Você está na página 1de 12

Base Flow Separation by the Conductivity Mass Balance Method, Southwest Florida Water

Management District, Florida


Mark Stewart and Darline Lott
University of South Florida
May, 2010
Background
Basic Objective
Total stream discharge (discharge) can be considered to be the sum of two principal stream
flow components, base flow (ground-water inflow) and run off, or quick flow (Stewart et al.,
2007). Ground-water models predict ground-water discharge to streams, so if base flow is
known, it can be used as a calibration target for ground-water flow models. This study presents a
method for estimating base-flow contributions to stream flow for selected stream gages within
the SWFWMD.
Methods
Several classes of methods are currently used for base flow separation. These include analytical
methods, such as the USGS program HYSEP (Sloto and Cruse, 1996), and the program WHAT1
(Lim et al., 2005). Analytical methods use some form of mathematical, low-pass filtering to
smooth the stream hydrograph. Base flow is assumed to be the low-frequency component.
Commonly used analytical methods are arbitrary smoothing functions, unrelated to actual basin
measurements other than discharge. HYSEP sets the degree of smoothing of the three smoothing
algorithms it uses based on basin area. WHAT1 does not directly incorporate basin variables, and
also uses only discharge as a basin variable.
HYSEP is a low pass filter that incorporates three similar methods for smoothing the discharge
hydrograph. All involve low-pass filtering accomplished by passing a time-specified window
along the discharge record. All three methods can be incorporated in a spreadsheet solution, and
the local minimum and moving window methods are particularly easy to use in a spreadsheet.
The degree of smoothing is determined by the length of the time window. The window length is
termed 2N*, where N is the presumed time, in days, after a peak in the hydrograph that all direct
runoff ceases. HYSEP determines N from N = A0.2, where A is basin area in square miles. 2N* is
the odd integer closest to 2N.
WHAT1 ( Lim et al., 2005) uses a version of the recursive digital filter proposed by Nathan and
McMahon (1990), Chapman (1991) and modified by Eckhardt (2005). It uses two smoothing
parameters, , an arbitrary smoothing parameter set to a default value of 0.925 in WHAT1, and
BFImax , the maximum long-term value of the ratio of base flow to discharge. This value is set to
a default value of 0.80 in WHAT1, although it can be changed by the operator. While BFImax is
defined as a basin variable, it is used as an arbitrary smoothing variable in WHAT1 and is not
actually determined or calibrated from basin variables. Lim et al. (2005) and Eckhardt (2005)
assess the accuracy of the recursive digital filter used in WHAT1 by comparing results to another

analytical filter method, BFLOW (Arnold et al., 1995), which has been compared to manually
applied, analytical separation methods and base flow estimated from basin hydrologic budget
studies (Arnold and Allen, 1999). It is reasonable to suggest that the accuracy of WHAT1 for
determining actual base flow values has not been established. As stated by Lim et al. (2005),
The digital filter method has no physical meaning, but it removes the subjective aspect from
manual separation, and it is fast, consistent, and reproducible (italics added).
Another base-flow separation method is based on the slope of the presumed base-flow recession
curve after a runoff peak on a hydrograph. A version of this method, developed by Rorabaugh
(1964) has been automated by Rutledge and Daniel (1993) and is the basis for the USGS
program RORA. An extensive, thorough, and entertaining dismantling of the recession curve
method was performed by Halford and Mayer (2000).
The Conductivity Mass Balance Method
Stewart et al. (2007) propose using the conductivity mass balance (CMB) method to separate
base flow from quick flow or run off. In their article, they discuss and test the assumptions
behind the CMB method. They propose to use stream-flow conductivity data as a proxy for ionic
concentration of stream flow. The use of concentration to separate the base-flow component from
run off or quick flow was proposed by Steele (1969) and Pinder and Jones (1969). Stewart et al.
use the CMB-derived base flow to calibrate an analytical method to reproduce the CMB results
by varying the smoothing constant in the analytical method. In their example they use the
analytical methods used in HYSEP, but in a spreadsheet solution. Once the 2N*, or time window,
of a HYSEP method, such as the local minimum method, has been calibrated from CMB-derived
base flow, the calibrated analytical method can be used to separate base flow over the entire
available discharge record. This is useful, as discharge data are typically available over a much
longer period of record than conductivity data. A similar procedure was used to calibrate the
recursive digital filter of Eckhardt (2005) by Gonzales et al. (2009). However, as demonstrated
by Stewart et al. (2007), low-pass filter methods yield arbitrary and unsubstantiated results
unless they are calibrated with direct or mass-balance measurements of base flow over a
significant period of record (typically 2+ years for the CMB method).
As the basic assumption of the conductivity mass balance method is that higher conductivity
flow components are ground water, the method cannot distinguish between natural stream-flow
contributions and anthropogenic contributions. For example, in the southern ground-water basin
of the SWFWMD, substantial quantities of irrigation tail water contribute to stream flow in
basins such as the Myakka River. As these irrigation waters are derived from the Floridan
Aquifer, and may have been further concentrated by evaporation, they will be counted as part of
the base-flow component of stream flow. A similar effect is caused by addition of discharges
from waste-water treatment plants. If the fluid conductivity and volumes of these anthropogenic
discharges are known, an additional, known, term can be added to the mass balance equation to
correct for anthropogenic discharges. For similar reasons, the CMB method may produce
erroneous results downstream of major impoundments.

