GIS Modeling of Soil Loss to Enhance Watershed Management Strategies
Brendan Belby
Department of Geography, Geographic Modeling Systems Laboratory, University of Illinois at Urbana-Champaign
The Pilot Watershed Program supported through the State of Illinois involves the interaction of multiple natural resource agencies with the purpose of supporting community efforts at sustainable landuse management of their watersheds. Court Creek is one of four watersheds in Illinois selected for the Pilot Watershed Program. Each pilot is paired with a “reference” watershed of similar landuse, hydrology, geology, and physical stream characteristics. The reference watershed for Court Creek is Haw Creek, also located in Knox County, Illinois (Figure 1). Court Creek will be the object of extensive Best Management Practice (BMP) implementation whereas the reference, Haw Creek, will receive minimal attention – allowing for a comparison of effectiveness at reducing levels of soil loss from the watersheds. The Geographic Modeling Systems Laboratory at the University of Illinois at Urbana-Champaign is conducting computational research to investigate improved methods of identifying spatial patterns of soil loss that will enable the watershed community to better predict the effect of various landuse alternatives with the goal of reducing soil erosion.
Digital Elevation Models from the USGS, at both 30m and 10m resolution, and Landsat satellite imagery serve as the base of the erosion analysis. A combination of GRASS5.0 and ARC/INFO software is being utilized in an innovative new modeling approach, the Modified 3D Universal Soil Loss Equation (USLE) with a new Length Slope factor based on upslope area and flow convergence (Mitasova et al. 1998). Traditional approches to landscape scale erosion modeling have been based on the watershed/subwatershed hydrologic and sediment transport, for example, ISWSBL (Borah 1999), a model that applies a homogenous value for a USLE variable throughout the entire hydrologic unit to compute average soil loss values. For example, if row crops are the dominant land cover for a particular unit, the entire area will be classified with this variable, regardless of additional landuses present. The new approach is a distributed modeling of erosion by overland flow. This method takes advantage of floating-point raster data to limit the generalizations of the older model and enable site-specific analysis throughout the entire watershed at various scales. A multiscale approach enables digital landscape characterization and visualization of the Court Creek Watershed at a hierarchy of scales: landscape/watershed (50-30m), subwatershed (10-5m), and a complex field (1m). Erosion risk analysis at varying scales will identify potential sediment sources and sinks throughout the watershed, and enable watershed managers, from the local farmer to State and Federal agencies, to make scientifically informed decisions as to the most effective and efficient approach for sustainable watershed management.
The Modified 3D RUSLE method was utilized to determine levels of soil loss throughout Court Creek Watershed, in Knox County, Illinois. It is an adaptation of the original USLE, an empirical equation designed for the computation of average soil loss in agricultural fields. This equation was developed for detachment capacity limited erosion in fields with negligible curvature and no deposition and represents soil loss averaged over time and total area. The equation has the following form (Wischmeier and Smith 1978, Renard et al. 1991),
where E [ton/(acre.year)] is the average soil loss, R [hundreds of ft.tonsf.in/acre.hr.year] is the rainfall intensity factor, K [tons per acre per unit R] is the soil factor, LS [dimensionless] is the topographic (length-slope) factor, C [dimensionless] is the cover factor and P[dimensionless] is the prevention practices factor. Various modifications of this equation are often applied to the estimation of soil loss using GIS (Warren et al. 1989). Revised USLE - RUSLE uses the same empirical principles as USLE, however it includes numerous improvements, such as monthly factors, incorporation of the influence of profile convexity/concavity using segmentation of irregular slopes, and improved empirical equations for the computation of LS factor (Foster and Wischmeier 1974, Renard et al. 1991).
To incorporate the impact of flow convergence, the hillslope length factor was replaced by upslope contributing area A (Moore and Burch 1986, Mitasova et al. 1995, 1996, Desmet and Govers 1996). The modified equation for computation of the LS factor in GIS in finite difference form for erosion in a grid cell representing a hillslope segment was derived by Desmet and Govers (1996). A simpler, continuous form of equation for computation of the LS factor at a point r=(x,y) on a hillslope (Mitasova et al. 1996), is
LS(r)
= (m+1) [ A(r) / a0 ]m [
sin b(r) / b0 ]n
where A[m] is upslope contributing area per unit contour width, b [deg] is the slope, m and n are parameters, and a0 = 22.1m = 72.6ft is the length and b0 = 0.09 = 9% = 5.16deg is the slope of the
standard USLE plot. This modification better reflects the impact of concentrated flow on increased erosion. It has been shown that the values of m=0.6, n=1.3 give results consistent with the RUSLE LS factor for slope lengths <100m and slope angles <14 deg (Moore and Wilson 1992), for slopes with negligible tangential curvature. Exponents m and n can be calibrated if the data are available for a specific prevailing type of flow and soil conditions.
