Study on improving short-range flash flood forecasts using data assimilation in the Hiwasa river, Japan

The increasing frequency of flood damage worldwide due to climate change underscores the urgent need for effective early warning systems. In Japan, flash floods triggered by typhoons or frontal heavy rainfall occur annually in small and medium-sized rivers. This study investigates the improvement of short-range flood forecasts for the Hiwasa River, a typical small mountain river with a 71 km2 catchment area. We conducted forecasting experiments for 11 historical flood events and quantitatively evaluated prediction accuracy up to six hours ahead. A Rainfall–Runoff–Inundation (RRI) model was employed to simulate hydrological processes, and a particle filter (PF)-based data assimilation (DA) approach was introduced to reduce uncertainty in the initial conditions, e.g. soil moisture, river water depth. Results indicate that integrating PF with the RRI model (PF-RRI) improves water level forecasts at all forecast times, particularly within the first three hours. Beyond four hours, the effect of DA diminishes, and underprediction becomes longer. These findings demonstrate that the proposed PF-RRI method enhances short-term flood forecasting accuracy and can contribute to more reliable early warnings for flash floods, supporting timely and safe evacuation.