Data Assimilation Approach for Flood Level Estimation using State Space Model for Urban Internal Flooding

This paper proposes a method for estimating flood levels by data assimilation using a state space model to determine the spatial-temporal flood expansion process. The method incorporates flood simulation values to analyze the causal relationship with observation data of river water levels. First, we simulate flood scenarios using a flood simulator with various precipitation patterns to construct time series datasets with high spatial resolutions (5-10 m). Then, after estimating the water level in the channels using an auxiliary particle filter, we analyze the inflow and the outflow of flood water for each grid element by improving the state space model to add spatial variables. We evaluated the performance of the proposed method using observation data in Aichi Prefecture, Japan. While the conventional method had an error of approximately 60 cm compared to the observation value, our estimation method using the proposed state space model showed a significant improvement, with an error of less than 9 cm.