Designing probabilistic AI monsoon forecasts to inform agricultural decision-making
创建于 更新于
摘要
We develop a decision-theory framework and an “evolving-expectations” Bayesian baseline to design and evaluate probabilistic monsoon onset forecasts that farmers can use despite heterogeneous constraints and priors [page::0][page::1]. By blending AI weather-prediction outputs (NGCM, AIFS) with the evolving-expectations statistical model via a multinomial logistic calibration, the blended system improves Brier Score by ~5–10%, RPSS by ~20–25%, and AUC by 3–5 percentage points relative to static climatology, and proved operationally useful in a 2025 dissemination to millions of farmers [page::3][page::11].
