Forecast error calculator

Paste one pair per line: forecast, then observed. An optional third column is a reference forecast, such as persistence, for a skill score. Use one unit throughout.

Pairs (forecast, observed, optional reference) Normalise nRMSE by Skip rows where forecast and observed are both zero (night rows would otherwise shrink every error)

Paste at least two pairs.

The examples are one Colorado NSRDB site, 20 June 2017, forecasts issued 1 hour ahead for 09:00 to 14:00 local time, in W/m². MAE is the mean absolute error, RMSE the root mean square error, MBE the mean bias error. Skill = 1 minus RMSE divided by the reference RMSE. A design aid: use at least a few hundred daylight pairs for a real result.