1. INTRODUCTION

This research is focused on the transport of nutrients and herbicides in surface waters of the Midwest and the extent to which Geographic Information System (GIS) technology can be used to automate and improve the computation of spatio-temporal distribution of selected agricultural chemicals. Up to now, there have been elaborate flow models and pollutant transport models in use at the watershed scale. Yet, these models focus mainly on the detailed description of the small-scale physical phenomena and they would require huge computer and data resources if implemented on the scale considered here. In the light of the existing projects, it is important that the aim of this project is to describe the distribution of agricultural chemicals over such a large region as the Upper Mississippi River basin. Here, the Upper Mississippi River basin refers to the Mississippi River watershed which has its outlet at the confluence point of the Mississippi River and the Missouri River. The watershed area of 450,000 km² extends over eight states: Illinois, Indiana, Iowa, Michigan, Minnesota, Missouri, and South Dakota.

In this work, the words river, stream, and stream/river reach are used as words with the same meaning. Similarly, drainage area, watershed and basin are considered here as equivalent terms. Sometimes one word nitrate is used to represent nitrate plus nitrite as nitrogen. Unit watershed, elementary watershed, and modeling unit refer to the smallest drainage area or partial drainage area into which the region under investigation is divided. Each modeling unit is considered as lumped system. Names of: maps, computer files, database fields, as well as computer commands are printed in Courier fonts.

1.1 Motivation

It is commonly known that the agricultural activity endangers the quality of the surface waters. Farmers apply chemical nutrients to increase soil fertility and use pesticides to control unwanted plants and destructive insects. About 60 percent of the pesticides (Kolpin et al., 1991 after Gianessi and Puffer, 1990) and nitrogen fertilizers (Kolpin et al., 1991, after US Environmental Protection Agency, 1990) used in the USA are applied to cropland in twelve Midcontinental States (Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, North Dakota, Ohio, South Dakota, Wisconsin). Four major herbicides, alachlor, atrazine, cyanazine and metolachlor account for about 73 percent of the pesticides applied (Scribner et al., 1994 after Gianessi and Puffer, 1990).

Numerous studies have been performed on a small scale. But until recently, few studies have been conducted to evaluate the contamination of rivers the Upper Mississippi-Missouri and Ohio River basins.

In 1990, the United States Department of Agriculture (USDA) initiated a Management System Evaluation Areas (MSEA) program. The major purpose of this program is to determine the influence of the agricultural practices on the water quality in the Midwest and to identify management systems that can protect water quality. Ten study areas were established to gather data for a better understanding of the factors and processes that control the fate and transport of agricultural chemicals (Hatfield et al., 1993). Also, the United States Geological Survey (USGS) conducts studies to determine the distribution, transport and persistence of selected herbicides, insecticides, and inorganic nutrients in the Mississippi River and tributaries (Battaglin et al., 1993).

The Geographic Information System (GIS) offers a unique opportunity to formulate more objective and consistent methods to utilize collected data and to assess water quality over large areas ( Maidment,1996). This work explores the applicability of GIS technology to regionalize to large basins the results of small scale water quality studies, and to evaluate the temporal and spatial distribution of the loads and concentrations of selected agricultural chemicals in the surface waters of the Midwest region.

1.2 Objectives

The general objective of this research is regionalization of watershed scale measurements. The original approach to achieve this goal assumed that models of two important processes would be developed using data collected in the Walnut Creek watershed near Ames, Iowa: (1) the relationship between the chemical concentration in a stream and such parameters as the chemical application on the field, water runoff, precipitation, time of the year, and watershed characteristic and (2) the model that describes chemical losses as the chemical is carried in a water parcel downstream (Maidment and Mizgalewicz, 1995). Daily flow rates (water years 1991-93) and weekly observations of nitrate plus nitrite as nitrogen and atrazine concentrations are available for four sites: site 220 (drainage area = 3.6 km², July 1990 - April 1994), site 310 (25.4 km², March 1992 - April 1994), site 320 (~38km², July 1990 - April 1994), and site 330 (51 km², July 1990 - April 1994) (Soenksen et al., 1993, Soenksen, 1994). Unfortunately the author of this dissertation have been unsuccessful in obtaining the information about chemical application on the field for the years 1990 - 1994 and thus the proposed methodology of regionalization watershed scale studies, and some of the original sub-goals had to be modified.

The general objective can be divided into the following goals:

1. To formulate statistical models capable of representing the spatio-temporal variability of agricultural chemicals in surface waters;
2. To evaluate the applicability of the GIS technology for deriving stream and watershed characteristics that have influence on the chemical transport process;
3. To develop a methodology for calculating the average monthly flow rate in ungauged streams;
4. To build a model for predicting concentrations and loads under different hydrologic conditions and for different agricultural chemical application rates.

