AIFS ENS Debuts at Number 1 Skill Scores U.S. Regions Days 6-10

Climate Impact Company AI Research

Issued: Sunday, July 13, 2025

Highlight: AIFS ENS debuts at number 1 with skill scores for U.S. regions in the 6-10-day period.

Fig. 1: Regions of the U.S. 2-meter temperature forecasts for the 6-10-day period comparing ECM ENS, routinely the number 1 dynamic forecast model versus AI. The brand new AIFS ENS (began July 1^st) was evaluated.

Discussion: On July 1^st, the AIFS ENS is introduced to operational weather forecasts. The 6-10-day 2-meter temperature forecasts for regions of the U.S. indicate a skill consensus score that ranks the debut of AIFS ENS at number 1 with AIFS a close second and the routinely top dynamic model, ECM ENS, at number 3 (Fig. 1). GC ECM ENS has routinely been the top AI forecast slips to number 4 during early July. Regional scores (by ECM ENS and AI models) are best in the Southwest U.S. (where daily temperature fluctuations are smallest) and least skillful in the Pacific Northwest where cool marine air masses off the Pacific and hot continental tropical air extending north from U.S. Deserts are constantly at battle.

During the past 60 days, prior to the release of the AIFS ENS, the ECM ENS routinely produced the best skill scores in the 6-10-day period (Fig. 2). Surprisingly, AIFS edges GC ENS at number 2, which is rare. Other AI models are routinely lagging. Regionally, skill scores are best in the Southwest U.S. and least successful in the Central Plains, possibly due to the unusually cloudy/wet weather.

In the 11-15-day period, during the past 60 days, 2-meter temperature forecasts by the ECM ENS ranked number 1 followed by GC ECM ENS and then AIFS (Fig. 3). Prior to the launch of AIFS ENS on July 1^st, AIFS produced higher skill scores and are competitive with the top dynamic models (ECM, GFS ECM, and CMC ENS).

Despite improving AI skill scores, maintained is the forecast guidance which states AI models should not be used solely when issuing an operational medium range forecast rather in conjunction and/or context with dynamic models. This guidance is especially important to commodities markets which are concerned with BOTH the forecast and the confidence of the forecast. Dynamic models can be explained; AI models cannot.

Fig. 2: Regions of the U.S. 2-meter temperature forecasts for the 6-10-day period comparing ECM ENS, routinely the number 1 dynamic forecast model versus AI.

Fig. 3: Regions of the U.S. 2-meter temperature forecasts for the 11-15-day period comparing ECM ENS, routinely the number 1 dynamic forecast model versus AI.