When is each model good?

Skill Lab

Understand when and where each forecast source performs best. Filters here drive the rest of the application.

Variable

Region

Weather regime

Lead time

Season

Southwest Monsoon

Date range

Rolling 90 days to 28 Sep 2026

Skill insight

BharatFS has been the strongest source for 24-hour heavy rainfall forecasts in Telangana over the last 90 days (skill 80), while GFS shows the largest bias (+1.93 mm).

Generated from the deterministic demo skill dataset — no language model involved.

Skill matrix

Composite skill score (0–100) per source and region for the selected regime, variable and lead time. Hover a cell for sample count and trend; click to change region.

BharatFS

NCUM

GFS

GEFS

AI Forecast Proxy

Skill by lead time

Skill decays with lead time at different rates.

Skill by weather regime

Regime is the strongest driver of relative skill.

Model disagreement

Spread across the five sources in this context.

Low disagreementHigh disagreement

Spread index 38moderate divergence

1BharatFSskill 80 · bias -0.7
2NCUMskill 76 · bias -1.45
3GEFSskill 66 · bias -0.75
4GFSskill 56 · bias +1.93
5AI Forecast Proxyskill 49 · bias +0.63

Skill history is generated from a fixed seed, so the same context always yields the same ranking.

Forecast spaghetti

Each source traced against the blended forecast and its uncertainty band.

6 valid times · Southwest Monsoon climatology assumed