Dataset for Shipwreck Susceptibility Predictive Modeling

README 1. DATASET OVERVIEWThis dataset is compiled for advanced Shipwreck Susceptibility Mapping research using Machine Learning (ML) models (ANN, RF, SVM) in the Chinese adjacent seas. It provides a balanced set of historical shipwreck locations and non-shipwreck locations, alongside 16 high-resolution marine environmental conditioning factors. The primary objective is to offer a standardized, model-ready resource for researchers to train, validate, and benchmark novel spatial prediction algorithms. 2. FILE ORGANIZATIONThe dataset is organized into the following main folders: shipwreckData: Contains the historical shipwreck inventory.conditioningData: Stores the 16 conditioning factors and the geographic extent of the study area.susceptibilityMapping: Includes the final susceptibility maps and model assessment metrics.code: Contains the Python scripts used for data processing and analysis.README:Guide to file organization, providing concise descriptions of each file's purpose and usage. 3. DATA CONTENT DESCRIPTION A. shipwreckData: Format: CSV and GeoDatabase (.gdb). Content: Critical variables for 856 historical shipwreck sites, including vessel name, nationality, lost date, longitude, and latitude. B. conditioningData: Format: GeoDatabase (.gdb) folders and GeoTIFF raster files. Structure: Organized into four GDBs: studyArea.gdb: Contains the geographic extent of the dataset. geospatial conditions.gdb: Includes Underwater depth, distance to mainland coastline, and ship density. hydrodynamic conditions.gdb: Includes Rotation of surface wind stress, Atmosphere relative vorticity, Swell wave direction, Swell wave height, and Wave height. depositional conditions.gdb: Includes Net Primary Productivity (Nppv), pH, temperature, Salinity, Total zooplankton, Total chlorophyll, Dissolved oxygen, and Total phytoplankton. C. susceptibilityMapping: Format: Raster maps (e.g., GeoTIFF) and tabular metrics (e.g., CSV). Content: Contains three subfolders (one for ANN, RF, and SVM results). Each subfolder includes the final Susceptibility Map and a file listing model assessment metrics (Accuracy, Precision, Sensitivity, Specificity, Kappa, and AUC) for training and validation data. 4. CODE EXECUTION (code folder)The code folder provides Python scripts for the three predictive algorithms (ANN, RF, SVM). Prerequisites: You must first install the necessary Python environment listed in the requirements.txt file. Input Data: The scripts utilize the pre-processed and generalized raster data located in the training2025new_FR.gdb GeoDatabase, along with the trainingset and validatingset CSV files. Purpose: The code ensures the replicability of the susceptibility analysis and serves as a baseline benchmark for comparing new algorithms.