分布型流出モデルの誤差構造分析と多地点水位データ同化手法

抄録

This study evaluates the performance of Optimal Interpolation (OI) as a data assimilation with observed water levels at multiple-sites for a distributed rainfall-runoff model. The main objective of the assimilation is to improve the accuracy of water level estimations at ungauged river sections. The advantage of the OI is its simplicity without model linearization or ensemble simulations, but it requires to identify an error covariance matrix in advance for all prediction variables. This study analyzes the error structure of the RRI Model applied to the Chikusa River Basin based on past seven flood events. The results of cross-validation and twin-experiment show that by removing local simulation biases at gauged sites beforehand, the assimilation can improve the estimation of water levels even at ungauged sites. Furthermore, on-line assimilation with OI, which updates the model initial conditions regularly, show better performance in the estimation compared to the off-line assimilation.