A Survey of AI and IoT-Based Approaches for Climate-Smart Rice Disease and Yield Prediction

Climate variability poses an increasing risk for rice cultivation while operational climate products are still too coarse (tens to hundreds of kilometres) to be used for making decisions on field level regarding diseases and yields. Despite significant progress in AI-IoT disease detection, the disease detection systems are primarily image-based and are independent from the climatic drivers, and digital twin (DT) applications have been limited to the monitoring and synchronising of data in agriculture so far, but not closed-loop forecasting. This paper introduces an AI-IoT enabled spatial modelling framework of rice disease dynamics and yield prediction based on machine-learning climate downscaling, using a unified architecture that synchronises them. We review the latest literature on both agricultural digital twins and ML-based climate downscaling, as well as AI-IoT rice disease detection, to find the gap between them that motivates this work, and introduce the proposed five-layer framework, goals, and scope. The framework is designed to be a decision-support architecture applicable to climate-smart rice farming, especially in data and resource-limited rice production areas.