| Abstract: |
Shipwrecks hold dual significance as cultural time capsules and ecological refugia that enhance marine biodiversity. However, systematic, large-scale methods for locating them are still limited. This study presents an innovative approach to map shipwreck susceptibility in Chinese adjacent seas by integrating remote sensing data with machine learning techniques. We assembled a historical shipwreck inventory and analyzed 16 conditioning factors, grouped into geospatial, hydrodynamic, and depositional categories. These factors were processed using Frequency Ratio (FR) values, which served as inputs for three ensemble models: Multi-Layer Perceptron (MLP-FR), Random Forest (RF-FR), and Support Vector Machine (SVM-FR). Model performance was evaluated through statistical metrics and ROC-AUC curves, with the RF-FR model outperforming others, achieving an AUC of 0.995 for training and 0.901 for validation. The resulting susceptibility maps identify priority areas for archaeological exploration. Feature importance analysis revealed proximity to the coastline, chlorophyll concentration, and oceanographic conditions as the primary factors influencing shipwreck occurrence. This scalable, cost-effective framework offers a valuable tool for directing underwater heritage surveys and has potential applications in marine conservation and tourism planning.
|