Both the standard and modified equations can be properly applied only to areas experiencing net erosion. Depositional areas should be excluded from the study area because the model assumes that transport capacity exceeds detachment capacity everywhere and erosion and sediment transport is detachment capacity limited. Therefore, direct application of USLE/RUSLE to complex terrain within GIS is rather restricted. In case that the RUSLE is applied without excluding the depositional areas, the results should be interpreted as an extreme case with maximum spatial extent of erosion possible, or as a map of soil detachment, rather than net erosion.
Digital Elevation Models (DEMs)
The initial step in the calculation of soil loss was the download and preparation of 1:24,000 scale DEMs. The USGS website http://edc.usgs.gov/doc/edchome/ndcdb/ndcdb.html offers free downloadable DEMs for the entire United States. After decompressing and converting the files from USGS format to ASCII grids, they were imported into ARC/INFO, whereupon the 6 DEMs that cover the Court Creek Watershed were merged into one file in ARC-GRID. Since this research is focusing on modeling hillslope erosion, the sinks in the DEM were not filled to form a depressionless file. By not filling in the sinks, water flow stops in the model when it reaches areas of flat terrain, and will not continue on toward the stream channels. This method gives a more accurate representation of hillslope processes within the watershed. A polygon boundary coverage of the Court Creek Watershed was used to clip the merged file, producing an output DEM of the watershed itself. The horizontal resolution of the DEM is 30m and the vertical is 1m (Figure 16). This DEM was exported into GRASS5.0 and reinterpolated to an enhanced 20m resolution, using the following procedure.
The command g.region was used to set the resolution and map extent to the clipped file, and a MASK was generated in GRASS so that all subsequent analysis would be confined to within the boundary of Court Creek Watershed. A sites elevation file was created from the 30m DEM using the command r.random. The program allows the user to create a raster map layer and a site list file containing geographic coordinates of points whose locations have been randomly determined. After creating a sites file, the program s.surf.rst - interpolation and topographic analysis from given site data to GRASS floating point raster format using regularized spline with tension (GRASS Site Program) - was utilized to smooth out the vertical resolution from meters to centimeters, and increase the horizontal resolution to 20m. Figure 2 is a 3D image of the DEM for the Court Creek Watershed.
Flow accumulation for the watershed was calculated with the r.flow command (Figure 3). The DEM serves as the input file, and one of the options is used to generate flowline densities (which are equal upslope contributing areas per unit width, when multiplied by resolution). The value in each grid cell is the number of flowlines which pass through that grid cell, that means the number of flowlines from the entire map which have segment endpoints within that cell. Next, slope values in decimal degrees were calculated from the input DEM file using the command r.slope.aspect (Figure 4).
With the creation of flow accumulation and slope files, it was possible to compute the length slope factor, as previously discussed, for Court Creek Watershed, using the r.mapcalc (Grass raster program) (Figure 5).
r.mapcalc length-slope = “1.6*exp(flow accumulation*resolution/22.1,0.6)*exp(sin(slope)/0.09,1.3)”
Since the resolution of the DEM is 20m, a value of 20 was used for resolution in the command.
Soil loss was calculated within r.mapcalc using the following syntax:
r.mapcalc soilloss=R*K*C*P*length-slope
- R = rainfall energy factor. The product of rainfall energy and maximum 30 minute intensity divided by 100. An average annual R value of 150 is appropriate for the Court Creek Watershed region (Haan, et al. 1994).
- K = soil erodibility factor. Measures the soil’s resistance to the erosive powers of rainfall energy and runoff. A value of 0.3 was used for the study area (Haan, et al. 1994).
- C = land cover factor. Accounts for the effects of cover above the ground, ground cover, root mass, incorporated residue, surface roughness, and soil moisture on soil erosion. A landuse map derived from Landsat satellite imagery served as the basis for determining the C-Factor values. Table 1 lists the C-factor values for the landuses categories. Figure 6 is the landuse for Court Creek Watershed, and Figure 7 displays the C-Factors derived from the Landsat imagery (Haan, et al. 1994).
Cover
Factor
|
Landuse Value
|
Category
|
|
0.0 |
0 |
other |
|
0.0001 |
13, 15 |
dense forest |
|
0.001 |
14 |
open forest |
|
0.003 |
11, 12 |
grass |
|
0.01 |
10 |
orchards |
|
0.3 |
9 |
small grains |
|
0.5 |
8 |
row crops |
Table 1
- P = prevention practice factor. The ratio of soil loss from any conservation support practice to that with up- and downslope tillage. No information on conservation practices is currently available from landuse maps, so a value of 1 was assigned, assuming no preventative measures were undertaken.
- Length-Slope Factor (Figure 8).
