Expanding Performant Deep Learning NWP Models from Medium-Range to Subseasonal Forecasts

WindBorne Systems Inc. has developed and deployed WindBorne WeatherMesh, the most accurate AI-based weather model in the world based on current benchmarks. WeatherMesh leverages live GSB data and already delivers real-time, medium-range global weather forecasts, and we use WeatherMesh wind forecasts in our daily operations to autonomously guide our worldwide constellation of GSBs. WeatherMesh currently executes weather forecasting calculations more than 1,000x faster than physics-based meteorological models, and can compute a ten-day forecast in under 60 seconds on a single GPU. We have also proven its ability to accurately predict the behavior of extreme weather phenomena, such as hurricanes. WeatherMesh is capable of medium-range forecasts, and can execute these forecast more than 1,000x faster than traditional weather models and on high-powered laptops running on a single GPU. In this Phase I, we will discover the needs of the DON with regard to both medium-range and subseasonal forecasts, and determine the variable and dataset augmentations required to expand WeatherMesh capabilities to include subseasonal forecasts. WindBorne's team is uniquely suited to this work, as we are comprised of best-in-class expertise in both atmospheric scientific research and AI development and optimization. We also represent an extraordinary low-risk Phase I award, because we have already proven, within publicly published benchmarks, that WeatherMesh outperforms dominant AI weather forecasting models, including Google's GraphCast.  In our Phase I Option, we will continue building upon our Base period research by integrating high priority variables into WeatherMesh, thereby lengthening forecast times. As we expand the forecasting capabilities of WeatherMesh, we will continually benchmark against actual weather data and existing forecasts, and will also continually share our research results with the DON. It is this transparency that will enable us to ensure alignment with DON objectives and will enable the DON to ensure performant medium-range and subseasonal forecasts that support integrated deterrence.