A modified response matrix method to approximate SWAT for computationally intense applications

Abstract

The computational burden of running a semi-distributed hydrological model numerous times, such as for optimization applications, can be exorbitant. This study provides a surrogate model to estimate streamflow, nutrient, and sediment export under spatially distributed management decisions. Specifically, we surrogate the Soil and Water Assessment Tool (SWAT) using a modified response matrix (RM) approach. A traditional RM approach applied to SWAT assumes hydrological responses are approximated by linear functions of management decisions, and falls short in accounting for in-stream and reservoir processes. Here, we explain and illustrate three key modifications that address interaction effects between co-located conservation practices and in-stream and reservoir processes affecting nutrient and sediment loads. The modified RM approach provides excellent estimation (Nash-Sutcliffe Efficiency, NSE >0.95) for streamflow and nutrient export throughout the stream network and provides very good estimation (NSE >0.85) for sediment export at most, though not all, points in the stream network.