AGNPS Simulation of Runoff and Sediment Loss from

 the Bugenhagen basin, Clayton County, Iowa

Mary P. Skopec

Graduate Assistant

Geological Survey Bureau

 

Robert D. Libra

Hydrogeologist

Geological Survey Bureau

  

A report on Cooperative Agreement #68-6114-516 between the

Iowa Department of Natural Resources and the

U.S. Department of Agriculture-

Natural Resources Conservation Service

December, 1995

Introduction

he Bugenhagen basin is a 1,105 acre watershed located in Clayton County, northeast Iowa. The climate is humid continental, and precipitation averages about 32 inches per year. Relief within the basin is about 145 feet. Wisconsinan loess and loess-derived alluvium are the surficial geologic materials. Downs, Luana, and Otter-Worthen soils predominate. Landuse is agricultural; typically 65% to 75% of the watershed cropped to corn, and 10% to 15% to alfalfa. About 10% of the watershed is used for pasture. Typically, flow is perennial in the downstream part of the basin, with the exception of extended dry periods and some winter conditions. Flow typically terminates at a set of sinkholes. Water quality has been monitored within the basin since 1981 (Hallberg et al., 1983, 1984; Rowden et al., 1995).  In mid-1986, a standard U.S. Geological Survey gauging/sediment/precipitation station was installed to monitor precipitation inputs, surface-water discharge, and suspended sediment concentrations from the upper 736 acres of the watershed (Kalkhoff, 1989). The Bugenhagen basin was selected in the Big Spring Basin Demonstration Project (BSBDP) to be a model area for monitoring the effects of implementation of improved farm management and soil conservation on water quality (Littke and Hallberg, 1991). A goal for the watershed was to reduce soil erosion on all cropland to less than "T", the tolerable soil loss defined as the maximum erosion rate for a given soil to maintain the soil's long-term productivity. Detailed information on land use and erosion control measures are available because of this effort. Terracing was the major practice selected to protect soil resources on 80% of the cropland within the sub-basin. The majority of the erosion control practices were installed during 1986 and 1987. Discharge and sediment concentration data are available for few small (<1,000 acre) watersheds. In this investigation, discharge and sediment data, along with land use and erosion control practices from the Bugenhagen basin were used to evaluate the computer model AGNPS (AGricutural Non-Point Source; Young et al., 1987) version 4.02 (Young et al., 1994). AGNPS is an event-and watershed-based model that simulates runoff, sediment, nutrient, and chemical losses from agricultural lands. The model was developed by the U.S. Department of Agriculture-Agricultural Research Service. Eleven rainfall events from Water Years (WYs) 1990 and 1991 were simulated using AGNPS, to allow comparison of the model’s output with field data collected at a standard U.S. Geological Survey gauging/sediment station. An additional five events from WY 1992 were also simulated; measured discharge, but not sediment, data are available for comparison with the results of the WY 1992 simulations. Finally, AGNPS was used to simulate runoff and sediment loss volumes for the WY 1990 through WY 1992 rainfall events, but under the conditions existing prior to the implementation of extensive erosion control practices.  

Bugenhagen Basin

Physical Description

 

Figure 1 shows the location of the Bugenhagen Basin, and Figure 2 shows the basin’s topography. The axis of the basin runs generally northwest to southeast. Total relief is 140 feet above the gauging station. Slopes range to over 14%. The uplands consist of Wisconsinan loess, over remnants of Pre-Illinioan glacial till and Ordovician carbonate bedrock (Littke and Hallberg, 1991). The drainageways are filled with loamy alluvium that progressively thins, down drainage. Downs, Luana, and Otter-Worthen soils predominate (Figure 3). Sixty percent of the basin is mapped as Downs series soils, developed in greater than 5 feet of loess; 10% as Luana series soils, developed in 20-40 inches of loess over weathered carbonate bedrock; and 9% as Otter-Worthen series soils, developed in loess-derived alluvium (U.S.D.A.-Soil Conservation Service, 1982). Water discharging from the basin enters a complex of sinkholes and enters the groundwater system of the underlying Ordovician bedrock aquifer. The sinkholes are often clogged with soil and only occasionally act as open funnels into the bedrock. Following major rains or snow melt, some runoff overflows the sinkholes, beyond which there is no distinct channel (Hallberg et al., 1983). These sinkholes are sufficiently far downstream as to have no effect on discharge rates at the gauging station (Figure 2).
 

