HydroBound-ML: Automated Water-Surface-Mask Generation for Airborne Lidar Bathymetry Processing Using Hybrid Machine Learning and Object-Based Image Analysis

Airborne Lidar Bathymetry (ALB) requires accurate water-surface masks for precise bathymetric target extraction from full-waveform data, such as via Exponential Decomposition, preventing false target extraction over adjacent land masses. HydroBound-ML is an image-based tool developed to automatically generate accurate water-surface masks for subsequent Airborne Lidar Bathymetry (ALB) processing. Accurate water-surface masks are essential for reliable bathymetric target extraction from full-waveform data, such as via Exponential Decomposition, as they prevent false target extraction over adjacent land areas. Since static topographic databases are frequently obsolete due to river dynamics and manual delineation is prohibitively time-consuming, this study automates the water boundary vectorization process. We introduce HydroBound-ML, an open-source, hybrid Geographic Information System (GIS) framework utilizing simultaneous high-resolution multi-spectral imagery. By decoupling semantic classification from geometric vectorization, a Random Forest classifier operates within a high-dimensional spatial–spectral feature space to identify core water masses. Precise physical boundaries are subsequently delineated using Simple Linear Iterative Clustering (SLIC) superpixels via Object-Based Image Analysis (OBIA). The system integrates advanced Big Data operationalizations, including Cloud Optimized GeoTIFFs (COGs) and hardware-accelerated Boolean topology editing. Quantitative benchmarking on complex, turbid river environments demonstrates that HydroBound-ML decisively resolves the spectral ambiguity between water and gray infrastructure. Achieving exceptional accuracy (Precision = 0.99, F1-Score = 0.95) and outperforming standard thresholding techniques (NDWI, OTSU), HydroBound-ML provides a highly precise, scalable alternative to manual masking for ALB processing.