Dynamic monitoring of fine-grained ecological vulnerability in dryland urban agglomeration integrating novel remote sensing index and explainable machine learning

Rapid urbanization and climate change are exacerbating ecological vulnerability in dryland urban agglomerations, threatening ecosystem stability and sustainability. However, persistent technological gaps in large-scale, fine-grained and long-term monitoring hinder a comprehensive understanding of vulnerability patterns in these fragile regions. To address this, a novel Dryland Ecological Vulnerability Index (DEVI) is proposed by integrating six key indicators and combining remote sensing and machine learning to simplify the complex vulnerability scoping diagram (VSD). The Hohhot-Baotou-Ordos-Yulin urban agglomeration, a typical fragile region in China’s drylands, was selected as a case study to examine the spatiotemporal dynamics of ecological vulnerability from 1986 to 2024. SHapley Additive exPlanations (SHAP) coupled with XGBoost revealed the impact mechanisms and nonlinear interactions of natural and anthropogenic drivers. Results showed that DEVI effectively captured the surface ecological features, such as mobile sand, with high correlations (>0.79) to indicators. Over nearly 40 years, DEVI initially increased, then decreased, indicating ecological conditions first deteriorated and then improved, with an overall improvement of 14%. Vulnerability notably reduced in the Loess Hills but remained high in the Ordos Plateau and northern Inner Mongolia, with degradation areas still exceeding improvements by 4.5%, reflecting an imbalance in current ecological governance. Fortunately, improved areas have been increasing, with spatial sustainability reaching 82.8%, largely driven by land cover restoration (46.6%) and socioeconomic factors (27.0%). The study also identified the thresholds and interaction effects of key drivers, delineated new ecological management zones, and proposed targeted improvement suggestions. This study provides a novel index for ecological vulnerability monitoring and further offers practical guidance for sustainable development in dryland urban agglomerations.