Assessment of runoff and sediment yield using remote sensing, GIS, and AGNPS.
Watershed models can be valuable tools to aid in the study of the relationships between watershed conditions and the quantity and quality of water in the watershed. Models are being used for watershed management, planning and development, and best management practice (BMP) implementation. Acquiring detailed and accurate field data is one of the most important ways to improve the prediction capabilities. Recent use of remote sensing and GIS in modeling can simplify the process and add confidence in the representation of watershed conditions (Jensen 2000; Basnyat et al. 2000; Liao and Tim 1997). Hence, combining the capabilities of remote sensing, GIS, and hydrologic data analysis will improve the prediction capabilities of the hydrologic models. It is also important to evaluate the capabilities of these processes to estimate runoff and sediment outputs accurately during various storm events.
In 1998, the Kansas Water Office and Cheney Reservoir Water Quality Project (CRWQP) Office (Natural Resources Conservation Service (NRCS) in Reno County) initiated a research project, Evaluation of AGNPS in Cheney Lake NonPoint Source Management, because of growing concern about overloading of sediment and nutrients in the Cheney Reservoir, a major source of drinking water for the city of Wichita, Kansas. The researchers selected event-based model AGNPS version 5.0, which can easily be used to study the effects of the changing conditions in a watershed (Tim et al. 1995, Lee and White 1992). Current conditions can also be evaluated and compared with results from past or future scenarios for the watershed (Mankin et al. 1999) The AGNPS model has been applied in number of watersheds in Kansas to study the impacts of different landcover scenarios (Koelliker and Humbert 1989a, 1989b; Mankin et al. 1999). However, the model results have never been assessed against the measurements taken in the field. The AGNPS-ARC/ INFO interface (Liao and Tim 1997) used in this study was designed to organize data from a variety of digital sources, including landcover, soils, elevation, and hydrography and then to create an AGNPS input file. Therefore, the interface would enable us to evaluate the process of obtaining landcover information about watersheds from satellite images. In addition, the interface enables users to study the critical areas in the watershed and the effect of changing landcover scenarios for BMP study easily and efficiently. During the time of this study the annualized version of AGNPS was still in a beta version.
The application of the Soil Conservation Service Curve Number (SCS-CN) method (SCS 1968) in the AGNPS model with improved estimation of antecedent moisture condition (AMC) ratios is another important aspect of this research. The SCS-CN method is one of the most popular methods to estimate surface runoff from watersheds. Its ease of application and lack of competition has resulted in a dramatic expansion of the scope of its application in hydrologic modeling (Hjelmfelt 1991). Watershed condition has been defined by AMC with three numbers, I, II, and III, with AMC II considered an average watershed condition. The relationships among the CNs in these conditions were shown as discrete, not continuous, thus implying sudden shifts in CN (Hawkins 1978). Therefore, using an average AMC II condition regardless of the watershed's condition and location may not be reasonable and could substantially underestimate or overestimate the runoff from the watershed. Accurately estimating AMC and using reasonable CNs for watersh eds would improve the runoff estimation in the SCS-CN method and the AGNPS model. This would subsequently affect the estimation of peak discharge, sediment yield, and nutrient loading from the watershed.
Research objectives. The objectives of this study were:
1. To develop a modeling process using remotely sensed data, GIS, and AGNPS to assess runoff and sediment yield from various sub-watersheds of Cheney Reservoir watershed;
2. To evaluate the method of obtaining watershed landcover information from satellite images. Use of satellite images for discriminating cropland with different residue cover and rangeland with different cover amounts was another important aspect;
3. To estimate AMC ratios for the watershed and adjust CN values of various landcover classes in the AGNPS model during various runoff events; and
4. To evaluate the modeling process as a "post-season assessment" of selected storm events against measured data of stream flow and suspended sediment on several sub-watersheds of the Cheney Reservoir watershed.
Satisfying these objectives, the results of this project will help the CRWQP office develop its action plan for land-use decisions, management strategies, and adoption of BMPs within the watershed.
The AGNPS model. The AGNPS model is a single-event, distributed-parameter model developed by scientists and engineers at the U.S. Department of Agriculture's Agricultural Research Service (Young et al. 1987, 1989, 1994). It can be used to estimate surface runoff, soil erosion, sediment yield for different particle sizes, and associated chemical properties from an agricultural watershed. It accepts inputs from point pollution sources, such as feedlots and wastewater treatment plants, and from nonpoint sources generated from landcover types to estimate the overall pollutants. The distinct feature of this model is that it divides the watershed into small, discrete, square cells to represent the effects of agricultural management practices on sediment and nutrient loading. Each cell is characterized by 22 input parameters, including cell number, cell division, receiving cell number, aspect/flow direction, CN, slope, slope shape, slope length, Manning's roughness coefficient, soil erodibility factor (K-factor), C- factor, conservation practice factor (P-factor), surface condition constant, soil texture, fertilizer indicator, fertilizer availability, pesticide indicator, point-source indicator, gully indicator, additional erosion, impoundment indicator, and channel indicator.
Sediment yield is calculated from a modified form of the USLE (Wischmeier and Smith 1978), and runoff volume is calculated by the SCS-CN method (SCS 1968). Peak runoff rate for each cell is estimated using an empirical relationship proposed by Smith and Williams (1980). The chemical transport part of the model estimates the chemical properties throughout the watershed. Relationships used to calculate pollutant levels are taken from the Field-scale Model for Chemicals, Runoff, and Erosion from Agricultural Management Systems (CREAMS) (Smith and Williams 1980).
AGNPS-ARC/INFO interface. The AGNPS-ARC/INFO interface has four functional modules: data generation, input-file creation, program execution, and AGNPS output-file extraction. A detailed description of these modules can be found in Liao and Tim (1997) and Liao (1996, 1997). The data-generation module creates a fishnet coverage for the entire watershed. The fishnet includes AGNPS required parameters, which are extracted from various GIS data layers. The input-file creation module creates an AGNPS input file from the fishnet. The AGNPS input file can be then run by the program execution module, which creates the output file. The output-file extraction module can be used to extract, analyze, and display the model results.
The AGNPS-ARC/INFO interface has been used in few earlier studies to test its applicability (Liao and Tim 1997) and to study the impacts of BMPs in the watershed (Tim et al. 1995; DeAussen et al. 1998).
The AGNPS model has also been used in various studies to find critical areas and effects of BMPs on the water quality of watersheds (Young et al. 1989; Hession et al. 1989; Prato and Shi 1990).
Previous study with AGNPS. A few researchers have used AGNPS for water--quality modeling to evaluate the impacts of agriculture in Kansas. Koelliker and Humbert (1989a, 1989b) applied the AGNPS model in five Kansas watersheds to evaluate the sediment and nutrient outputs with the present landcover conditions. They also evaluated various landcover scenarios appropriate for these watersheds. The landcover information for this study was gathered from aerial photography and a reconnaissance survey. The model-predicted results were compared with the study done on sedimentation in small reservoirs in Kansas by Holland (1971). The predicted results compared quite well with the values recommended by Holland.
Mankin et al. (1999) also reported the application of the AGNPS model in the Melvern watershed in Kansas with a large area of 900 sq km (347 sq mi). They applied the annualization procedure (Koelliker and Humbert 1989b) with the AGNPS--estimated results of 15 various storms to estimate the annual loadings in Melvern Lake with various land-use scenarios. The landcover information for the watershed was also gathered by a combination of field reconnaissance and 1:4,800-scale aerial photographic maps from each county of the watershed.
