitude, timing and regularity indicators are of
limited value by themselves; their strength lies in comparing different hydro-climatic
scenarios, and subsequently inferring conclusions from their relative changes (i.e., will
extreme flows become larger/smaller, occur sooner/later, be more/less frequent, or
be more/less regular?).
In the analysis of flooding conditions (shown Figure 55), annual maximum daily
flows are selected from the model output, and fitted to the Gumbel and Log Pearson
III statistical distributions. The outputs in Figure 55 are shown as probability paper
plots, where model data (annual extremes) is compared with its fitted distributions.
The probability paper plots are then converted to return period plots, to show the
return period versus annual maximum daily flow, extended to an arbitrary return
period of 500 years (see Figure 55).
The historically identical scenario shows the state of flooding conditions as they
are represented by the model in Table 20 and Figure 55. All measures are based on
annual maximum flows of daily duration, simulated for an arbitrary period of 100
years. The mean day of flood within this period occurs on 37th day of the year, with
a regularity of 0.56. The interpretation of the numbers implies that the largest annual
daily flows are likely to occur both during snowmelt in early spring, and as a result of
265
intense precipitation during later times of the year (particularly later on in the fall).
Daily flood flow with a return period of 100 years, taken at the Byron stream gauge
station just outside of London, is estimated as 714 m3 /s using the Log Pearson III
statistical distribution.
It is recognized that hydrologic data of daily duration is of limited use in the design
of flood management infrastructure, as instantaneous peak flows are often larger and
thus more critical. Even though daily peak flows are different in magnitude than in-
stantaneous peak flows, studying flood flows of daily duration is still beneficial as daily
flows can reveal patterns of change applicable to instantaneous peak flows as well.
For example, methods exist that are able to convert daily peak flows to instantaneous
flows (Fill and Steiner, 2003). Alternatively, if flooding is the only issue of interest,
use of event-based rainfall-runoff models may be justified (Prodanovic and Simonovic,
2006b).
Drought conditions are similarly analyzed (shown in Table 20 and Figure 56),
with the only exception being that low flows are considered. As droughts normally
represent longer term phenomena, flow is aggregated to 7-day and monthly durations
before analysis. Again, the same statistical procedures are used, where annual mini-
mum low flows are selected from the model output, and fitted to a Weibull probability
distribution. This distribution is typically used in analyzing minimal extremes (i.e.,
low flows).
Average seven day and monthly annual minimum flows occur at the beginning
of August (on 213th and 218th day of year, respectively). The regularity indices of
these flows have values of 0.69 and 0.73, respectively, thus being more regular than
flood flows previously analyzed. In the Upper Thames River basin, July and August
have typically been the driest months of the year, sometimes bringing forth intense
periods of drought. Water quality during the dry season is of particular concern to
266
residents of the basin, as baseflow in area rivers can consist of a large percentage of
treated sewage effluent. Water quality during dry seasons is also of concern, as the
basin watercourses have consistently been below provincial standards for phosphorous
concentration. The parameter often used in design of municipal infrastructure is the
7Q20 —a seven day average flow with a return period of 20 years is estimated as 1.61
m3 /s, using the Weibull statistical distribution. Similarly, monthly averaged flow
with a return period of 20 years is approximated as 3.37 m3 /s, using the same low
flow statistical distribution.
Figures 57 and 58 show the evolution of the state of urban and rural conditions in
Middlesex County. The output is presented in terms of two columns of plots: the left
column shows the major state variables used in the socio-economic model component
(business structures, housing and population for urban and farm units, and population
for the rural sector), while the right column shows the variables influencing the state
variables (such as availability of land, labour, water, crops and housing).
In the base case scenario, economic growth and expansion is seen as occurring
at a staggering pace. For example, the number of business structures in the City of
London and Middlesex County in general, is shown to increase by some 60% within
the next decade. Given the current strength of the Canadian economy, the rapid
increase in population growth and immigration, together with the wide availability
of water, labour force, and land, this is not an unreasonable assumption. For the
purposes of this study, the chance of such scenario occuring is only important in that
it provides a base case (or a reference) mode of behaviour, upon which alternate
policies and management strategies can be tested.
As expansion of business structures continues, so must the population needed to
sustain it. Rates of in migration (and possibly even birth rates) are expected to
increase, mainly due to the attractiveness of social and economic well being. During
267
1.4
400
1.2
300 1
0.8
200
0.6
100 0.4
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Houses Land Housing Avail.
Population Attractivness (-)
800 1.4
Population (1000’s)
1.2
700
1
600 0.8
500 0.6
0.4
400
0.2
300 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
People Jobs Housing Water
1 1.6
0.8 1.2
1
0.6
0.8
0.4 0.6
0.4
0.2 0.2
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Farms Crop Labour Water
Population Attractivness (-)
60 1.4
Population (1000’s)
1.2
50
1
40 0.8
0.6
30
0.4
20 0.2
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
People Jobs Water
the period of growth, jobs are expected to be plentiful (and sometime even exceed the
actual labour force), thus attracting even more people. Due to such high economic
well being in both the urban and rural sectors of the basin, the stock of housing must
inevitably increase to support the rapid pace of population growth.
areas most likely to experience such development are the ones closest to existing
developed centres, as access (such as roads, highways, rail lines) and services (water,
sewer, electricity, telecommunications) are already widely available. However, such
a rapid rate of growth places additional stress on an already stressed environment.
The stock of forested and agricultural lands will likely decline, as the the urbanized
landscape increases (businesses, shops, malls, parking lots, homes).
As more land is appropriated for urban use (and impervious or paved areas in-
crease), ground water recharge is expected to decrease. Ground water recharge is
defined here as the fraction of precipitation that enters the ground water aquifers
any given year. In this study, it is assumed that ground water in use can only be as
large as the recharge rate, so that mining ground water supplies is avoided (extract-
ing more from the reserves than is naturally replenished). Decreased ground water
recharge rates are a direct consequence of increased urbanization (i.e., paved and par-
tially paved surfaces are unable to absorb precipitation into the soil that eventually
recharges the aquifers). As a result of this change, the precipitation will be forced
into channels, storage ponds, reservoirs, rivers and streams almost immediately after
falling in the surface, instead of slowly infiltrating into the ground. As more water is
diverted from infiltrating into the ground, less water will recharge into the aquifers
270
that many municipalities use as their water supplies (see Figure 59). Of course, as
the area’s available water resources decline, future growth and expansion may slow
(or indeed stop altogether).
Of all factors limiting growth that are considered (land, labour force, jobs, water
availability, housing), water turns out to be the most important. In other words,
due to high usage rates and declining supplies, water is observed as a limiting factor
to future growth and expansion. At such rapid rates of growth, development and
continual expansion (of population, business structures, farm units, housing) can not
be sustained past two to three decades. After reaching such a time, despite plentiful
housing conditions, availability of land, labour force and jobs, growth simply will not
be able to take place, because water will not be available (or if it is available, it might
be too polluted and costly to treat).
The equilibrium population reaches about 650,000 for London and its surrounding
urban areas, while rural population is seen to grow to 50,000 thus bringing the total
population to 700,000 (see Figures 57 and 58). This population size is more than
double the 2001 census amounts. The region’s housing stock is expected to reach close
to 500,000 (a three fold increase since 2001). Much of the housing stock is assumed
to have been built during times of rapid expansion, when families concentrated on
accumulating capital and wealth, rather than on raising children.
Dynamics of land cover, flood and drought conditions, as well as availability and
use of water are shown in Figure 59. The top left and top right plots show the state
variables depicting land use, and thus show the decline of agricultural and forest land
as a result of increasing residential and business land use. The middle two plots in the
same figure present signals received by the socio-economic model component from its
hydrologic counterpart; these plots show the water use reduction ratio (determined
from drought conditions) and a flood damage investment multiplier (calculated from
271
160 800
Land Area (km2 )
60 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Water Use Reduction Ratio (-)
120 100
100 80
80 60
60 40
40 20
20 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
GW Recharge SW Pumping GW SW
Figure 59: State of land and water use conditions in Middlesex County
272
perceived state of flooding). The variables showing drought and flood conditions
reveal that drought has the potential to reduce water use significantly (that is, if
residents actually conserve water during periods of drought), while flooding is not
expected to pose a great risk. In other words, flood damage is restricted only to those
living in close proximity to low lying areas prone to flooding; such damage alone is
not enough to majorly impact the region’s overall economic well being.
