A Critical Synthesis of Remote Sensing and Machine Learning Approaches for Climate Hazard Impact on Crop Yield

This review critically assesses the application of machine learning (ML) algorithms and remote sensing (RS) products in detecting and predicting climate hazards, as well as their impacts on crop yields. Using the PRISMA approach, it examines 177 studies on climate hazards and 197 on RS–ML applications in crop yield modeling. Research is most concentrated in Asia, followed by Africa and the Americas, with agricultural drought emerging as the most frequently studied hazard. Statistical approaches, such as the coefficient of variation, remain the dominant methods for analyzing climate variability. For drought detection, Random Forest (RF) was the most used machine learning (ML) algorithm (17%), followed by Support Vector Machines (SVM, 11%), Artificial Neural Networks (ANN, 8%), Adaptive Neuro-Fuzzy Inference System (ANFIS, 5%), and XGBoost (5%). For drought impacts on crop productivity, RF dominated (39%) followed by LASSO (11%), while for climate variability impacts, RF (21%) led alongside SVM (10%), ANN (9%), Long Short-Term Memory (LSTM) (8%), Multiple Linear Regression (MLR) (8%), and Convolutional Neural Network (CNN) (7%). Asia leads in the integration of advanced ML/DL techniques, namely RF, SVM, XGBoost, LSTM, and high-resolution RS datasets across multiple spatial scales. In contrast, due to infrastructure and data limitations, Africa predominantly employs simpler and more interpretable models. RS products, namely MODIS, TRMM, CHIRPS, and ERA5, are widely used due to their accessibility. However, their limited spatial resolution restricts their effectiveness in assessing localized climate impacts, especially within smallholder farming systems. The review recommends hybrid modeling frameworks that integrate process-based and data-driven methods, broader spatial and crop coverage, standardized protocols, and real-time, microclimate-aware monitoring systems to improve model reliability and applicability in underrepresented, data-scarce regions, such as Sub-Saharan Africa. This approach is expected to strengthen climate-resilient agriculture and global food security.