Multisensor Precipitation Fusion Based on Random Forest for Enhancing Flood Prediction Lead Time

Accurate precipitation data plays a central role in ensuring the reliability of flood prediction, especially in catchments where observation networks are sparse. When relying on a single measurement source such as rain gauges, weather radar or satellite products surface runoff predictions often become less reliable. This limitation largely arises from topographic effects such as radar beam blockages as well as the inherent constraints of each sensor in capturing extreme rainfall events.

To address these challenges, this study proposes a multisensor precipitation data fusion framework based on the Random Forest machine learning algorithm with the aim of improving flood prediction lead times in the Busan city, South Korea. The approach combines weather radar observations, GPM IMERG satellite estimates, and ground-based gauge data along with topographic information derived from a Digital Elevation Model.

The resulting fused precipitation product was evaluated using a range of statistical metrics and further tested within a hydrological modeling framework. The results show that integrating remote sensing data with ground observations can substantially improve predictive performance. In particular, the Random Forest model demonstrates strong capability in capturing complex nonlinear relationships and in reducing orographic biases, leading to more accurate spatial rainfall estimates.

When applied to hydrological simulations, the multisensor framework enables earlier detection of upstream rainfall signals, which in turn supports longer and more reliable evacuation lead times. This improvement is also reflected in more stable simulations of hydrograph peak timing. Overall, the findings highlight the practical value of multisensor data integration for supporting more informed and timely decision-making in flood risk management, particularly in regions with complex terrain.

 

Acknowledgments: The research for this paper was carried out under the KICT Research Program (Project no. 20260161–001, Development of Digital Urban Flood Control Technology for the Realization of Flood Safety City) funded by the Ministry of Science and ICT.