Lo (1995) investigated the Tsengwen watershed in Taiwan to estimate the yearly sediment yield. The estimated sedimentation rate was calculated to be 4.2 mm/yr (0.2 in /yr) compared with the measured of sedimentation rate of 6.3 mm/yr (0.3 in/yr). Perrone and Madramootoo (1997, 1999) calibrated and then validated the AGNPS model developed for the St. Esprit watershed in Quebec, Canada. For calibration, they adjusted the curve number, C-factor, and hydrograph shape factor to estimate runoff, peak runoff rate, and sediment yield. The coefficient of performance (CP), as defined by the following equation, was used to compare measured and estimated values (James and Burgess 1982): CP = [summation over (n/i=1)] [[M(i) - E(i)].sup.2]/[summation over (n/i=1)] [[M(i) - [M.sub.mean]].sup.2], (1)
CP = [summation over (n/i=1)] [[M(i) - E(i)].sup.2]/[summation over (n/i=1)][[M(i) - Mmean].sup.2], (1)
where M(i) is the ith measured value, E(i) is the ith estimated value, and [M.sub.mean] is the mean of the measured values. CP approaches zero as estimated approach measured values. During calibration, the range of CP and percent error values were to be 0.05 to 0.43 and 6.2% to 44.3%, respectively However, during validation, CP and percent error values ranged from 0.01 to 2.07 and 21.7% to 117.8%, respectively.
Landcover classification using TM imagery. Jensen (2000) discussed the use of remote sensing for identifying nonpoint-source pollution potential through the discrimination of landcover types and integrating these data with GIS modeling in the AGNPS model. Jensen indicated that modeling with remote sensing and GIS offered a method of identifying nonpoint-source pollution potential in a way that is easy for the general public to understand and apply Schultz (1988) discussed the potential use of satellite imagery in hydrological modeling for discrimination of landcover classes based on reflectance characteristics. King and Delpont (1993) also discussed the importance of using remote-sensing data to assess the spatial and temporal variability of certain factors that increase the potential for soil erosion. DeGloria et al. (1986) studied monitoring land area and spread of conservation tillage practices (with stubble residue cover) using Landsat MSS (multi-spectral scanner) data in San Luis Obispo County, Californi a. A classification accuracy of 81% suggests that the use of Landsat imagery can he used effectively to monitor conservation tillage. In addition, McNairn and Protz (1991, 1993) illustrated that Lansdat TM imagery can provide measurements of crop residue, which is important because residue affects the amount of runoff, sediment, and nutrients that run off agricultural fields.
Methods and Materials
Description of the study area. The Cheney Reservoir watershed is in south central Kansas and covers a 2,404 sq km (928 sq mi) area within Reno, Kingman, Pratt, Stafford, Kiowa, and Sedgwick counties (Figure 1). More than 99% of this rural watershed is agricultural land. It contains 1,000 farms with a population of 3,300. About 41% of the watershed is rangeland, primarily tall-grass prairie. Cropland covers 58% of the watershed, and 0.5% is residential area. Farming practices within the watershed vary greatly, from small dairies and diversified crop and livestock farms, to rangeland and large cropland acreages under center-pivot irrigation (personal communication with CRWQP Office 1999). Average rainfall of this sub-humid watershed is 695 mm/yr (27.4 in/yr), with substantial year-to-year variability. The USGS has divided the watershed into five sub-watersheds. They are:
ub-watersheds USGS gaging Drainage area, station sq km (sq mi) 1. West Ninnescah sub-watershed 07144601 1,223 (472) 2. Silver Creek sub-watershed 07144660 497 (192) 3. Goose Creek sub-watershed 07144680 130 (50) 4. Red Rock Creek sub-watershed 07144730 145 (56) 5. East Ninnescah sub-watershed 07144780 419 (162)
The Cheney Reservoir was constructed between 1962 and 1964 by the U.S. Bureau of Reclamation. The design incorporates a 100-year project that acts as a water supply for the city of Wichita, as well as provisions for wildlife/recreation area and flood control. Wichita draws 40% to 60% of its daily water supply from the reservoir. Thus, this modeling process is important to provide estimates of the sediment load entering a major water-supply reservoir.
The following sections describe preparation of the various GIS data layers, including the satellite image classification procedure to obtain the landcover layer. These GIS layers are required by the data-generation module of the interface to extract the AGNPS model parameters and then to generate input files. After creating the GIS layers for the entire watershed, the data layers for sub-watersheds were cut out using their boundary layers.
Creating and editing GIS layers. Landsat TM images were obtained for Aug. 1, 1997, and July 19, 1998. Advance processing of the satellite images involved registration of the 1997 image to an ARC/INFO land-use coverage (UTM coordinate system, zone 14, datum NAD27, scale 1:4,800) provided by the CRWQP Office. The 1998 image was then registered to the rectified 1997 image. The study watershed was then cut out, using the Cheney Reservoir watershed boundary.
A combined unsupervised and supervised classification procedure was used to discriminate the landcover classes for the watershed using a procedure described in Marzen et al. (2000). Landcover data is used to provide input parameter values for AGNPS, assigned for C-factor, CN, surface condition constant, overland Manning's coefficient, COD, and fertilizer application levels (Young et al. 1994).
An unsupervised Iterative Self-Organizing Data Analysis Technique (ISODATA) (Jensen 1996) algorithm was used to group pixels with similar reflectance characteristics into 30 statistical clusters. The clusters were grouped into four initial landcover classes: rangeland, water, and two classes of cropland (wheat [Triticum aestivum L.] stubble and nonwheat areas). Wheat (a cool-season crop) and nonwheat (warm-season crops) areas were discriminated based on a crop calendar. The summer image was created after wheat had been harvested and when other crops (primarily corn [Zea mays L.], sorghum [Sorghum bicolor (L.) Moench], and soybean [Glycine max (L.) Merr.]) were in the prime of the growing season. A supervised classification was then used to add both an urban/built-up class and a class for woodland. Sample areas of known woodland and urban cover were obtained using information provided by the CRWQP office. A base landcover layer consisted of six classes, including water, other crops, wheat stubble, rangeland, b uilt-up, and woodland. Accuracy of this classified image was assessed against a 1997 landcover layer provided by the CRWQP office that had been produced over 6 months using digital ortho-photography. Overall classification accuracy was 89.8%, with the woodland and water class accounting for the greatest error. In some places, it was evident that the classified image picked up more detail than the generalized landcover layer provided by the CRWQP office.
To increase the landcover detail for rangeland and wheat-stubble areas, a masking and cluster-busting procedure was used (Jensen 1996). The rangeland and wheat-stubble classes were isolated from the classified image for this purpose. To subdivide the rangeland class, a normalized difference vegetation index (NDVI) was calculated. NDVI uses the reflectance in the infrared and red bands to estimate vegetation cover (Lyon et al. 1998). Pixels with low NDVI numbers indicate areas of relatively low vegetative cover, and pixels with high NDVI numbers have greater vegetation amounts. A histogram of the NDVI values and ground-based observations were used to help determine thresholds of low, medium, and high amounts of vegetative cover for rangeland areas. A rangeland specialist from the CRWQP office provided information on range cover for a sample of 43 fields. The locations of these 43 sites were received as an ARC coverage, and NDVI values of the pixels within each polygon were checked against field observations. B ased on this analysis, the NDVI values were subdivided into three rangeland classes for low, medium, and high amounts of cover.
