Y. R. Li¹, G.H. Huang¹, Y. F. Li² , J. Struger³ and J. D. Fischer⁴

¹ Department of Environmental Engineering, University of Regina, Regina, Sask., Canada

² Meteorological Service of Canada, Environment Canada, Toronto, Ont., Canada

³ EHD, Environment Canada, Burlington, Ont., Canada

⁴ CWS, Environment Canada, Burlington, Ont., Canada

Abstract: An integrated modeling system was developed to predict runoff losses of pesticides from agricultural lands. The system is an integration of a mathematical model, a database system, and a geographic information system. Information of soil type, land use, land slope, watershed boundaries, precipitation, pesticide usage, as well as physical and chemical properties of pesticides have been input to a GIS, managed through a database, and used for further modeling studies. The developed modeling system can simulate pesticide losses due to runoff based on the consideration of emission, degradation, adsorption and desorption of pesticides, as well as their movement in dissolved and adsorbed phases. The modeling outputs were in turn put into the database, such that runoff patterns along with pesticides losses could be further simulated by using a database management system. The final results could then be visualized through GIS. The developed modeling system was applied to the Kintore Creek Watershed, Ontario, Canada, for simulating losses of atrazine from agricultural lands. A water quality monitoring project was carried out from 1988 to 1992 in the watershed to detect conditions of surface water pollution due to the use of pesticides. The modeling outputs were verified through the monitoring data, demonstrating reasonable prediction accuracy. The result indicated that the model provides an effective means for forecasting pesticide runoff from agriculture lands.

Keywords: agriculture, environment, modeling, monitoring, nonpoint source, pesticide, runoff, watershed

Runoff losses of pesticides and their transport in surface waters are principal processes leading to their widespread dispersion in the environment (Baker, 1988; Leonard 1988; Burgoa et al., 1995). These processes are main causes for nonpoint source pollution of pesticide from agricultural lands. Runoff losses of pesticides are generally affected by interactions among pesticide properties, soil and weather conditions, and pesticide-application patterns.

The existing methods for estimating pesticide runoff from agricultural lands include lump-parameter and distributed-parameter approaches (Crawford et al., 1973; Haith, 1980; Donigian et al., 1983; Young et al., 1987). In the lump-parameter method, it was assumed that compartments (air, water, soil) in a watershed are homogeneous in composition and properties. Thus, spatial variations of model parameters are neglected, resulting in a single soil or water environment for the entire watershed.

In the distributed-parameter approach, the parameters’ spatial variations were considered. However, the method required large amounts of data that were often unavailable. One potential approach for improvement is to link pesticide runoff model with a GIS and a database system. Thus, spatial data can be organized in the GIS and activated in the database system, such that a variety of system parameters can be well prepared and organized for further modeling computation. The objective of this study is to develop such an integrated modeling system for simulating pesticide runoff losses in watersheds. The developed model will then be applied to a case study in the Kintore Creek Watershed, Ontario, Canada, for demonstrating its practical applicability.

Figure 1 shows the fate of pesticides after being applied to agricultural fields. Pesticides leaving a treated area through runoff constitutes only a small percentage of that applied, since most of the applied pesticides are dissipated by other processes. These processes must be conceptualized before a simulation model can be developed.

Methods of pesticide application will affect the amount of pesticides reaching the top layer of soils. Pesticides applied by ground equipment mainly reach the soil surface if that is the intended target. Aerially applied pesticide may surfer significant drift and volatilization losses and be distributed between both plant and soil surface depending on the degree of canopy density (willis et al., 1980). Once a pesticide reaches either the plant or soil surface, it is degraded or transformed by chemical, biological, and photochemical processes, and is also subject to volatilization losses. Although dissipation is by a combination of processes, pesticide persistence with time on foliage and soil may be approximated by a first decay function (Nash, 1980; Willis et al., 1980). Pesticide half-lives based on the decay function vary from a few hours to months depending on the pesticide and the local environment.

