Skip to content
Eric Hare

2026 — Present · Creator

Aeroza

Programmable weather intelligence: real-time radar, geospatial queries, streaming APIs, signed webhooks, and probabilistic nowcasting.

PythonFastAPITypeScriptNext.jsPostGISRedisNATS

Aeroza is an open-source platform for applications that need to understand and react to weather in real time. It turns live weather data into a queryable API: radar, alerts, surface observations, nowcasts, verification metrics, streams, and signed webhooks.

The system currently ingests NWS alerts, MRMS reflectivity, and METAR observations. It exposes point, polygon, and tile reads over that data; runs persistence, pySTEPS Lucas-Kanade, and lagged-ensemble nowcasting; and scores forecasts continuously with continuous, categorical, and probabilistic metrics.

Architecture

The backend is a FastAPI modular monolith with extracted workers for ingest, materialization, nowcasting, verification, streams, and webhooks. Storage is PostgreSQL with PostGIS, Redis, and Zarr-style raster storage; NATS JetStream handles streaming and worker coordination.

The web surface is Next.js and includes a landing page, live map, Storm Replay demo, calibration view, developer console, and docs. The developer console is driven through the @aeroza/sdk TypeScript client, which keeps the public API and UI in close contact.

Engineering focus

The project is built around accountable forecasts rather than just attractive weather imagery. Short-horizon predictions are compared against observed radar as reality arrives, with MAE, bias, RMSE, POD, FAR, CSI, Brier, and CRPS available through the verification surface.

Recent public work has focused on making the live map production-grade: WebP tile negotiation (#84), cache prewarming and adaptive radar loops (#82), and pre-staged MapLibre radar frames for smooth playback (#85).