Remote-Sensing-Supported Transboundary Ecosystem Service Mapping: A Robust Integrated Index for the Heilongjiang–Amur River Basin

Comparable ecosystem service mapping across international borders requires inputs and choices that remain consistent across the study area. We harmonized satellite-derived and geospatial datasets for the Heilongjiang–Amur River Basin to a common 1 km grid for 2000, 2010, and 2020. Landsat-derived land cover, MODIS burned area, harmonized nighttime lights, terrain, climate, and modeled soil products supported four InVEST indicators: Habitat Quality, Carbon Storage, Annual Water Yield, and Soil Retention. Pooled p01–p99 normalization and equal weighting produced a dimensionless, relative, modeled integrated ES index. Robustness was assessed using alternative normalization, leave-one-service-out tests, bounded weight perturbation, land-cover-excluded XGBoost–SHAP models, and spatially blocked cross-validation. The basin-wide mean index values were 0.444, 0.461, and 0.451. The Chinese-side means were 0.357, 0.365, and 0.361, compared to 0.528, 0.555, and 0.538 on the Russian side; the Russia–China contrast remained positive across tested construction alternatives. The random-split R2 values were 0.819, 0.806, and 0.793, whereas the spatially blocked R2 values were 0.651, 0.582, and 0.606. SHAP analysis characterized associations in fitted predictions rather than causal effects. The three benchmark-year differences are indicative and remain sensitive to Annual Water Yield, interannual climate, and index construction.