Você está na página 1de 9

c 


 
    
Regardless of criteria, landslides can be considered one of the most costly natural hazards. They do
enormous damage to infrastructure, water and electricity networks and there exists the ever-
present possibility to cause loss of life. Whilst not the focus of this essay, it is worth noting that the
causes of landslides are exceptionally complex and incorporate geological, geomorphologic and
meteorological processes (Sarkar and Kanungo, 2004). Unlike some hazard events landslides are
world-wide phenomena and as such the approaches to mapping them vary enormously. Mapping
landslides is very useful in managing and mitigating the impacts of landslides (Anbalagan et al, 1993
and Kienholz, 1978).

GIS is used extensively to map landslide events, it is possible to identify three main approaches to
this. These broad categories tend to have differing focuses and work best at a specific scale or
location. The first of these approaches is known as heuristic mapping. Heuristics comes from the
Greek ȵºÑ , meaning to find or discover, in the context of GIS it refers to an experience or
knowledge based methodology. Heuristic approaches can take two forms, either direct or indirect
mapping. Direct mapping involves the production of a field map based on observations conducted
by a professional. The study area is divided into polygons based on geological characteristics. These
polygons can then be assessed by a specialist. The results of this analysis usually take the form of a
series of qualitative statements e.g., highly susceptible to landslides, somewhat susceptible to
landslides etcetera. These results are mapped using GIS to give the final output. The second form of
heuristic mapping is known as indirect mapping. In this an expert assesses the likelihood of a
landslide based on existing data. Of all methods studied this is the most subjective.

The second method identified from literature is known as deterministic mapping. This method
involves the use of detailed geological data and physical characteristics as well as hydrological data
for the study area. This data is placed within a modelling environment, commonly a dedicated piece
of software, to calculate values based on slope stability models (Vanackera et al, 2003). The most
widely used slope model is known as the ͞infinite slope model͟.
ü  ʹ Framework of a typical deterministic landslide mapping project (Dahal et al, 2008)

These stability thresholds can also take into account rainfall data and thus provide a much more
realistic and detailed mapping method than the heuristic type. Another key improvement on the
previous method is that likelihood is expressed in probabilities rather than qualitative terms (Varnes,
1984). An example of this is shown in figure 2 over leaf.




ü  ʹ Slope failure thresholds based on 2 year rain averages. (Dahar et al, 2008).

The final method of mapping is a statistical approach. The starting point for this method involves
locating the position of past landslide events, these are then referenced in relation to a whole host
of variables. These variables incorporate any number of geological and geomorphic features but can
generally be assigned into a typology of terrain, soil and land cover. Each of these is mapped as a
separate layer within the GIS and the magnitude of these at the landslide locations are noted. These
layers are then merged into a composite which aims to explain the presence of landslides based on
the various input layers (e.g., Siddle et al, 1991; Atkinson and Massari, 1998).
ü  ʹ Layers used in the statistical methodology (Chau et al, 2004, Fig. 15)

T1 ʹ Elevation
T2 ʹ Slope
E1 ʹ Rainfall
L1 ʹ Surface Run off
L2 ʹ Landslide Inventory
G1 ʹ Soil Deposition
G2 ʹ Geology

Each of these ways of producing landslide susceptibility maps have their relative merits and
associated negatives. Heuristic models take the fullest advantage of expert knowledge and
consultations in their predictions. However this means they are the most subjective and it is not
possible to apply the work of one area or analyst to another. Deterministic models take advantage
of improvements in computing power to utilise increasingly complex mathematical models of slope
stability. It is widely accepted that they provide the most accurate and most useful landslide
susceptibility maps, however the process is reliant on the availability of high resolution data for a
number of variables. The effect of using lower quality data is a vastly inferior accuracy. Statistical
models are, at least theoretically, the fastest and cheapest to produce of these three methods. They
can also operate over much larger scales since they do not necessarily require field work. There are
negative aspects however; firstly there is a need for an enormous amount of data covering a broad
range of variables and all needing to be compatible. Secondly there is an inherent assumption that
is often overlooked or simply discarded. The location of prior landslides are used to determine the
locations of future landslides (Chau et al, 2003), thus it is assumed the causes of previous landslides
are the same as the causes of future ones. This may not always be the case; the features of one
location cannot be applied universally to find others. Perhaps because it is perceived as the easiest
option, the statistical method outlined above is the most widely adopted.