This report uses a modification of the method of Stewart et al. (2007). If the CMB-derived base
flow is plotted against discharge on a log-log plot, for most streams a straight line, log-linear,
relationship is apparent (Figure 1).

A simple log-linear equation can be fitted to the CMB base flow versus discharge plot, of the
form
QB = aQb,

(1)

or log QB = log a + b*logQ,

(2)

where QB is base flow (L3/t), Q is discharge (L3/t), and a and b are the coefficients of the loglinear plot (Figure 2).

The a and b coefficients determined from CMB-derived base flow can then be used to derive
estimated base-flow values from the entire available discharge record. A major advantage of this
method is that it is based entirely on values of stream-flow conductivity and discharge obtained
at the gage, and does not rely on any arbitrary smoothing coefficients. In a calibration procedure,
the values of a and b are varied until the difference between the cumulative CMB base flow and
the cumulative base flow estimated by the log-linear equation (1) is minimized (Figure 3).

A simple linear, least-squares regression procedure can also be used to estimate a best-fit line on
the log-log plot, but a better agreement between the CMB data and the log-linear estimates can
be obtained if the values of a and b are adjusted by the operator to minimize the difference
between the cumulative CMB and log-linear base flows over the period of record of conductivity
data (Figure 3).
Results
Types of relationships between base flow and discharge
In the SWFWMD, there are two principal types of log-linear plots of CMB-determined baseflow discharge versus stream discharge. The first is the most common relationship, both in
Florida and the Southeast (Stewart et al., 2007). This is the straight line relationship shown in
Figure 1. In this case, base flow becomes an increasingly smaller percentage of total discharge as
discharge increases, and is related to discharge by an exponent <1. The second type of plot is
illustrated by the Hillsborough River near Zephyrhills gage (Figure 4). In this case, the plot has
two segments. At low discharges, flow is dominated by spring discharge, and nearly all stream
flow is base flow. Above a critical discharge, QC, base flow becomes a smaller percentage of
stream flow as discharge increases. This is best illustrated by a log-log plot of stream flow
conductivity versus discharge (Figure 4).

The spring-flow component is the horizontal part of the plot, and the critical discharge is about
125 cfs. At discharges below 125 cfs at this gage, most of the stream flow is base flow. For the
two-segment plots, two forms of the log-linear equation are fitted to the data, one for discharges
<Qc, and one for discharges >Qc. At the Hillsborough River gage, the value of the exponent b is
1.0 for Q<125 cfs, suggesting that most stream flow at this gage at lower discharges is base flow.
High Flow Suppression of Base Flow
At very high discharges, stream stages can get high enough so that the hydraulic gradient
between the aquifer supplying base flow and the stream is greatly reduced or even reversed. This
results in a substantial decrease in base flow and deviation from the log-linear relationship at
very high discharges. As the log-linear equation calculates some base flow at all discharges, at
very high discharges the log-linear method may overestimate base flow. However, as the
percentage of base flow of total discharge at very high discharges is typically a few percent, the
effect on calculations of monthly or annual base flow is small. The suppression of base flow at
high discharges can be observed on log-log plots of CMB base flow versus discharge (Figure 5).

