Accurate default risk prediction for agricultural loans is a prerequisite for balancing financial inclusion and safety in rural finance, yet traditional assessment methods have a limited capacity to capture exogenous variables such as climate risk. To this end, this paper constructs a credit risk prediction framework integrating multi-source remote sensing data with explainable machine learning to optimize the credit profile of agricultural loans. Controlling for conventional agricultural loan variables, a logistic regression model examines the statistical association between multi-source remote sensing features and farmers’ default risk. A comparative analysis of multiple machine learning models further demonstrates that incorporating remote sensing data helps improve prediction accuracy, with temperature and precipitation volatility emerging as the most important remote sensing predictors, capturing the predominant climate-related variations in default prediction. Analysis using the Explainable Boosting Machine (EBM) quantifies the contribution of these variables to default risk prediction and identifies notable interaction patterns between remote sensing indicators and traditional agricultural loan variables.
Wang et al. (Wed,) studied this question.