Flood Estimation and Prediction Using Particle Filters

Abstract

Data assimilation methods have received increased attention to accomplish uncertainty assessment and enhancement of forecasting capability in various areas. Despite their potential, applicable software frameworks for probabilistic approaches and data assimilation are still limited because most hydrologic modeling software are based on a deterministic approach. In this study, we developed a hydrologic modeling framework for data assimilation, namely MPI-OHyMoS. While adapting object-oriented features of the original OHyMoS, MPI-OHyMoS allows user to build a probabilistic hydrologic model with data assimilation. In this software framework, sequential data assimilation based on particle filtering is available for any hydrologic models considering various sources of uncertainty originating from input forcing, parameters, and observations. Ensemble simulations are parallelized by a message passing interface (MPI), which can take advantage of high-performance computing (HPC) systems. Structure and implementation processes of data assimilation via MPI-OHyMoS are illustrated using a simple lumped model. We apply this software framework for uncertainty assessment of a distributed hydrologic model in synthetic and real experiment cases. In the synthetic experiment, dual state-parameter updating results in a reasonable estimation of parameters to cover synthetic true within their posterior distributions. In the real experiment, dual updating with identifiable parameters results in a reasonable agreement to the observed hydrograph with reduced uncertainty of parameters.