Spatio-temporal heterogeneity and driving mechanisms of RSEI in the north-south sections of the Beijing-Hangzhou grand canal: an empirical study using GEE and XGBoost-SHAP

Abstract

Abstract The Beijing-Hangzhou Grand Canal, a vital ecological corridor and cultural heritage site, requires a comprehensive understanding of the spatio-temporal evolution and driving mechanisms of its ecological environment to support sustainable regional development. This study leveraged the Google Earth Engine cloud platform and MODIS growing-season imagery (May-September, 2000–2020) to assess the spatiotemporal dynamics of ecological quality along the entire canal using the Remote Sensing Ecological Index (RSEI). An explainable machine learning framework (XGBoost-SHAP) was further applied to quantitatively disentangle the contributions of natural and anthropogenic drivers underlying the observed spatial heterogeneity in RSEI. The results revealed that: (1) A pronounced and persistent north-south gradient in RSEI values was identified, with ecological quality consistently higher in southern regions compared to northern regions over the two-decade period; and (2) the driving mechanisms demonstrated distinct differences between sections-the ecological quality in the northern section was primarily shaped by natural factors such as precipitation and temperature (“natural factor-dominated” regime), whereas in the southern section it was mainly driven by nighttime light intensity, indicative of urbanization (human activity-dominated” regime). This study elucidates the differential causes of ecological quality divergence between the north and south sections of the canal. The integrated GEE and XGBoost-SHAP framework provides a robust and interpretable approach for attribution analysis in complex environmental systems. This approach has the potential to be extended to other large linear ecosystems and provides a scientific basis for region-specific ecological protection and restoration strategies.