Discussion of Base Flow Separations at Stream Gages Within SWFWMD


Unfortunately, most stream gages operated by the USGS do not have available stream-flow
conductivity data. Within the SWFWMD, 23 gages have sufficient conductivity data to attempt a
base-low separation using the CMB method. In addition, a CMB analysis was performed at
several gages that are not appropriate for the typical application of the CMB method, such as
major springs. However, the CMB method does appear to provide interesting insights into the
hydrology of these spring systems, although the analysis is not a base-flow separation.
The base-flow separation for each of the 23 gages where sufficient data exist to conduct a CMB
analysis is briefly described below. For each gage, a spreadsheet was prepared that contains the
CMB analysis, the calibration plots, the a and b coefficients, and an estimate of base flow from
the available discharge record. The spreadsheets are contained in the digital files that accompany
this report.
Buckhorn Creek
The conductivity vs. Q plot is not noisy, and the CMB base flow vs. Q plot is linear. The loglinear model fit is good.
Charlie Creek near Gardener

The record at this gage is short. The apparent fit of the log-linear model is good, but the value of
the a coefficient is higher than at most other gages.
Coker Creek
The conductivity vs. Q plot is quite noisy. The a and b values of the log-linear model are
reasonable when compared to the other gages.
Hillsborough River nr Zephyrhills (State Park)
This gage has a long period of conductivity data. It exhibits a pronounced two-segment
conductivity vs. Q plot, indicating that most stream flow below 125 cfs is spring flow or base
flow
Little Charlie Bowlegs
The conductivity vs Q plot is not overly noisy, and the log-linear fit is good. The a and b values
look to be reasonable compared to the other gages..
Long Creek, Myakka City
This record has a noisy conductivity vs. Q plot. The log-linear fit to the CMB data looks
reasonable. This gage is affected by irrigation tail water at some times of the year.
Maple Creek
The stream-flow conductivity record at this gage is short, providing limited data. There is some
suppression of base flow evident at higher discharges. The fit of the log-linear model to the CMB
data is good.
Myakka River, Myakka City
The CMB base flow vs. Q plot is fairly linear, however there are some low conductivity events
apparent on the plot.
Myakka River: Youngs Creek
This gage produces a very noisy conductivity vs Q plot. The value of the b coefficient suggests a
strong influence of irrigation water on conductivity values.
Oklawaha nr Conner
This gage has a short conductivity record. The stream flow conductivity varies over a small
range, suggesting that the conductivity record is not long enough for a valid CMB analysis. The
conductivity vs. Q plot is noisy at higher Q values. The a and b coefficients are far out of range
when compared to the values at other gages.

Ogleby Creek
The conductivity vs. Q plot is very noisy. The a and b values are somewhat lower than at most of
the other gages, but the log-linear fit produces a reasonably good agreement with the cumulative
CMB base-flow data.
Prairie Creek
The conductivity vs. Q plot is somewhat noisy at higher Q values. The a parameter value is
significantly higher than at most other gages; the b parameter is within range.
Rocky Creek
The conductivity values at this gage have a narrow conductivity range, suggesting possible
anthropogenic influence. The log-linear fit matches the CMB-derived data well, the b value is
reasonable, but the a value is on the low side of the range of a values.
Shell Creek
This gage may have a relatively constant conductivity below about 20 cfs of about 1000 S/cm,
but a one-segment log-linear fit works well. The relatively high conductivity of the <20 cfs
component suggests an anthropogenic influence. The values of a and b are within normal range.
St Johns Rvr
The conductivity vs. Q plot looks reasonably linear. The CMB-derived base flow vs. Q plot
indicates a strong suppression of base flow at discharges above about 8,000 cfs.
Withlacoochee at Croom
This gage has limited conductivity data. The a value is too large, suggesting that the CMB
analysis may not be valid.
Withlacoochee at Cumpressco
This gage and the gage at Eva produce the best-fit CMB analyses of the Withlacoochee gages,
although the a value at this gage is low.
Withlacoochee at Dade City
This gage has a reasonably good fit of the log-linear model to the CMB data, however the a
value is nearly 1, higher than at most other gages
Withlacoochee at Eva

