Increasing the Accuracy of Two-Hour-Ahead Lightning Threat Predictions Using a 1-km-Resolution Numerical Weather Prediction Model: A Decision-Support Approach with a Spatial Maximum Filter

To bridge the 1-2 h forecast gap in short-range lightning threat prediction, we developed a method that applies a lead-time-dependent spatial maximum filter (SMF) to instantaneous forecast fields to mitigate forecast skill degradation caused by spatial displacement errors. This filter was applied to a numerical weather prediction (NWP) model (Cloud-Resolving Storm Simulator; CReSS), run at 1-km resolution with assimilated convective-scale observations, and benchmarked against the Japan Meteorological Agency (JMA) Lightning Nowcast. We evaluated four intense thunderstorm events from May to June 2022, using total lightning (intracloud and cloud-to-ground) data from the JMA Lightning Detection Network. Applying SMF (CReSS_MaxF) not only increased false alarms but also substantially increased the detection rate. Assuming a 15-km influence range for the observed lightning, its Critical Success Index (CSI) at a 1-h lead time improved markedly from 0.05 (unfiltered) to 0.18. The CSI of CReSS_MaxF is comparable to that of the JMA Lightning Nowcast at a 1-h lead time and, crucially, decreases by only 0.03, from 1 to 2 h. This spatially filtered NWP-based approach offers a decision-support tool that can complement the operational nowcast, which covers only the first hour, bridging the 1-2 h forecast gap in short-range lightning predictions.