Assessing disaster resilience in mountain villages using an improved DPSIR-A framework and multi-model machine learning

Abstract

Abstract Amid the dual challenges of global climate change and frequent geological hazards, evaluating the disaster resilience of mountainous villages is crucial for sustainable regional development. This study proposes an integrated resilience assessment framework specifically designed for high-altitude, tourism-dependent ethnic villages. The framework extends the classical DPSIR model by incorporating an Adaptability (A) dimension, which quantifies traditional ecological knowledge and community learning capacity. To address the class imbalance typical of small-sample geological hazard datasets, the study integrates the SMOTE oversampling technique with a multi-model evaluation chain (IVM-SVM-RF). This method overcomes the generalization limitations of machine learning models in small-sample environments. The results show: (1) Model Performance Breakthrough: The SMOTE-enhanced Random Forest (S-RF) model outperforms both SVM and IVM models, with the highest performance (AUC = 0.753, Kappa = 0.754). It is particularly effective in identifying low-resilience areas under small-sample conditions, confirming that SMOTE augmentation corrects class imbalance and improves model accuracy in capturing marginal low-resilience zones. (2) Spatial Differentiation: Zhangzha Town exhibits distinct “topography-constrained economic clustering” patterns of resilience. The high-resilience core zone (13.05%) is concentrated in the Ganhaizi sector, driven by socio-economic adaptability, while the extremely low-resilience zone (9.42%) is scattered along the southern periphery, influenced by steep terrain and delayed responses. (3) Nonlinear Drivers: Feature importance analysis identifies Fractional Vegetation Cover, Rainfall, and Distance from Roads as key resilience determinants. Additionally, “soft resilience” factors, such as Building Disaster Resistance and Villagers’ Disaster Awareness, play significant roles in mitigating physical risks.