The field of landslide mapping has seen widespread adoption of GIS methods. The interest in using
GIS in this area is perhaps not surprising given the many positive impacts it can have. GIS allows the
handling of large amounts of data that simply is not possible with paper mapping. This enlarging of
datasets has been a threefold process, an increase in the number of variables studied; the resolution
they are studied at, as well as the size of the study area. Traditional modelling focussed singularly on
fieldwork for data collection and could only cover a very small area. As well as increases in the size
of the study area, the speed of the process is also vastly improved. Data collection/collation is still
the most time consuming phase of most GIS projects but as with many of the aspects, this is
improving all the time. Whilst access to the specialist datasets required is relatively easy in the
western world, it is considerably more difficult in developing countries. There are also issues of
compatibility arising from using a multitude of data from various different sources. Once compiled
into a GIS the analysis itself is comparatively rapid.

A further benefit to using GIS is the ability to apply models to various locales with little need to
change them. Whereas with traditional approaches there was a need for site-specific field work to
determine landslide likelihood, GIS models can simply be re-run using new parameters. There is
clearly some loss of accuracy in using a generalised model however for larger scale regional
applications this is minimal.

There are still a large number of issues both in the methodology and in the deployment of GIS as a
solution to landslide mapping. As was briefly mentioned, data collection is a vital aspect of any GIS
application and this is certainly the case with landslide susceptibility mapping. Whilst most often
attributed to deterministic modelling approaches, quality of data is an issue that plagues GIS
applications more generally. With landslides being commonplace across the world the issue of data
availability is faced by both richer and poorer nations alike. The statistical method requires as much
data as possible for the study area, the exact number of layers varies study to study but it is not
uncommon to see as many as thirty or forty being used. It is not likely these datasets are held by a
common body and as such various intuitions and organisations need to be approached in order to
acquire them. Where data is unavailable extensive field work is required before the GIS can be used;
effectively nullifying a key benefit. Even where data is available there is no guarantee of its
suitability. One of the major issues facing GIS is the Multiple Areal Unit Problem or MAUP, whereby
data at one scale is used to infer outcome at another scale. If data is of a lower or higher resolution
than its application it is either interpolated or extrapolated; this can lead to a significant reduction in
accuracy. Additionally by using a bivariate (Dai et al., 2001) or multivariate (Van Westen, 2000)
regression the key assumption of variable independence is broken where continuous data is used.
By applying an arbitrary grid cell layout to a region, the probability of cell independency is highly
improbable (Carrara et al. 1991, 1995). This affect is known as spatial autocorrelation (Anselin,
2001) and can be defined as the coincidence of value similarity with locational similarity.

It is very difficult to calculate the accuracy or validity of a susceptibility map. The output of the GIS is
a visual representation of the likelihood of a polygon to suffer a landslide, there is no predictive
element within the work. Thus it falls to prove or disprove susceptibility, a near-impossible task.
Often the only estimation of a model͛s accuracy is to compare a GIS result to the same task
completed by a specialist.
ü  ʹ Comparison of specialist produced (top) and GIS produced (bottom) susceptibility maps
(Rowbotham and Dudycha, 1998, Figs 7,8)

As illustrated in figure 4, there is a tendency to overestimate the likelihood of a landslide when using
a GIS. The results will vary between papers but in this case the concordance between expert
produced and GIS produced mapping was approximately 89.8% (p166). This value is only based on
the bivariate distinction of stable and unstable, if a multivariate discrimination was used the
concordance is significantly reduced. This exemplifies one major issue of GIS applications, their
assumed ability to remove ambivalence and personal opinion and to replace them with scientific
rigour. This is clearly not the case when the accuracy of a computer produced model is tested in
relation to a human produced map. This demonstrates an apparent hypocrisy within GIS, whilst on
the one hand it is lauded for removing human bias and replacing it with technical competency and
irrefutable scientific legitimacy; at the same time it is continually compared to the very method it
aims to render redundant.
The efforts to remove human-induced bias from the system are themselves only a token gesture. As
the various layers of a statistical approach are compiled it is common to apply a system of
weightings to them; in other words to place greater importance on certain features and to relatively
devalue others. For example in Sarkar and Kanungo (2004) slope angle and lithology are given
greater weighting than the slope aspect. This method introduces an element of human choice and
therefore bias to the GIS.

