Bathymetrix-AI
An advanced Machine Learning pipeline for Satellite-Derived Bathymetry (SDB). Features ICESat-2 integration, In-Situ Data filtering, Spatial Residual Stacking, and Multi-Year Coastal Dynamics.
Bathymetrix-AI is a QGIS research toolkit designed to derive high-precision Satellite-Derived Bathymetry (SDB) from multispectral satellite imagery. It integrates physics-based corrections, Machine Learning, and temporal monitoring to overcome traditional SDB limitations.
CORE WORKFLOW (5-PHASE SYSTEM)
1. Advanced Pre-Processing:
- Physics-based sun-glint correction (Hedley)
- 3-Index Water Masking (NDWI, MNDWI, NWI) or automated Otsu thresholding
- Log-Ratio features: NDWI, Log(Green)/Log(NIR), Log(Red)/Log(NIR)
- Deep Water Filter with dynamic Elbow Point Detection & Polygon masking
2. Robust Filtering:
- ICESat-2 noise removal via Linear RANSAC, LS Variance Fit, or Huber Variance Fit
3. Global Auto-ML & Feature Analysis:
- Feature correlation (Auto-RANSAC, Auto-Random Forest)
- Benchmarking of 15+ ML algorithms (Ex. XGBoost, LightGBM, CatBoost, etc.)
- Spatial Cross-Validation & hyperparameter tuning (Random, Grid, Bayesian)
- Chunk-based processing for memory management on large rasters
4. Adaptive Refinement:
- Localized spatial bias correction & residual analysis (Alevizos, 2020)
5. Validation & Reporting:
- Independent accuracy assessment on unseen test data
STANDALONE MODULES
- Coastal Dynamics: Multi-year volumetric erosion/accretion trends, Morphological Stability Index (MSI), and shoreline migration polygons with automated HTML reports.
- ICESat-2 Downloader: Directly queries and downloads ATL24 LiDAR data from NSIDC for in-situ calibration.
- Tidal Converter (FES2014): Hydrographic vertical datum correction and tidal height modeling utilizing local FES2014 ocean tide grid data.
KEY REFERENCES
- Otsu, N. (1979) | Threshold selection method from gray-level histograms
- Stumpf et al. (2003) | Log-Ratio SDB inversion
- Hedley et al. (2005) | Sun-glint correction
- Fischler & Bolles (1981) | RANSAC algorithm
- Zhang et al. (2021) | LS/Huber Variance Fit filtering
- Alevizos (2020) | Spatial residual refinement
- Bergstra & Bengio (2012) | Hyperparameter search
- Parrish et al. (2025) | Global ICESat-2 bathymetry
- Wheaton et al. (2010) & Lane & Chandler (2003) | Fluvial/DEM uncertainty
Developed for scientific research and hydrographic applications.
Code & documentation optimized using Google Gemini AI.