An Evaluation of Riverbed Roughness Metrics Derived from UAV–SfM Point Clouds and Their Relationships with Grain Size Distribution in Mountain Rivers

Understanding sediment grain size distribution in riverbeds is fundamental to analyses of sediment transport, riverbed morphology, and ecological habitats. Recent advances in unmanned aerial vehicle (UAV)–Structure-from-Motion (SfM) photogrammetry have enabled indirect characterization of sediment grain size (D) using surface roughness (R) derived from point cloud analyses. However, the relationships between grain size and roughness, as obtained using different roughness metrics in mountain rivers, remain insufficiently investigated.

In this study, manual sediment sampling and high-resolution UAV surveys were conducted across multiple mountainous river reaches in Taiwan, characterized by coarse bed materials and wide grain size distributions. SfM-derived point clouds were used to compute three roughness metrics: roughness height (RH), standard deviation of elevations (σ), and detrended standard deviation (σd). Linear relationships were established between local grain sizes (Di, where i = 16, 25, 50, 75, and 84) and their corresponding percentile roughness values (Ri). In addition, integrated power-law relationships were developed by pooling all paired Di–Ri data across the study reaches.

The results indicate that all three roughness metrics (RH, σ, and σd) exhibit strong correlations with grain size in gravel-bed rivers when analyses are conducted within the same river reach. The linear Di–Ri relationships show moderate to strong correlations (R² = 0.57–0.95), with the D₅₀–R₅₀ relationship demonstrating the highest consistency across all three metrics. Similarly, the integrated power-law relationships derived from the three roughness metrics yield high correlations (R² = 0.89–0.93). However, notable differences emerge when these relationships are applied to other river reaches. The RH-based relationship maintains more consistent predictive performance, whereas relationships derived from σ and σ_d exhibit larger deviations. These results suggest that RH-based roughness metrics offer superior applicability for estimating sediment grain size in mountain rivers. Overall, this study provides practical insights into the selection of suitable roughness metrics for grain size estimation in coarse-grained riverbeds.