Application of Multimodal Optimization for Uncertainty Estimation of Computationally Expensive Hydrologic Models

Abstract

The generalized likelihood uncertainty estimation (GLUE) framework has been widely used in hydrologic studies. However, the extensive random sampling causes a high computational burden that prohibits the efficient application of GLUE to costly distributed hydrologic models such as the soil and water assessment tool (SWAT). In this study, a multimodal optimization algorithm called isolated-speciation-based particle swarm optimization (ISPSO) is employed to take samples from the search space. A comparison between the ISPSO-GLUE, proposed here, and traditional GLUE approaches shows that the two approaches generate similar uncertainty bounds, but that the convergence rate to stable uncertainty bounds is much faster for ISPSO-GLUE than for GLUE. That is, ISPSO-GLUE needs a much smaller number of samples than GLUE to arrive at a very similar answer. Although ISPSO-GLUE slightly underestimated the prediction uncertainty and missed a number of observed values, the proposed approach is considered to be a good alternative to the typical GLUE approach that employs random sampling.