Post-Model Validation of a Deterministic Watershed Model Using Monitoring Data

. Data are often collected during or after hydrological and water quality (H/WQ) model development, thus limiting the ability for direct comparison or use in calibration and validation. In this study, we demonstrate a way to validate the performance of three previously performed watershed simulations by comparing SWAT-simulated loads prior to 2011 with loads estimated from monitoring data collected after the SWAT models were completed. Water quality data, including total suspended solids, total phosphorus, and nitrate-nitrogen concentrations, were collected from the three watersheds from 2011 through 2013. Calibration and validation for WQ was accomplished for the Poteau River watershed using historical monitoring data. However, similar data were not available for the Strawberry and Upper Saline River watersheds, and their subsequent SWAT models were not calibrated or validated for WQ. A combination of regression, residual, and ANOVA analyses was used to post-validate the SWAT models beyond their simulation periods. Results indicated that the SWAT-simulated loads were in general agreement with the loads estimated from monitoring data for the Poteau and Saline River watersheds, but not for the Strawberry River watershed. A comparison of constituent loads modeled using SWAT and estimated from monitoring data for the Strawberry River watershed resulted in statistically different slopes and residuals and dissimilar ellipse orientation, range, and variance for each constituent. Load comparisons for the Saline and Poteau River watersheds using these same techniques resulted in variable slope and residual significance but improved ellipse orientation, range, and variance for each constituent. The approach used in this study was sufficient for analyzing the adequacy of simulated loads and could inform users concerning model performance when an uncalibrated model is used to make policy decisions and allocate resources.