A full-chain assessment and prediction of social-ecological system vulnerability and its driving factors in multi-ethnic regions: Evidence from southwestern Gansu, China,

Multi-ethnic regions are intricate human–environment coupled systems. Their pronounced internal heterogeneity engenders distinctive social–ecological system vulnerability (SESV) under the twin pressures of global change and regional transformation. Yet existing studies rarely deliver systematic, full-chain analyses of SESV in such settings and often overlook the decisive influence of social-cultural and demographic factors. Targeting three representative multi-ethnic areas in southwestern Gansu, China—Dingxi City, Linxia Hui Autonomous Prefecture, and Gannan Tibetan Autonomous Prefecture—this study builds on the Vulnerability Scoping Diagram (VSD) to construct an integrated full-chain SESV assessment-and-forecasting framework that incorporates natural, social, economic, demographic, and cultural dimensions. We quantify the spatiotemporal evolution of county-level SESV during 2000–2020, identify dominant drivers with an Elastic Net model (mean R2 = 0.9981), and project SESV trajectories to 2035 using a Back-propagation (BP) neural network (mean MSE = 0.0026). The results reveal marked spatiotemporal differentiation and regional disparity: overall vulnerability declines, yet pockets of resurgence risk persist. Driver analysis shows that SESV dynamics arise from complex, spatially heterogeneous interactions among multiple factors, some of which exert counter-intuitive effects at specific development stages. Social-cultural capital and demographic structure are especially influential: demographic shifts hamper vulnerability reduction in Dingxi City; ethnic diversity bolsters resilience in Linxia Prefecture; and ecological constraints, together with industrial structure, influence Gannan Prefecture’s relative stability. By highlighting the complexity and context dependence of SESV drivers in multi-ethnic regions, this research advances theoretical understanding, offers robust empirical evidence for fine-scale vulnerability modelling, and provides a scientific basis for designing sustainable, culturally attuned risk-management strategies.