Artificial Intelligence Framework for Predictive Sediment Dynamics and Environmental Vulnerability Assessment in Data-Limited Coastal Systems

Abstract

This study proposes a novel Artificial Intelligence–Driven Coastal Vulnerability and Sediment Dynamics Framework (AI-CVSD) designed explicitly for data-limited coastal systems. Unlike traditional process-based and purely data-driven approaches, the framework integrates machine learning (ML), deep learning (DL), remote sensing, and GIS within a hybrid predictive architecture capable of operating under sparse and heterogeneous data conditions. The proposed model incorporates feature engineering from proxy environmental variables, multi-source data fusion, and physics-informed constraints to improve generalizability and interpretability.

A simulated validation experiment demonstrates that the AI-CVSD framework achieves improved predictive performance for suspended sediment concentration (SSC) and shoreline change dynamics, with reductions in RMSE of up to 18–27% compared to baseline empirical models. Additionally, the framework enhances coastal vulnerability classification accuracy through integrated socio-environmental indicators.

The study further introduces an explainability layer using SHAP-based feature attribution and a probabilistic uncertainty module for decision support integration. Results highlight the capability of AI-driven approaches to extract meaningful patterns from limited datasets while maintaining robustness and transferability.

This research provides a scalable AI-based methodological framework for predictive coastal intelligence, with direct implications for climate adaptation, hazard mitigation, and sustainable management in data-constrained coastal environments.