This chapter emphasizes research that is focused on the GIS application in the agricultural non-point source pollution modeling. Only models that include nutrients or pesticides are presented. Soil erosion and sediment transport applications are mentioned here for the completeness of discussion.
Application of the GIS in modeling the non-point source pollution can be grouped into three categories:
The hybrid models, GIS <=> Pollutant Model, are dominant systems in modeling non-point source pollution. The GIS derives the data required for the model then executes the model. After calculations, the GIS is used for visual analysis of results. The most popular pollution model that is linked with a GIS software is AGNPS (Agricultural Non-Point-Source).
AGNPS is an event-based model. It calculates runoff from agricultural watershed and transport processes of sediment, nitrogen, phosphorous, and COD. Watershed is represented by square cells of 0.4 - 16 ha. Each cell is characterized by twenty-two parameters that include: SCS curve number, terrain description, channel parameters, soil-loss equation data, fertilization level, soil texture, channel and point source indicators, oxygen demand factor. Sediment runoff is estimated from the modified version of USLSE (Universal Soil Loss Equation) and its routing is performed for five particle size classes. Calculations of the nutrients transport are divided into soluble and sediment-absorbed phases. ). The application of AGNPS is limited to about 200 km² watersheds ( Young et al., 1989, DeVries and Hromadka, 1993, Engel et al., 1993).
At least three interfaces between AGNPS and GRASS (Geographical Resources Analysis Support System) have been constructed: (1) in Michigan State University (He et al., 1993), (2) by Srinivasan and Engel (Engel et al., 1993, Mitchell et al., 1993), and (3) in Soil Conservation Service (Cronshey et al., 1993). GRASS is the major public domain GIS. It is widely used by many federal and states agencies. The access to the source code provides the flexibility to modify existing GRASS procedures or to add new ones. This GIS software has a considerable ability to support hydrologic analysis.
AGNPS has also been linked to other GIS programs, such as:
The last interface has been used to evaluate erosion and sediment yields in a lover alpine drainage basin of area of 65 ha (located in Austria). The interface contained EPIC (Erosion/Productivity Impact Calculator, Williams et al., 1990) a field scale comprehensive model developed to predict the long-term relationship between erosion and productivity. EPICs components include weather simulation, hydrology, erosion-sedimentation, nutrient cycling, plant growth, tillage, soil temperature, economics, and plant environment control.
Cronshey et al. (1993) report interface that includes GRASS and a watershed scale water quality model SWRRB (Simulator for Water Resources in Rural Basins). SWRRB (Arnold et al., 1990) uses daily time step for calculations of sediment yield, routing, as well as pesticide and nutrient fate. Basins are subdivided to account for differences in soils, land use, crops, topography, weather. Soil profile can be divided into ten layers. Basins of several hundred square miles can be studied, but number of sub-basins is limited to 10.
Engel (1993) discusses the application of GRASS-ANSWERS (Aerial Nonpoint Source Watershed Environment Response Simulation) interface. ANSWERS (Beasley et al., 1982 after Engel, 1993) calculates runoff, erosion, sedimentation and phosphorus movement from watersheds. The watershed is divided into a grid cells. Runoff, erosion, sedimentation, and water quality related to sediment associated chemicals are computed for each cell and routed.
In 1993 Arnold, Engel and Srinivasan (from Mamillapalli et al., 1996) developed a new version of the SWRRB--Soil Water Assessment Tool (SWAT). In SWAT, the watershed can be divided into practically unlimited number of cells and/or subwatersheds. New features have been added such as routing of the flow through the basin streams and reservoirs, simulating lateral flow, groundwater flow, stream routing transmission losses, modeling sediment and chemical transport through ponds, reservoirs, and streams. The major components of the SWAT include weather, hydrology, erosion, soil temperature, crop growth, nutrients, pesticides, subsurface flow, and agricultural management. The QUAL2E (Enhanced Stream Water Quality Model) water quality component has been incorporated into SWAT. First-order decay relationship for algae, dissolved oxygen, carbonaceous biochemical oxygen demand, organic nitrogen, ammonium nitrogen, nitrate nitrogen, nitrite nitrogen, organic phosphorus, and soluble phosphorus used in QUALE2E were adopted in SWAT with necessary adjustments (Ramanarayanan et al., 1996). In 1994, a GRASS GIS - SWAT interface was developed by Srinivasan and Arnold (1994). In 1996 Bian et al. linked SWAT with Arc/Info.
