Integrating artificial intelligence and climate data for predicting rice brown spot disease dynamics

Abstract

Rice Brown Spot Disease (RBSD), caused by episodic epidemics of the airborne pathogen Bipolaris oryzae, contributed to the Bengal Famine of 1942 and continues to threaten food security, highlighting the urgent need for effective forewarning systems. This study developed and validated predictive models for RBSD using AI/ML approaches based on comprehensive field data collected over four cropping seasons (2019–2023, excluding 2020) in Southern India. Two complementary modeling approaches were developed and compared: pure machine learning models, including XGBoost, Random Forest, ANN, and others, for maximum predictive accuracy; and a hybrid hierarchical clustering-regression framework for interpretability. Both approaches couple meteorological, phenological, and pathogen data to forecast aerospora concentration (A) and disease severity (PDI). XGBoost emerged as the best-performing model, achieving an R² of 0.66 for aerospora prediction and 0.70 for PDI, with RMSE values below 8 under cross-validation. These models were applied to gridded CMIP6 climate data under SSP2–4.5 and SSP5–8.5 scenarios to map spatio-temporal RBSD risk dynamics across Peninsular and Central India. Results identified clear vulnerability patterns, with high-humidity regions including Telangana, Chhattisgarh, and coastal Andhra Pradesh emerging as critical hotspots. Under the high-emission SSP5–8.5 scenario, severe airspora risk is projected to expand from 7% of the study area in 2050 to over 51% by 2100, with PDI values escalating markedly in the latter half of the century. Building on these findings, we developed rbsdPredict, an open-source R package for modeling, predicting, and visualizing RBSD spatio-temporal risk, with future scope for integration into operational early-warning and decision-support systems.