AI-Enhanced Coastal Ecosystem Monitoring for Abrasion and Mangrove Decline Detection Using NDVI and CNN Models

Abstract

Coastal ecosystems in Indonesia are increasingly threatened by accelerating abrasion and severe mangrove degradation, especially in Mangunharjo, Semarang, where shoreline retreat continues to endanger local communities and ecological stability. This study aims to develop an AI-driven monitoring framework for detecting coastal abrasion and mangrove loss using Normalized Difference Vegetation Index (NDVI) combined with a Convolutional Neural Network (CNN) classifier. Multispectral data from Sentinel-2 imagery were processed to extract NDVI time-series from 2015 to 2025, followed by image preprocessing, normalization, and CNN-based classification. The model identifies abrasion-affected zones and declining mangrove cover, while the geospatial dashboard visualizes risk levels and restoration priority areas. Experimental results show that the CNN–NDVI model achieves high accuracy in distinguishing stable and abrasion-prone regions, with clear detection of vegetation loss patterns along the western coastline of Mangunharjo. The developed dashboard successfully integrates prediction output, interactive mapping, and AI-assisted recommendations for mangrove restoration. In conclusion, this system demonstrates the potential of combining satellite data, CNN-based analysis, and geospatial visualization to support data-driven decision-making for coastal ecosystem management and sustainable environmental planning.