Rain yes/no scores (3-hour totals at or above each threshold)
POD: share of observed rain events forecast. FAR: share of forecast events that didn't happen. CSI: hits ÷ (hits + misses + false alarms). Frequency bias above 1: rain forecast more often than it happens.
Forecast and observed
Error by hour of day (Maldives time)
Error by lead time
Forecast against observed
Daily bias
By station (click a station to look at it alone)
Suggest a correction
Coefficients are fitted on the earlier part of the selected period and judged on the held-out rest, so the improvement shown is what to expect on new forecasts. Errors are forecast − observed.
Bias by hour of day, before and after (test period)
RMSE by lead time, before and after (test period)
Coefficients corrected = a × forecast + b
Groups with fewer pairs than the minimum use the all-pairs value (marked “fallback”). You can edit a and b, then re-check.
Save as a proposal
Correction proposals
Active corrections (JSON) ↗Approving a proposal retires the parameter's previously approved one. Post-processing can read the approved set at /api/corrections/active.