Wheat stubble was also separated into low, medium, and high cover based on satellite estimates of the amount of crop residue left in the field after harvest. NDVI has not been shown to differentiate different levels of cover for senescent vegetation. Some researchers have studied the use of some bands of Landsat TM image to separate the amounts of crop residue (Biard and Baret 1997; Leblon et al. 1996). Our study, however, involved the use of a cluster-busting procedure (Jensen 1996) plus ground truth to differentiate between wheat stubble. The wheat-stubble image was grouped into 30 clusters using the ISODATA unsupervised algorithm. Ground-based data for percent residue covers were used to differentiate wheat stubble into estimates of low, medium, and high cover amounts. These field data were also received as ARC polygons and were cross-referenced against reflectance values to assist in determining the cover amounts for the cluster-busting classification procedure.
The cluster-busted images for rangeland and wheat stubble were finally overlaid with the original classified image to produce a GIS coverage with 10 landcover classes. This landcover layer included high, medium, and low cover classes for wheat stubble and rangeland, other cropland (including corn and sorghum), woodland, water, and residential (Figure 2).
Additional GIS coverages required for AGNPS-ARC/INFO interface include soils, elevation, conservation measures, streams, locations of point sources, and impoundments. The Certified Soil Survey Geographic (SURGO) soil coverages and soil database files for different counties of the watershed were obtained from the Kansas Data Access Support Center (DASC). The values of the K-factor, hydrologic soil group, and texture were then added as attributes to the polygonattribute table of the soil coverage for each map unit identification symbol. USGS digital elevation model (DEM), 30 m (98.4 ft), files were converted to a triangular irregular network (TIN) as required by the interface for extracting slope, length of slope, and flow direction for streams and overland flow (Liao 1997). A coverage showing the location of conservation practices, obtained from the CRWQP office, was combined with the landcover layer. USLE P-factors were then assigued to the attribute table for each land-- cover class based on expert opinion f rom the CRWQP office. The interface requires the boundary and stream coverages to be converted to a grid with user-desiguated cell size. For this study, a grid cell size of 65 ha (160 ac) was selected to produce input files for AGNPS. The information on feedlots and wastewacer treatment plants were collected from a field survey conducted by the CRWQP office. Basic information regarding soil, channel characteristics, and fertilizer application were given in the form of lookup tables (INFO files) as required by the interface to make the AGNPS input file (Young et al. 1994; Liao 1997). This original information was also obtained from a field survey of the watershed.
Development of GIS layers for cropping management factor (C-factor). C-factor is one of the most sensitive parameters in the AGNPS model. It changes with tillage and management practices applied in the watershed throughout the year (Young et al. 1987). Because the AGNPS model is an event-based model that was used to estimate the runoff and sediment yield during different runoff events, it was necessary to generate appropriate C-factors for the watershed when the selected events occurred. Values for C-factors of cropland throughout the year were provided by the CRWQP office (Frees 2001). These C-factors were generated using the Revised Universal Soil Loss Equation (RUSLE) (Renard et al. 1991) program on the basis of tillage and management practices applied on wheat and other cropland areas during 1997 and 1998. The values of C-factors in other cropland areas were estimated as an average for corn, sorghum, and soybean fields. Based on the information from the CRWQP office, it was determined that variation in C- factor values for all noncropland classes was not substantial during the period considered for this study, and therefore C-factor values for these classes were kept constant. As a result, several landcover layers were created with appropriate C-factor input parameters for the runoff events selected.
Rainfall data collection and estimation of energy intensity (EI) values. Daily rainfall data from 11 weather stations located within and near the Cheney Reservoir watershed were obtained from the Weather Data Library at Kansas State University and from the CRWQP office. Of the 11 weather stations, only four were inside the watershed (Figure 1). Average rainfall over a sub-watershed was computed using the Thiessen Polygon Method (Thiessen 1911), considering the weather stations, which were within and near the sub-watershed. Only daily rainfall data at the weather stations were available. To estimate El values for each storm, the probability method developed by Koelliker and Humbert (1989a, 1989b) was followed. This method was also applied in the water-quality assessment study in Melvern Lake, Kansas (Mankin et al. 1999).
In that study, rainfall amounts for a 24 hr storm with return periods of 100-, 50-, 25-, 20-, 10-, 5-, 2-, and 1- year were obtained from Hershfield (1961) for the watershed area and were assumed to follow a log-normal relationship. The total annual average rainfall for the area was 695 mm/yr (27.4 in/yr), which was computed from the historical rainfall data of the weather stations between 1960 and 1997 (rainfall data were obtained from the National Oceanic and Atmospheric Administration). The rainfall amount for the 200-year storm was assumed to complete this probability curve and was extrapolated from the curve. The rainfall amounts for 0.5-, 0.25-, 0.10-, 0.05-, and 0.025-year storms were then developed based on analysis of normal precipitation at recording stations in the area. A storm of 0.0125-year (80 days per year) was taken to be 0 mm to complete the area under the probability curve on the lower end (Mankin et al. 1999).
Similarly, for estimation of EI values, the EI values of 20-, 10-, 5-, 2-, and 1-year probability storms for the nearest station to the watershed were used (Wischmeier and Smith 1978). For the Cheney Reservoir watershed, the total annual EI value is 2,861 MJ/ha-mm/hr (165 ft-tons/ac-in/hr) (Wischmeier and Smith 1978). The relationship between EI and probability also follows a log-normal distribution. Accordingly, EI values for 200-, 100-, 50-, and 25-year and for less than 1-year probability storms were estimated from this distribution. The EI value for the 0.05-year storm was assumed to be 0.0 (Mankin et al. 1999).
The EI value for each individual storm was estimated by assuming that the probability for the EI value had the same probability of the rainfall event. The relationship between rainfall and EI values could be represented by the following equation with a correlation coefficient of 0.9943 (Figure 3):
EI = 0.026 [P.sup.2.450] (2)
where EI is the energy intensity value, MJ/ha-mm/hr, and P is the amount of rainfall, mm.
Identification of different storms for the AGNPS run. Storms chosen for running the AGNPS model for each sub-watershed were all those with surface runoff hydrographs which gave at least 0.25 mm (0.01 in) of runoff and for which measured sediment data were also available from the USGS. These storms were sufficient to total about 75% of the total annual surface runoff from each sub-watershed.
Observed stream-flow and water-quality data were obtained from the five USGS gaging stations installed within the watershed (Figure 1). The East Ninnescah sub-watershed was not considered for this study because stream-flow and water-quality data that represented only this area were not available.
Best estimates of surface runoff quantity and quality for different storms. Base-flow values were separated from the total stream flow to get best estimates of surface runoff amount from the daily stream flow record using the USGS Hydrograph Separation Program (HYSEP 2.2) model (Sloto and Crouse 1996). Daily base flow was taken as an average of base flow, estimated using the three methods of fixed interval, slide interval, and local minimum of the HYSEP 2.2 program. For each event, the surface runoff portion was summed from the date of storm until the surface runoff was less than 0.025 mm/day (0.001 in/day) to get the total surface runoff for that event. When two runoff events occurred on consecutive days, it was difficult to separate surface runoff hydrographs from each day because the precipitation was available only as daily totals. In such cases, both the storm and the surface runoff from it were considered a single event. The separated daily surface-flow amounts were then converted into depth of surface runoff water by dividing by drainage area. The gaging station of Red Rock Creek sub-watershed has a drainage area of 137 sq km (53 sq mi) (Putnam et al. 1998) and was used to compute the measured surface runoff depth. Accordingly, the AGNPS model outputs were observed for this location of this sub-watershed to compare with the measured data.