Once a rainfall event begins, pesticides sprayed or dusted on plants can be washed into soils. Obviously, the amount washed off depends upon the sorption of the pesticide by the plant, the type of plant surface, and the intensity and length of the storm. However, scarcely any data are available to provide estimates of this process, much less a quantitative expression. At the present time, all of any compound applied to plants is assumed to be washed immediately to the soil surface.

The residue of pesticide available in a given rainfall event will greatly influence the concentration of pesticide in runoff. However, this contribution is highly variable, depending on pesticide chemistry and formulation. Most of pesticides are adsorbed to the soil to some degree. A linear relation between the concentrations in the soil and water is often found (Evans and Duseja, 1973). The adsorption of pesticides to soil particles is a reversible process. Generally, adsorption involves the formation of a monolayer of pesticides around a soil particle. This is followed by the accumulation of multilayers of pesticides (Haque and Freed, 1974).

The general behaviors of pesticides in soil can be quantified by considering the following processes: emission from soil, chemical, biological and photo degradation, dissolution from granule into water, adsorption and desorption, runoff, and leaching. In this study, the process of pesticide uptake by plants is not considered.

A watershed is usually composed of fields of different crops, soil types, and land slopes, leading to various of ways for pesticide runoff. To reflect this spatial distributive, the watershed can be conceptually divided into several grid cells, with each of them being uniform with respect to crop species, soil type, hydrological conditions, and topological characteristics. Thus, each grid cell can be treated as a unit for pesticide behavior with inflows from and outflows to other grid cells. A grid cell structure is a discrete representation of a terrain, based on identical square cells arranged in rows and columns. Grids are used to describe spatially distributed parameters. The number of grid cells in a watershed varies with the watershed’s size and the cells’ dimensions, but should be large enough to account for the watershed’s spatial variability.

Output from a grid cell at the top of the watershed is routed to cells below it and/or to the stream channels, and finally to the watershed’s outlet. Water and sediment from cells without pesticide application can dilute pesticide concentrations in solid and liquid phases. In developing the model, the following assumptions were made: (1) The pesticide transport through sediment was insignificant since dissolved pesticide runoff losses greatly exceed solid-phase losses (Haith, 1980; Huber et al., 1998); (2) a thin layer of surface soil (less than 1.0 cm) mixes completely with rainwater in the process of pesticide transfer from soil to runoff (Donigian et al., 1977; Frere et al., 1980; Leonard and Waucope, 1980); and (3) a unique slope direction and magnitude within a single grid cell must be assigned to avoid ambiguous flow directions.

Fig. 1  Fate of pesticides after applied to agricultural fields

(1) Hydrological system

A model developed by the US Soil Conservation Service (SCS) was utilized to estimate runoff during rainfall events with a known precipitation volume. The volume of runoff (Q) depends on the volume of precipitation (P) and the initial abstraction (I). The initial abstraction (I) is a fraction of total rainfall that does not appears as runoff (McCuen, 1981):

                              (1)

The initial abstraction (I) consists mainly of interception, infiltration and surface storage, all of which occur before runoff begins. This can be described as follows:

                                    (2)

                                    (3)

where CN is an index representing the combined effects of soil hydrologic characteristics, land use and antecedent moisture conditions (McCuen, 1981).

(2) Pesticides

Residues of atrazine were described as follows (Li et al., 2000a):

                                      (4)

where R_t is the residue of pesticide in time t (days) after pesticide application; U is the total application of pesticide; F_t is the daily emission factor of pesticide; t_1/2 is the half-life of pesticide in the soil. For atrazine t_1/2 is approximately equal to 60 days (Mackay et al., 1997). The residues left in soils due to the use of atrazine in the previous years were not considered here. The daily emission factors F_t (Li et al., 2000b) were obtained based on the annual emission factors (Scholtz et al., 1997).