Total available water, from ground and surface supplies, is shown on the bottom
left of Figure 59. The plots show ground water recharge, as well as maximum wa-
ter amounts that may be allowed to be extracted from the Great Lakes for human
consumption. An arbitrary maximum is set as twice the amount pumped in 2001—
implying that, at most, people of Middlesex County can have twice what they had in
2001. Such an assumption illustrates that future international agreements between
Canada and United States will cap the amount of water allowed to be drawn from
the Great Lakes for use by humans. The bottom right plot of Figure 59 shows total
water use from ground and surface sources; this plots illustrates that growth can be
sustained only as long as available water exists. Sometimes large fluctuations in total
water use are observed, resulting from an optimistic assumption that urban and rural
water users actually reduce their consumption during times of drought conditions.
The model assumes that intense drought conditions strengthen conservation efforts,
thus reducing total water use.
This section presents potential changes of hydrologic conditions (flood and drought
regimes) in Middlesex County (and the Upper Thames River basin in general) as a
result of altered climatic signals. All conclusions are derived from Table 21 and Figure
60. 21 shows changes in timing, regularity and magnitude of hydrologic extremes for
273
three different climate change scenarios (historic, dry and wet). Figure 60 shows
frequency plots for high and low flows for the same three scenarios. A potentially
altered flooding regime is examined, by analyzing the wet climate change scenario; this
scenarios represents a plausible future where intensity and magnitude of precipitation
increases. Remarks regarding drought conditions are based on dry climate scenario,
formulated with the intent of studying a possible future where higher temperatures
and dry spells become more frequent.
Impacts on flooding
In assessing floods under changed climatic conditions, information from Table 21 and
from the Log Pearson III statistical distribution (see top right plot in Figure 60).
This indicator is selected because a majority of water management structures (such
as dams, levees, dykes and diversions) are designed to withstand impacts of 100-yr
events. In examining the 100-yr daily peak flow magnitude for the historic scenario
at the Byron stream gauge in Middlesex County is 714 m3 /s, while under the wet
scenario (more intense precipitation regime), it is noted that the flow of the same
return period can increase to 885 m3 /s. This is a significant finding, signalling the
need for possible revision of existing flood management guidelines and municipal
design practices. The dry scenario, on the other hand—representing higher incidents
1200 1200
1000 1000
800 800
600 600
400 400
200 200
1 10 100 1000 1 10 100 1000
Return Period (yrs) Return Period (yrs)
Hist Dry Wet
Weibull Monthly Flow (m3 /s)
6 12
5 10
4 8
3 6
2 4
1 2
0 0
1 10 100 1000 1 10 100 1000
Return Period (yrs) Return Period (yrs)
Hist Dry Wet Hist Dry Wet
Figure 60: Effect of altered climate on floods and droughts in Middlesex County
275
Analyses of possible adverse changes in the frequency of peak flows can be also
performed by using Figure 60. For example, if the top right graph is selected (showing
the Log Pearson III fit) the peak flow of 714 m3 /s for the historic scenario has a return
period of 100 years, while the same flow under the wet scenario may occur with a
return period of 30 years. This finding means that a flow which currently has an
annual exceedence probability of 0.01 (1 in 100), may, under future conditions have
an annual exceedence probability of only 0.03̇ (1 in 30). Such a finding implies that
incidents of flooding in Middlesex County may occur more frequently in the future.
This is true not only for large daily peak flows, but for lower flows as well.
The timing of annual maximum daily peak flows is also affected by the changing
climate. For the historic scenario, the mean day of flood occurs on the 37th day of the
year, with a regularity index of 0.56 (see Table 21); this finding implies that floods
are likely to occur due to snowmelt in the spring, and due to intense rainfall later
in the fall. Under the changed climate, the mean annual daily peak flow may occur
as late as the 55th day of the year (for the wet scenario), comparable to the historic
case. This finding implies that future flooding conditions (under the wet scenario) are
expected to be as variable as they have been in the past, but may occur on average
some 20 days later (i.e., floods from intense rainfall are expected to play more of a
dominant role than in the past).
Altered hydro-climatic conditions resulting from climate change imply that flood
flow frequency is altered as well, especially for the wet climate scenario. This sug-
gests that some of the current guidelines and management practices could be revised
accordingly. For example, the UTRCA, the government body responsible for estab-
lishing and mapping the basin’s floodplains, may need to revise and update their
floodplain maps in light of the changed climatic signal.
276
Low flow and drought conditions are similarly analyzed under the altered climate.
Hydrologic indicators of change are presented in Table 21 and Figure 60. As before,
the results presented here are for Middlesex County only; similar results for Oxford
and Perth Counties are presented in digital form in the Appendix.
The magnitude indicators selected for the study of hydrologic drought conditions
are the annual minimum seven day flows with a return period of 20 years (i.e., 7Q20 ).
These are flows defined by Ontario’s Ministry of Environment as ones providing a
limiting condition for sewage treatment and waste water disposal for a receiving water
body (Pryce, 2004). The 7Q20 flows are therefore used to measure a stream’s ability
to accept (and dilute) point source discharge, like treated sewage effluent; these flows
consequently evaluate a stream’s water quality. Monthly minimum flows of the same
return period are shown here for completeness, as they might be useful in studying
longer term trends.
The 7Q20 parameter for Middlesex County (taken at the Byron stream gauge)
is computed as 1.61 m3 /s for the historic climate scenario. Little variation of this
parameter is observed under an alternate climatic signal (see Table 21 and Figure
60), implying that variation in magnitude of such flows is not expected to change
significantly. Timing and regularity of low flows are also not significantly impacted by
the changes in climate; annual minimum low flows are seen to occur on the 213th day
of the year (end of July and/or beginning of August) for the historic scenario, while
the dry scenario (designed specifically to study impacts of drought) shows minimum
annual low flows occurring on the 215th day of the year. Confidence in the above
conclusions is strengthened by the fact that even the wet scenario (showing higher
precipitation regimes) depicts the mean day of drought as occurring on the 220th day
277
of the year, only a week later from the historic case. Regularity indicators also show
a consistent trend, and support the conclusion that large changes are not expected in
variation of timing of low flows.
Drought conditions are observed to be less frequent than those experienced in the
past. For example, taking a flow of 1.61 m3 /s on the bottom left plot of Figure 60,
showing the annual minimum seven day flow versus the Weibull probability fit, we
observe that under historically identical conditions the return period for such flow is
about 20 yrs. Under the changed climate scenario B21, the same flow has a return
period of 30 yrs, implying that such low flow is expected to occur (on average) less
often. This means that the probability of observing an annual minimum seven day
flow of 1.61 m3 /s or less any given year is currently 0.05 (1 in 20), but may be as
little as 0.03̇ (1 in 30) as a result of changed climate.
conclusion is reached that warrants further discussion prior to addressing the scenario
specific results of the model. This conclusion states that socio-economic policies and
management strategies are observed to have a greater impact on the long term socio-
economic character of a region than impacts of climate change alone. Of course this
conclusion does not suggest that climate change will have no impact on socio-economic
behaviour in the region; it undoubtedly will have an impact, but its impact may not
be as severe as those resulting from policies that have the potential to significantly
278
Table 10 for a quick summary of the scenarios considered in this work. All scenarios
studied in this report start with the same assumptions as those in the base case sce-
nario, and are subsequently appropriately adjusted. Therefore, all simulation results
are compared to the base case scenario.
Even though highly unrealistic, this scenario examines the behavioural implications
of the belief that water resources are limitless in the basin. This scenario shows
what might happen if people believe water will always be available and never in short
supply. This assumption reflects the present belief of many in the basin who think the
region’s close proximity to the Great Lakes makes it immune to any possible future
water shortages. The output of a number of selected variables for this scenario is
shown in Figures 61, 62 and 63.