The daily concentration values of TSS at different USGS gaging stations were translated into pollutant yield. We assumed that the TSS concentration in the base flow is negligible. Initially, the daily total volume of water was calculated. We multiplied this water volume by concentration values (mg/1) and then divided by drainage area at the gaging station. The pollutant yield, thus obtained, could be directly compared with the AGNPS output (Figure 4). An example for computing the depth of surface runoff and pollutant yield of TSS from daily stream flow and TSS concentration at the Red Rock Creek gaging station is shown in Table 1 for a runoff event.
Sediment-loss data available from the USGS was reported as TSS. However, the most significant part of this TSS is suspended sediment (USGS 2000). Therefore, in the absence of sediment data, AGNPS-estimated amounts of clay, silt, and small aggregate portions were considered to be the best estimates of the sediment loss for comparison with the reported TSS. For the conditions in these sub-watersheds, AGNPS model estimates that the clay, silt, and small aggregate portion of sediment losses average about 85% of the total sediment.
Runoff curve number coverages. The AGNPS model requires the CN values of various landcover to compute surface runoff amounts. These CN polygon coverages, required by the interface, for the sub-watersheds during various runoff events were made by combining the landcover and soil coverages. The CN for AMC II are available from several sources for landcover classes and hydrologic soil groups (SCS 1968; Koelliker et al. 1981; Novotny and Olem 1994). The growth of canopy over the growing period for different agricultural crops (wheat with low, medium, and high residue, plus other crops including corn, sorghum, and soybean) was provided by the CRWQP office. We assumed a linear relationship between the growth of canopy cover and the reduction of CN, and we developed the CN values during different storm events for each agricultural crop. The separation between other crops could not be made in this study. Therefore, the CN values considered for the other cropland area were the average for three crops (corn, sorghum, a nd soybean). Apparently, the variation of these CN values was between that for bare cropland and that for the full-grown crops. Variations in the CN values for the rangeland, woodland, and residential areas were, however, kept constant during the period of storm events considered for this study.
These assigned CN values were then adjusted for individual storms depending on the hydrologic condition or AMC before each storm event in the AGNPS input file and described in the following section. Note that the AGNPS-ARC/INFO interface computes and assigns an average CN value ([CN.sub.II] for AMC II), to each 65 ha (160 ac) cell in the input file.
Antecedent moisture condition (AMC) and curve number (CN) adjustment. AMC and CN are two very important parameters in the SCS-CN equation that affect the runoff depth for a storm. By the SCS-CN method, the AMC of a watershed is determined by the five-day antecedent rainfall amount (SCS 1968). AMC is divided into three categories: AMC I for dry, AMC II for average, and AMC III for wet conditions. In this study, we estimated the AMC ratios of the watershed during each storm event from the measured runoff, rainfall, and average curve number of the watershed. Then we changed CN values in the AGNPS input files based on the estimated AMC ratios. The process is described briefly below.
After assigning the [CN.sub.II] to each polygon in the runoff CN coverage, the [CN.sub.av] of each sub-watershed was also computed by weighting by the area of each polygon. The actual CN ([CN.sub.actual]) of each sub-watershed to produce the measured amount of runoff depth during each rainfall event was then determined by SCS-CN equation (SCS 1968; McCuen 1982). This [CN.sub.actual] may be between AMC I and II or between AMC II and III (Figure 5).Thereafter, the AMC ratio was estimated (AMCratio) using Equation 3 or 4, whichever was applicable, on the basis of [CN.sub.actual], [CN.sub.av], and CN for AMC I ([CN.sub.I]) or III ([CN.sub.III]). Note that the change in CN is not equal between AMC I and II range and AMC II and III range. Therefore,
for, [CN.sub.actual] < [CN.sub.av]
AMCratio = 1 + ([CN.sub.actual] - [CN.sub.I])/([CN.sub.av] - [CN.sub.I]) (3)
and for [CN.sub.actual] > [CN.sub.a]
AMCratio = 2 + ([CN.sub.actual] - [CN.sub.av])/([CN.sub.III] - [CN.sub.av]) (4)
This AMCratio was then used to adjust all of the CN values in the AGNPS input file for each event by linear interpolation between [CN.sub.II] and the appropriate [CN.sub.I] or [CN.sub.III] for each land use.
The schematic diagram for the modeling set-up process is shown in Figure 6. For calculation of peak rate of flow, SCS TR55 method (USDA 1986) was used. A default value of 484.0 was used as the hydrograph shape factor value for calculation of the triangular hydrograph (Young et al. 1994).
Comparison between measured and AGNPS-estimated values. For comparison of measured and model-estimated values for this study, CP (as defined earlier) and estimation difference (ED), as defined below, were calculated.
ED = \[summation over (n/i=1)] M(i)-[summation over (n/i=1)] E(i)/[summation over (n/i=1)] M(i)\ 100. (5)
Terms used in this equation were defined earlier. This ED is similar to the "percent error" used by Perrone and Madramootoo (1997,1999). The ED, for the same amount of difference between an observed value and estimated value, becomes larger in cases of overestimations compared to values with underestimations (because of a smaller value in the denominator).
The CP will be important to evaluate the event-to-event assessment capability of this modeling process. In terms of water quality of Cheney Reservoir, the total annual pollutant loading estimation also plays an important role for adopting the best management practices in the watershed. Therefore, ED will also be an important parameter to evaluate the overall performance of this modeling process.
Results and Discussion
Change of landcover in the watershed. Remote sensing data enables modelers to see the various landcover practices and variation over time in the watershed. Landcover is one of the most important pieces of information that a water-quality modeler needs. Results from the landcover classifications made from the Landsat images of 1997 and 1998 in the Cheney Reservoir watershed are presented in Table 2. This classification showed an increase in wheat area by 5% and decrease in other crop area by 6% during 1998. The total rangeland area during 1998, however, increased by 1% from the previous year. Changes in the percent of landcover classes during these years in the different sub-watersheds are shown in Table 2. Overall, the total area in rangeland and cropland remained about the same in all sub-watersheds. Red Rock Creek sub-watershed showed an increase in rangeland area of 1% and a decrease in cropland area of 2%. This change would decrease the runoff and soil loss potential of the sub-watershed somewhat, if all other conditions were constant. In the case of Goose Creek and Silver Creek sub-watersheds, a very small decrease in rangeland area was observed. The change in areas covered by the other classes of residential, woodland, and water were not substantial.
Water- quantity data analysis. Total depths of annual stream flow amounts for different gaging stations are shown in Figure 7. Total stream flow is the sum of base flow and separated surface flow. The average separated base flow was between 37% and 74% of the total annual stream flow. The smaller sub-watersheds, Red Rock Creek and Goose Creek, had a smaller portion of base flow of 37% and 55%, respectively. Thus, the total amount of separated surface runoff depth were 89 mm (3.5 in) and 78 mm (3.1 in) in Red Rock Creek and Goose Creek sub-watersheds, respectively, during these two years. On the West Ninnescah sub-watershed, the portion of the base flow in total flow was 74%. This is due to its soil texture and the relatively flat topography.
