Mapping subaerial sand-gravel-cobble fluvial sediment facies using airborne lidar and machine learning

Substrate facies monitoring is critical for the understanding of fluvial geomorphologic and ecohydraulic patterns processes. However, direct substrate measurement time-consuming subjected to data sparsity because small sample, size, limited collections within an area interest, which make it difficult capture patterns. Most new experimental studies focus on mapping based median grain size a specific class using automatic or semiautomatic photosieving techniques. This study aimed develop apply method accurately predict size-mixture exposed riverbeds with minimal ground truth plots (100) airborne lidar machine learning. The selected testbed river was 37.5-km stretch regulated lower Yuba River in California, USA, mapped at sub-meter resolution 2017. First, we designed grid-by-point sampling binned sizes into representative mixtures, such as fine large gravel, assign subaerial labels. Second, classified multivariate cluster analysis. Third, generated 15 lidar-derived topographic spectral predictors. Six distinct were identified from field seventh, pure sand facies, UAV data. A random forest predictive model 86% 10-fold cross-validation accuracy applied produce map 1.54 m pixel scale. detrended elevation most important variable predicting spatial patterning, followed by baseflow, wetted proximity, green intensity. We conclude that learning combined intensity highly effective distinguishing mixed classes substrates. Ultimately, mixture-binning approach also provides novel insights arrangement sediment