Operational/AI Model Verification Report, North America, Past 30 Days

Climate Impact Company Dynamic/AI Models Verification Report

Issued: Sunday, January 4, 2026

Highlight: Skill scores of North America 2-meter temperature forecasts for the medium/extended-range during the past 30 days.

Fig. 1: The North America medium range skill scores forecasting 2-meter temperature, all models, for the past 30 days. Source: CWG/SVWM.

Discussion: During the past 30 days, 2-meter temperature forecasts across North America in the 6-10-day and 11-15-day period are best performed by AIFS ENS (Fig. 1). Second best is ECM ENS. Each model has routinely finished No. 1 and No. 2 for the second half of 2025. The changeable GFS ranked next to last for the medium range while (surprisingly) AI 4Cast Net V2 GFS ENS was consistently last. Skill scores for days 1-5 are generally between 0.95 and 1.0. Therefore, reasonable is the assertion that 6-10-day outlooks are about 75% as good as short-range forecasts while 11-15-day projections are about 50% as good as short-range forecasts for North America entering the New Year. Of course, not factored in is the forecast difficulty of the weather pattern which was quite high in North America given the extreme East cold to start the month and the very warm finish. The convenient zero to one scale anomaly correlation skill score is used in this assessment.

In the extended range, interesting is the drop-off in skill from 0.25 to 0.45 in the 16-20-day period to 0.15 to 0.25 in the 25-30-day period (Fig. 2). In fact, the 25-30-day forecast skill of the popular ECM “weeklies” was matched by the 10-year climatology. Not commonly used for extended range forecasts, GFS ENS ranked first for the extended range outlooks for the past 30 days.

Fig. 2: The North America extended range skill scores forecasting 2-meter temperature, all models, for the past 30 days. Source: CWG/SVWM.