The 3D visualization program SG3D within GRASS was utilized to generate images of levels of soil erosion for the entire Court Creek Watershed. Figure 9 is an overview of the entire watershed, followed by three additional zoomed in images at a smaller scale - Figure 10, Figure 11, and Figure 12.
Implications of Research into GIS Modeling of Soil Loss – Alternative Conservation Strategies
Ongoing research into
improved GIS modeling of soil loss with more powerful predictive capabilities
ultimately aims to provide the watershed community with increased information
to make informed management decisions.
The Court Creek Watershed is a sub-watershed of the Illinois River with
the majority of its land in agriculture.
In a determined effort to reduce the sedimentation problem of the
Illinois River and improve water quality, the State of Illinois formed a
partnership with the federal government in March of 1998 and formed the
Illinois Conservation Reserve Enhancement Program (CREP). Over $500 million dollars has been budgeted
to lease riparian agriculture land from private landowners and take it out of
production. Two major objectives of CREP
are to reduce sediment loading of the Illinois River by 20 percent and nitrate
loading by 10 percent. Guidelines currently state that only riparian land
within 300 feet of a stream are eligible for the CREP program. Although the
creation of riparian buffers will very likely have a positive impact on
reducing soil transport into the Illinois River, it is not known whether this
is the most effective and efficient approach. Buffers may prevent the soil from
entering the river, but do not treat the problem of soil erosion at its
source. With so much money being
allocated to the program, it is vital to determine the best approach for
spending it. The GMS Laboratory has been modeling different landuse
alternatives within the Court Creek Watershed to assist in determining what the
best strategy may be at reducing soil loss.
Through a series of
computations using map algebra in ARC-GRID, tables have been generated that
describe what the effects of various conservation strategies will be. In each scenario, a new C-Factor map was
created with updated values. All land
taken out of production was replaced with a C-Factor value of 0.001, which
corresponds to open forest. Any land
that met the criteria to be taken out of production, but already had a C-Factor
of 0.001 or less, retained its current value.
For instance, a model was run according to the CREP guidelines of all
land within 300 feet of streams be converted to open forest. If a parcel of dense forest exists within
this buffer, it will remain so. Changing
it to open forest would have led to increased soil erosion. New soil erosion risk maps were created by
rerunning the model with the updated C-Factors. The tables show total acres and percent of land cover for each
category within the Court Creek Watershed and were produced with the r.report
command in GRASS5.0.
Under present conditions,
63 percent of the land in the watershed is classified as stable, while 24
percent is at a level of high erosion or greater. Row crops are the dominant landuse at 39 percent (24,181 acres)
and dense and open forest comprise 26 percent of the cover (Table
2, Figure
9). When the 300 feet buffers of
the CREP program were modeled, only a minimal reduction in higher levels of
soil erosion resulted with an increase of stability by 3 percent. A two percent reduction in row crops was
necessary to accomplish this (Table
3, Figure
13). If all land with slopes
greater than 5 degrees were to be replaced with open forest, a slight reduction
in extreme erosion was obtained – from 3.33 to 1.73 percent (Table
4, Figure
14). Stability increased by 1,842
acres. Grasses and small grains were
affected the greatest with this strategy.
Land that met the criteria of having slopes greater than 5 degrees and
was within 300 feet of a stream displayed nearly identical results as the
previous query (Table
5).
Instead of removing all
land within 300 feet of a stream from production, a scenario was tested of only
land within the buffer strip that also had a soil erosion risk factor greater
than 10 (high). Interestingly, this
produced results very similar to all land within 300 feet, indicating that
significant areas of higher levels of erosion are not located in riparian
areas, and thus will not experience reductions in soil loss from the creation
of buffer strips (Table
6). A final query was run which
converted all land throughout the Court Creek Watershed with a soil erosion
risk level greater than 10 into open forest, unless it already had a lower
C-Factor. This management technique had
the greatest impact on lowering soil loss.
Stable land cover increased from 63 to 80 percent, an increase of 10,187
acres. Such a large decrease in soil
erosion would come at the expense of agricultural land. Row crops would need to
be reduced by 7,358 acres (11.8 percent) (Table
7, Figure
15).
Research with GIS
modeling has shown that the highest areas of soil erosion predicted by the
model are located between the upland areas and the stream buffers. This information is being relayed to other
scientists and watershed managers so that the most effective Best Management
Practices will be implemented. It may
be the case that CREP needs to re-evaluate its guidelines for qualification of
funding if resources spent elsewhere in a different strategy will yield better
results.
10m DEM Analysis
When modeling soil loss
within a GIS framework, much of the power and accuracy in predictive
capabilities rests in the quality of the data.