QUAL2E model uses a finite-difference solution of the advective-dispersive mass transport, reaction, and sink/source equation. The stream network is divided into headwaters, reaches, and junctions. The changes in flow conditions are represented as a series of steady- flow water profiles. Such parameters as velocity, cross-sectional area, and water depth that are required for the mass transport calculations are computed from the flow rate. For each river reach, QUAL2E requires specification of as many as 26 physical, chemical, and biological parameters. (DeVries and Hromadka, 1993, Camara and Randal, 1984, Schoellhamer, 1988). Compiling such data at a regional scale would take a very great investment of time and resources.
Most of the GIS programs are equipped with a macro language that allows user to write models within application. For example, Arc/Info has very powerful macro language, AML (Arc/Info Macro Language). In addition, external procedures written in such programming languages as C/C++ or FORTRAN can be executed by macro, thus the modeling process can be very efficient. This Section discuss models of water pollution build using GIS tools.
White and Hofschen (1993) developed a spatial model for assessing nutrient loads in New Jersey rivers using Arc/Info. They used 3 arc-sec digital elevation models (DEM) to partition the study area (15,385 km²) into 2,893 drainage basins (polygons) with a network of 10,916 stream segments (arcs). Time of travel was assumed as the basis for calculating predictors of water quality. The simple formula v = 0.38 * Q 0.24, which was estimated for New Jersey, was used to estimate the flow velocity in each reach. A first-order decay reaction was assumed to calculate the non-conservative downstream transport. White and Hofschen attempted to improve the model by representing the decay constant as a function of stream slope, and the nonpoint source yields as a function of subbasin gradient. The nitrogen model performance showed no improvement with these refinements. White and Hofschen found that the time of travel, which was calculated from the exponential velocity formula, underestimated by a factor of 0.57 the time of travel of dye-tracer, that is, the dye took approximately twice as long to traverse the stream as the formula suggested. This travel time underestimation was accommodated by assignment of higher values of pollutant decay than those reported in the literature.
Smith et al., (1993) constructed a GIS model of total phosphorus concentrations in New Jersey streams. The core of this model was a regression equation that related transformed (natural logarithm) total phosphorus concentration measured at a given point to transformed concentrations resulting from exponentially decayed phosphorous loads in the upstream watershed. In this study, the classical approach of modeling first-order reaction was modified. Instead of using the time of travel and time decay coefficient, the travel distance and a distance decay coefficient for phosphorus were applied in the model respectively. The data from 104 long term sampling stations, collected in the period from 1982 to 1987 was utilized to estimate regression coefficients. The area of the studied region was 15401 km². The sources of phosphorous were represented by such variables as area of agricultural land, total human population, and total municipal effluent flow.
The most basic application of the GIS is spatial data manipulation, data extraction for further analysis, and presentation of results in the map form. This Section discusses work in which the GIS tools have been utilized to support statistical analysis of the surface water pollution.
Cressie and Majure (1994) used Arc/Info to determine explanatory variables for a statistical model of the variation in pollutant concentration from dairies in streams of the Upper North Bosque watershed located principally in Erath County, Texas. The Arc/Info GRID and the Digital Elevation Models (DEM) were used to determine drainage basins and the lengths along flow paths. Cressie and Majure assumed a spatially constant flow velocity (0.5 m/s), and using simple map algebra, they determined a 3-day flow-time area of influence for each stream measurement site. Seventeen explanatory variables including a number of dairies per acre, a number of heads per acre, lagoons per acre, waste application method, soil hydrologic code, average slope, distance to basin outlet, and precipitation were considered. All variables, except one (seasonal variation), were determined using the GIS. The authors concluded that the GIS was an important tool in observational studies due to its ability to construct explanatory variables at the appropriate scale.
Mueller et al. (1993) applied logistic regression to relate discrete categories of nitrate concentrations to such explanatory variables as land use in the drainage basins upstream from the sampling sites, percentile of streamflow at the time of sampling, acreage of the basin in corn, acreage in soybeans, density of cattle, and population density. They extracted data from GIS databases stored in 1:2,000,000- scale maps of the conterminous United States. The GIS software was used only to areally weight the extracted data and sum it by basin; their model did not include stream transport. Better classification of nitrate concentration was achieved by model that included the flow percentile, the areal extend of corn and soybean production, the density of cattle, and the density of population, then the model that contained percentile of flow, nitrogen fertilizer application, and population density. In addition, Mueller et al. found, that as the percentile of flow increased, the probability of nitrate concentration being in a higher category also increased. Low explanatory power of fertilizer application researchers explained by the fact, that the fertilizer use was approximated by the county level sales thus such nitrate sources as manure were not included.
