Media release
From: Springer NatureClimate: GenCast outperforms existing weather forecasting
A machine learning model that can produce a reliable probabilistic weather forecast, expressed as probabilities of various outcomes, based on current and future weather is reported in a paper published in Nature. The model, named ‘GenCast’, outperforms the best-performing traditional medium-range weather forecast and is also able to better predict extreme weather, tropical cyclone tracks and wind power production.
Accurate weather forecasts are essential for individuals, governments and organizations to make key decisions on a daily basis ranging from whether to carry an umbrella, assessing wind power production or extreme weather planning. Weather forecasting is traditionally based on numerical weather prediction methods, which estimate current weather and maps this to a forecast of future weather over time (known as deterministic forecasts). This generates numerous potential scenarios, which are combined to produce a weather forecast.
Ilan Price and colleagues present a machine learning weather prediction method named GenCast. The method is able to generate a probabilistic forecast, which predicts the likelihood of future weather based on the current and previous weather states. The authors trained GenCast on 40 years (1979 to 2018) of best-estimate analysis data of weather occurrences and it is able to generate 15-day global forecasts, at 12-hour steps, for over 80 surface and atmospheric variables, in 8 minutes. When compared to the European Centre for Medium-Range Weather Forecasts’ ensemble forecast (ENS) — a deterministic forecast and currently the top-performing medium-range forecast globally — they found that GenCast outperformed the ENS in 97.2% of the 1,320 targets used to summarize the performance. They also observed that GenCast is more effective at predicting extreme weather, tropical cyclone tracks and wind power production.
The authors suggest that GenCast may provide more efficient and effective weather forecasts to support effective planning.