The fraction of pesticides in runoff is quantified as follows:

                        (5)

where C_r is the pesticide soluble concentration in runoff, R is the residue of pesticide when rainfall begins, K_d is the distribution coefficient derived from linear adsorption coefficient for organic carbon K_oc, which can be calculated for different contents of organic carbon (OC) as follows:

                                 (6)

Leonard et al. (1987) developed a functional relationship to relate the extraction coefficient (B) to the distribution coefficient (K_d):

                              (7)

(3) Routing component

The routing of pesticide runoff can be analyzed through a continuity equation. The calculation process can be summarized as follows: (a) the amount of mass generated in each cell is first calculated for each time step; and (b) this mass generated in a cell is then transported to the next cells. The continuity equation for the pesticide runoff can be written as:

                                           (8)

where is the amount of pesticide leaving a cell through runoff (ug), is the sum of all pesticides entering a cell from other cells through runoff (ug), and is the amount generated within a cell.

The developed model was applied to the Kintore Creek area, Ontario, Canada. Two adjacent sub-watersheds in the area (Figure 2) were considered. They are equal in size, and both have highly erodible landscapes and similar cropping patterns. A study of the effects of farm conservation practices on pesticide transport to surface water was conducted by the Ecosystem Health Division and the Upper Thames River Conservation Authority of Environment Canada in 1988 and 1991. Runoff losses of atrazine, which is one of most heavily used pesticides in this area, was monitored. Further details of the watershed were provided in Merkley and Glasman (1984). Table 1 shows the detailed characteristics of the sub-watersheds.

In the western sub-watershed, landowners employed conservation techniques for reducing pesticide losses, which included the mulch-finishing of row crops, the planting of forage and cover crops, the no-till and reduced till practices, the installation of sediment control basins, the slope stabilization along stream banks, and the planting of trees. In comparison, landowners in the eastern sub-watershed used conventional tillage practices of fall moldboard ploughing with a corn-wheat-alfalfa rotation.

Fig. 2  Locations of the Kintore Creek and Thames River Watersheds

(1) Spatial database

A spatial database for the Kintore Creek Watershed was developed to support simulation modeling of pesticide runoff losses from agricultural lands. The Kintore Creek Watershed and its sub-watershed boundaries were obtained from the Upper Thames River Conservation Authority. The base-map scale is 1: 24,000, and the grid cell resolution is 10 hectares. All spatial data are digitized and geo-referenced using MapInfo.

Topography - Generally, characteristics of the terrain can be obtained from a digital elevation model (DEM). However, at the present time, DEM data with the required resolution for the Kintore Creek Watershed is unavailable. Thus, information of slope gradient was derived from a field survey, with the result shown in Figure 3a.

Soils - Soil information for the watershed is provided by the Oxford County (Charlton, 1995). Within the watershed there are four major soil associations which directly influence the fate and transport of the pesticides. The first association represents soils with high infiltration rates, including Fox sandy loam, Huron clay loam, and Guelph loam. The second association includes soils with moderate infiltration rates, such as Embro silt loam, and Tavistock silt loam. Soils of this association have moderate rates of water transmission, with moderately fine to moderately coarse textures. The third association represents soils with slow infiltration rates, such as Maplewood silt and Crombie silt loam, and the fourth association includes those with very slow infiltration rates (e.g. Muck soil). Figure 3b shows distribution of soils in the Kintore Watershed.

Land Use - Information of land use in the Kintore Creek Watershed was from the Upper Thames River Conservation Authority, as shown in Figure 3c. Agricultural practices occupy the majority of lands in the watershed, with beans and corn being the main crops.

(2) Pesticide monitoring program

A pesticide monitoring program was initiated at the watershed’s outlet to evaluate the modeling results. The monitored parameters include pesticide concentration, flow velocity, and water level. Pesticide concentrations during runoff events were monitored using ISCO automatic water samplers (model 2100), and those during storm events were monitored using ISCO liquid level sampler actuators (model 1640). The first five samples were collected with one-hour intervals; and the remaining three were with two-hour intervals. Each event, therefore, was monitored for a period of eight hours. The flow velocity was measured by a flow meter with an ultrasonic sensor, and the water level by a pressure gauge. Parts of the monitoring results are shown in Figure 4.