Figure 61 shows the evolution of the urban and rural sectors, resulting from the
comparison of the infinite water availability scenario with the base case. In this
scenario, tremendous growth is observed (even more than in the base case), as wa-
ter is no longer taken as the limiting factor. Equilibrium values are sometimes four
times higher than observed in the base case scenario. Urban population in Middlesex
279
60 2500
120
Farms (1000’s)
1.5
100
1 80
60
0.5
40
0 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base InfWat Base InfWat
600 600
400 400
200 200
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base InfWat Base InfWat
GW Recharge (106 m3 /yr)
100 60
GW Use (106 m3 /yr)
80
40
60
20
40
20 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base InfWat Base InfWat
SW Pumping (106 m3 /yr)
106 120
SW Use (106 m3 /yr)
100
105
80
104
60
103 40
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base InfWat Base InfWat
Figure 62: Middlesex land and water use conditions under InfWat
281
1200 5
1000 4
800 3
600 2
400 1
200 0
1 10 100 1000 1 10 100 1000
Return Period (yrs) Return Period (yrs)
Base InfWat
Weibull Monthly Flow (m3 /s)
Base InfWat
LP3 Peak Flow (m3 /s)
1000 12
10
800
8
600 6
4
400
2
200 0
1 10 100 1000 1 10 100 1000
Return Period (yrs) Return Period (yrs)
Base InfWat Base InfWat
County (London and its surroundings) is therefore seen to grow to about 2 million,
while the rural sector of the population is seen to grow to more than a 100,000. This
implies that the City of London could become a large metropolitan centre within
the next two to three decades. Tremendous development would be seen in the area,
living standard than originally observed in the base case scenarios. A large number
of businesses (and people) must be sustained on smaller land areas, consistent with
metropolitan like settings. Furthermore, per capita water use in all sectors is signif-
icantly reduced, as area businesses and residents can not use more than is currently
available (dictated by yearly ground water recharge rates and the amount pumped
from the Great Lakes). Ironically in this scenario, growth reduces the amount of
ground water recharge from the base case, thus forcing the area residents to increas-
ingly rely on more surface water supplies. However, the maximum surface water
drawn from the Great Lakes (still assumed to be capped at about twice the to-
tal amount pumped in 2001) is reached in about two decades, thus making further
growth absolutely senseless.
It is interesting to point out changes in regional hydrology that could be expected
from such high rates of growth. As more residential and business land becomes oc-
cupied (and subsequently paved over to make room for roads, houses, parking lots),
a larger portion of rainfall is anticipated to immediately end up in area rivers and
streams, thus increasing the amount of surface runoff. At times when area rivers are
flowing at or near capacity (such as during snowmelt for example), this additional
runoff (that could have infiltrated into the ground) is now added to the flood water,
283
thus further exacerbating flooding conditions. The result of this are increased fre-
quencies of flood flows, for all magnitudes of flow (see Figure 63). Drought conditions
under this scenario are unchanged, compared to those of the base case.
gram for all water users in the area. In testing this scenario, all urban and rural
residents are assumed to reduce their per capita water use by 30%; businesses and
farms are also expected to reduce water use per business structure and farm unit by
the same amount.
water use is indeed reduced. However, since the area now has a greater initial water
availability (people and businesses are using less), the restriction on future growth
due to water availability is not as severe (as more water is now available). This ap-
parent water availability means that higher growth rates in the regional economy,
and in the population are sustained, amounting to more growth than in the base case
scenario. Eventually, the state variables equilibriate with higher levels of population,
businesses, farm units and houses, and therefore increase the area’s overall water use
in comparison to the base case. A policy whose aim was to reduce total water use, in
the end accomplished the exact opposite.
35 1200
70
Farms (1000’s)
1
60
0.8
50
0.6
40
0.4 30
0.2 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWat Base RedWat
600
600
400
400
200
200 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWat Base RedWat
GW Recharge (106 m3 /yr)
100 60
GW Use (106 m3 /yr)
80
40
60
20
40
20 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWat Base RedWat
SW Pumping (106 m3 /yr)
106 100
SW Use (106 m3 /yr)
80
105
60
104
40
103 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWat Base RedWat
variable (such as per capita and per business water use) is often offset by changes
in other places (support for additional growth and development through increased
water availability), so that in the end make the performance of the initial policy
totally counteractive.
If reducing water use alone can not reduce the limiting impact of water availability,
perhaps a policy that restricts water use in combination with placing limits on land
rezoning will perform better. The current scenario investigates exactly this possibility.
This scenario is identical to the one previously discussed, with added policies that
the rural sector increase, as there is more freedom for those involved in agriculture
to expand without pressures from the urban community. The urban community is
now restricted in development opportunities, as emphasis is placed on environmental
preservation. The urban population grows to nearly twice that it was in 2001, while
expansion of residential land occurs at a much lower rate. As water and land conser-
vation programs dominate the management practice, the region is expected to adopt
an increasingly sustainable outlook for its future. Smarter homes are expected to be
designed and built, which require fewer resources (such as water, land and energy).
Careful land use planning leading to smart development and expansion can occur,
25 800
1.5 80
1.2
60
0.9
0.6 40
0.3 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWatLimLand Base RedWatLimLand
600
600
400
400
200
200 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWatLimLand Base RedWatLimLand
GW Recharge (106 m3 /yr)
100 60
GW Use (106 m3 /yr)
80
40
60
20
40
20 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWatLimLand Base RedWatLimLand
SW Pumping (106 m3 /yr)
106 100
SW Use (106 m3 /yr)
80
105
60
104
40
103 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base RedWatLimLand Base RedWatLimLand
while increasing residential land only by a modest amount. Having more agricultural
land available over time increases ground water recharge, meaning that more water
becomes available for use than compared to the base case scenario. Even though more
ground water is available, the land rezoning restriction policy successfully achieves a
reduction of total water use (both for ground and surface water).
Of course, such a future can not be expected to occur unless major shifts in atti-
tude occurs, from economic and social sectors within the region. Major efforts will be
needed to achieve public support for such policies, while educational efforts towards
The last of the socio-economic scenarios involves the testing of a policy where a shift
from ground to surface water use is made basin-wide. The simulation output for this
scenario is shown in Figures 68 and 69, and simply represents a policy change that
eliminates the use of ground water entirely, replacing it with water pumped from the
Great Lakes. It assumes that massive investment efforts are initiated by all levels
of business structures, when compared to the base case scenario. In the base case
scenario, the combination of surface and ground water provided more total water
than the increased supply of surface water alone available for this scenario, thus
290
25 800
120
Farms (1000’s)
2
100
1.5
80
1
60
0.5 40
0 20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base SwitchToSW Base SwitchToSW
600
600
400
400
200
200 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base SwitchToSW Base SwitchToSW
GW Recharge (106 m3 /yr)
100 60
GW Use (106 m3 /yr)
80
40
60
20
40
20 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base SwitchToSW Base SwitchToSW
SW Pumping (106 m3 /yr)
140 120
SW Use (106 m3 /yr)
100
120 80
60
100 40
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (yrs) Time (yrs)
Base SwitchToSW Base SwitchToSW
Figure 69: Middlesex land and water use conditions under SwitchToSW
292
lowering the population and the number of business units. The rural population and
the number of farm units grows modestly, mainly because of the wider availability
of land and water. Even though ground water use in this scenario is reduced to
zero, surface water use is correspondingly higher. Cost associated with converting an
entire basin to surface water supplies, especially when simulation shows that better
socio-economic conditions do not emerge, makes this scenario rather unattractive for
future consideration.
293
5 Conclusions
Conclusions provided here relate to three major themes of this study. Firstly, con-
clusions are outlined relating to changes in storm water management design practice
that result from changes in extreme rainfall based on the changing climate. Secondly,
remarks are presented regarding basin-wide flood and low flow management under
changing climatic conditions drawn from the inverse analysis. Thirdly, conclusions
from the integrated water resources management model are given, which outline the
combination of physical and socio-economic impacts of changing (climatic and socio-
economic) conditions.
As the climate changes in southwestern Ontario, rainfall patterns will most certainly
change from what they have been historically. This part of the report attempted to
measure these changes, and to estimate impacts of changed climatic conditions on de-
sign, operation and maintenance of municipal water management infrastructure (such
as roads, bridges, culverts, drains, sewer and conveyance systems, etc). The results
are derived from applying rainfall intensity duration frequency curves developed for
the City of London under climate change. The outputs of the study indicate that:
• The rainfall magnitude (as well as intensity) will be different in the future (8
to 23% higher, depending on duration and return period).