A further issue with the use of GIS comes not as a critique to the method, but rather the outcome.
The final product of many landslide susceptibility maps is a qualitative assessment of the likelihood
of a landslide at any given location. Whilst useful as a planning tool, it does not provide any
predictive qualities. It is not possible to determine when a landslide will occur, merely that it is more
likely in position x than position y. This is made worse by the categorical mapping used in the
majority of papers, terms such as ͞somewhat susceptible͟ or ͞highly likely͟ are used to describe the
polygons. These terms convey very little useable information, it is not possible to infer what the
implications are or what measures should be taken to reduce the potential impact of a landslide.
This is clearly an issue within a specific paper but is worse when considered in a broader context.
Where one author has defined ͞highly likely͟ to mean one thing, the same term is applied elsewhere
with an entirely different classification. This makes it impossible to compile a larger scale landslide
susceptibility map where differing methods are introducing conflict between seeming identical
classifications.

Furthermore the process of statistical mapping is potentially flawed in its use of historic events to
predict future ones, as briefly mentioned earlier. The route to improving the accuracy of landslide
modelling and eventually to landslide prediction lies in a greater understanding of the root causes of
landslides. By looking only to previous events and classifying the relative importance of a series of
variables statistical methods do not further the understanding of the causes of landslides
(Rowbotham and Dudycha, 1998).

It can be seen that there are significant issues within the current GIS techniques associated with
landslide susceptibility mapping, additionally though it is worth noting the methods used to diminish
these. It was identified that the collection of data is still a major issue for any GIS application. This
has been combated by an increase in the availability and resolution of civilian satellites and the
related increase in remote sensing as a source of data. It is not possible to get all the required data
from remote sensing; however it can prove to be increasing useful for land cover and land use
mapping as well as DEM generation.

Secondly it is possible to reduce the impact of human bias within the system of weightings found in
statistical methods. Several authors have used Artificial Neural Networks (ANNs) to attempt to do
this (Lee, 2007 and Ermini et al, 2005). ANNs aim to replicate the learning ability of the human
brain, to ͚learn͛ through repetition. By repeatedly running a model and having the output error
terms included as an input variable the model can improve in accuracy the more it is run. The
theoretical basis of using an ANN is relatively simple; however the actual process itself is somewhat
more complex. As such this has been used only very selectively and requires a high degree of
mathematical and technical ability.
To conclude, it is evident that GIS is found in a burgeoning range of applications of which landslide
susceptibility mapping is but one example. Of the three methods of utilising GIS in this context,
statistical methods are the most prevalent; this involves calculating the importance of a series of
layers before combining them to produce a hazard susceptibility map. Using GIS has several key
advantages over traditional mapping techniques, such as allowing for larger areas to be mapped and
much more quickly. Once collated the data analysis is relatively straight forward and by using GIS a
greater number of variables can be incorporated.

The ultimate outcome of the process is to map the susceptibility of a region to landslides. In this
vein it is questionable whether the results of this complex series of analyses gives any better
mapping than more traditional fieldwork approaches. Additionally GIS in its current utilisation does
not enhance understanding of landslides and in some cases its usefulness in mapping is
problematised through ambiguous terminology.

It can be seen that where issues, such as those around data collection, have been identified progress
has been made in negating them. Taking this one step further though, there are still several issues
with GIS methods currently, which once acknowledged and solutions sought may see GIS becoming
a more useful tool for landslide susceptibility mapping.

