Automated 3D extraction of hydromorphological metrics from LiDAR data

Abstract. Hydromorphological characterization is essential for understanding river functioning, ecological status, and sediment dynamics, particularly in the context of large-scale environmental monitoring programs such as the European Water Framework Directive. The increasing availability of high-density national LiDAR datasets offers new opportunities for automated and standardized river analysis. This paper presents an automated framework for extracting hydromorphological metrics from LiDAR point clouds, combining two complementary methods: RiverCourse, dedicated to river centerline delineation using a Random Forest model applied to pre-classified point clouds, and Bf3D, which estimates bankfull elevation and width through a reach-scale three-dimensional hydraulic depth analysis. Unlike conventional cross-section-based approaches, the methodology operates directly in 3D, reducing sensitivity to transect placement and local topographic variability. Applied to 1,440 river reaches from the French Carhyce database, RiverCourse achieved an 88% acceptable delineation rate, demonstrating strong robustness across a wide range of river morphologies. The main limitations were associated with low ground-point density under dense vegetation and classification errors affecting bridges or narrow streams. Bf3D produced bankfull width estimates in good agreement with field measurements, with a mean percentage deviation of 15.7%. Results also indicate that method performance is strongly dependent on centerline quality and is particularly effective in alluvial systems, while confined rivers remain more challenging due to the absence of clear floodplain signatures. Overall, the framework demonstrates the potential of national LiDAR datasets for automated, standardized, and large-scale hydromorphological assessment.