We observed that the runoff was not a linear function of rainfall. This was due to the effect of several parameters, including landcover, soil texture, and hydrologic condition of the watershed. For example, the largest storm on the Red Rock Creek sub-watershed was 51.8mm (2.0 in) on June 24,1997, which produced about 6.1 mm (0.2 in) of runoff. And the largest storm of 67.3 mm (2.6 in),which occurred in Goose Creek sub-watershed on September 23, 1997, produced only 2.1 mm (0.1 in) of runoff. However, as per the SCS-CN method, these storms with [CN.sub.av] for AMC II of 78 (as estimated) for the Red Rock Creek sub-watershed and [CN.sub.av] of 77 (as estimated) for the Goose Creek sub-watershed would produce about 12 mm (0.5 in) and 21 mm (0.8 in) of runoff, respectively, if AMC II were assumed. Therefore, estimation of continuous AMC ratio and adjustment of CN values during different storm events were necessary to improve runoff estimation.
The AMC ratios estimated during 67 storms that occurred in the five sub-watersheds were within the range of 0.4 to 2.6, the average being 1.5.These values are presented only for Red Rock Creek and Goose Creek sub-watersheds in Tables 3 and 4. This indicates that for water-quality modeling purposes for a sub-humid watershed in this area of Kansas, an average AMC ratio of 1.5 may be more appropriate.
Comparison between measured and AGNPS-estimated values in various sub-watersheds. Red Rock Creek sub-watershed. The AGNPS-estimated results in 1997 for the Red Rock Creek sub-watershed are presented in Table 3. Of the seven storms selected, three storms produced more than 2.2 mm (0.1 in) of surface runoff in each event. The model-estimated surface runoff produced ED and CP values of 3% and 0.03, respectively, and for TSS, they were 3% and 1.0, respectively. The event of June 16 was simulated poorly for TSS when compared with other storms. Two other storms occurred before this storm, which may have been a factor in producing higher amounts of TSS than estimated by the model. However, AGNPS is an event-based model and could not take this factor into account.
AGNPS was run with seven storm events in 1998, and these values are presented in Table 3. The result of AGNPS estimation on the same sub-watershed was also satisfactory. The surface runoff depth and TSS estimation produced the CP values of 0.03 and 0.22, respectively. The total amounts of measured and estimated values were close enough to make ED values only 5% and 6% for surface runoff depth and TSS yield, respectively.
Goose Creek sub-watershed. Testing of AGNPS was made on Goose Greek sub-watershed during different storm events of 1997 and 1998. The Goose Creek sub-watershed is located in the southeast corner of the Cheney Reservoir watershed and has the smallest drainage area, 130 sq km (50 sq mi). Both landcover and soil texture are similar to the Red Rock sub-watershed. No rainfall gaging stations are located within this sub-watershed. Therefore, three rainfall stations located nearest to the boundary of this sub-watershed were used to estimate average rainfall amounts, using the Thiessen Polygon method (Thiessen 1911) and also for computing El values. Seven storm events in 1997 were selected for AGNPS run on this sub-watershed. The measured and estimated results during different storms of 1997 and 1998 are presented in Table 4. AGNPS estimations were similar with estimations on the Red Rock Creek sub-watershed. For surface runoff, the ED and CP values were 2% and 0.1, and for TSS, ED and CP were 14% and 2.1, respectively.
Seven storms in 1998 were selected for AGNPS (Table 4). In terms of total amounts of measured and estimated values, the ED values were 8% and 34% for surface runoff depth and TSS, respectively. On an individual storm basis, AGNPS-estimated results for 1998 resulted in CP values of 0.0 and 0.6 for surface runoff and TSS, respectively, which were superior to the results for 1997.
Silver Creek sub-watershed. The Silver Greek sub-watershed is larger than the previous two sub-watersheds, with total drainage area of 497 sq km (192 sq mi). No rainfall stations are located within this sub-watershed. This sub-watershed was shaped differently, as well, being longer from east to west and shorter from north to south. This shape created a variation in rainfall amounts among the rainfall stations considered for estimation of average rainfall. Eight storm events were selected during 1997, of which five resulted in less than 1.0 mm (0.04 in) of surface runoff. The total AGNPS-estimated values and the total measured values are presented in Table 5. CP computed with eight storms of 1997 for surface runoff and TSS were 1.0 and 2.2, respectively. However, total estimated values were reasonably close to the measured values, with ED values of 37% and 9%. Most of the smaller events overestimated the measured values resulting in higher ED values.
During 1998, only six storm events were selected on this sub-watershed. The overall estimation for the total amounts of measured and estimated value was reasonably satisfactory with ED values of 15% and 9% and CP values of 0.1 and 0.3 for surface runoff and TSS, respectively (Table 5).
The estimates from AGNPS probably would improve with better estimation of rainfall amounts and further division of this larger sub-watershed into smaller units to handle the variation in rainfall.
West Ninnescah sub-watershed. The West Ninnescah sub--watershed is the largest sub--watershed in the Cheney Reservoir watershed, covering an area of 1,223 sq km (472 sq mi). This sub-watershed is also very long eastwest--about 90 km (56 mi).The topography of this sub-watershed is relatively flat and soil texture is relatively coarser when compared with the eastern part of the Cheney Reservoir watershed. The USGS reports that 186 sq km (72 sq mi) of this sub-watershed is noncontriburing (Christensen and Pope 1997). In an earlier study conducted at Wet Walnut Creek in Kansas, Ramireddygari et al. (2000) decreased the CN values of the watershed by 3 to improve the runoff estimation capability of POTYLDR (Koelliker et al. (1981). This was done because the topography of the watershed was flat and soil texture was coarse. In addition, the watershed also had a noncontributing area similar to the West Ninnescah sub-watershed. Hence, the CN values for this sub--watershed were also decreased by 3 to get better results in this modeling process. Of five rainfall stations selected for computing average rainfall, two stations are within the sub-watershed. Only five smaller storm events could be selected during 1997. In terms of total estimated values, TSS was highly underestimated with ED values of 66% (Table 5). The CP values were 0.3 and 1.2 for surface runoff and TSS, respectively. Similarly, during 1998, total TSS was highly underestimated during nine selected storm events with ED values of 82% and CP value of 1.5.
These results reveal that estimation of average rainfall amount did not work well on this bigger sub-watershed. The precipitation amounts in the eastern part were more than in the western part of this sub--watershed, which resulted in distortion of average rainfall estimates. The uncertainty caused by the noncontributing area of this sub-watershed also affected AGNPS estimation. Therefore, further division of this sub-watershed is needed to improve the estimation.
One-year storm over the entire watershed for BMP study. The GIS layers/coverages for the entire watershed were used to make the AGNPS input file by the AGNPS-ARC INFO interface. A total of 3,699 cells of 65 ha (160 ac) each was created by the interface for the AGNPS input file. To identify the potential areas for adopting BMPs in the watershed, AGNPS was run with a 1-year storm during 1997 and 1998 considering CN values for AMC ratio of 1.5. The 1-year storm amount was 63.5 mm (2.5 in) with an El value of 653.7 MJ/ha-mm/hr (37.7 fttons/ac-in/hr). Based upon the size of the storm actually measured in the watershed, a 1-year storm is representative of those that created a substantial amount of sediment and nutrient loss from the watershed. The 1-year storm produced 8.6 mm (0.3 in) and 8.9 mm (0.4 in) depth of surface runoff and 6,358 tonne (7,010 tons) and 6,934 tonne (7,645 tons) total soil-loss values coming out of the entire watershed during 1997 and 1998, respectively. The increase in runoff and soil loss during 1998, compared with 1997, is due to the increase in the amount of low-cover wheat area in the watershed, as reported earlier. This increase will also increase the nitrogen and phosphorous loss from the watershed.
