GeoRRDI: An explainable and multi-source data-driven framework for rural revitalization assessment with spatial heterogeneity consideration

• Proposed GeoRRDI, a multi-source and spatially interpretable framework. • Developed a four-dimensional, 19-indicator RRDI system with SHAP-based selection. • Designed a triple evaluation framework and identified the optimal GWRF-RF model. • Revealed spatial drivers and evolution through global–local importance and prediction. Current assessments of rural revitalization at the municipal scale and village-level granularity face persistent challenges, including limited data integration, weak spatial heterogeneity modeling, and unclear variable interaction mechanisms. These issues constrain the identification of development gaps and the design of targeted interventions. To address these challenges, this study proposes GeoRRDI (Geographically Explainable Rural Revitalization Development Index), an interpretable assessment framework integrating satellite-derived and multi-source geospatial data. Using 3093 administrative villages in Luoyang, Henan Province, China, as the study area, we construct a 19-indicator system across four dimensions: Ecological Conditions, Land Use Structure, Settlement Dynamics, and Service Accessibility. SHAP (SHapley Additive exPlanations) is used to interpret variable importance and perform feature selection, enhancing the interpretability and applicability of the indicator system. For modeling, a GWRF-RF model—combining geographically weighted mechanisms with random forest—is applied, effectively capturing spatial heterogeneity and nonlinear patterns. Results show Land Use Structure as the most influential factor. Fragmented land at urban fringes and limited accessibility in remote areas significantly impact development levels. Rural revitalization exhibits spatial clustering and regional polarization. Projections for 2024 indicate overall improvement, with notable gains in southern mountainous areas and urban fringe zones, narrowing regional disparities. Compared with OLS, GWR, RF, and GWRF models, the GWRF-RF achieves the best performance (RMSE = 1.5241; R 2 = 0.7506), demonstrating strong accuracy and spatial adaptability. This study advances a closed-loop approach from problem identification to mechanism explanation and practical guidance, offering a scalable pathway and empirical reference for fine-scale rural revitalization assessment.