During Calendar Summer 2026. Operational Models Edge AI in Skill

Climate Impact Company Research Report

Issued: Tuesday, September 22, 2026

Highlight: Operational vs. AI models, 2-meter temperature forecasts for North America, during calendar summer 2026. Operational models edge AI in skill.

Discussion: Comparing 5 operational models with 7 AI models forecasting 2-meter temperature across North America utilizing CWG/SVWM anomaly correlation skill score (0 to 1 scale) during calendar-summer 2026 indicates operational models out-performed AI models (Fig. 1) by a small margin (Fig. 2). In the 6-10-day period, AIFS ENS and ECM ENS tied for best skill score (0.81) while 4CastNetV2 GFS ENS was lowest (0.65). In the 11-15-day period, ECM ENS achieved top skill (0.70) by a small margin over AIFS ENS (0.69) and GFS ENS (0.68). The lowest skill score in the 11-15-day period was AI GFS (0.48). The high visibility GFS OP scored next to last in both the 6-10-day and 11-15-day period. Interestingly, one top performer was CMC ENS finishing 3^rd in skill score days 6-10 and 4^th days 11-15. The average anomaly correlation skill score of 5 operational models forecasting 2-meter temperature in the 6-10-day period was 0.75 compared to 0.73 by 7 AI models. In the 11-15-day period, operational models edged AI models 0.62 to 0.59.

During the past 30 days, week 3-6 North America 2-meter temperature forecasts (also) utilizing anomaly correlation skill scores reveals AI models finished “middle of the road” compared to the top 2 models (GFS ENS and ECM). CFS V2 finished last (Fig. 3).

Summary: As stated during the infancy of introducing AI models to weather forecasting, in this case medium range temperature anomaly forecasts, use in conjunction with operational models is the preferred methodology to convey a weather forecast. The initial recommended use of AI models remains consistent as skill scores by AI models are competitive with operational models but not clearly more skillful.

Fig. 1: Comparing 5 operational and 7 AI models forecasting 2-meter temperature across North America anomaly correlation skill scores for calendar-summer 2026.

Fig. 2: Summary of 5 operational and 7 AI models forecasting 2-meter temperature across North America anomaly correlation skill scores for calendar-summer 2026.

Fig. 3: Comparison between 3 operational and 3 AI “Weeklies” forecast models of North America 2-meter temperature for the past 30 days for weeks 3-6 ahead.