• The wet climate scenario (recommended for use in future storm water man-
agement design standards) reveals a significant increase in rainfall magnitude
(and intensity) for a range of durations and return periods. For example, de-
sign of small storm drains (with times of concentrations less than 90 minutes,
294
of storm water management ponds, detention basins and other storm water
management facilities will require 10 to 23% increases in volume to accomodate
future flows under climate change (requiring detention between 6 and 24 hours,
with return periods between 2 and 100 years).
• The increase in rainfall intensity and magnitude has major implications for the
ways in which current (and future) municipal water management infrastructure
will be designed, operated, and maintained.
• The design standards and guidelines currently employed by the City of London
The main objective of this part of the project is to assess risk and vulnerability
of high and low flows to changing climatic conditions, through use of the inverse
approach. Of the many different accounts, vulnerability in this project is defined
private sector officials, as it gives them insight into what the changed climate may
bring.
A new approach for water resources risk assessment is outlined, and a number
of tools are developed, which can be used as aids in quantifying effects of changed
who will have to act and/or formulate new management guidelines in light of this
information. Too often, reports and assessments have been made without consulting
end users; as a result, the recommendations have rarely been implemented.
One of the benefits of the proposed inverse approach is the ease with which it can
be employed by an end user of the basin. Although illustrated with few examples (the
problem of over topping of dykes, and the occurrence of a level II drought), interested
readers can readily use the tables and charts provided to perform analyses of their
own, and thus estimate risks and vulnerability to climatic change for problems of
greatest interest to them. This is indeed encouraged.
Two of the climate scenarios considered show different possible futures in terms of
hydro-meteorologic conditions in the basin, but emphasis should be placed on the
wet scenario. The wet scenario is selected specifically to study the impact of flooding,
296
resulting from changed climatic conditions. With this information, it is believed that
better and more flexible management practices can emerge, which should ultimately
minimize the regions risk and vulnerability. As an example, the flow causing the
dykes in West London (i.e., Byron) to over-top is approximately 990 m3 /s, with a
return period of 33 years under the historic scenario (representing the chance of 1 in
33 any given year that a flow of 990 m3 /s will be equaled or exceeded). Under the wet
climate change scenario, the return period of dykes over-toping reduces to 17 years,
implying a significantly lower levels of safety offered by the dyke. Similarly, the flow
of 840 m3 /s represents a 100 year return period design event at the Fanshawe dam
under the historic climate scenario; under the wet climate change scenario, the return
period for this design event reduces to 50 years, similarly reducing the level of safety
offered by the dam and increasing the risk of flooding significantly.
ture, and increased maintenance costs of present and future infrastructure, can lead
to higher operating budgets), and (iii) engineering (where a review of current design
standards of critical infrastructure is recommended).
The impact of changed hydro-climatic conditions on current rules and regulations
shall be multi-fold. First, the Upper Thames River Conservation Authority (UTRCA)
should seriously consider reviewing its operating guidelines, especially those that are
sensitive to changes in flood magnitude and frequency; the UTRCA should particu-
larly consider the operation of its three reservoirs, use of river and its surrounding
park land use, patrolling and monitoring of critical infrastructure during periods of
high flow, and its guidelines of allowing permits for floodplain development. Second,
297
the increasing magnitude and frequency of flood flows are expected to have an impact
on operating budgets that authorities must secure for safe operation and maintenance
of its existing flood management works (including both structural and non-structural
measures). Third, a recommendation for a full review of engineering design standards
Similarly, the application of the inverse approach to low flow and droughts is consid-
ered here. Even though climate change scenarios show less frequent occurrence under
drought conditions (using the dry scenario and compared to the base case scenario),
they still occur often enough to cause serious concern. As an example, level II drought
for the West London (i.e., Byron) occurs when the flow in the river is 5 m3 /s or lower.
Under the historic scenario, level II drought occurs at a recurence interval of 6.3 years
for the historic scenario (meaning that there exists a chance of 1 in 6.3 any given year
that a flow of 5 m3 /s or less will exist in the river). Under the dry climate scenario,
the return interval for the 5 m3 /s flow remains unaltered, therefore implying that fu-
ture drought characteristics are likely to remain close to their current levels. Because
practices can emerge, which should minimize the region’s risk and vulnerability to
drought. Based on the analysis presented, recommendations are made for future work
which include the following: (i) drought impact studies (where impacts on agriculture,
recreation, wetlands, reservoir operation, ground water withdrawal and streamflow
quality are assessed), (ii) definitions of local drought triggers (including guidelines on
subwatershed scale, as well as monitoring how drought triggers can change over time),
(iii) water quality management (setting in place practises that enhance water quality
over the short and long term), and (iv) implementation of education programs (to
bring current scientific knowledge to all those who stand to be adversely impacted by
drought).
Since water resources problems can not be studied without the socio-economic con-
text in which they are embedded, this part of the report presents a framework able to
implement the idea of integrated water resources management. The model is general
enough to be able to test a wide range of policies and management strategies (like
changes in demographics, housing, jobs and labour force, land use practices, etc); the
299
model can also produce detailed hydrologic output on a daily time scale, which can
be used for impact analyses and/or a variety of engineering design purposes.
Hydrologic extremes (such as floods and droughts) are assessed using different
climate signals with a continuous hydrologic model component, while socio-economic
impacts are evaluated using a model component of the same name. Simulation scenar-
ios are used to investigate effects of changed climatic and socio-economic conditions
on overall basin behaviour.
After simulating a number of different climate change and socio-economic scenar-
ios and evaluating their impacts, a number of concluding remarks are made regarding
the possible future impact of physical and socio-economic well being of the Upper
Thames River basin. These conclusions relate to possible changes in hydrologic ex-
tremes (floods and droughts), together with changes in regional economic activity
(urban and rural population, business and farm units, land use development, water
availability and use, etc). The conclusions outline ways in which climatic change,
coupled with policies and management strategies have the potential to significantly
alter the physical and socio-economic landscape of the basin.
Climate change is expected to intensify flooding in the basin, thus bringing flows
of higher magnitudes with greater frequency. Such conditions may demand additional
investment in flood management infrastructure, and may require complete revision
of budgets for flood management and yearly maintenance. Such conditions may
even necessitate retrofitting and replacing current (or even building additional) flood
stability, severe droughts in the region are common; therefore this eventually requires
300
serious attention from water resources managers, stakeholders and users. This is of
particular importance, especially since drought impact assessments in the basin are
altogether lacking. Commissioning such studies forms one of the major recommen-
dations regarding future drought initiatives in the Upper Thames River basin. Such
initiatives are needed, because deteriorating water quality during summer periods
may become a serious concern to the health of both humans and ecosystems.
In addition to droughts, socio-economic characteristics of the Upper Thames River
basin have the potential to become significantly altered as a result of both climatic
recharge rates are significantly lower). Situations like this bring about conditions
of increased long term average water use, while at the same time reducing available
ground water availability. In other words, of all the factors postulated to place limits
on economic development (availability of water, land, labour force, jobs and housing),
the availability of water turned out to be the most important. Water is important
because residents and businesses in the basin use such staggering amount of water
that currently known sources are expected to become either exhausted or unavailable
within the next two to three decades. If such a time is reached, economic growth may
slow, or altogether stop.
sidered which implements a strict, water conservation policy with the aim of reducing
total water use and elevating regional socio-economic well being in the long term.
Simulation reveals that implementation of such a policy actually has negative long
term consequences on regional socio-economic behaviour. This counter intuitive find-
ing is nevertheless accurate, because reducing per capita (and per business/farm)
water use lowers total water use in the short term, while increasing overall water
availability (i.e., if less gets used, more becomes available). However, higher wa-
ter availability motivates the region to actually increase its economic base, therefore
making additional business investments more lucrative. Over time, such increased
economic activity eventuates in greater water use, eventually reaching use patterns
comparable to the base case scenario.
A possible way to keep total water use at acceptable levels may be to implement a
combination policy, where a reduction in per capita water use is required alongside a
policy that primarily limits urban expansion and development (or allows only smart
development). Such a case is tested in the simulation model and encouraging results
are obtained. Total water use can indeed be lowered, while keeping most forest and
agricultural land intact. Ground water recharge rates increase when compared to the
base case scenario, while expansion of existing residential and business lands occurs
at a modest rate, without negatively impacting the population.
The main findings revealed by simulating different scenarios include the following:
(i) climate change has the potential to significantly alter flooding characteristics of the
region by increasing risk levels of extremes (and their corresponding frequencies), (ii)
the frequency of extreme drought conditions is likely to remain at current levels and,
(iii) the most significant regional socio-economic factor is water availability, shown to
limit the growth of the population and the regional economy.
