Verify are the forecasts any good?
Pick a past window and see what every source predicted, laid over what actually happened. Then the scoreboard: MAE (how far off, on average), bias (does it lean warm or cold), and RMSE (does it blow the occasional call) — so you know which lines to trust.
What am I looking at?
The thick white line is the actual observation from the truth station. Every colored line is a different forecast source's prediction for the same period. The closer a colored line tracks the white line, the better that source forecasted.
MAE (Mean Absolute Error) — average size of error in the variable's units. Smaller is better. For temperature, MAE under 2°F is genuinely good.
Bias — average signed error (forecast minus actual). Positive = source forecast too warm/high. Negative = too cold/low. The bias-correction model will learn to subtract this.
RMSE (Root Mean Square Error) — like MAE but punishes big misses more.
Sources prefixed OM_HIST_ are Open-Meteo's archived per-model forecasts.
NWS_GRID is the human-edited official forecast.
PIRATE wraps NOAA's National Blend of Models (NBM).