Watershed-scale Mapping of Braided River Substrate using Multimodal Airborne Remote Sensing and Machine Learning

The presence of excess surficial and interstitial fine sediment in gravel- and cobble-bed rivers may create a cascade of impacts that degrade ecosystem health, recreational value and natural character. Measurements of morphologically dynamic river substrates are time-intensive and synoptic methods must balance resolution and cost.

A method to map, reproducibly, the spatially-explicit and longitudinal distribution of fine sediment has been developed for the piedmont section of the Rangitata River, Canterbury, NZ. Dense airborne lidar (~120 pts/m²) and RGB orthophotography (35 mm/pix) was acquired in March 2021 over the entire Rangitata fairway below the gorge (reach length 55 km). These data were combined with a boosted random forest method to discern surficial facies using the local elevation distribution and multi-scale elevation and colour texture with low training effort. The method is validated to field mapping and terrestrial lidar patches, and the predictors are summarized via permuted feature importance.

Detailed substrate facies maps allow an understanding of longitudinal and relative elevation trends in surficial fine sediment relative to abstractions and changes in hydraulic geometry.

Survey design choices affect the utility of a lidar and orthophoto dataset. This poster supplements the results of this high-density survey with an analysis of progressively thinned data, evaluating the balance of cost and accuracy in substrate mapping.

___________________________________________________________________________