Forecasts stop being just a grid: WindBorne turns heatwaves, coldwaves and fronts into trackable events

The output of a weather forecast shifts from a global grid to a set of numbered weather events that can be followed through time. A gridded model only gives temperature, wind, humidity and other variables per cell, and never says where a heatwave is; judging whether heat is unusual, where it is heading and how severe it will get takes a meteorologist reading map after map. Events are detected on the WeatherMesh-6 ensemble mean across the full 15-day forecast and refresh hourly with each new model run, and each event keeps one ID across forecast hours and model runs.

All four event types pass through one four-stage pipeline: per-cell detection, grouping qualifying cells into polygons or lines, linking objects across forecast hours within a run, and matching tracks across runs. Detection is set per event type: atmospheric rivers use integrated water vapor transport following Guan and Waliser's tARget v4 method, heatwaves and coldwaves use 2 m daily maximum or minimum temperature against local percentiles, and fronts use the FrontFinder neural network to assign cold, warm, stationary and occluded probabilities. A heatwave is triggered when daily highs exceed the local 85th percentile for that calendar day for at least three consecutive days, with percentiles from ERA5 1991 to 2020; heat and cold also have three stricter tiers at the 90th, 95th and 98th and the 10th, 5th and 2nd percentiles.

Events can be subscribed by region and type, with notification when a run's event products are ready, and then followed by ID as the forecast is revised. This tracking was previously an internal tool used by WindBorne's own meteorologists and is now available through a REST API in JSON and GeoJSON formats. Endpoints return all events in a run, an index of events across runs, the history of a single event, and the events affecting up to ten coordinates; the webhook filters by event type, time window and region.

The demonstrated scope is detection and tracking of four event types (atmospheric rivers, heatwaves, coldwaves and fronts) on the WeatherMesh-6 ensemble mean across the 15-day forecast, plus access through a REST API and webhook; the material says more event types and features are planned for future versions.

The material labels Weather Events an experimental v1 and calls fronts the most experimental of the four types; each event is the output of a versioned detection and tracking method applied to the WeatherMesh-6 ensemble mean.