49. Weather-source selection¶
GAP draws on many weather-source modules, but for any given country it selects one named model and one statistic and serves that directly. It does not blend models into a served value. This is one of the most consequential decisions in the platform.
49.1 Why not blend¶
Blending several models produces a number that no single model actually predicted, which makes the resulting advisory impossible to attribute or validate. A single named model per country can be scored against observations, and when it is wrong the failure is legible. The guiding trade-off is that auditability beats marginal accuracy, because an advisory that cannot be defended cannot be deployed.
flowchart TD
C[Country] --> SEL{Forecast selector}
SEL --> M[One model<br/>+ one statistic]
M --> ADV[Served advisory]
B[Blend path] -.->|"labels only,<br/>no served value"| ADV
49.2 Governed selection¶
The per-country choice of model and statistic is a governed decision rather than a per-request accident: it is chosen deliberately and applied consistently for that country. Different model families serve different variables — for example, one family for precipitation and another for temperature, humidity, wind, and solar — and they are reached through the GAP backend rather than called directly from client surfaces.
Sources fall into two states that matter operationally:
- Serving — sources currently selected and used to produce advice.
- Retired — former sources kept for historical context but no longer served.
A source list therefore changes over time; treat the live configuration as authoritative rather than any fixed list in documentation.
49.3 Visible fallback on the sovereign path¶
Sovereignty also appears in the forecast layer: an agency can run its own model, which becomes the primary source for that instance once it has been validated against observations. Until then, selection falls back to an already-validated source, and the fallback is recorded rather than hidden — consistent with the platform's honest-degradation commitment described in Observability and degradation.
For the data products underneath this selection, see the developer architecture and the SPW documentation.