An AI-Driven Framework for Climate-Aware Rice Disease Prediction Using Machine Learning-Based Climate Downscaling and Spatial Modeling
Contributors
Dr Santhosh Kumar S
Dr Abeer Ahmad Hamad Aljohani
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Outbreaks of rice disease depend on weather that varies over a few kilometres, yet the climate products usually available to agronomists are gridded at several kilometres or coarser, which blurs the microclimate signals that drive infection. This paper presents a two-stage framework that first downscales coarse temperature, humidity and rainfall fields onto a grid eight times finer using gradient-boosted regression, and then passes weekly climate summaries, spatial covariates and neighbouring-field disease status to a classifier that predicts field-level outbreak risk. Because no public dataset pairs fine-scale climate with field-level rice disease records, experiments use a controlled synthetic landscape with known ground truth, and all figures come from five independent realisations. Learned downscaling reduced RMSE by 11 to 20 percent against bilinear interpolation. Under spatially blocked, season-ahead validation the proposed model reached a ROC-AUC of 0.925, against 0.726 for a coarse-climate random forest, and matched an oracle given the true fine-scale climate. Downscaling added little over a spatially aware coarse-input model (mean AUC gain 0.014); spatial context, rather than climate resolution, carried most of the predictive skill.