302
This section presents general conclusions regarding the implications of climate change
on water resources management. Although specific to the Upper Thames River basin
and the City of London, the conclusions are applicable to other basins and munici-
palities facing similar changed climatic conditions.
Results stemming from this research, together with growing awareness of climate
change and its potential impacts, have lead senior local water management officials
to accept the fact that the climate is changing. The major result of this realization is
that current water resources management strategies will have to be changed and/or
updated to fully account for the effects of changed conditions. The traditional ap-
proach, in which future scenarios are extrapolated from the historic record assuming
statistical stationarity, is no longer valid. All global circulation models are showing
that future warming is likely (for the air, earth and water); this warming will lead
to an intensified hydrologic cycle. Such intensification is expected to increase the
frequency of extreme events, which may alter the risks associated with floods and
droughts on basin and catchment scales. The risks associated with such changes may
not be uniform across all scales, and will probably be different for large and small
drainage areas. As an aside, flood risks for larger catchments in southwestern Ontario
are typically governed by snow melt induced floods, whereas extreme summer rainfall
actually happens, future conditions demand a risk based approach to planning that is
303
require a continual investment of funds to survey channel cross sections, and monitor
changes in sediment transport and erosion processes. Future flooding may occur from
possible combinations of weather events that have not been observed previously. For
example, frontal type storms may start to encroach on other seasons, thus allowing
for combination of heavy rainfall events on top of spring snow melt, or heavy rainfall
on frozen ground. In order to investigate events such as these (and their associated
flood risks), different techniques than are presently available are needed to capture
such conditions.
In the Upper Thames River basin, more detailed work is needed to determine the
design of future flood control structures (dams, dykes, channels, etc.). If the hydro-
climatic characteristics of the region are changing, the safety of existing structures
may be affected. Alternate operating protocols, guidelines and strategies will need to
be fully revised and/or updated. The same argument applies to recreational structures
currently in operation.
Relatively little work has been done on the impact of ground water resources and
its associated future planning. Ground water characteristics are not easily under-
stood, as limited data sets are available for the region. The lack of knowledge is
further exacerbated by the lack of public interest in the matter (i.e., it is perceived
as “out of site, out of mind”). Studying the interactions of surface and ground wa-
ter is extremely complex, especially since the two do not share the same physical
boundaries. Current provincial source water protection legislature is closing some of
the data gaps, but it is not addressing how ground water supplies may change in
304
More sophisticated modeling tools need to be developed for the study of low flow and
droughts, and especially those combining effects of surface and ground water. When
such studies have been conducted, true climate change impacts will be ascertained.
The implications of changed climatic conditions on water resources planning and
be revised. As a result of the research presented in this report (on changes in rainfall
IDF values), the staff of the City of London Engineer’s Office have recommended that
two different climate change adaptation strategies be developed and implemented.
The first strategy includes a transitional adaptation strategy to be completed during
– Reviews and updates IDF curves currently used by the City for design of
municipal infrastructure;
• Reviews (and discusses) with various levels of municipal, provincial and federal
agencies, regarding such increases as adaptation strategies;
Long term climate change strategies may include a radically different approaches
to storm water management. Changes to storm water modeling and rainfall IDF
values will require updates to subwatershed studies, thus leading to new and more
efficient storm water management policies. Consideration of a green infrastructure
plan, to minimize storm water conveyance and detention (thus leading to a reduction
in the number of storm water facilities and land requirements), is one such strat-
egy. Policy initiatives like these will rely on the natural features of the landscape to
store and retain storm water, instead of forcing it to run off to pipes, storm water
management ponds and/or watercourses. Implementing green strategies like these
will undoubtedly improve the management of the City’s water resources systems, and
ensure that the public, property and watercourses are adequately protected.
306
In the last section of this report, possible extensions and future research directions
stemming from this work are discussed. Theoretical and application extensions are
therefore covered.
The first model developed in this study is the spacio-temporal storm model, used as
alternative to current procedures of generating rainfall patterns for basin scales. This
the results of the model to actual storms (during both summer and winter seasons)
will increase the model’s overall applicability. This could be achieved by comparing
storms captured with point rainfall, radar, and/or satellite data, to those constructed
with the spacio-temporal model. This type of analysis is necessary, before enough
require rainfall with small time steps (i.e., of order of minutes). Further extension
and refinement can be achieved by comparing Probable Maximum Precipitation and
the spacio-temporal model. It is speculated that the spacio-temporal model is able
to estimate rainfall fields of comparable quality to Probable Maximum Precipitation,
307
The K-Nearest Neighbour weather generator is the current state of the art model in
statistical weather data generation. The K-NN weather generator is able to reproduce
both spacial and temporal characteristics of observed weather records that have been
applied in impact assessment studies. Despite their popularity and wide applicability,
some limitations still remain. The application of weather generators requires use of
neighbouring stations, which are used to generate data and ensure reproduction of
observed spacial and temporal characteristics. The process of selecting stations for
use in a weather generator algorithm is still not clear. Research should be undertaken
to assess the sensitivity of data selection, and to establish a clear set of criteria and
guidelines that could be used by future users. Such clarification is needed, because
having either too many (or too few) stations influences simulation results (through
the calculation of regional means used to select sets of closest neighbours). Clustering
and other statistical analyses should be used to aid in the process of data selection.
Another enhancement suggested for weather generators involves developing mean-
ingful, locally applicable climate scenarios. The methods applied in this study repre-
sent only one set among many. Biasing the observed record for wetter, drier, cooler or
other climates is one way of developing scenarios (Yates et al., 2003), while consider-
ing El-Niño and El-Niña cycles represents another (Clark et al., 2004). Other means
308
with which to generate locally applicable and meaningful climate scenarios should be
further investigated.
Since weather generators are becoming increasingly popular, assessment studies
should be undertaken to investigate how generated weather data is used, and for
what purposes. Examples of applications in this work include those for basin-wide
continuous hydrologic modeling and local extreme rainfall analysis. Weather genera-
tor use for other applications should be summarized, limitations noted, and research
recommended to further develop and enhance the available tools.
One of the current limitations of weather generator algorithms is that they are
applied to daily data. Development and application of weather generators able to
produce similar data for finer temporal time scales should be undertaken. Currently,
the only way to proceed is by employing disaggregation algorithms.
The disaggregation algorithm used in this report draws on the success of the non-
parametric K-Nearest Neighbour weather generator algorithm. The algorithm’s main
advantage is that parametrization is not required, thus extending the algorithm to
any region of the world with few adjustments. However, the disaggregation algorithm
has its downside. Currently, daily rainfall is disaggregated by using shorter duration
rainfall at the same station, thus not taking into account region’s (disaggregated)
inter-station characteristics. It is recommended that research be initiated to inves-
tigate ways of preserving inter-station characteristics in the disaggregation process.
Some research in this direction has been undertaken by Wey (2006), but more is
needed. A suggested approach could include the use of Mahalanobis distance, similar
to the K-Nearest Neighbour weather generator algorithm.
An output of the disaggregation analysis is used in this study for the construc-
tion of rainfall IDF curves. Such curves are used as a basis for engineering design
standards, and are of interest to municipal and provincial governments (as these in-
309
stitutions are responsible for setting design standards). Even though IDF analysis is
relatively straightforward, it contains a number of drawbacks that should be further
explored. In the analysis of extreme rainfall for use in engineering design, frequencies
of short-duration rainfall are required (at time steps as small as five minutes). De-
tailed records for such small duration often do not exist; or if they exist, have large
portions of record missing. In IDF analysis, estimates of rainfall for durations of less
than one hour are based on peak hourly values; these estimates may not represent
observed conditions. Other methods of estimating short-duration rainfall frequencies
should be investigated.
The inverse analysis presented in this report is a general methodology used for analyz-
ing the risk and vulnerability of water resources systems. The methodology depends
on the ability of weather generators and hydrologic models to reasonably reproduce
agencies have customized operating rules and procedures that are not always easy to
capture with methods available in hydrologic models. It is recommended that further
enhancements in this area be pursued, for both event and continuous models.
