|Hydrological model||First description|
|CREAMS||Rudra et al. (1985*)|
|AGNPS||Young et al. (1989*)|
|HSPF||Chew et al. (1991)|
|PRMS||Yan and Haan (1991*)|
|ACRU||Kienzle and Schulze (1992)|
|CASC2D||Julien et al. (1995*)|
|SWAT||Rosenthal et al. (1995*)|
|WASMOD||Schimming et al. (1995)|
|DHSVM||Nijssen et al. (1997)|
|SWIM||Krysanova et al. (1998*)|
|MIKE SHE||Bøggild et al. (1999*)|
|HMS||Yarnal et al. (2000)|
|MONERIS||Behrendt and Opitz (2000*)|
|WASIM||Rode and Lindenschmidt (2001)|
|ARCEGMO||Klöcking and Haberlandt (2002*)|
|DRIPS||Röpke et al. (2004*)|
|DWSM||Borah et al. (2004)|
|MARTHE||Thiéry and Amraoui (2001*)|
|TRACE||Herbst et al. (2005b*)|
|MIKE BASIN||Ireson et al. (2006*)|
MONERIS predicts diffuse emissions of nutrients for mid-sized to large catchments and includes a module that describes the emission path runoff. The temporal resolution of MONERIS is low amounting to one year and until present, the model fails to provide good estimates of nutrient emissions for catchments smaller than 50 km2. Correlations between modelled and measured loads were significantly well, but total nitrogen loads are generally overestimated (Behrendt and Opitz, 2000).
CREAMS (Rudra et al., 1985) is composed of three modules: hydrology, erosion, and chemistry and was created to predict diffuse emissions via runoff. CREAMS is a precursor of the leaching model GLEAMS. Hence, runoff is similarly modelled in both models and is based on the Soil Conservation Society (SCS) curve number approach. Results of a simulation on a field scale generally matched the observed order of magnitude (Yoon et al., 1992).
The GIS-based hydrological model SWAT (Rosenthal et al., 1995) has a modular structure and consists of hydrological, sedimentological, and chemical subroutines applicable to watershed-scales. The hybrid model spatially based on hydrological response units includes both, conceptual and physical approaches. A central part of SWAT is the general water balance equation. Surface runoff is determined by the SCS Curve Number approach. Frede et al. (2002) found that physical soil properties affect total runoff moderately, but highly influence surface runoff in SWAT. The model was found to be less efficient in predicting runoff in relation to land cover in a semi-arid watershed, therefore calibration was strongly recommended (Hernandez et al., 2000). Nonetheless, SWAT (Borah and Bera, 2004*) was found suitable for predicting annual flow volumes, sediment, and nutrient loads. Monthly predictions were generally good, except for months with extreme storm events and hydrologic conditions (Borah and Bera, 2004*).
Similar to SWAT, MIKE SHE (Bøggild et al., 1999) has a modular structure and calculates 3D surface, sub-surface, and stream flow involving distributed grid points. In a case study in an arctic environment, the model was found to overestimate measured runoff, because modelled surface retention of melting water was too low. However, there has been little information on how well MIKE SHE works simulating the transport of pesticides.
MIKE BASIN (Ireson et al., 2006*) is another product within the MIKE family and functions as an extension of ArcView. The water resources management tool is raster-based and works on a basin scale. In a case study, the main flaw of MIKE BASIN was that it failed to simulate high water flow, but otherwise satisfactory results were achieved (Ireson et al., 2006).
Further watershed-scale models, such as AGNPS (Young et al., 1989), CASC2D (Julien et al., 1995), and PRMS (Yan and Haan, 1991) were found to be eligible to simulate diffuse pollutant loads to surface waters (Borah and Bera, 2004). Muleta et al. (2006) tested AGNPS simulating soil erosion and nutrient transport in an Ethiopian catchment and succeeded in identifying hot spots of sediment and nutrient release.
The 2D raster-based model MARTHE (Thiéry and Amraoui, 2001) has successfully been tested to predict salinity in groundwater (Weinthal et al., 2005) and may be applied to simulate transport and fate of pesticides, as well. However until present, there are scarce published results of such simulations using MARTHE as platform.
In the model SWIM (Krysanova et al., 1998), a three-level scheme of spatial disaggregation from basins to sub-basins and to hydrotopes is used. The processes of transpiration, evaporation, and percolation within soils are implemented in SWIM. Retention of water and solutes is described by means of a dimensionless retention coefficient. SWIM was successfully validated for the Elbe catchment (Hattermann et al., 2005), but these authors recommend to accompany macroscale simulations of runoff with empirical investigations in small catchments, in order to identify the dominant hydrological processes.
TRACE is a recent development in hydrological modelling documented by Herbst et al. (2005b*). The Richards equation based numerical model calculates the three-dimensional saturated/unsaturated water flow. For the modeling of regional scale pesticide transport TRACE was combined with the plant module SUCROS and with 3DLEWASTE, a hybrid Lagrangian/Eulerian approach to solve the convection/dispersion equation (Herbst et al., 2005b*). A first-step application of TRACE/3DLEWASTE to a 20 km2 test area for a ten-year period was used to identify hot spots of isoproturon in groundwater. In general, the model results were consistent and reasonable.
Röpke et al. (2004*) developed a simple model (DRIPS) on horizontal pesticide transport. In this model, surface runoff is described as a function of rainfall and water infiltration. In contrast to the majority of hydrological models, other parameters such as slope and surface roughness are disregarded in this model. Horizontal attenuation is considered by implementing partitioning between the soluble and solid phases (KD) and degradation of pesticides. The authors found a good correlation between measured and modelled pesticide concentrations, but in an uncertainty analysis, the confidence interval spanned several orders of magnitude. Therefore, the results of DRIPS have to be evaluated cautiously, although this model seems to be an efficient alternative to more elaborate hydrological approaches.
All hydrological models explicitly describe water runoff, but they were seldom created to model exclusively transport of pesticides. Although hydrological models often use the same SCS curve number approach to relate land use to runoff as do leaching models including a runoff component, the former provide more realistic results because of their larger horizontal resolution. In addition, hydrological models differentiate between surface and subsurface runoff and thereby their performance is improved again. However, surface and subsurface attenuation processes of pesticides are often insufficiently described and for peak flow modelled water runoff did not always match measured results. Therefore, in order to calculate more realistic results, hydrological models need to be augmented in temporal resolution.
Most hydrological models can account for changes in land use. For example, Wang et al. (2005) reported a successful test of AnnAGNPS combined with a lake model, when several scenarios of sediment and nutrient loadings were calculated for different land use scenarios. In contrary, Klöcking and Haberlandt (2002) tested the model ArcEGMO for changes in land use and found that problems of impact studies in large river basins resulted mainly from a huge spatial heterogeneity of land use and a rough input database. These authors stated that simple approaches are needed to setup possible land use changes on the basis of easily available spatial data.
The role of crops for the fate of pesticides has been described in leaching models, but hydrological models only consider the effect of vegetation on surface roughness, rather than of pesticide export by harvesting. This deficit is easy to remove. In contrary, it remains doubtful if a more detailed description of retention and detention by nonlinear sorption and desorption would improve the performance of hydrological models. At least, an elaboration of sorption processes would also increase the number of input parameter required, which in turn would be barely available in high resolution at large scales.