 

 

 

Land Use and Erosion Control Practices

Figures 4 and 5 show basin land use for years 1990 and 1991, respectively; Table 1 summarizes these data. During these years, respectively, 72% and 65% of the watershed was cropped to corn, and 11% and 13% to alfalfa. About 9% of the watershed is used for pasture. Significant erosion control practices were put in place in the watershed in 1986 and 1987, as part of the Big Spring Basin Demonstration Project (Littke and Hallberg, 1991). Terracing was the major practice selected to protect soil resources on 80% of the cropland within the sub-basin. The terrace systems, shown on Figure 6, include approximately 21.6 miles of terraces, 15.1 miles of tile lines, about 250 tile inlets, and several major tile outlets. Contour strip cropping was used on 15% of the cropland acres. The remaining 5% of the basin’s cropland is bottomland with little sheet and rill erosion potential. The installed practices have resulted in the entire basin having a soil loss of less than “T”, the tolerable soil loss will not degrade the soil’s long-term productivity (NRCS Clayton County Field Office Files).

 

 

Monitoring System

Water quality has been monitored within the basin since 1981 (Rowden et al., 1995). In mid-1986, a standard U.S. Geological Survey gauging/sediment/precipitation station was installed to monitor precipitation inputs, surface-water discharge, and suspended sediment concentrations from the upper 736 acres of the watershed (Kalkhoff, 1989). Samples for sediment analysis were collected during runoff events, during the period 1988-1991, by an ISCO water sampler that was activated by changes in flow. The sediment samples were supplemented with periodic and event-related sediment samples collected by local observers. Difficulties were often encountered while utilizing the flow-activated sampler on this small stream, and therefore many of the event-related samples were collected by local observers (Steve Kalkhoff, U.S. Geological Survey, personal communication).

Hydrology

Table 2 summarizes hydrologic and sediment loss data for the basin for WYs 1990 and 1991 (Kalkhoff et al., 1992; Kalkhoff and Kuzniar, 1994; Rowden et al., 1995). The proceeding period, 1988 and 1989, was the driest two-year period in Iowa’s recorded history. Discharge from the basin was a rare occurrence from February 1988 through WY 1989, occurring only after large rains or snow melt periods. The dry antecedent conditions continued to affect discharge from the basin in WY 1990 (Rowden et al., 1995). While precipitation was above average, at 37.87 inches (average precipitation is 32 inches), discharge was 2.82 inches, or only 7% of precipitation. Measured sediment loss was about 45 tons, or 0.06 tons/acre. In contrast, precipitation was far above average in WY 1991, at 47.27 inches. Discharge totaled 16.65 inches, or 35% of precipitation. Much of the excess precipitation, about 13 inches, occurred during one extraordinary 24-30 hour rain in mid-June. The maximum instantaneous discharge measured during this event was 880 cubic feet/second, which has a calculated return interval of 25 years (O’Connell et al., 1991). Measured sediment loss for WY 1991 was 2,707 tons, or 3.7 tons/acre. Almost 80% of the measured sediment loss, 2,090 tons, occurred during the major mid-June runoff event; losses for the remainder of the year were equivalent to 0.85 tons/acre. Figures 7 and 8 show the discharge hydrographs for the basin for WYs 1990 and 1991, respectively.  Note that flow was rare during the first 8 months of WY 1990, again reflecting the dry antecedent conditions; measurable flow occurred during 44% of the water-year. In contrast, flow was essentially continuous during WY 1991, with the exception of a three-month period during the winter.

 

 

 

AGNPS Model

Eleven runoff events from 1990 and 1991 were selected for simulation with AGNPS (Young et al., 1987) version 4.02 (Young et al., 1994). These include virtually all events, occurring during these two years, with the exclusion of those affected by snow melt or data collection problems. Discharge and sediment data are also available for WYs 1986 through 1989. However, the main erosion control efforts in the basin were not completed until 1987, and 1988 and 1989 were periods of extreme drought, when little runoff occurred in the watershed. Therefore events from WYs 1990 and 1991 were chosen for the simulation. The events are indicated by inverted triangles on Figures 7 and 8. Gage and sediment data for these events are summarized in Table 3. A wide range of runoff conditions are included in these events. Precipitation varies from 0.5 to 14 inches; total measured runoff from about 90 thousand to almost 14 million cubic feet; and sediment loss from 1 to 2,100 tons (the latter figure is about 3 tons/acre).

Figure 9 shows the grid used for the AGNPS simulations. Eighty-three 10-acre cells were utilized, with 20 cells along the periphery of the basin subdivided into 2.5 acre sub-cells. The simulated area totaled 744 acres. Model inputs are listed in Appendix A. Table 4 summarizes the sources used for the model inputs.