The way the continuous model currently captures the process of evapotranspi-
ration is quite rudimentary (it uses monthly average evaporation values, multiplied
by a pan coefficient to obtain evapotranspiration). This problem is further compli-
cated by land use and climate changes. The same values of evapotranspiration are
310
currently used, regardless which climate and/or land use change scenario is being ex-
amined. Refinements in the computational engine of the continuous hydrologic model
could include such features, and therefore produce a model able to capture hydrologic
conditions more accurately.
When it comes to the application of the continuous hydrologic model to the Up-
per Thames River basin, additional work should be done on the model’s seasonal
parameter sets. The model is currently used in such a way that an abrupt change
of parameters is made between summer and winter conditions; in future versions of
the model, the change need not be so abrupt, but may be a function of temperature.
Such a refinement may ensure that high flood peaks during the first storm of the
winter season may not appear so pronounced. Lastly, since other continuous hydro-
logic models (i.e., Hydrologic Simulation Program FORTRAN, HSPF) exist in source
code, they should be investigated, as they might be applicable to both high and low
flows, thus eliminating the need for two different models (event and continuous).
clients will learn how to embed and test alternate assumptions, model structures, and
test parametric and other sensitivities. Doing this will provide clients with a clear
understanding of the limitations of the model, as well its powerful features. In a sense,
the goal is to have clients “own” part of the model. Second, more elaborate models
should be built, based on previous work and gaps that have been filled in regards to
model structures, functional relationships, and parameters.
In this study, simple models were never developed, nor have clients ever tested the
311
explore relationships between variables omitted in this study (water quality issues),
as well as engaging in extensive interviews that attempt to capture the rules that
decision makers use to advance (and change) current management practice. Group
model building exercises could reveal a number of relationships that were assumed
in the current study, and on which the current socio-economic model structure was
built.
Following such changes to model structure, questions regarding excluded model
sectors could be addressed. For example, the current model assumes that water
quality plays no role in long term decision making, despite the fact that water quality
is on the mind of all major decision makers and stakeholders. For the purposes of this
study, water quality characteristics were not considered, but could easily be embedded
in the model as an additional sector. Other features neglected in the current model
could also be included. With more information, alternate model structures could
conditions.
The application of the integrated model to other regions is possible, and indeed
encouraged. As different regions have different hydro-climatic characteristics, the
hydrologic component of the model will therefore require revision. New hydrologic
component will have to be assembled, and properly calibrated and verified. Note
that methods adopted in this study to quantify hydrologic processes apply for other
regions, with the only difference that different model parameters will have to be speci-
fied. Furthermore, different regions also have different socio-economic characteristics,
that, unlike the hydrologic component, will require the development of new feedback
type systemic structures. New systemic structures are needed as different regions have
different socio-economic characteristics with respect to economic drivers, urbaniza-
tion and development, water supply and use practices, cultural values, environmental
sensitivity, etc. Regardless of the region the model is being considered, basic physical
(hydro-climatic) and socio-economic (census) data need to be collected, synthesized,
and analyzed. Based on all available data, feedback type systemic structures need
to be proposed, and verified with regional stakeholders before simulation models are
first proposed.
It is important to mention that surface water drawn from the Great Lakes is not
part of the continuous hydrologic model, only the ground water (and its accompanying
recharge) is. One of the reasons for this ommission is because the effect of surface
water pumped from the Great Lakes for human consumption (and eventually returned
to the lakes) is simply too small to contribute to the study of extreme flood conditions.
During spring and summer flooding, the flood water volumes are significantly higher
than flows resulting from water originally extracted from the Great Lakes. During low
flow and drought conditions however, it is recognized that a portion of water volume
in the area streams consist of treated sewage effluent (a fraction of which has been
313
originally pumped, and is now on its way back to the Great Lakes). Since one of the
purposes of this work is to assess changes in regional low flow extremes in response
to climate change, omission of water originally pumped from the Great Lakes is seen
as conservative. In other words, the inclusion of this water would increase steamflow
volumes and lessen the effects of droughts. Addition of the water originally pumped
from the Great Lakes to the continuous hydrologic model is recommended as a future
model enhancement.
314
References
Ahmad, S. and Simonovic, S. P. (2000). “System dynamics modeling of reservoir op-
erations for flood management.” ASCE Journal of Computing in Civil Engineering,
14(3), 190–198.
Argyris, C., Putnam, R., and Smith, D. M. (1985). Action Science: Concepts, Meth-
ods and Skills for Research and Intervention. Jossey-Bass Publishers, San Francisco,
California.
Bicknell, B. R., Imhoff, J. C., Kittle, J. L., Jobes, T. H., and Donigian, A. S. (2001).
Hydrological Simulation Program-FORTRAN (HSPF), User’s manual for version
12. Aqua Terra Consultunts, Mountain View, California.
Bruce, J., Burton, I., Martin, H., Mills, B., and Mortsch, L. (2000). Water Sector:
Vulnerability and Adaptation to Climate Change. Global Strategies International
Inc., and Meteorological Service of Canada, Ottawa, Canada.
Brutsaert, W. (1982). Evaporation into the Atmosphere: theory, history and applica-
tions. Kluwer Academic Publishers, Dordrecht, Netherlands.
Cai, X., McKinney, D. C., and Lasdon, L. S. (2002). “A framework for sustainability
analysis in water resources management and application to the syr darya basin.”
Water Resources Research, 38(6), 21.1–21.14.
315
Cheng, C. S., Li, G., Li, Q., Auld, H., and MacIver, D. (2007). “Climate change and
extreme rainfall-related flooding and surface runoff risks in Ontario—plain lan-
guage summary report.” Report No. A901, Environment Canada, Climate Change
Impacts and Adaptation Program, Toronto, Ontario, Canada.
City of London (2003). Sewer Design Standards for the City of London. Environmen-
tal and Engineering Services Department, City of London, Ontario, Canada.
Clark, M. P., Gangopadhyay, S., Brandon, D., Werner, K., Hay, L., Rajagopalan, B.,
and Yates, D. (2004). “A resampling procedure for generating conditioned daily
weather sequences.” Water Resources Research, 40(W04304), 1–15.
Cole, H. S. D., Freeman, C., Jahoda, M., and Pavitt, K. L. R. (1973). Models of
Doom: A Critique of the Limits to Growth. Universe Books, New York.
Connolly, R. D., Schirmer, J., and Dunn, P. K. (1998). “A daily rainfall disaggregation
model.” Agricultural and Forest Meteorology, 92, 105–117.
Conway, D., Wilby, R. L., and Jones, P. D. (1996). “Precipitation and air flow indices
over the British Ises.” Climate Research, 7, 169–183.
Fill, H. D. and Steiner, A. A. (2003). “Estimating instantaneous peak flow from mean
daily flow data.” ASCE Journal of Hydrologic Engineering, 8(6), 365–369.
Fleming, M. and Neary, V. (2004). “Continuous hydrolgoc modeling study with the
hydrologic modeling system.” Journal of Hydrologic Engineering, 9(3).
Giorgi, F., Hewitson, B., Christensen, C., Fu, R., and Jones, R. (2001). Regional
Climate Information - Evaluation and Projections. Climate Change 2001: The Sci-
entific Basis. Contribution of Working Group I to the Third Assessment Report of
the Intergovernmental Panel on Climate Change, 585–683. Cambridge University
Press, Cambridge, United Kingdom.
GWP (2003). Global Water Partnership Toolbox: Sharing knowledge for eq-
uitable, efficient and sustainable water resources management. Global Water
Partnership Secretariat, Stockholm, Sweeden. Last accessed: 14 April 2007,
http://www.gwpforum.org.
GWP (2008). Climate Change Adaptation and Integrated Water Resources Manage-
ment - An Initial Overview. Policy Brief 5, Technical Commitee of the Global
Water Partnership Secretariat, Stockholm, Sweeden. Last accessed: 26 Jan 2008,
http://www.gwpforum.org.
Interagency Committee on Water Data (1981). Guidelines for determining flood flow
frequency. Bulletin 17B, U.S. Geological Survey, Reston, Virginia.
IPCC (2001). Climate Change 2001: Scientific Basis. Contribution of the Working
Group I to the Third Assessment Report of the Intergovernmental Panel on Climate
Change. Cambridge University Press, Cambridge, United Kingdom.
IPCC (2007). Climate Change 2007: Synthesis Report. Fourth Assessment Report
of the Intergovernmental Panel on Climate Change. Cambridge University Press,
Cambridge, United Kingdom.
