I’m refining catchment-specific rainfall IDF thresholds to drive actions like road closures and pump staging, and need solid methods and datasets to cut false alarms without missing high-consequence events. If you’ve blended MRMS radar with USGS gages to validate 3–6 h exceedance triggers, what performance metrics did you rely on (e.g., CSI vs FAR) and how did you handle spatial bias in dense urban basins?
, this drives me nuts too! I’ve found that blending MRMS with USGS gages has really helped, but I had to tweak the IDs based on specific catchment responses — using a combined threshold helped cut down on false alarms. Also, keep an eye out for spatial variability; sometimes it masks the actual rainfall impact.
I’ve had luck using ensemble forecasting alongside the radar data; it feels like having multiple GPS navigating through the stormy weather. It can add some complexity, but I’ve noticed it significantly reduces both false alarms and misses. @sking, have you experimented with any ensemble techniques?