 

 

Measured versus Modeled Results

 

TABLE 5 summarizes measured and modeled results for the selected runoff events; five events from WY 1992 are included.  Figures 10 and 11 are cross-plots of measured vs. modeled runoff and sediment, respectively (WY 1992 events are included on the runoff plot).  These plots are logarithmic, because both measured and modeled runoff and sediment values span several orders of magnitude.  For perspective, the plots show the 1:1 line (i.e., if modeled values equal measured values, they fall along this line).  While measured and modeled runoff volumes compare relatively well, one very anomalous point is evident in this data (Figure 10), from an event that occurred on 8/8/91 (Tables 3 and TABLE 5). Four and a half inches of rain fell in the Bugenhagen basin and generated only 0.05 inches of runoff.  The AGNPS-generated runoff, simulated under dry antecedent conditions, was 2.35 inches.  Other stream and rain gages in the area showed similar results as those measured at the Bugenhagen basin (Rowden et al., 1995).

 

 

An additional minor, but consistent, discrepancy between measured and modeled runoff volumes occurs with some very small (0.03-0.15 inches) runoff events (TABLE 5). AGNPS generated 0.0 to 0.01 inches of runoff for four of five of these small events (modeled zero runoffs are plotted as 0.01 inches on Figure 10).  This discrepancy may result from the fact that measured “runoff” volumes include some “baseflow”--shallow groundwater/tile drainage water--that AGNPS doesn’t account for.  For larger events, this baseflow component would become insignificant, relative to the event-related runoff volume.

Modeled sediment losses exceed measured losses for most events (Figure 11; the data point which shows a measured loss of 1 ton and a modeled loss of over 400 tons is from the anomalous 8/8/91 event).While some of this excess may result from aspects of AGNPS, or of this particular use of AGNPS, collection of the sediment data may have contributed to this result.As previously noted, automated sample collection at the Bugenhagen site--a small stream--was only intermittently successful; therefore, much of the sediment data was collected by local observers.Given the short duration of runoff--and therefore significant sediment loss--from a 740 acre watershed, the sediment samples may not have always been collected adequately across the “peak” of the hydrograph to capture the full magnitude of sediment losses.

Figure 12 shows measured runoff versus sediment loss.While sediment loss does increase with greater runoff Figure 12 shows the concerns with the measured sediment data. For example, several events generated 0.60-0.70 inches of runoff.The measured sediment loss with these events, which caused a similar amount runoff, varied from 2 to 254 tons. Figure 13 shows, for comparison, modeled runoff versus modeled sediment loss. The modeled results show a more constant increase of sediment loss with increased runoff, relative the measured results.

TABLE 5 also gives measured versus modeled peak discharges for events from WYs 1990 and 1992 (data for WY 1991 are not currently available). The peak discharge numbers compare poorly, with the AGNPS peak discharges typically exceeding the gauged peaks. Modeled peaks exceed measured peaks even for events where modeled runoff volumes are smaller than those measured 1/17-18/91, but the greatest discrepancies occur when modeled runoff is greater than measured (e.g., 11/1/91).

 

Statistical Analysis

Two types of statistical analyses were used on the measured and modeled data from this project.  First, the mean discharge and sediment yield measured at the U.S.G.S. gauging station and mean discharge and sediment yield predicted from AGNPS were compared using the non-parametric equivalent of a T-test.  The use of a non-parametric test was warranted given the small sample size (16 modeled precipitation events for runoff analysis; 11 modeled precipitation events for sediment analysis).   The non-parametric test chosen was the Wilcoxon signed ranks test (Hays, 1973).  The Wilcoxon test is the non-parametric equivalent of the paired T-test and accommodates data that is not statistically independent.  When the measured and AGNPS modeled discharge were compared by the Wilcoxon test the two-tailed p value was 0.212.Given this result, it appears that there is no significant difference between the mean discharge measured at the gage and the mean discharge predicted by the model. A Wilcoxon test of the measured and predicted sediment yield resulted in a two-tailed p value of 0.033.This result suggests that there may be a significant difference between the mean measured sediment yield from Bugenhagen basin and the mean sediment yield predicted by AGNPS.

Secondly, the measured and predicted values were compared using regression techniques.  Both the measured and predicted values for discharge and sediment yield show a strong correlation (Pearson correlation coefficient of 0.98 and 0.99 for discharge and sediment, respectively).  A linear regression of the modeled and measured discharge showed a significant linear relationship (R² =0.97, p=0.000).  A linear regression of the modeled and measured sediment yield also exhibited a significant linear relationship

(R² =0.99, p=0.000).