From the observations made on the Mississippi River and four tributaries during a one - year period (from April 1, 1991 to March 31, 1992), Battaglin et al. (1993) estimated a single relationship between the annual use of nitrogen and nitrate transport: Ntransport = -0.2 + 0.1547 Nuse and a linear relationship between annual atrazine use and the atrazine transport: Atransport = -12 * + 0.0156 * Ause, (in metric tons). They used a GIS to estimate the nitrogen and atrazine use within gauged watershed from the county level sales of nitrogen fertilizer and atrazine herbicide. in addition, Battaglin et al. estimated that 321 Mg of atrazine and 33.7 Mg of alachlor were discharged from the Mississippi River basin to the Gulf of Mexico in streamflow (from April 1, 1991 to March 31, 1992), while the amounts of these herbicides used in the basin were approximately equal.
Moody and Goolsby (1993) report the results of a large scale USGS study of herbicide transport in the Lower Mississippi River. Although, they didn't use a GIS, this work is mentioned here since it is one of the scarce large scale studies. During May 26-29, 1990, water samples for triazine herbicide analysis were collected every 16 km from Baton Rouge in Louisiana, upriver to the Mississippi-Ohio River confluence (distance of 1900 km). The measurements showed the background level of ~2.7 µg/L of triazine herbicides and an upriver concentration gradient of 0.2 µg/L per 100 km (concentration decreased going downstream). The authors suggest that the longitudinal spatial variability in concentration is a result of cross-channel gradients and the addition of 'slugs' of water from various upriver tributaries. A routing scheme was used to predict the location of water masses. This routing method was tested by using the measurements of the specific conductance. The average flow velocity was v = 6 km/h. It is interesting that the measurements show about 50% decrease in the load whereas the reported atrazine half-life in water is about 140 days (Thurman et al., 1992).
Presented in this dissertation model is designated to represent average monthly values of flow rate, agrichemical concentration, and chemical load in streams of the Midwest. Introduction of a seasonal component to the model, fills the gap in existing GIS models of pollutant transport. Most of the hybrid models that have been introduced in Section 2.1, are capable to perform continuous-time simulations (SWAT-GIS), event - related calculations (AGNPS), or daily computations (SWRRB, ANSWERS-GIS). The models of surface water pollution that are constructed within GIS (discussed in Section 2.2) estimate average chemical concentrations for the studied period. Even the USGS studies do not evaluate changes in chemical concentration in the Midwest surface water on the monthly basis. The seasonal variations are represented by usually three terms of the year: pre-planting, post-planting, and Fall lowflow (Goolsby et al., 1993a, Goolsby and Battaglin, 1993, Scribner et al., 1993). The USGS model of the agricultural chemicals transport in the Midwest rivers relates the annual chemical load with annual agrichemical use (Battaglin et al., 1993).
The model is designated to predict the loads and concentrations in such large basins as the Upper Mississippi River basin (490,000 km²), the Ohio River basin, or the Upper Missouri-Mississippi-Ohio River basin (above Ohio-Mississippi River junction, about 2,800,000 km²). It has been applied for evaluation of nitrate and atrazine concentrations in the Iowa-Cedar River basin, Iowa of 32,000 km² - area larger that reporting limits of hybrid models discussed in Section 2.1 and models constructed within GIS presented in Section 2.2.
Krysanova et al. (1996), specifies limitations of selected pollution models: for example, AGNPS and ANSWERS are limited to watersheds of about 200 km², SWRRB was developed for agricultural basins as large as 600-800 km², and SWAT is intended to be applied in watershed up to 25,000 km².
The only known to the author models that can be applied for a such large area as the Midwest region are annual agrichemical load functions presented in section 2.3, which has been estimated by Battaglin (1993).
The model can be classified to the second category, i.e., models constructed within GIS, that have been presented in Section 2.2. Since (1) estimated concentrations and loads estimates are spatially and temporally distributed, and (2) one of the explanatory variables of the regression equation is flow length--an attribute of a distributed system, it can be characterized as a distributed system. On the other hand the agrichemical concentration or load is calculated at a given location by applying the values that characterize the total upstream drainage area to the regression equation. Thus the model can be considered as a lumped system.
The major differences between existing GIS models and the presented
one result from the spatial extend for which the model has been
developed. It utilizes data available for the whole US, thus the
model contains a limited number of parameters, e.g., it does not
contain neither time-decay nor length-decay coefficients. On the
other hand, presented model includes such components that has
not been utilized in GIS pollutant transport models as seasonal
index and selected watershed morphometry estimators.