Jackson, M. C. (2003). Systems Thinking: Creative Holism for Managers. John Wiley
& Sons, Ltd, Chichster, United Kingdom.
Keifer, C. J. and Chu, H. H. (1957). “Synthetic storm pattern for drainage design.”
ASCE Journal of the Hydraulics Division, 83(HY4), 1–25.
Kell, R. (2004). Middlesex Elgin Groundwater Study Final Report. Dillon Consulting,
London, Ontario, Canada.
Kundzewicz, Z. W., Mata, L. J., Arnell, N. W., Doll, P., Kabat, P., Jimenez, B.,
Miller, K. A., Sen, T. O. Z., and Shiklomanov, I. A. (2007). Freshwater resources
and their management. Climate Change 2007: Impacts, Adaptation and Vulnera-
bility. Contribution of Working Group II to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change, 173–210. Cambridge University Press,
Cambridge, United Kingdom.
Lall, U. and Sharma, A. (1996). “A nearest neighbour bootstrap for time series
resampling.” Water Resources Research, 32(3), 679–693.
Li, L. and Simonovic, S. P. (2002). “System dynamics model for predicting floods
from snowmelt in North American prairie watersheds.” Hydrological Processes, 16,
2645–2666.
Linsley, R. K., Kohler, M. A., and Paulhus, J. L. H. (1958). Hydrology for Engineers.
McGraw-Hill Book Company, Inc., New York.
Loucks, D. P. and van Beek, E. (2005). Water Resources Systems Planning and
Management: An Introduction to Methods, Models and Applications. United Na-
tions Educational, Scientific and Cultural Organization, Paris, France and Delft,
Netherlands.
Maass, A., Hufschmidt, M. N., Dorfman, R., Thomas, H. A., Marglin, S. A., and
Fair, G. M. (1962). Design of Water-Resources Systems. Harvard University Press,
Cambridge, Massachusetts.
Markus, M., Angel, J. R., Yang, L., and Hejazi, M. I. (2007). “Changing estimates of
design precipitation in northeastern illinois: Comparison between different sources
and sensitivity analysis.” Journal of Hydrology, accepted for publication.
Meadows, D. H., Meadows, D. L., Randers, J., and Behrens, W. M. (1973). A response
to Sussex. In Models of Doom: A Critique of the Limits to Growth, Universe Books,
New York.
Meadows, D. H., Randers, J., and Meadows, D. L. (1972). The Limits to Growth:
A Report for the Club of Rome’s project on the predicament of Mankind. Universe
Books Publishers, New York.
Meadows, D. H., Randers, J., and Meadows, D. L. (2004). Limits to Growth: The 30
year update. Chelsea Green Publishing Company, Vermont.
Meadows, D. H., Richardson, J., and Bruckman, G. (1982). Groping in the dark: the
first decade of global modelling. John Wiley & Sons Ltd., New York.
Mearns, L., Hulme, M., Carter, T., Leemans, R., Lai, M., and Whetton, P. (2001).
Climate Scenario Development. Climate Change 2001: The Scientific Basis. Con-
tribution of Working Group I to the Third Assessment Report of the Intergovern-
mental Panel on Climate Change, 739–769. Cambridge University Press, Cam-
bridge, United Kingdom.
Mehdi, B., Mrena, C., and Douglas, A. (2006). Adapting to climate change: An intro-
duction for Canadian Municipalities. Canadian Climate Impacts and Adaptation
Research Network (C-CIARN), http://www.c-ciarn.ca/.
Merry, A. (2003). Perth County Groundwater Study Final Report. Waterloo Hydro-
geologic, Inc., Waterloo, Ontario, Canada.
MMAH (2002). London Regional Analysis. Planning Policy Section, Provincial Plan-
ning and Environmental Services Branch, Ministry of Municipal Affairs and Hous-
ing, London, Ontario, Canada.
NWS (1978). Probable Maximum Precipitation Estimates, United States East of the
105th Meridian. NOAA Hydrometeorological Report No. 51, National Weather
Service, US Department of Commerce, Washington, District of Columbia.
OLWR (2003). Ontario Low Water Responce Manual. Ontario Ministry of Natural
Resources, Queen’s Printer for Ontario.
Palmer, R. N., Clancy, E., VanRheenen, N. T., and Wiley, M. W. (2004). The Impacts
of Climate Change on The Tualatin River Basin Water Supply: An Investigation
into Projected Hydrologic and Management Impacts. Department of Civil and En-
vironmental Engineering, University of Washington, Seattle, Washington.
Postel, S. and Richter, B. (2003). Rivers for life: Managing water for people and
nature. Island Press, New York.
Prodanovic, P. and Simonovic, S. P. (2006b). “Inverse flood risk modelling of the Up-
per Thames River basin.” Report No. 052, Facility for Intelligent Decision Support,
Department of Civil and Environmental Engineering, The University of Western
Ontario, London, Ontario, Canada.
SCS (1986). “Urban hydrology for small watersheds.” Report No. 55, Engineering
Division, Soil Conservation Service, US Department of Agriculture, Washington,
District of Columbia.
Senge, P. M. (1990). The Fifth Discipline: The Art & Practise of the Learning
Organization. Doubleday Currency Press, New York.
Senge, P. M., Kleiner, A., Robers, C., Ross, R. B., and Smith, B. J. (1994). The
Fifth Discipline Fieldbook: Strategies and tools for building a learning organization.
Doubleday Currency Press, New York.
Sharif, M. and Burn, D. H. (2004). “Assessment of water resources risk and vulner-
ability to changing climatic conditions: Development and application of a K-NN
weather generating model.” Report No. III, Department of Civil and Environmental
Engineering, The University of Western Ontario, London, Ontario, Canada.
Sharma, A., Tarboton, D. G., and Lall, U. (1997). “Streamflow simulation: A non-
parametric approach.” Water Resources Research, 33(2), 291–308.
Simonovic, S. P. and Fahmy, H. (1999). “A new modling approach for water resources
policy analysis.” Water Resources Research, 35(1), 295–304.
Simonovic, S. P., Fahmy, H., and El-Shorbagy, A. (1997). “The use of object-oriented
modeling for water resources planning in Egypt.” Water Resources Management,
11, 243–261.
Sivakumar, B., Sorooshian, S., Gupta, H. V., and Gao, X. (2001). “A chaotic approach
to rainfall disaggregation.” Water Resources Research, 37(1), 61–72.
Snyder, P. K., Delire, D., and Foley, J. A. (2004). “Evaluating the influence of different
vegitation biomes on the global climate.” Climate Dynamics, 23, 279–302.
Soong, D. T., Straub, T. D., and Murphy, E. A. (2005). “Continuous hydrologic sim-
ulation and flood-frequency, hydraulic, and flood-hazard analysis of the Blackberry
Creek Watershed, Kane County, Illinois.” Report No. 2005-5270, United States
Geological Survey, Reston, Virginia.
Southam, C. F., Mills, B. N., Moulton, R. J., and Brown, D. W. (1999). “The
potential impact of climate change in Ontario’s Grand River basin: Water supply
and demand issues.” Canadian Water Resources Journal, 24(4), 307–330.
USACE (2006). Hydrologic Modelling System HEC–HMS, User’s Manual for version
3.0.1. United States Army Corps of Engineers, Hydrologic Engineering Center,
Davis, California.
UTRCA (2001). The Upper Thames River Watershed: Report Cards 2001. Upper
Thames River Conservation Authority, London, Ontario, Canada.
Veneziano, D. and Villani, P. (1999). “Best linear unbiased design hyetograph.” Water
Resources Research, 35(9), 2725–2738.
Vincent, L. A. and Mekis, E. (2006). “Changes in daily and extreme temperature and
precipitation indices for Canada over the thentieth century.” Atmosphere-Ocean,
44(2), 177–193.
Waters, D., Watt, W. E., Marsalek, J., and Anderson, B. C. (2003). “Adaptation of
a storm drainage system to accommodate increased rainfall resulting from climate
change.” Journal of Environmental Planning and Management, 46(5), 755–770.
Watt, W. E., Chow, K. C. A., and Lathem, K. W. (1986). “A 1-h urban design storm
for Canada.” Canadian Journal of Civil Engineering, 13, 293–300.
Watt, W. E., Lathem, K. W., Neill, C. R., Richards, T. L., and Rousselle, J. (1989).
