This software archive supports the study “Machine Learning–Based Satellite-Derived Bathymetry in Optically Complex Coastal Lagoons: Model Performance, Transferability, and Explainability.” It contains the custom Python workflow used for spectral-predictor generation, predictor screening, stratified 70/30 data partitioning, Z-score standardization, five-fold cross-validation, final model fitting, independent validation, spatial residual export, SHAP interpretation, and longitudinal bathymetric-profile visualization for Lake Burullus and Lake Manzala, Egypt. The workflow includes a Multiple Linear Regression baseline and five machine-learning algorithms: Random Forest, XGBoost, CatBoost, Support Vector Regression, and Multilayer Perceptron. Lagoon-specific configurations, environment specifications, documentation, and input-data templates are included. Original field bathymetric observations and processed Sentinel-2 calibration datasets are not redistributed in this software record and are governed by the Data Availability statement of the associated article.