Statistical Verification of a New Blending Forecast with a Spatial Maximum Filter for “Senjo-Kousuitai” in Japan

Abstract

This study proposed a new blending approach for forecasting “senjo-kousuitai,” combining extrapolation-based nowcasting (EXT) and numerical weather prediction (NWP), to support early decision-making by municipalities regarding evacuation. A major deficiency in the short-term (1–2 h) operational forecasts of the Japan Meteorological Agency is underestimation of the precipitation area associated with senjo-kousuitai formation, which is mainly attributed to the EXT component. To address this problem, our blending approach emphasizes NWP using a cloud-resolving model for the 2-h forecast. A notable aspect of our approach involves incorporating a spatial maximum filter (MF) to account for spatial displacements between the EXT and the NWP outputs, replacing forecasted rainfall with the maximum value in the surrounding area. Compared with conventional blending methods, statistical verification of the results obtained using our proposed approach revealed marked improvements in both underestimation bias and probability of detection during the senjo-kousuitai formation stage. These findings highlight the potential of the MF-based approach for reducing forecast misses and facilitating timely municipal decision-making. The simplicity of the method also underscores its value as an urgently required disaster mitigation strategy against the increasing occurrence of senjo-kousuitai. However, the rise in false alarms, as a trade-off for fewer misses, implies an increase in the cost associated with protective actions. Although the proposed method entails increased costs, adopting this approach can be a cost-effective strategy for preserving lives by mitigating misses.