Hydrology of Floods in Canada: A Guide to Planning and Design. National Re-
search of Canada, Ottawa, Ontario, Canada.
326
Watt, W. E., Waters, D., and McLean, R. (2003). “Climate change and urban
stormwater infrastructure in Canada: Context and case studies.” Report No. 2003-
1, Toronto-Niagara Region Study Report and Working Paper Series, Meteorological
Service of Canada, Waterloo, Ontario, Canada.
Wilcox, I., Quinlan, C., Rogers, C., Troughton, M., McCallum, I., Quenneville, A.,
Heagy, E., and Dool, D. (1998). The Thames River Watershed: A Background
Study for Nomination under the Canadian Heritage Rivers System. Upper Thames
River Conservation Authority, London, Ontario, Canada.
WMO (1981). “Selection of distribution types for extreme precipitation.” Report No.
WMO-No.560, World Meteorological Organization, Operational Hydrology Report
No. 15, Geneva, Switzerland.
WMO (1994). “Guide to hydrological practice: data acquisition and processing, anal-
ysis, forecasting and other applications.” Report No. WMO-No.168, World Meteo-
rological Organization, Geneva, Switzerland.
Wood, M. and Goldt, R. (2004). Reference Manual for the Use of Precipitation Design
Events in The Upper Thames River Watershed. Upper Thames River Conservation
Authority, London, Ontario, Canada.
Yates, D., Gangopadhyay, S., Rajagopalan, B., and Strzepek, K. (2003). “A tech-
nique for generating regional climate scenarios using a nearest-neighbor algorithm.”
Water Resources Research, 39(7), SWC7.1–SWC7.15.
Yen, B. C. and Chow, V. T. (1980). “Design hyetographs for small drainage struc-
tures.” ASCE Journal of the hydraulics division, 106(HY6), 1055–1076.
328
Appendix
This manual is meant to provide technical details necessary for reproducing results
developed in this study. Therefore model descriptions, justifications, methodologies
and techniques are not provided here; these are covered in the main body of the
report, which this appendix supplements.
The computer code needed to simulate the models developed in this work is not
listed, but is rather provided on the accompanying DVD. All models are developed in
the Java programming language, and are thus able to run on any platform (Windows,
Mac, Unix/Linux). Extensive documentation of the code is provided for each source
file in the Java Documentation Standards (javadoc), thus providing the user with
comprehensive documentation for all models developed in this work. The reader
is referred to Sun’s website http://java.sun.com for latest versions of the Java
compiler, documentation syntax, as well as installation instructions.
The Java code was written in a program called JEdit, a simple and powerful Java
text editor. It is available from http://www.jedit.org/ free of charge. The code
written in this research utilizes many of JEdit’s features (such as syntax highlighting,
code folding etc), and is therefore recommended for viewing, editing and modifying
the source code of the models. However, since *.java files are in essence text files,
they may be viewed and edited with any available text editor program.
In Java, the file that links all components of the project is referred to as the main
file. All project files (referred to as objects or classes), model parameters, as well
as all input and output files/directories are specified within this file. The main file
is compiled, and then executed to simulate the model. Following the execution of
the main file, all model outputs (simulation results and its post processing analysis)
are stored as text files in appropriate output directories (see in line javadoc code
documentation for further details).
If the main file within a project is called mainProject.java, typing the following
at the command prompt will compile the main file:
javac mainProject.java
The javac compiler is referred to as the Java Compiler, and creates binary exe-
cutable files of *.class type for the entire project. The execution of the project is
accomplished by typing the following at the command prompt:
java mainProject
where the program java is used to execute the compiled main file. All models devel-
oped in this work are compiled and executed using the above two commands.
The models are organized in separate directories (their numbering corresponds
with numbering of this Appendix). Therefore, each model has its own directory,
where all its input files are located. All data and simulation results are likewise
placed in appropriate files/directories, and are available to the interested reader for
further investigation. Note that the code written contains extensive documentation
and commenting for every object, class and method according to the Java Documen-
329
tation Standard. Descriptions provided in the appendix are given only at a general
level; description of programming and other implementation details are provided in
model documentation, which the interested reader is encouraged to investigate fur-
ther. The embedded documentation within each file is completed according to the
Java Documentation Standard, and is readily available to the programmer compiling
the code using the javadoc compiler.
330
• DataWriter.java. Similar to above, with the only exception that data are
written externally instead being read as input.
• WGUtils.java. This is the class that contains a collection of utilities used to pre
and post process inputs and outputs required by the weather generator model.
All methods contained in this class have a short description, followed by its
inputs and outputs.
• mainWG.java. This is the main file of the weather generator algorithm. This
file links the above files, as well as specifies input data (daily observed data),
outputs (synthetic daily data) and parameters (length of temporal window)
used. This is the file that should be modified if the model is to be applied for
different locations, or if other stations should be included.
• mainDissagregate.java. This is the main file that takes the synthetic data
generated by the daily weather generator, and disaggregates it into hourly rain-
fall. The inputs are the daily stations to be disaggregated, together with hourly
332
observed record. The outputs of the program are hourly disaggregated time
series data for each station considered.
• mainAnalysis.java. This is the main file that performs the statistical analyses
on the generated data. Inputs are time series data (hourly and/or daily), and
outputs are various statistical summaries of time series data.
Before the above model files can be compiled, directories containing data sets
need to assembled. All input data is stored in the directory labelled Data, for hourly
and daily time steps. Both hourly and daily sub directories are further divided into
directories containing data that is formatted and temporally interpolated (i.e., put
through an algorithm to estimate missing values). The formatted and interpolated
data is then used for the construction of scenarios, which are obtained by taking the
data and applying change fields from global circulation model outputs. The scenarios
are labelled as B11 (dry), B21 (wet) and Historic, and contain daily rainfall data used
as inputs to the weather generator model.
Simulation results are stored in a directory labelled Output, which are further
divided into hourly and daily durations. Model results are generated using stations
within 200, 100, and 0 km of London, for B11 (dry), B21 (wet), and Historic climate
scenarios. (The case using stations within 0 km of London is one that uses only
London data for the simulation.) Each combination of results is found in the provided
directories.
333
• mainAnalysis.java. This main file is used for pre and post processing of
simulation model results. The pre processing includes various data formatting
methods, while post processing includes performing flood flow frequency anal-
ysis of peak flows obtained through simulation.
Similar to the weather generator model, all input data used by the model is stored
in a directory called Data, where rainfall data for the same three simulation scenarios
are found B11 (dry), B21 (wet) and Historic. The model output data is likewise
stored in a directory labelled Output, and contains detailed simulation model results,
as well the post processed statistical summaries.
335
• Clark.java. This is the method that convolutes a unit hydrograph, and thus
computes direct runoff of a subcatchment from surface excess and catchment
time-area map as input.
• contUtils.java. This class is a utilities class that contains all pre and post
processing of results. It contains a similar set of methods and algorithms to one
used by the event model. It is important to note that the snow accumulation
and melt algorithm is found in this class.
• DataReader.java. This is the class that is used to read data from text files
into the simulation model.
E Integrated model
The integrated model files and directories are identical to those found in the con-
tinuous hydrologic model. This is because the continuous model described above is
one of the major components of the integrated model. Therefore, all file names de-
scribed above are used, and so will not be repeated here. The only difference is that
additional code in HydModel object is used (see comments within the file for details)
to make the hydrologic model component depend on the socio-economic component.
The files part of the integrated model (in addition to all continuous model files) are:
• SysModel.java. This is the file that contains all sectors of the socio-economic
model component. All model parameters, and assumed relationship are found
within this file.
• mainIntegrated.java. This is the main file of the integrated model. As before,
this is the file that lists the locations of all inputs, outputs, and analyses to be
performed. It consists of a combination of SysModel and HydModel objects,
linked dynamically via feedback.
Base
InfWat
Hist RedWat
RedWatLimLand
SwitchToSW
Base
InfWat
Scenario Dry RedWat
RedWatLimLand
SwitchToSW
Base
InfWat
Wet RedWat
RedWatLimLand
SwitchToSW
Integrated model results are available for all combinations of climatic and socio-
economic scenarios, for three spacial units (Middlesex, Oxford and Perth Coun-
ties). They are found in a directory labelled Results, for all combination scenarios.
340
Curriculum Vitae