All landowners in the watershed were investigated for information of pesticide application. Visits were made in late summer of each year. Information gathered included type of crops, number of acres allocated for each crop, type of pesticide, number of acres applied with the pesticide, and rate and time of pesticide application. Table 2 shows parts of pesticide-application information. Figure 5 illustrates the pesticide application pattern in 1989 and 1990.

Fig. 4  Runoff discharges and atrazine concentrations and loads in the west sub-watersheds (July 30, 1991)

Fig. 5  Pesticide application pattern in 1989

First, spatial data related to soil, land use, slope, pesticide application pattern were loaded into MapInfo. With an area of 1300 ha, the watershed was divided into 175 computational elements with each representing approximately 10 hectares. The overlay of the obtained spatial data was realized through Map Basic and MapInfo. The results were summarized into new tables and exported to a database managed by an interface coded with Visual Basic. Many input parameters for the model could then be estimated using a query function within the database; moreover, input files for the model could be organized in the database as well. After the input parameters were formatted for further simulation, thirty-five rainfall events were input to the model, with the outputs being managed by the database and graphically presented by the MapInfo.

Spatially, modeling outputs (including pesticide concentration and runoff volume) for each grid cell can be analyzed to graphically present the response of the watershed. Figures 6a to 6d illustrate the watershed’s dynamic responses to a rainfall event of 1.2 inches that occurred on August 28, 1990. The results would facilitate identification of critical areas for mitigation actions. The time series of modeling outputs in the watershed’s outlet were compared with those of monitored data, with results displayed in Figures 7a to 7b. Figure 8a is a plot of the predicted and observed peak discharges for 35 rainfall events, with a high correlation level (r = 0.979). Figure 8b shows a plot of predicted and observed peak pesticide loads, with a correlation level of r = 0.923.

As the results indicate, a correlation level of r = 0.979 existed between the observed and predicted peak runoff discharges, indicating that the model provides a reasonable prediction accuracy. Prediction errors existed in several individual runoff events, which may be attributed to a number of causes, such as errors in input data, assumptions associated with the model, and uncertainties of sub-system interrelationships. For example, the method used for predicting runoff is distributive by nature (Michaud and Sorooshian, 1994; Srinivasan and Arnold, 1994), while it requires aggregation of information such as land use and soil type for describing the runoff process. Although the distributive method is considered more accurate in representing spatial characteristics of a watershed, its accuracy in simulating runoff processes does not significantly get improved (Michaud and Sorooshian, 1994). Also, the method assumes that soil moisture conditions are homogeneous throughout the watershed while, in reality, the soil moisture levels vary seasonally.

Spatially variable rainfall intensity can greatly influence the variations associated with the predicted pesticide runoff patterns (Amorocho, 1979; Holbert, 1989). Unfortunately, due to the limitation of data availability, the rainfall data for this modeling study were from the London Station that is ten miles from the study watershed. This could be a major contributing factor that is responsible to the prediction errors.

A number of factors contributed to the errors in predicting pesticide runoff losses. Pesticide loads from agricultural lands are related to a number of factors that are often non-quantifiable. For example, farmers who were investigated should report the types of crops they plant, the area of their pesticide application, the time when they apply the pesticides, and the amount of pesticide application. These require that the farmers keep precise records of their croping and management activities. In reality, however, most of the farmers are unable to provide exact records of them. Consequently, the exact pesticide application time was unknown in this modeling study, replaced by an interval value (i.e. between late May and early June). The potential for pesticide runoff loss is strongly affected by the timing of pesticide application in relation to runoff-producing rainstorms (Leonard et al., 1979). Therefore, uncertainties in the timing would significantly affect the modeling results.