This gage produces a good log-linear fit, with reasonable a and b values.
Withlacoochee at Holder
This gage has a very noisy conductivity vs. Q plot. However, the a and b values are reasonable.
Withlacoochee at Nobleton
This gage has limited conductivity data. There are a few dominant events that produce a noisy
conductivity vs. Q plot. The a value is large and out of range compared to other gages, and the b
value is low.
Withlacoochee at Trilby
This gage produces a two-segment plot, with one linear relationship for discharges <about 400
cfs, and a second linear relationship for discharges >400 cfs. While the log-linear models
produce a very good agreement with the CMB-derived base flow values, the a values for both
segments are greater than 1.

References
Arnold, J.G. and P.M. Allen. 1999. Automated methods for estimating baseflow and ground
water recharge from streamflow records. Journal of the American Water Resources Association
35(2): 411-424.
Arnold, J.G., P.M. Allen, R. Muttiah, and G. Bernhardt. 1995. Automated base flow separation
and recession analysis techniques. Ground Water 33(6): 1010-1018.
Chapman, T. G., 1991: Comment on Evaluation of automated techniques for base flow and
recession analyses by R. J. Nathan and T. A.McMahon, Water Resour. Res., 27, 17831784.
A. L. Gonzales, J. Nonner, J. Heijkers, and S. Uhlenbrook, 2009. Comparison of different base
flow separation methods in a lowland catchment. Hydrol. Earth Syst. Sci., 13, 20552068.
www.hydrol-earth-syst-sci.net/13/2055/2009/
Eckhardt, K.: How to construct recursive digital filters for base flow separation, 2005. Hydrol.
Processes, 19, 507515.
Halford, K.J., and G.C. Mayer. 2000. Problems associated with estimating
ground water discharge and recharge from stream-discharge records. Ground
Water 38, no. 3:
331342.
Kyoung Jae Lim, Bernard A. Engel, Zhenxu Tang, Joongdae Choi, Ki-Sung Kim, Suresh
Muthukrishnan, and Dibyajyoti Tripathy, 2005. Automated web GIS based hydrograph
analysis tool, WHAT1. Jour AWRA, Paper 04133, 41, 6, p. 1407-1416.

Nathan, R.J., and T.A. McMahon. 1990. Evaluation of automated techniques


for base flow and recession analysis. Water Resources Research 26, no. 7:
14651473.
Pinder, G.F., and J.F. Jones. 1969. Determination of the groundwater
component of peak discharge from chemistry of total runoff. Water
Resources Research 5, no. 2: 438445.
Rorabaugh, M.I. 1964. Estimating changes in bank storage and ground water
contribution to stream flow. International Association of Scientific Hydrology,
Publication 63:
432441.
Rutledge, A.T., and C.C. Daniel III. 1994. Testing an automated method to
estimate ground-water recharge from streamflow records. Ground Water 32,
no. 2: 180189.

Sloto, R.A., and M.Y. Crouse. 1996. HYSEP: A computer program for streamflow hydrograph
separation and analysis.USGS Water Resources Investigations Report 96-4040.Washington,
D.C.: USGS.
Steele, T.D. 1968. Digital-computer applications in chemical quality studies of
surface water in a small watershed. International Association of Scientific
Hydrology, Publication
80: 203214.
Stewart, M., J. Cimino, M. Ross, 2007. Calibration of Base Flow Separation Methods with
Streamflow Conductivity. Ground Water, 45, 1, 17-27

Você também pode gostar