While measured and modeled runoff and sediment loss volumes show reasonable statistical relationships, summations of measured and modeled volumes for all events compare poorly (Table 6).  The total measured runoff from all 11 events was 8.2 inches, while AGNPS predicted 15.3 inches, or 187% of the measured runoff. Worse, the predicted sediment loss, 11,202 tons (15.2 tons/acre), was four times greater than the measured loss of 2,540 tons. Most of the discrepancy between measured and modeled results arise from two events: 1) the extremely large June 1991 rainfall, and 2) the previously discussed anomalous event of August 1991 (TABLE 5 Measured runoff from the 14-inch June 1991 event was 5.18 inches, while the modeled runoff was 11.19 inches. Measured sediment loss was 2,090 tons, compared to a modeled loss of 9058 tons. While there may be questions concerning how well AGNPS, or any watershed model, can reproduce extreme events, gauging such events is also problematic.Stream stage varies rapidly and at times far exceeds any established stage-discharge relationships. The June 1991 event at the Bugenhagen basin is itself an example, as measured precipitation and runoff volumes suggest the watershed “retained” almost 9 inches of rainfall.Collection of representative sediment samples may be physically impossible during such extreme events, during which significant erosion occurs.

Excluding the June and August 1991 events results in more comparable measured and modeled volumes (Table 6). The modeled runoff from the nine remaining events, 2.87 inches, is 97% of the measured runoff.  A discrepancy remains for the sediment loss volumes.  Modeled losses are 1,665 tons (2.26 tons/acre), which is 370% of the measured sediment loss of 449 tons (0.6 tons/acre).

 

 

Chemical and Nutrient Output

 

TABLE 7 summarizes the chemical and nutrient outputs from the AGNPS simulations for Wys 1990 through 1992.  Water quality analyses are available for comparison with AGNPS outputs, for several parameters, for five events; these are compared on Table 8. Samples were analyzed for commonly-used herbicides, and for ammonia-and organic-nitrogen. Samples were not collected for phosphorus or COD analysis. Two of the most commonly used herbicides in the area, atrazine and alachlor, were simulated with AGNPS.

Measured ammonia-plus organic-nitrogen concentrations are far greater than modeled soluble nitrogen concentrations.  However, some of the measured nitrogen may be what AGNPS would consider “sediment” nitrogen, so the measured and modeled results may not be comparable.  Modeled and measured atrazine and alachlor concentrations are typically within a order of magnitude of each other.  Note that for both nitrogen compounds and herbicides, an insufficient number of samples were collected  to calculate an average “event” concentration.  Typically, only one sample per event was collected.  Concentrations in small watershed vary rapidly across runoff events (Hallberg et al., 1984), and numerous samples would be needed  to reasonably calculate an average concentration for comparison with AGNPS outputs.

 

Comparison of Pre-and Post Erosion Control Simulations

 

 

TABLE 9 summarizes the results of AGNPS simulations of the WY 1990-1992 rainfall events under pre-and post-erosion control conditions (“pre-and post”).  Simulated  post-erosion control runoff volumes and peak discharges are show reductions relative pre-erosion control conditions, particularly for smaller events. The simulations also indicate significant sediment loss reductions as a result of the implementation of erosion control practices (Figure 14).  Table 10 totals pre-and post sediment losses for all rainfalls, and for all rainfalls excluding the extreme 6/14/91 event. The model suggests a 67% reduction in sediment loss, totaling over 22,000 tons, or 30 tons/acre, for all rainfall events during the three-year period.  If data from the extreme 6/14/91 event are excluded, a 53% reduction in soil loss is indicated, from 4,555 tons to 2,144 tons, or from 4.0 to 2.9 tons/acre.

Average annual sheet and rill erosion (S & R erosion) were estimated for the watershed, using the Universal Soil Loss Equation (USLE) prior to implementation of erosion control practices (NRCS-Clayton CountyThese estimates were made for an 899-acre parcel of the basin (Big Spring Demonstration Project, unpublished data).  Adjusting these to the 736-acre modeled area indicates an estimated decline in average annual S & R erosion of 64%, from 8,900 to 3,230 tons, or from 12 tons/acre to 4.4 tons/acre (Table 10).