The proposed model assumes that a thin layer of surface soil (less than 1.0 cm) mixes completely with rainwater in the process of pesticide transport from soil to runoff (Donigian et al. 1977; Frere et al., 1980; Leonard and Waucope, 1980). In other words, the model does not consider the contribution of pesticide loss from soils in the deeper layer. Ahuja et al. (1983) reported that the pesticides may be transferred to runoff from a soil depth of 2.0 cm, and that the degree of mixing between soil and rainwater (and thus the rate of pesticide transfer) decreases exponentially with depth below the surface. They indicated that the transfer of pesticide from below soil surface to the surface was a result of turbulent mixing of soil water caused by a raindrop impact. Thus, an effective depth of complete mixing may be assumed for soils with high infiltration rates, but not for those with low rates.

A simple mixing cell approach was used to calculate the transport of pesticides across the watershed. This approach is subject to numerical dispersion affected in part by the cell size and the assumption of complete mixing. As the watershed is conceptually divided into several small elements, its response at the outlet would be an integration of multiple individual responses. The shape of the concentration vs. time curve at the watershed outlet would depend on (a) the traveling time of runoff from each element, (b) the degree of mixing during the transport, and (c) the effects of channel processes such as deposition and reentrainment. With the above complexities, the simplifications made in the model could cause prediction errors.

Many modeling parameters are uncertain, which may lead to variations in the modeling outputs. Therefore, in addition to testing the model against observed data, it is important to know the sensitivity of modeling outputs to variations in its inputs. Such information is useful for identifying critical factors in mitigating the pesticide losses. Figure 9 shows results of the sensitivity analyses. It is indicated that the model is most sensitive to variations of precipitation and CN number, while the other factors are much less sensitive.

In this study, integration of GIS and database with simulation models demonstrates an effective approach for conducting complicated, large-scale watershed modeling works. The system allows users to conveniently manage vast amounts of modeling inputs and outputs, and present them graphically. It also facilitate effective model-based scenario analyses of the watershed system. The developed system was applied to simulate atrazine losses from agricultural lands in the Kintore Creek Watershed, Ontario. The modeling outputs were verified through monitoring data, demonstrating reasonable prediction accuracy. The result indicated that the model provides an effective means for forecasting the pesticide runoff from agriculture lands.

Ahuja, L.R. Release of a soluble chemical from soil to runoff. TRANSACTIONS of ASAE. 1982, 25(4), p948-956.

Amoracho, J. Spatially distributed variables in hydrologic modeling. In Lloyd, E.H., O’Donnell, T., Wilkinson, J.C. (ed.) The Mathematics of Hydrologic and Water Resources. Academic Press, 1979. San Francisco, CA.

Burgoa, B.; Wauchope, R.D.. Environmental Behaviour of Agrochemicals. Edited by Roberts, T.R. and Kearney, P.C.. 1995, John Wiley & Sons Ltd., p223-349.

Baker, D.B. Sediment, Nutrients and Pesticide Transport in Selected Lower Great Lakes Tributaries. USEPA-905/4-88-001. Great Lakes National Program Office, Chicago,IL.

Charlton,D Upgrade of Soil Survey Information for Oxford County. Ecological Services For Planning, Agricultural Canada. COESA 1995. Report No.: RES/MON-005/95.

Donigian, A.S., Jr., D.C. Beyerlein, H..H.Davis, and N.H.Crawford. 1977. Agricultural Runoff Management (ARM) model version II: refinement and testing. EPA 600/3-77-098. Environmental Research Laboratory, U.S. EPA, Athens, GA.

Donigian, A.S., Jr., J.C. lmhoff, B.R. Bichnell. 1983. Predicting water quality resulting from agricultural nonpoint source pollution via simulation—HSPF p. 200-249. In F.W. Schaller and G.W. Bailey (ed.) Agricultural management and water quality. Iowa State Univ. Press, Ames.

Evans, J.O., and D.R. Duscja. 1973. Herbicide concentration of surface runoff waters. Environ. Protec. Technol. EPA-R2-73-266.

