A high-resolution coastal risk assessment framework: Integrating knowledge driven and machine learning models for the Andhra Pradesh coastline

Abstract

Coastal regions face increasing threats from climate change and anthropogenic activities worldwide, and more sophisticated risk assessment tools are needed. This study introduces and applies a novel hybrid framework for coastal risk assessment, integrating a knowledge driven analytic hierarchy process (AHP) model with a data driven eXtreme Gradient Boosting (XGBoost) predictive model. The framework was applied to 55 coastal villages along the Andhra Pradesh coastline in India, a region frequently exposed to hydrometeorological hazards. A comprehensive coastal risk index (CRI) was computed using 18 variables encompassing geophysical hazards, socioeconomic vulnerabilities, and adaptive capacities. The AHP model, which is based on expert-weighted pairwise comparisons, produces internally consistent and reliable variable weights. The XGBoost model was subsequently trained to predict the AHP-derived risk classes, serving as a data-driven validation of the expert-based framework. Statistical validation through ANOVA confirmed that the five risk classes (very low to very high) were statistically distinct. Multiple linear regression revealed shoreline change, coastal slope, and population growth as significant positive drivers of risk. Crucially, the analysis revealed that improving accessibility to cyclone shelters and road networks offered the most substantial potential for risk reduction. While strong agreement (Cohen's kappa = 0.83) was observed between the AHP and XGBoost classifications, indicating a strong alignment between expert knowledge and data patterns, the XGBoost model's low predictive accuracy (0.41) on a limited validation set suggests the need for more extensive data for developing a standalone predictive tool. The outputs of this study, including a high-resolution risk map, provide actionable intelligence for policymakers to formulate targeted interventions. This research contributes a replicable methodology for monitoring progress towards UN Sustainable Development Goals 13 (Climate Action) and 14 (Life Below Water) by providing a quantifiable measure of coastal resilience.