The variation of the soil loss out of 65 ha (160 ac) cells in the entire watershed for 1997 and 1998 are shown in Figure 8. The highest soil loss leaving a 65 ha (160 ac) cell was 963 kg/ha (860 lb/ac).
The landcover practices during 1998 in the areas of higher soil loss--675 to 963 kg/ha (603 to 860 lb/ac))--were studied. The low-residue wheat covers 76% of these areas, followed by low-quality rangeland and medium-residue wheat by 7% and 6%, respectively Adopting high-residue wheat instead of low-residue or converting parts of these areas into conservation reserve program (CRP) will help reduce soil loss. The eastern part of the watershed contributes more soil loss than the western part of the watershed. The higher soil losses were produced from the Red Rock Creek, the eastern part of East Ninnescah, and a small portion from the north part of the West Ninnescah and Silver Creek sub-watersheds. Appropriate BMPs may be adopted in these areas to reduce the soil loss and other nutrients to maintain the total maximum daily load in the Cheney Reservoir.
Summary and Conclusions
Developing a modeling process for assessment of water quality of watershed is always a challenging task when availability of data is limited and different parts of the watershed have distinct conditions. In this study, a modeling process for assessment of runoff amount and sediment loss from a watershed during different storm events was accomplished using remotely sensed data, GIS, and the AGNPS model. The landcover classification derived from satellite images enabled us to prepare the GIS layer of watershed with the USLE C-factor, P-factor, and other AGNPS model parameters on the basis of various landcover conditions, which were used in AGNPS modeling to compute runoff and sediment loss from a watershed. Using remotely sensed data, researchers will he able to observe temporal variation in landcover practices within the watershed.
Application of continuous AMC ratios instead of discrete AMC ratios was found to be useful to improve the runoff estimation. This technique can be used in post-assessment for water quality of watersheds and in suggesting BMPs. Based on the measured runoff that actually occurred from storms during the period of this study, it was found that, for water-quality modeling purposes for a sub-humid watershed in this area of Kansas, an average AMC ratio of 1.5 may be more appropriate than an AMC ratio of 2.0. The AGNPS-estimated values during various events in the sub-watersheds of the Cheney Reservoir watershed are satisfactory on the basis of CP and ED values, compared with earlier studies conducted by Perrone and Madramootoo (1997, 1999). The estimations on the Silver Creek and West Ninnescah sub-watersheds, however, were not as good as estimations on smaller sub-watersheds. The landcover practices and soil texture of the western portion of the watershed is different than the eastern portion of the watershed.
Uncertainties are always involved in the modeling process. The noncontributing nature of the watershed was also another factor that could affect the estimation. AGNPS is a single-event model, which assumes a uniform rainfall event over the entire watershed. This assumption was certainly not true in the larger watersheds. Overall, the greater variation of landcover and soil texture, uncertainties involved in the modeling process, and nonuniform rainfall in cases of larger watersheds may contribute to poor modeling performance.
Hence, the accurate estimation of the average rainfall over the watershed is an important factor and should be supported by adequate rainfall data collected from various rain gages within the watershed. In larger sub-watersheds, it is recommended that they be divided into smaller sub-watersheds before using the estimation methods discussed in this paper. Another interesting aspect of this application would be a scaling-up process on a bigger watershed. The results from a smaller watershed may be scaled up with runoff depth to estimate the sediment yield from a larger watershed. The variation of landcover and soil texture, however, could affect this scaling-up process. Using the assessment technique discussed in this paper, we were able to identify the critical areas of the sub-watersheds, and the CRWQP Office is working on adoption of BMPs in those areas.
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
[FIGURE 6 OMITTED]
[FIGURE 7 OMITTED]
able 1 Example showing the computation of surface runoff depth and TSS yield from daily stream flow and TSS concentration. (Date of storm: June 24, 1997 at Red Rock Creek gaging station with drainage area of 137 sq km.) Average daily Daily surface runoff Stream flow, Base, flow, Date [m.sup.3]/s [m.sup.3]/s [m.sup.3]/s mm 24-June 6.9 0.1 6.8 4.3 25-June 2.4 0.1 2.3 1.4 26-June 0.6 0.2 0.4 0.3 27-June 0.3 0.2 0.1 0.1 Total 6.1 TSS in Amounts of TSS yield stream TSS in surface in surface Date flow, mg/l runoff, Mg runoff, kg/ha 24-June 278 163.3 11.9 25-June 189 37.6 2.7 26-June 234 8.1 0.6 27-June 234 2.0 0.2 Total 15.4 able 2 Landcover classes, in percentage of area, made from satellite image for the Cheney Reservoir watershed during 1997 and 1998. Percent landcover in the entire watershed Landcover classes 1997 1998 Rangeland Low cover 7.9 9.4 Medium cover 28.0 29.5 High cover 4.7 2.9 Other crop 24.0 17.8 Wheat Low cover 19.7 24.6 Medium cover 10.4 11.1 High cover 3.5 3.0 Woodland 0.5 0.4 Water 0.7 1.0 Residential area 0.6 0.5 Percent change in amount of landcover in 1998 from 1997 in various sub-watersheds Landcover classes RRC GC SC Rangeland Low cover -3.2 +0.7 +4.7 Medium cover +4.5 -0.3 -3.6 High cover -0.1 -2.3 -2.6 Other crop -2.1 -2.2 -5.8 Wheat Low cover -2.1 +10.6 +6.4 Medium cover +2.5 -5.4 +1.9 High cover -0.2 -1.4 -0.7 Woodland +0.9 -0.4 -0.6 Water +0.1 +0.6 +0.6 Residential area -0.4 0.0 -0.2 Percent change in amount of landcover in from 1997 in various sub-watershe ds Landcover classes WN Rangeland Low cover +1.6 Medium cover +1.1 High cover -1.1 Other crop -8.5 Wheat Low cover +5.8 Medium cover +1.2 High cover 0.0 Woodland -0.2 Water +0.1 Residential area -0.1 RRC: Red Rock Creek GC: Goose Creek SC: Silver