SAIL

SAIL delivers sub-kilometer wind predictions by fusing multiple data streams through physics-informed machine learning. Our multi-source fusion engine combines:

• NWP Models — HRRR (3km), GFS (25km), and WeatherFlow Beta-WRF (1km) with automated bias correction
• Surface Observations — MesoWest stations, NDBC buoys, and WeatherFlow Tempest sonic anemometers
• Novel Sources — Geostrophic winds from PurpleAir pressure networks and ADS-B aircraft telemetry
• Physics Constraints — Coastal PINN modeling thermal boundary layers and land-sea roughness transitions

Weighted by inverse-variance, temporal freshness, and spatial proximity using Ensemble Kalman Filtering with Gaspari-Cohn localization.