18. Forecast availability¶
Model run time, provider publication time, GAP ingestion time and data availability are different events. A fixed clock-time schedule does not prove that a particular run is ready. Use available valid times and returned metadata when selecting a forecast.
18.1 Nominal horizons¶
| Product | Catalogue horizon / time support |
|---|---|
| NextGen daily, including Nigeria | 10-day forecast, daily values |
| NextGen hourly, including Nigeria | 4-day forecast, hourly values |
| Google WeatherNext 2 | 15-day forecast; six-hour rainfall intervals |
| Daily precipitation forecast | 14-day daily rainfall statistics |
| Kenya rainfall forecast | 42-day daily rainfall |
| NeuralGCM S2S | Historical catalogue labelled 2021–2025 and a separate realtime product; inspect available times |
These labels are not guarantees of complete date or spatial coverage. The public options endpoint lists fields, not completed ingestion runs, and does not report freshness.
18.2 Choosing a run¶
- Check the live catalogue and account access.
- Request a small point and date range within the forecast horizon.
- Where forecast archives are supported, set
forecast_dateexplicitly. The API accepts a date or hourly initialisation such asYYYY-MM-DDTHH; support depends on the reader and archive. - Inspect the returned dates and metadata and retain the original request with your analysis.
- If there is no data, distinguish an unavailable run, out-of-coverage location, delayed ingestion and missing measurements. Do not substitute zero.
When comparing forecasts, align both their valid times and lead times. Do not apply the old GenCast/GraphCast timestamp adjustments to WeatherNext 2.
For a confirmed publication schedule or archive-retention window, contact the GAP service administrator with the product and requested run. This guide does not publish unverified ingestion completion times.