AI Models Out-performing Operational Models During Past 30 Days

Climate Impact Company Operational AI/Research Report

Issued: Monday, September 15, 2025

Highlight: AI models out-performing operational models during the past 30 days forecasting the weather pattern in the medium range for North America and Europe.

Fig. 1-2: 500 MB anomalies correlation skill scores for North America and Europe from the past 30 days for 6-10/11-15-day forecasts.

Discussion: A review of 500 MB height anomaly (“the weather pattern”) forecasts across North America and Europe during the past 30 days is provided. Skill scores are calculated by CWG/SVMW.

In North America, the late summer 2025 weather pattern forecasts were most effective during the 6-10-day utilizing ECM ENS, traditionally the best operational model. In second place, AIFS ENS continued to score well since the July 1^st debut. ECM ENS and AIFS ENS have routinely scored no. 1 and 2 for medium-range forecasts during mid-to-late calendar summer 2025. The 11-15-day forecasts provided similar results with AIFS ENS edging ECM ENS. Note that the last place finisher was GFS. Also note that 6 AI models out-performed 5 operational models for both the 6-10-day and 11-15-day North America forecasts during the past 30 days.

In Europe, the 6-10-day 500 MB anomaly correlation skill score is best utilizing AIFS ENS followed closely by ECM ENS (Fig. 2). However, an impressive debut was notable by the NOAA AI Eagle model producing best skill scores in the 11-15-day period. Note that traditional operational models GFS and ECM finished last with 6-10-day and 11-15-day forecasts. Similar with North America, the AI forecasts outperformed the operational models for Europe medium range outlooks during the past 30 days.