Bathymetrix-AI
An advanced Machine Learning pipeline for Satellite-Derived Bathymetry (SDB). Features ICESat-2 integration, In-Situ Data filtering, and Spatial Residual Stacking.
Bathymetrix-AI is an advanced QGIS research toolkit for deriving high-precision Satellite-Derived Bathymetry (SDB) from corrected multispectral satellite imagery (e.g., Sentinel-2 Level-2A). It combines physics-based image processing with modern Machine Learning to improve bathymetric accuracy while addressing many of the limitations of traditional SDB methods.
Core Workflow (Five-Phase Framework)
Phase 1 – Advanced Pre-processing
Prepare satellite imagery and generate optimized input features by applying:
-Hedley et al. (2005) sun-glint correction
-Advanced water masking
-Physics-based Log-Ratio feature generation
-Deep-water filtering
Phase 2 – Robust Data Filtering
Improve training data quality by removing noisy or unreliable observations using:
-Linear RANSAC
-Least Squares (LS) Variance Fit
-Huber Variance Fit
These filtering methods help produce cleaner ICESat-2 training datasets before model development.
Phase 3 – Global AutoML & Feature Analysis
Automatically evaluate and optimize Machine Learning models through:
Optional feature correlation analysis
Benchmarking of 15+ Machine Learning algorithms, including:
-Random Forest
-XGBoost
-LightGBM
-CatBoost
and additional regression models , Independent spatial cross-validation
Hyperparameter optimization using:
-Random Search
-Grid Search
-Bayesian Optimization
Phase 4 – Adaptive Spatial Refinement
Reduce localized prediction errors through:
-Spatially adaptive residual correction
-Independent spatial cross-validation
-Residual analysis to identify and correct local biases in shallow-water predictions
Phase 5 – Validation & Reporting
Evaluate final model performance using independent unseen test data and generate comprehensive accuracy reports for scientific and hydrographic applications.
Scientific Foundations
Bathymetrix-AI builds upon well-established methods from the remote sensing and hydrographic literature, including:
-Stumpf et al. (2003) – Log-Ratio algorithm for Satellite-Derived Bathymetry
-Hedley et al. (2005) – Physics-based sun-glint correction
-Fischler & Bolles (1981) – RANSAC algorithm
-Zhang et al. (2021) – Least Squares and Huber Variance Fit for ICESat-2 filtering
-Alevizos (2020) – Residual analysis and spatial refinement
-Bergstra & Bengio (2012) – Random Search for hyperparameter optimization
-Parrish et al. (2025) – Global assessment of ICESat-2 bathymetry
Bathymetrix-AI is developed for scientific research and hydrographic applications.
The software architecture, source code, and documentation were developed by the author and further optimized with assistance from Google Gemini AI.