2





Anbalagan, R., Sharma, L. and Tyagi, S., 1993, Landslide hazard zonation (LHZ) mapping of a part of
Doon Valley, Garwhal Himalaya, India. In: Chowdhury, R.N., Sivakumar, M. (Eds),
*  
     
 *   , Balkema, Rotterdam,
pp. 253 ʹ 260

Anselin, L., 2001, Spatial econometrics. In Baltagi, B., (ed) 



 
 

 , Basil
Blackwell, Published by Oxford Press

Atkinson, P.M. and Massari, R., 1998, Generalized linear modelling of landslide susceptibility in the
Central Apennines, Italy, 
  
  , Vol. 24, No. 4, pp. 373 ʹ 385

Carrara, A., Cardinali, M., Detti, R., Guzetti, F., Pasqui, V. and Reichenbach, P., 1991, GIS techniques
and statistical models in evaluating landslide hazard, *
  
,
Vol. 16, pp. 427 ʹ 445

Carrara, A., Cardinali, M., Guzetti, F. and Reichenbach, P., 1995, GIS technology in mapping landslide
hazard, In: Carrara, A. and Guzetti, F. (Eds.), 
  

  
 , Kluwer Academic Publishers, Netherlands, pp. 135 ʹ 175

Chau, K.T., Wong, R.H.C., Lui, J. and Lee, C.F., 2003, Rockfall hazard analysis for Hong Kong based on
rockfall inventory, 
   
*   , Vol. 36, No. 5, pp. 383 ʹ 408

Chau, K.T., Sze, Y.L., Fung, M.K., Wong, W.Y., Fong, E.L. and Chan, L.C.P., 2004, Landslide hazard
analysis for Hong Kong using landslide inventory and GIS, 
  
  ,
Vol. 30, pp. 429 ʹ 443
Dai, F.C., Lee, C.F., Li, J., Xu, Z.W., 2001, Assessment of landslide susceptibility on the natural terrain
of Lantau Island, Hong Kong, *  
  

Vol. 40, pp. 381 - 391

Dahal, R.K., Hasegawa, S., Nonomura, A., Yamanaka, M., Masuda, T. and Nishino, K., 2007,
GIS-based weights-of-evidence modelling of rainfall-induced landslides in small catchments
for landslide susceptibility mapping, *  
  

, Vol. 54, No. 2, pp. 311 ʹ 324

Ermini, L., Catani, F. and Casagli, N., Artificial Neural Networks applied to landslide susceptibility
assessment, 



Vol. 66, No. 4, pp. 327 ʹ 343

Lee, S., 2007, Landslide susceptibility mapping using an artificial neural network in the Gangneung
area, Korea,   

 

  , Vol. 28, No. 21, pp. 4763 ʹ 4783

Kienholz, H., 1978, Maps of geomorphology and natural hazards of Grindelwald, Switzerland: scale
1:10,000,    , Vol. 10, No. 2, pp. 169 ʹ 184

Rowbotham, D.N. and Dudycha, D., 1998, GIS modelling of slope stability in Phewa Tal watershed,
Nepal, 



Vol. 26, pp. 151 ʹ 170

Sarkar, S. and Kanungo, D.P., 2004, An integrated approach for landslide susceptibility mapping using
remote sensing and GIS, 

 *    
  , Vol. 70,
pp. 617 ʹ 625

Siddle, H.J., Jones, D.B., Payne, H.R., 1991, Development of a methodology for landslip potential
mapping in the Rhondda Valley In: Chandler, R.J. (Ed 
  *   , Thomas
Telford, London, pp. 5 ʹ 58

Vanackera, V., Vanderschaeghea, M., Govers, G., Willems, E., Poesen, J., Deckers, J. and De Bievre,
D., 2003, Linking hydrological, infinite slope stability and land-use change models through
GIS for assessing the impact of deforestation on slope stability in high Andean watersheds,




Vol. 52, pp. 299 ʹ 315

Van Westen, C.J., 2000, The modelling of landslide hazards using GIS,  
 Vol. 21,
pp. 241 - 255

Varnes, D.J., 1984, Landslide hazard zonations: a review of principles and practice, commission on
landslides of the IAEG, UNESCO, , No. 3, p.61

Você também pode gostar