Fast and efficient optimization of hydrologic model parameters using a priori estimates and stepwise line search

Abstract

Summary Routine availability of hydrological, geological, and other physiographic data today allows us to obtain a priori estimates of hydrologic model parameters prior to explicit model calibration. When informative a priori estimates of model parameters are available, the problem of hydrologic model calibration becomes one of filtering, i.e. improving the a priori estimates based on observations of input and output to and from the hydrologic system, respectively, rather than one of bounded global optimization based solely on the input and output data as in traditional model calibration. Given that global optimization is computationally very expensive and does not, in general, transfer the spatial patterns of soil and land surface characteristics to the model parameters, the filtering approach is particularly appealing for automatic calibration of distributed hydrologic models. Toward that ultimate goal, we explore in this work calibration of a lumped hydrologic model via limited optimization of a priori estimates of the model parameters. The technique developed for the purpose is a simple yet effective and efficient pattern search algorithm called the Stepwise Line Search (SLS). To evaluate the methodology, calibration and validation experiments were performed for 20 basins in the US National Weather Service West Gulf River Forecast Center’s (NWS/WGRFC) service area in Texas. We show that SLS locates the posterior parameter estimates very efficiently in the vicinity of the a priori estimates that are comparable, in terms of reducing the objective function value, to those from global minimization. A cross validation experiment indicates that, when parametric uncertainty due to lack of calibration data is considered, limited optimization of a priori parameters using SLS may be preferred to global optimization.