Comparing AI Vs. Dynamic Models 2-Meter Temperature Forecasts North America and Europe

Climate Impact Company Dynamic/AI Models Verification Report

Issued: Thursday, May 7, 2026

Highlight: AI out-performs dynamic models for 2-meter temperature forecasts during the medium range for North America; Dynamic models win in Europe.

Fig. 1: Anomaly correlation skill scores provided by CWG/SVWM for 6-10-day/11-15-day medium range forecasts from the past 30 days for North America and Europe.

Discussion: A comparison of anomaly correlation skill scores (convenient zero-worst to 1.0 best scale) between North America and Europe during the 6-10-day/11-15-day (medium range) timescales from the past 30 days reveals the ongoing success of AIFS ENS and AI Graph Cast ECM ENS (Fig. 1). CMC ENS surprises in the 6-10-day period with the best 2-meter temperature forecast skill for Europe, the highest skill score in this review. ECM ENS, typically the best dynamic model, outperforms AIFS ENS and AI Graph Cast ECM ENS in the 11-15-day period for North America. Conventional dynamic models GFS and ECM are generally least skillful in this review. AI models (7) outperform dynamic models (5) in both the 6-10-day and 11-15-day timescales for North America by small margins (Fig. 2). In Europe, AI models edge-out dynamic models in both time periods. Climate Impact Company regards 0.75-0.80 as a good skill score for 6-10-day forecasts and 0.55-0.60 as desirable in the 11-15-day period. Anomaly correlation skill scores provided by CWG/SVWM indicate results slightly below desired levels.

Fig. 2: Anomaly correlation skill scores provided by CWG/SVWM for 6-10-day/11-15-day medium range forecasts from the past 30 days for North America and Europe summarizing the dynamic versus AI models.