Creek WN: West Ninnescah able 3 Comparison between total measured and AGNPS-estimated values for Red Rock Creek sub-watershed with average C-factor and CN values during 1997 and 1998. Date of Weighted Average AMC Rainfall, Runoff, mm storms, 1997 C-factor CN ratio mm meas. 11-Apr 0.114 65 1.1 46.7 2.2 13-Jun 0.049 61 0.9 44.2 0.8 16-Jun 0.049 83 2.4 34.5 7.8 24-Jun 0.049 69 1.5 51.8 6.1 09-Jul 0.113 65 1.1 36.8 0.6 22-Aug 0.140 69 1.4 28.9 0.3 23-Sep 0.168 70 1.4 49.8 5.6 Total 292.7 23.4 Date of Runoff, mm TSS, kg/ha storms, 1997 est. meas. est. 11-Apr 2.0 4.0 11.2 13-Jun 1.0 2.7 4.1 16-Jun 8.9 29.3 6.2 24-Jun 5.3 15.4 10.7 09-Jul 0.8 2.9 3.6 22-Aug 0.3 0.7 2.3 23-Sep 5.8 13.1 27.6 Total 24.1 68.1 65.8 ate of Weighted Average AMC Rainfall, Runoff, mm storms, 1998 C-factor CN ratio mm meas. 17-Mar 0.118 77 2.0 32.0 2.8 07-Apr 0.109 68 1.4 14.0 0.5 28-Apr 0.089 68 1.5 31.0 0.4 23-Jun 0.156 65 1.4 54.6 4.4 07-Jul 0.136 74 1.8 32.5 2.0 10-Jul 0.136 76 1.9 31.0 2.3 28-Sep 0.163 62 1.0 41.4 0.8 Total 236.5 13.2 Date of Runoff, mm TSS, kg/ha storms, 1998 est. meas. est. 17-Mar 3.0 6.9 6.4 07-Apr 0.0 0.9 0.0 28-Apr 0.5 0.3 2.3 23-Jun 4.5 19.3 24.9 07-Jul 2.3 5.4 6.2 10-Jul 2.5 10.1 5.8 28-Sep 1.0 7.0 7.3 Total 13.9 50.2 53.0 Meas.: measured Est.: estimated able 4 Comparison between measured and AGNPS-estimated values for Goose Creek sub-watershed with average C-factor and CN (AMC II) values during 1997 and 1998. Date of Weighted Average AMC Rainfall, Runoff, mm storms, 1997 C-factor CN ratio mm meas. 11-Apr 0.135 70 1.5 44.7 4.0 30-May 0.096 69 1.6 32.0 0.6 24-Jun 0.105 78 2.2 32.0 3.6 21-Jul 0.111 68 1.4 34.8 0.8 11-Aug 0.161 76 1.9 37.8 4.8 22-Aug 0.170 87 2.7 20.6 3.3 23-Sep 0.207 53 0.4 67.3 2.1 Total 269.2 19.2 Date of Runoff, mm TSS, kg/ha storms, 1997 est. meas. est. 11-Apr 3.8 8.8 8.7 30-May 0.5 0.8 2.3 24-Jun 2.8 6.9 3.9 21-Jul 0.8 1.0 3.0 11-Aug 4.6 14.7 8.1 22-Aug 3.8 7.2 3.1 23-Sep 2.5 1.8 18.0 Total 18.8 41.2 47.2 ate of Weighted Average AMC Rainfall, Runoff, mm storms, 1998 C-factor CN ratio mm meas. 08-Mar 0.174 87 2.5 11.7 0.4 17-Mar 0.174 80 2.3 33.0 4.7 31-Mar 0.173 89 2.7 11.9 1.0 07-Apr 0.146 87 2.5 12.2 0.6 29-Apr 0.079 72 1.9 28.9 0.8 11-Jul 0.082 73 1.7 23.1 0.3 28-Sep 0.235 66 1.3 35.5 0.6 Total 156.3 8.4 Date of Runoff, mm TSS, kg/ha storms, 1998 est. meas. est. 08-Mar 0.7 0.2 1.3 17-Mar 4.8 22.3 6.9 31-Mar 1.3 2.3 1.5 07-Apr 0.5 0.7 1.2 29-Apr 0.8 0.6 2.2 11-Jul 0.3 0.2 1.3 28-Sep 0.7 0.6 3.3 Total 9.1 26.9 17.6 able 5 Comparison between total measured and total AGNPS-estimated values for different sub-watersheds of Cheney Reservoir watershed during 1997 and 1998. Year: 1997 Sub-watersheds Sum of storm Total runoff, mm. Total TSS, kg/ha rainfall, mm meas. est. meas. Silver Creek 288.8 8.7 11.9 22.8 West Ninnescah 141.2 3.3 3.6 21.7 Sub-watersheds Total TSS, kg/ha est. Silver Creek 24.8 West Ninnescah 7.3 Year: 1998 Sub-watersheds Sum of storm Total runoff, mm. Total TSS, kg/ha rainfall, mm meas. est. meas. Silver Creek 118.9 6.8 7.8 11.3 West Ninnescah 184.9 5.9 8.1 65.1 Sub-watersheds Total TSS, kg/ha est. Silver Creek 10.3 West Ninnescah 11.4
This research was approved as Contribution No. 01-32-J, dated June 26, 2000, of the Kansas Agricultural Experiment Station. Financial assistance for this project was provided in part by the Kansas Water Office, the Cheney Reservoir Water Quality Project (CRWQP) office, the city of Wichita, and Kansas State Research and Extension. We also acknowledge the assistance provided by Lyle Frees, water quality coordinator, and Chad Volkman, GIS specialist, in the CRWQP office in South Hutchinson, Kansas.
Basnyat, P, L. D. Teeter, B. G. Lockaby. K.M. Flynn. 2000. The use of remote sensing and GIS in watershed level analyses of non-point source pollution problems. Forest Ecology and Management 128 (2000): 65-73.
Biard, F. and F. Baret. 1997. Crop residue estimation using multiband reflectance. Remote Sensing of the Environment 59(3): 530-536.
Christensen, V.G. and L.M. Pope. 1997. Occurrence of dissolved solids, nutrients, atrazine, and fecal coliform bacteria during low flow in the Cheney Reservoir watershed, South Central Kansas. 1996. USGS Survey Water-Resources Investigation Report 97-4153. 13 pp.
DeAussen, R., P.K. Kalita, J.K. Koelliker, and K.R. Mankin. 1998. Application of AGNPS-ARC/ INFO interface model to an agricultural watershed of Kansas. American Society of Agricultural Engineering Paper No. 983139. ASAE, St. Joseph, MI.
DeGloria, S.D., S.L. Wall, A.S. Benson, and M.L. Whiting. 1986. Monitoring conservation tillage practices using Landsat multispectral data. Journal of Soil and water Conservation, May-June: 187-190.
Frees, L. 2001. Cheney Reservoir Water Quality Project Office, U.S. Department of Agriculture-Natural Resources Conservation Service, Reno County. Personal communication. Jan. 19.
Hawkins, R.H. 1978. Runoff curve numbers with varying site moisture. Journal of the Irrigation and Drainage Division 104(4): 389-398.
Hershfield, D.M. 1961. Rainfall frequency atlas of the United States. Tech paper No. 40, U.S. Department of Commerce, Weather Bureau. U.S. Government Printing Office, Washington, D.C.
Hession, W.C., K.L. Huber, S. Mostaghimi, V.O. Shanholtz, and P.W McClellan. 1989. BMP effectiveness evaluation using AGNPS and GIS. ASAE paper No. 89-2566. ASAE, St. Joseph. MI.
Hjelmfelt, A.T. 1991. Investigation of curve number procedure. Journal of Hydraulic Engineering 117(6): 725-737.
Holland, D.D. 1971. Sediment yields from small drainage areas in Kansas. Bulletin No. 16. Kansas water Resources Board, Topeka, KS.
James, L.D. and S.J. Burgess. 1982. Selection, calibration, and testing of hydrologic models. In: Hydrologic modeling of small watersheds. C.T. Haan, H. P. Johnson, and D.L. Brakensiek (eds.). ASAE, St. Joseph, MI.
Jensen. J.R. 1996. Introductory digital image processing: A remote sensing perspective. Prentice Hall, Upper Saddle River, NJ.
Jensen, J.R. 2000. Remote sensing of the environment: An Earth resource perspective. Prentice Hall, Upper Saddle River, NJ.
King, C. and G. Delpont. 1993. Spatial assessment of erosion: Contribution of remote sensing, a review. Remote Sensing Reviews 7(3-4) 223-232.