Initial models have relied on 30m DEMs - reinterpolated to 20m, and 30m
resolution Landsat imagery. The GMS
Laboratory has recently acquired 10m DEMs from the USGS and 10m satellite
imagery of landuse. Preliminary
analysis and modeling with this higher resolution data has produced excellent
results. Distributed modeling of
erosion by overland flow using 10m floating point raster data significantly
increases the detail and level to which soil loss can be predicted. With the increased resolution, flowline
densities improve. Areas that are
displayed as overland flow in 30m data exhibit channel formation with the new
10m DEMs. The three following figures
are used as a comparison between different resolution data mapping flowline
densities.
Figure
16 - 30m DEM, meter vertical
resolution
Figure
17 - 20m reinterpolated DEM, centimeter
vertical resolution
Figure
18 - 10m DEM, decimeter vertical
resolution
Conclusion
The presented study
focused on spatial distribution of soil detachment by overland flow has shown
that the current CREP criteria do not address the reduction of sediment
sources. Riparian buffers may prevent
sediment from reaching the stream, but new management techniques will need to
be undertaken if total soil loss is to be reduced from upland areas. The GMS Laboratory at the University of
Illinois at Urbana-Champaign is continuing with research into better methods of
GIS modeling of soil loss. As higher
quality data is obtained and procedures are modified, the results from these
computations will only improve. Results
are being compared with other researchers with the intention of making
adjustments that will allow the watershed community to make scientifically
informed decisions as to the best conservation strategies to implement. Some limitations have yet to be
overcome. In this study we have focused
only on soil detachment by overland flow.
Future modeling will evaluate the capacity of landscapes with buffers to
deposit the detached soil and provide estimates of net erosion/deposition and
sediment delivery to streams. Better
estimates of prevention practice factors already in place throughout Court
Creek Watershed need to be included into the model, and additional on-site
field reconnaissance should be undertaken to calibrate the models. It is the eventual goal that local farmers
will be able to compute for themselves at the field level areas on their
property most prone to soil erosion.
Acknowledgments
Helena Mitasova, William
M. Brown
Geographic Modeling
Systems Laboratory, University of Illinois at Urbana-Champaign
Endnotes
GMS Laboratory – Erosion Modeling Web Page http://www2.gis.uiuc.edu:2280/modviz//
GRASS5.0 website http://www.baylor.edu/~grass/grass5/index.html
Due to their large file
size, images contained in this paper are located on the GMS web server.
References
Borah, D. K., and Bera,
M., Hydrologic Modeling of the Court Creek Watershed – ISWS Contract Report
2000-04. Published by the Illinois State Water Survey, Champaign, Illinois.
Desmet, P. J. J., and G.
Govers, A GIS procedure for automatically calculating the USLE LS factor on
topographically complex landscape units, J. Soil and Water Cons., 51(5),
427-433, 1996.
Foster, G.R., and
Wischmeier, W.H.,1974, Evaluating irregular slopes for soil loss prediction.
Transactions of the ASAE, 12, 305-309.
Haan, C. T., B. J.
Barfield, and J. C. Hayes, 1994, Design Hydrology and Sedimentology for Small
Catchments, pp. 242-243, Academic Press.
Mitasova, H., Mitas, L., Brown, W. M., Johnston, D., 1998, Multidimensional Soil Erosion/deposition Modeling and visualization using GIS. Final report for USA CERL. University of Illinois, Urbana-Champaign, IL.
Mitasova, H., J. Hofierka, M. Zlocha, and R. L. Iverson,
1996a, Modeling topographic potential for erosion and deposition using GIS,Int.
Journal of Geographical Information Science, 10(5), 629-641.
(reply to a comment to this paper appears in 1997 in Int. Journal of
Geographical Information Science, Vol. 11, No. 6).
Mitasova, H., Mitas, L., Brown, W.M., Gerdes, D.P., Kosinovsky, I., 1995, Modeling spatially and temporally distributed phenomena: New methods and tools for GRASS GIS. International Journal of GIS, 9 (4),special issue on integration of Environmental modeling and GIS, p. 443-446.
Moore I.D., and Burch G.J., 1986b, Modeling erosion and deposition: Topographic effects. Transactions ASAE, 29, 1624-1640.
Moore, I.D., and Wilson, J.P., 1992, Length-slope factors for the Revised Universal Soil Loss Equation: Simplified method of estimation. Journal of Soil and Water Conservation, 47, 423-428.
Renard G.K., Foster G.R., Weesies G.A., Porter J.P., 1991, RUSLE - Revised universal soil loss equation. Journal of Soil and Water Conservation, v. 46, p.30-33.
Warren, S.D., Diersing, V.E., Thompson, P.J., and Goran, W.D., 1989, An erosion-based land classification system for military installations. Environmental Management, 13, 251-257.
Wischmeier, W.H., and Smith, D.D., 1978, Predicting rainfall erosion losses, a guide to conservation planning. Agriculture Handbook No. 537, US Department of Agriculture, Washington D.C.