 

 

The estimated soil loss and S & R erosion from AGNPS and USLE, respectively, compare relatively well.  For all AGNPS-modeled events, an average annual pre-implementation loss of 15.2 tons/acre is suggested, compared to 12 tons/acre of S & R erosion  for USLE.  Post-implementation losses from AGNPS are 5.0 tons/years, while USLE suggests S & R erosion of 4.4 tons/year.  Note that the estimates provided by AGNPS and USLE are not strictly comparable. AGNPS simulates soil loss past a point in a watershed. This estimate includes sheet and rill erosion and channel erosion, minus sediment deposited within the watershed.  USLE estimates sheet and rill erosion.  Runoff events associated with snow melt, which are often quite significant, were not simulated with AGNPS.  In addition, the USLE estimates are for a climatically “average” year.  However, the estimates provide a perspective of soil  losses from a small watershed, particularly regarding the range of losses that occur year-to-year, as compared to estimates for erosion during a climatically “average” year.

 

Summary

Eleven runoff events that occurred in the 736-acre Bugenhagen basin during WYs 1990-1991 were simulated using AGNPS version 4.02.  Modeled results were compared to runoff and sediment volumes measured by a standard U.S. Geological Survey gauging/sediment station.  Individual and summed modeled and measured runoff volumes compared relatively well, with a notable exception during an extremely large event. Modeled sediment losses typically exceeded measured sediment losses.  This may result from an inadequate number of sediment samples being collect collections during the short-duration runoff events that characterize small drainage basins.  Peak discharges estimated by AGNPS compare poorly with measured peaks, and are typically larger.

Simulations of pre-versus post-erosion control implementation conditions suggest the erosion controls have resulted in a 65% in reduction in soil loss for non snow-melt runoff events.  Pre-and post sediment losses estimated by AGNPS for the simulated events compare relatively well with average S & R erosion calculated with the USLE.

  

References

 

 

Hallberg, G.R., Hoyer, B.E., Bettis, E.A., III, and Libra, R.D., 1983, Hydrogeology, water quality, and land management in the Big Spring basin, Clayton County, Iowa: Iowa Geological Survey, Open-File Report 83-3, 191 p.

 

Hallberg, G.R., Libra, R.D., Bettis, E.A. III, and Hoyer, B.E., 1984, Hydrologic and water-quality investigations in the Big Spring basin, Clayton County, Iowa: 1983 Water-Year: Iowa Geological Survey Open-File Report 84-4, 231 p.

 

Hays, W.L., 1973, Statistics, Holt, Rinehart, and Winston, Chicago, IL, Third addition, pp. 590-591.

 

Kalkhoff, S.J., 1989, Hydrologic data for the Big Spring Basin, Clayton County, Iowa, Water Year 1988: U.S. Geological Survey Open-File Report 89-230, 44 p.

 

Kalkhoff, S.J., Kuzniar, R.L., Kolpin, D.L., and Harvey, C.A., 1992, Hydrologic data for the Big Spring Basin, Clayton County, Iowa, Water Year 1990: U.S. Geological Survey Open-File Report 92-67, 80 p.

 

Kalkhoff, S.J. and Kuzniar, R.L., 1994, Hydrologic data for the Big Spring Basin, Clayton County, Iowa, Water Year 1991: U.S. Geological Survey Open-File Report 94-56, 87 p.

 

Littke, J.P., and Hallberg, G.R., 1991, Big Spring basin water-quality monitoring program; design and implementation: Iowa Department of Natural Resources-Geological Survey Bureau, Open-File Report 91-1, 19 p.

 

O’Connell, D.J., Lambert, R.B., Matthes, W.J., and Sneck-Fahrer, D., 1991, Water resources data for Iowa water year 1991: U.S. Geological Survey Water-Data Report IA-91-1, 385p.

 

Rowden, R.D., Libra, R.D., and Hallberg, G.R., 1995, Surface water monitoring in the Big Spring basin, 1986-1992: A Summary Review: Iowa Department of Natural Resources-Geological Survey Bureau, Technical Information Series 33, 109 p.

 

U.S. Department of Agriculture-Soil Conservation Service, 1982, Soil Survey of Clayton   County, Iowa, 238 p.

 

Young, R.A., Onstad, C.A., Bosch, D.D., and Anderson, W.P., 1987,  AGNPS, Agricultural non-point source pollution model, a watershed analysis tool: U.S. Department of Agriculture-Agricultural Research Service, Conservation Research Report 35, 77 p.

 

Young, R.A., Onstad, C.A., Bosch, D.D., and Anderson, W.P., 1994, Agricultural non-point source pollution model, version 4.02: AGNPS user’s guide: U.S. Department of Agriculture-Agricultural Research Service.