Koelliker, J.K., J.J. Zovne, J.M. Steichen, M.W. Berry. 1981. Study to assess water yield changes its the Solomon Basin, Kansas. Part I - Final Report. Kansas water Resources Research Institute, Manhattan, KS.
Koelliker, J.K. and C.E. Humbert. 1989a. Applicability of AGNPS model for water quality planning. ASAE Paper No. 89-2042, ASAE, St Joseph, MI.
Koelliker, J.K. and C.E. Humbert. 1989b. Application of AGNPS model to watersheds in Northeast Kansas. Complete Report to Kansas Soil Conservation Service, Civil Engineering Department, Kansas State University, Manhattan, KS. 167pp.
Leblon, B., L. Gallant, F. Bonn, and A. Peasant. 1996. Corn residue assessment from optical and thermal infrared ground-based measurements. Canadian Journal of Remote Sensing 22(2): 198-207.
Lee, M.T. and D.C. White. 1992. Application of GIS databases and water quality modeling for agricultural nonpoint source pollution control. Research Report 014. Water Resources Center, University of Illinois at Urbana-Champaign, Urbana, IL.
Liao, Hsiu-Hua. 1996. Nonpoint source pollution modeling of an agricultural watershed within a geographic information system. Ph.D. thesis. Iowa State University, Ames, IA.
Liao, Hsiu-Hua. 1997. AGNPS-ARC/INFO user interface manual. Agricultural and Biosystems Engineering Department. Iowa State University, Ames, IA.
Liao, Hsiu-Hua and U.S. Tim. 1997. An interactive modeling environment for non-point source pollution control. Journal of the American water Resources Association 33(3): 1-13.
Lo, K.F. 1995. Erosion assessment of large watersheds in Taiwan. Journal of Soil and Water Conservation 50(2): 180-183.
Lyon, J.G., D. Yuan, R.S. Lunetta, and C.D. Elvidge. 1998. A change detection experiment using vegetation indices. Photogrammetric Engineering and Remote Sensing 64(2): 143-150.
Mankin, K.R., J.K. Koelliker, and P.K. Kalita. 1999. Watershed and lake water quality assessment: An integrated modeling approach. Journal of the American Water Resources Association 35(5): 1069-1080.
Marzen, L.J., J.A. Harrington Jr., S.J. Bhuyan, and J.K. Koelliker. 2000. Use of satellite imagery to determine land cover input variables in an AGNPS water quality model. ASAE Paper No. 002200. ASAE, Milwaukee, WI.
McCuen, R.H. 1982. A guide to hydrologic analysis using SCS method. Prentice-Hall, Englewood Cliffs, NJ.
McNairn, H. and R. Protz. 1991. Evaluation of the use of satellite imagery to measure crop residue cover on fields in the Lake Erie Watershed. Proceedings, Remote Sensing and GIS Applications to Non-Point Source Planning 1991: 115-124.
McNairn, H. and R. Protz. 1993. Mapping corn residue cover on agricultural fields in Oxford County, Ontario, using thematic mapper. Canadian Journal of Remote Sensing 19(2): 152-159.
Novotny, V., H. Olem. 1994. Water quality: Prevention, identification, and management of diffuse pollution. Van Nostrand, Reinhold, NY.
Perrone, J. and C.A. Madramootoo. 1997. Use of AGNPS for watershed modeling in Quebec. Transactions of the ASAE 40(5): 1349-1354.
Perrone, J. and C.A. Madramootoo. 1999. Sediment yield prediction using AGNPS. Journal of Soil and Water Conservation 54(1): 41S-419.
Prato, T. and H. Shi. 1990. A comparison of erosion and water pollution control strategies for an agricultural watershed. Water Resources Research 26(2): 199-205.
Putnam, J.F., D.L. Lacock, D.R. Schneider, and M.D. Carlson. Water resources data, Kansas water year 1998. Water data report KS 98-1. U.S Department of the Interior, U.S. Geological Survey.
Ramireddygari, S.R. M.A. Sophocleous, J.K. Koelliker, S.P. Perkins, and R.S. Govindarajan. 2000. Department and application of a comprehensive simulation model to evaluate impacts of watershed structures and irrigation water use on streamflow and ground-water: Tire case of Wet Walnut Creek Watershed, Kansas. Journal of Hydrology 236 (3-4): 223-246.
Renard, K.G., G.R. Foster, G.A. Weesies, and J.P. Porter. 1991. RUSLE: Revised Universal Soil Loss Equation. Journal of Soil and Water Conservation 46(1): 30-33.
Schultz, G.A. 1988. Remote sensing in hydrology. Journal of Hydrology 100: 239-265.
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. 46 pp.
Smith, R.E., and J.R. Williams. 1980. Simulation of the surface hydrology. In Knisel, W. (ed.). CREAMS, A field scale model for Chemicals, Runoff, and Erosion from Agricultural Management Systems. Agricultural Research Service, USDA. Conservation Research Report 26, vol. 1., ch. 2. 15pp.
Soil Conservation Service. 1968. Hydrology, supplement A to sec. 4, Engineering Handbook, USDS-SCS, Washington, D.C.
Thiessen, A.H. 1911. Precipitation for large areas. Monthly Weather Review 39: 1082-1084. July.
Tim, U.S, R. Jolly, and H.S. Liao. 1995. Impact of landscape feature and feature placement on agricultural nonpoint source pollution control. Journal of Water Resources Planning and Management 121(6): 463-470.
USDA. 1986. Urban hydrology for small watersheds, TR 55, Technical release 55, June. Conservation Engineering Division, USDA-NRCS.
USGS. 2000. USGS Office, Lawrence, KS. Personal communication. June 2.
Wischmeier, W.H. and D.D. Smith. 1978. Predicting rainfall erosion losses, Agriculture Handbook No 537. USDA-ARS, Washington, D.C.
Young, R.A., C.A. Onstad, D.D. Bosch, and W.P. Anderson. 1989. AGNPS: A nonpoint-source pollution model for evaluating agricultural watersheds. Journal of Soil and Water Conservation 44(2): 168-173.
Young, R.A., C.A. Onstad, D.D. Bossch, and W.P. Anderson, 1994. AGNPS, Agricultural Nonpoint Source Pollution Model, Version 4.03: AGNPS user's guide, July. USDA-NRS-NSL, Oxford, MS.
Young, R.A., C.A. Onstad, D.D. Bosch, and W.P. Anderson. 1987. AGNPS, Agricultural Nonpoint Source Pollution Model: A large watershed analysis tool. Cons Research Report 35. USDA-ARS, Washington, D.C. 77pp.
Samar J. Bhuyan is a hydrologist with the Arizona Department of Environmental Quality in Phoenix, Arizona. Luke J. Marzen is an assistant professor in Auburn University's Geography Department in Auburn, Alabama. James K. Koelliker is a professor and head of the Biological and Agricultural Engineering Department, John A. Harrington Jr. is a professor and head of the Geography Department, and Philip L. Barnes is en assistant professor in the Biological and Agricultural Engineering Department at Kansas State University In Manhattan, Kansas.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Agricultural Nonpoint Source Pollution Model|
|Author:||Bhuyan, S.J.; Marzen, L.J.; Koelliker, J.K.; Harrington, J.A., Jr.; Barnes, P.L.|
|Publication:||Journal of Soil and Water Conservation|
|Date:||Nov 1, 2002|
|Previous Article:||Soil quality of two Kansas soils as influenced by the Conservation Reserve Program.|
|Next Article:||Modification and performance of the Coshocton wheel with the modified drop-box weir.|