Large area water resources development and management require an 
   understanding of basic hydrologic processes and simulation 
   capabilities at the river basin scale.  We define large areas as 
   river basins of thousands or tens of thousands of square kilometers.  
   Current concerns that are motivating the development of large area 
   hydrologic modeling include climate change, management of water 
   supplies in arid regions, large scale flooding, and offsite impacts 
   of land management.  Recent advances in computer hardware and 
   software including increased speed and storage, advanced software 
   debugging tools, and GIS/spatial analysis software have allowed 
   large area simulation to become feasible.  The objective of this 
   overview is to briefly describe the history, an overview of model 
   operation, and a description of model components of a river basin 
   scale model called SWAT (Soil and Water Assessment Tool).

   SWAT is the continuation of a long-term effort of nonpoint source 
   pollution modeling with the USDA-Agricultural Research Service (ARS).  
   In the early to mid-1970's, in response to the Clean Water Act, ARS
   assembled a team of interdisciplinary scientists from across the U.S. 
   to develop a process-based, nonpoint source simulation model.  From 
   that effort, a model called CREAMS (Chemicals, Runoff, and Erosion 
   from Agricultural Management Systems) was developed (Knisel, 1980).  
   CREAMS is a field scale model developed to simulate the impact of 
   land management on water, sediment, nutrients, and pesticides leaving
   the edge of a field.  By the early and mid-1980's, several models 
   were being developed with origins from the original CREAMS model.  
   GLEAMS (Groundwater Loading Effects on Agricultural Management Systems)
   (Leonard et al., 1987) concentrated on pesticide and nutrient 
   groundwater loadings.  A model called EPIC (Erosion-Productivity Impact 
   Calculator) (Williams et al., 1985) was originally developed to 
   simulate the impact of erosion on crop productivity and has now 
   evolved into a comprehensive agricultural management, field scale, 
   nonpoint source  loading model. 

   Other efforts involved modifying CREAMS to simulate complex watersheds 
   with varying soils, land use, and management.  One effort was the 
   AGNPS (Agricultural Nonpoint Source) (Young et al., 1987) model. 
   AGNPS is a spatially detailed, single event (storm) model that 
   subdivides complex watersheds into grid cells.  A model called SWRRB 
   (Simulator for Water Resources in Rural Basins) (Williams et al., 
   1985; Arnold et al., 1990) was developed to simulate nonpoint source 
   loadings from watersheds.  SWRRB is a continuous time (daily time 
   step model) that allows a basin to be subdivided into a maximum of 
   ten subbasins.

   To create SWRBB, the CREAMS daily rainfally hydrology model was 
   modified for application to large, complex, rural basins.  The major 
   changes involved were (a) the model was expanded to allow simultaneous
   computations on several subbasins to predict the basin water yield; 
   (b) a return flow component was added; (c) a reservoir storage 
   component was added for use in determining the effects of farm ponds 
   and reservoirs on water and sediment yield; (d) a weather simulation 
   model (rainfall, solar radiation, and temperature) was added to 
   provide for longer-term simulations and more representative weather 
   inputs, both temporally and spatially; (e) a better method was 
   developed for predicting the peak runoff rate; (f) a crop growth 
   model was added to account for annual variation in growth; (g) a 
   simple flood routing component was added; (h) components were added 
   to simulate sediment movement through ponds, reservoirs, streams, 
   and valleys; and (i) transmission losses were calculated.

   In the late 1980's, most of the SWRRB model development has been 
   focused on problems involving water quality.  Example additions 
   include the GLEAMS (Leonard et al., 1987) pesticide fate component, 
   optional SCS technology for estimating peak runoff rates, and newly 
   developed sediment yield equations. These and other less significant 
   developments extended SWRRB's capabilities to deal with a wide 
   variety of watershed management problems.

   Also in the late 1980's, the Bureau of Indian Affairs needed a model 
   to estimate the downstream impact of water management within Indian 
   reservation lands in Arizona and New Mexico.  SWRRB was utilized for
   smaller watersheds within the basin (up to a few hundred square 
   kilometers), but it was necessary to simulate streamflow from much 
   larger basins (several thousand square kilometers).  This required 
   the basin to be divided into several hundred subwatersheds.  SWRRB 
   was limited to ten subbasins and also had a simplistic routing 
   structure with outputs routed from the subbasin outlets directly to 
   the basin outlet.  This problem led to the development of a model 
   called ROTO (Routing Outputs to Outlet) (Arnold et al., 1995). ROTO 
   was developed to take output from multiple SWRRB runs and route the 
   flows through channels and reservoirs.  ROTO provided a reach routing 
   approach and overcame the SWRRB subbasin limitation by "linking" 
   multiple SWRRB runs together.  Although this approach was effective, 
   the input and output of multiple SWRRB output files was cumbersome 
   and required considerable computer storage.  Limitations also occurred 
   because all SWRRB runs had to be made independently, and then input 
   to ROTO for the channel and reservoir routing.  Thus, the SWAT model 
   was developed by merging SWRRB and ROTO into one basin scale model.  
   SWAT allows a basin to be divided into hundreds or thousands of grid 
   cells or subwatersheds.  SWAT is still a continuous time model (daily 
   time step) that is required to look at long-term impacts of management 
   (i.e., reservoir sedimentation over 50-100 years) and also timing of 
   agricultural practices within a year (i.e., crop rotations, planting 
   and harvest dates, irrigation, fertilizer, and pesticide application 
   rates and timing).


                         MODEL OPERATION

   SWAT is a continuous time model that operates on a daily  time  step.  
   The objective in model development was to predict the impact of 
   management on water, sediment and agricultural chemical yields in 
   large ungaged basins.  To satisfy the objective, the model (a) is 
   physically based (calibration is not possible on ungaged basins); 
   (b) uses readily available inputs; (c) is computationally efficient 
   to operate on large basins in a reasonable time, and (d) is continuous 
   time and capable of simulating long periods for computing the effects 
   of management changes. 

   SWAT uses a command structure for routing runoff and chemicals through 
   a watershed similar to the structure of HYMO (Williams and Hann, 1973).  
   Commands are included for routing flows through streams and reservoirs, 
   adding flows, and inputting measured data or point sources. Using a 
   routing command language, the model can simulate a basin subdivided 
   into grid cells or subwatersheds.  Additional  commands have been 
   developed to allow measured and point source data to be input to the 
   model and routed with simulated flows.  Also, output data from other 
   simulation models can be input to SWAT.

   Using the transfer command, water can be transferred from any reach 
   or reservoir to any other reach or reservoir within the basin.  The 
   user can specify the fraction of flow to divert, the minimum flow 
   remaining in the channel or reservoir, or a daily amount to divert.  
   The user can also apply water directly to a subbasin for irrigation.  
   Although the model operates on a daily time step and is efficient 
   enough to run for many years, it is intended as a long term yield 
   model and is not capable of detailed, single-event, flood routing.