Artificial intelligence (AI) weather models have made significant progress in the past 18 months, with companies like Google, Microsoft, NVIDIA, and Huawei claiming that their AI models perform as well as the widely regarded “European model” in weather forecasting. Start-ups such as Atmo, Excarta, and Jua are also developing AI weather models. These AI models are faster and cheaper to operate than conventional government-run weather models. While AI models still require improvement to provide all the capabilities needed for operational forecasting, their emergence signals a potential shift in the weather forecasting industry.
AI weather models have recently predicted the track of Hurricane Lee accurately, estimating it would make landfall in Nova Scotia a week later. The models fluctuated slightly in the following days but consistently forecasted landfall between Cape Cod, Massachusetts, and eastern Nova Scotia. This success suggests that AI weather prediction has emerged as a legitimate competitor to conventional models.
The European Centre for Medium-Range Weather Forecasts, which operates the European model, began publishing forecasts from AI models developed by Google, NVIDIA, and Huawei on its website. These models use current conditions from the European model as a starting point to produce a 10-day forecast in approximately one minute. The AI models were able to accurately predict the potential path of Hurricane Lee, even hinting that it could veer close to New England.
While some experts caution that AI models need further evaluation and testing before being fully adopted, their potential benefits are clear. AI models can process data much faster and at a lower cost compared to conventional models. They have the potential to provide more accurate and detailed predictions, particularly for extreme weather events. AI-based ensemble modeling, where the same model is run multiple times with slightly tweaked initial conditions, could lead to more useful forecasts and risk assessments.
The National Oceanic and Atmospheric Administration (NOAA) and the European Centre are rival agencies in weather modeling, with the European model generally considered more accurate. NOAA recently established its Center for Artificial Intelligence and held its fifth annual AI workshop. The agency plans to release a website similar to the European Centre’s, displaying forecasts from AI models that start with current conditions from the American model. NOAA’s interest in AI models reflects their potential to improve weather forecasting accuracy.
While conventional models will likely still be needed to train AI models, AI technology could significantly transform the weather forecasting industry. Faster and more efficient AI models could generate forecasts more frequently and at higher resolutions, without straining computing resources. AI models could eventually generate the forecast, with conventional models only used for training purposes.
Despite their recent progress, AI weather models still have limitations, such as the inability to produce forecasts for certain parameters like precipitation and clouds. The trust and understanding of forecasters who have predominantly used conventional models will also need to be earned. However, with the fast pace of innovation, meteorologists are optimistic about the future of AI in operational forecasting.
In conclusion, AI weather models have made substantial advancements in recent years and are emerging as potential competitors to conventional models. AI models offer faster and cheaper forecasting capabilities, and their performance has impressed scientists and experts in the field. While further evaluation and improvements are needed, the future of AI in weather forecasting looks promising.