1.3 Scope of study

The following restrictions define the scope of this research:

1. The analysis is limited to two selected agrichemicals - the nutrient, nitrate plus nitrite as nitrogen, and the herbicide, atrazine. These two chemicals are chosen because they are representative of nutrients and herbicides, respectively, and because they are present in measurable quantities in many Midwest streams and rivers. Although the research is performed for two agrichemicals, a similar procedure could be used for other herbicides and nutrients.
2. Since the model is constructed using GIS technology, and since the available data are limited, the concentrations and loads of the agricultural chemicals in surface waters are described by regression equations that have limited application for making predictions.
3. Observed loads and concentrations as well as chemical application rates are published in the USGS reports. The watershed and stream parameters are derived from the Digital Elevation Model that is available on Internet. Flow rate and the precipitation depth are extracted from CDROMs (Compact Disk Read Only Memory) published by Hydrosphere (Data utilized in this study is described in Section 3.)
4. Although the regression equations are estimated from data gathered in 151 sampling sites located over the Midwest region, the GIS model is developed for the Cedar-Iowa River basin.
5. Model formulation and model parameter estimation are constrained by the available computer resources (limited processor time and disk storage on SUN Sparc Station IPX).

1.4 Project summary

This research can by divided into the following steps:

1. Preparation of measurement data for statistical analysis. This step involves data entry from printed USGS reports, filling in missing values and data correction, conversion of data into common units.
2. Development of the general form of a equation that can explain spatial and temporal variation of the chemical concentration and load in the Midwest rivers.
3. For each watershed associated to measurement point, determining the drainage-basin morphometry and the chemical application rate. The 15second (500 m resolution) digital elevation model is utilized in the following steps:
   • delineation of the stream network from the digital elevation model DEM,
   • location of stations in which the water samples were collected on the delineated river network,
   • estimation of the drainage area associated to each measurement point,
   • calculating parameters that characterize this drainage area, and
     
   • calculating watershed-averaged chemical application rate.
4. Preliminary analysis of the spatio-temporal distribution of the constituent loads and concentrations in the Mississippi-Missouri and Ohio River basins. This step contains development of the regression equations that relate measured agricultural chemical loads and concentrations in a given point to such explanatory variables as the annual agricultural chemical application and the flow rate.
5. Formulation and application of the methodology for estimation of monthly flow rate in ungauged streams. This methodology is utilized for a selected watershed, i.e., the Iowa-Cedar River basin. The 3 arc-second (100 m resolution) DEM is chosen as a base map for determining the flow direction and for subdivision of the Iowa-Cedar watershed into small drainage units (mostly 20 - 50 km²). This step is comprised on the following:
   • constructing a spatio-temporal database of monthly average flow rate and monthly average precipitation rate for period from 1960 to 1992,
   • dividing the Iowa-Cedar River into small drainage units, converting resulting map from raster into vector format, and building "flow" topology,
   • interpolating or extrapolating recorded flow rate over ungauged drainage units using precipitation depth, estimated flow rate, average runoff coefficient, and drainage area as weighting factors.
6. Estimation of the spatio-temporal distribution of the agricultural chemical concentration and load. In this step the flow rate identified in step 5 is applied in the regression equation developed in step 4.

1.5 Contributions of study

This research has the following contributions to the knowledge:

1. The development of a method for application of the GIS technology to determine factors that influence the process of mobilization and transport of agricultural chemicals;
2. The estimation of the general temporal (monthly) pattern of the average atrazine and nitrate loads as well as concentrations in the Midwest region;
3. The application of the GIS technology to store time series of recorded monthly flow rate and precipitation depth (creation of spatio-temporal database);
4. The development of a routine for estimating the expected flow rate in ungauged streams;
5. The formulation of a GIS spatial model from which concentrations as well as loads of nitrate plus nitrite and atrazine can be calculated for different hydrologic scenarios;

In addition to the contributions listed above, during this research a set of new GIS tools supporting hydrologic modeling has been introduced. The following procedures have been developed:

1. Automated watershed division into hydrologic sub-units;
2. Improvement of the major flow paths delineated from a digital elevation model (DEM), and therefore enhancement of the delineated stream network and watershed boundaries.
3. Calculation of a new stream order system (topologic characteristics of a stream network similar to the Shreve or Strahler ordering methods) that makes flow and transport calculations very efficient; and
4. Hydrologic modeling tools that do not exist in ArcView GIS but they are supported in other GIS software ( e.g. flowaccumulation--accumulating an entity when traveling downstream, determining drainage area upstream of a given location, identification of the transport path).