Loading weather…
Hour by hour
Daily forecast
Why did the forecast change?
Radar timeline
Model comparison
Our Forecast
Combining available models with neutral weights until verified local performance is strong enough to adjust them.
What the models disagree about
Who is actually getting it right?
V2.1 stores forecast snapshots in a persistent SQLite database, verifies them against nearby official NWS station observations, and automatically reconstructs missing model runs and observations from the previous 72 hours whenever the PC comes back online.
How scoring works
Mean absolute error plus signed bias. V2 can correct a model that repeatedly runs warm or cold.
0–3h, 4–12h, 13–24h, and 25–72h are scored separately so short-range skill does not automatically control longer forecasts.
Wind MAE, rain hit rate, and Brier probability score are used when the official station observation contains enough data.
20 verified samples are required before learned weighting begins. More samples progressively increase confidence.
Forecast preferences
What powers this build
Never use this app as a replacement for official emergency instructions. Severe-weather decisions should follow NWS/NHC/local authorities.