Sensitivity Analysis of National Water Model Using DELSA

Each upgrade of the National Water Model (NWM, operational from 2016) involves a calibration procedure which has been expanded every year for the past 3 years. Here we aim to perform sensitivity analysis on the parameters already being used in the NWM calibration exercise as well as some of the potential parameters that the next calibration exercise could benefit from, such as the channel properties which control the lag and timing of the peak flow. We used the Distributed Evaluation of Local Sensitivity Analysis (DELSA), a hybrid local-global sensitivity analysis (SA) method, for extracting useful information on the importance of each parameter, with the added advantage of being relatively low computational cost compared to other common SA methods such as Sobol.

Different parameters have an impact on different aspects of the model, such as the baseflow, runoff-partitioning, etc. Some of the parameters used in the calibration are soil (bexp, smcmax, dksat, slope, lksatfac), surface storage and infiltration (refkdt, retdeprtfac), deep groundwater (zmax, expon), vegetation (cwpvt, vcmx25, mp), and snow (mfsno) parameters. We will also look at new additional parameters which have not been calibrated before. More than 50 basins from the GAGES II reference basins scattered throughout the continental United States have been selected for this study to cover a wide range of flow and climate regimes. This work evaluates the degree of streamflow variation with respect to the change in the parameter values as well as variation in several metrics such as bias, correlation coefficient, RMSE, NSE, etc. to cover different aspects of flow. The study has been performed at both hourly and daily time scale.