Arawave is an AI precipitation forecasting system that combines multiple state-of-the-art AI weather models into a calibrated probabilistic forecast, optimized specifically for Paraguay.
It's not a global product with Paraguay cropped out. It was built for Paraguay, validated on Paraguayan data, and every claim is published with its confidence interval.
Every number verified, red-teamed, and reproducible.
The pattern: The model wins where Paraguay's observation network is thinnest — the Chaco and transitional regions. It loses where convection dominates — the southeast soybean belt.
The answer to the heavy-rain weakness: a calibrated probability product (AUC 0.884). For extreme-weather decisions, calibrated probabilities beat deterministic forecasts.
The Chaco is where Paraguay has the fewest weather observations, the highest climate vulnerability, and the greatest adaptation need.
And it's exactly where the model works best.
The AI model compensates for sparse observations. The next step: deploy weather stations to make it even better. That's the case for fundable adaptation infrastructure.
Every claim carries a confidence interval. Losses are published alongside wins. The validation data is reproducible.
The −14% on heavy rain is not hidden — it's the strategy.
Climate adaptation projects routinely cite inflated skill claims. Publishing where you lose is the credibility signal that cuts through the noise. If a reviewer questions "+32%," they get a reproducible JSON and bootstrap method — not a sales pitch.