Frere, M.H., J.D. Ross, and L.J. Lane. 1980. The nutrient submodel. Chap.4. In Knisel, W.G. (ed.) CREAMS: A field scale model for chemicals, runoff, and erosion from agricultural management systems. U.S. Dept. Agr. Conservation Res. Rep, 26,p65-87.

Haith, D.A. A mathematical model for estimating pesticide losses in runoff. J.Environ. Qual., 1980,7(1), p63-68.

Haque,R., V.H. Freed. Behavior of pesticides in the environment: Environmental chemodynamics. In: Residue Reviews. F.A.Gunther, ed. New York, 1974,52, p89-116.

Holbert, S.B. Development of a geographic information system based hydrologic model for stormwater management and landuse planning. Masters’Thesis. The University of Arizona, Tucson, Arizona. 1989,P8-37.

Huber, A.; M., Bach; H.G., Frede. Modeling pesticide losses with surface runoff in Germany. The sci. of Total Envion. 1998,233, p177-191.

Leonard, R.A., G.W. Langdale, and W.G. Fleming. Herbicide Runoff from Upland Piedmont Watersheds: Data and Implications for Modeling Pesticide Transport. J.Environ. Qual., 1979,8(2), p223-228.

Leonard, R.A., and R.D. Wauchope. 1980. The pesticide submodel. Chap.5, In Knisel, W.G. (ed.) CREAMS: A field scale model for chemicals, runoff, and erosion from agricultural management systems. U.S. Dept. Agr. Conservation Res. Rep, 26,p88-112.

Leonard, R.A., W.G. Knisel, and D.A. Still. 1987. GLEAMS: Groundwater loading effects of agricultural management systems. Trans. ASAE 30:1403-1418.

Leonard, Ralph A.. Pesticides in Surface Waters. Environment Chemistry of Pesticides. 1988, CRC, p46-p82.

Li, Y. F., M. T. Scholdz, and B.J. van Heyst, 2000a, “Global gridded emission inventory of a - hexachlorocyclohexane”, J. Geophys. Res., 105, D5, 6621-6632.

Mackay, D., Shiu, W.Y., Ma, K.C.. Illustrated handbook of physical-chemical properties and environmental fate for organic chemicals. Volume V. 1997, Lewis Publishers, New York. P67-75.

McCuen RH. A guide to hydrologic analysis using SCS methods. Engelwood Cliffs: Prentice Hall, 1981: 62.

Merkley, C., B.Glasman. Kintore Creek Watershed Study: Background Report. 1984. Upper Thames River Conservation Authority.

Michard, J. and S.Sorooshian. Comparison of simple versus complex distributed runoff models on a midsized semiarid watershed. Water Resources Research, 1994,30(3),p593-605.

Nash, R.G. Dissipation rate of pesticides from soils. CREAMS: A field scale model for chemicals, runoff, and erosion from agricultural management systems. USDA Conserv.Res.Rep. 26. U.S.Gov. Print. Office, Washington, DC. 1980, p595-606.

Scholtz, M.T., A.C. McMillan, C. Slama, Y.F. Li, N. Ting, and K. Davidson, Pesticide emissions modelling: development of a North American pesticide emissions inventory, CGEIC Final Report # CGEIC-1997-1, May 1997.

Srinivasan, R. and J.G. Arnold. Integration of a basin-scale water quality model with GIS. Water Resources Bulletin, 1994, 30(3),p453-462.

Willis, G.H., W.F. Spencer, and L.L. McDowell. The interception of applied pesticide by foliage and their persistence and washoff potential. 1980, CREAMS: A field scale model for chemicals, runoff, and erosion from agricultural management systems. USDA Conserv.Res.Rep. 26. U.S.Gov. Print. Office, Washington, DC. p595-606.

Young, R.A., C.A. Onstad, D.D. Bosch and W.P. Anderson. Agricultural nonpoint surface pollution models (AGNPS) I and II model documentation. USDA, Agricultural Research Service, Conservation Research Report 35, 1987, Washington.