WindBorneの「WeatherMesh 6」、EUのゴールドスタンダードAI予報を上回る
WindBorne SystemsがAIモデル「WeatherMesh 6」をリリース。翌日予測並みの精度で5日間予報を提供し、ECMWFの最先端システムを上回ると報じられている。政府系物理モデルの支配に挑み、AI気象学における独自データ優位への転換を示唆する。
windborne systems released weathermesh 6, a transformer-based weather model that reportedly matches the accuracy of the european centre for medium-range weather forecasting's next-day forecasts five days in advance, specifically on surface temperature. the model runs at 3km resolution with hourly updates, outpacing traditional physics models that require supercomputers and run every six hours. this performance stems from direct data assimilation using a proprietary network of 400 balloons, reducing reliance on government datasets from ecmwf and noaa. this shift implies that private companies with unique data streams can disrupt the public sector monopoly on high-fidelity weather forecasting, potentially offering more granular predictions for energy, logistics, and aviation sectors. windborne plans to further reduce dependency on government initial conditions, positioning its data advantage as the core moat for its ai forecasting business.