Machine learning-informed vulnerability assessment of coastal erosion: A case study of the Minjiang River Estuary

Study region: Minjiang River Estuary, China. Study focus: Reliable assessments of coastal erosion risk are essential for sustainable urban planning. However, existing methods often fail to capture the dynamic, nonlinear interactions of natural and anthropogenic factors, and the ''black-box'' nature of many machine learning models limits their practical application. Addressing this gap, we developed a dynamic framework to assess long-term coastal erosion vulnerability in the Minjiang River Estuary. Our study integrated multi-temporal data from 16 key erosion-inducing factors over a 30-year period (1990–2020) and employed five machine learning algorithms to enhance both the predictive accuracy and interpretability of the model. New hydrological insights for the region: New Hydrological Insights for the Region: Results reveal a generally weak erosion trend along the estuary, punctuated by zones of intense local degradation. The Random Forest model achieved the highest accuracy (≥0.92) and AUC (≥0.97), enabling reliable identification of high-risk areas for targeted coastal management interventions, such as shoreline protection and urban planning adjustments. Feature importance analyses indicate watershed-scale land cover and land use (LCLU) dynamics are the dominant drivers of long-term erosion vulnerability, while short-term patterns are shaped by temporally variable factors. These findings highlight the critical value of integrating time-sensitive drivers into coastal risk assessments and underscore the importance of model selection for adaptive urban and environmental management. The proposed approach offers a scalable and transferable methodology for supporting climate-resilient planning in vulnerable coastal cities.