How we measure quality
To assess the quality of our weather models, we use data from >50,000 weather stations to compare our model's preditions to the actual weather conditions at those stations. We update our evaluations daily, and report our quality over the last 15 days.
We measure forecast quality at all forecast depths from hour 1 to day 14 to test how accurate our models are when forecasting weather at different horizons. We similarly evaluate the GFS forecast for NOAA, to compare how well we do against a premier weather model. Forecast quality is reported using two metrics:
- RMSE: RMSE (root mean squared error) indicates the error between forecasted and actual weather, averaged over all stations.
A lower RMSE indicates a better forecast - Bias: Bias indicates whether a forecast is systematically under or overpredicting a parameter. E.g., a bias of +0.5C indicates the forecast is typically 0.5C warmer than the actual weather.
A bias of 0 is ideal.