Satellite sensors, machine learning, and river channel unit types: A review

Understanding spatial arrangement and linkages of distinct morphological groups within a larger landscape mosaic will lead to a more generalized conceptual framework for interpreting fluvial systems. The focus of this chapter is on the mapping and interpretation of river channel unit types using satellite sensors and machine algorithms. Studies on river channels using remote sensing are beginning to emerge; however, very few to no studies have focused on medium- to small-scale basins. This is understandable due to the spatial resolutions of freely available Earth Observation products. However, with various pan sharpening techniques, issues around spatial resolution are being resolved albeit not without care and caution. The top five selected features for channel unit types prediction using random forest are cross-sectional area, total stream power, wetted perimeter, band 5, and discharge, while for kNN are slope, discharge, hydraulic radius, cross-sectional area, and depth. The choice of the Near-Infrared Band in random forest suggests the possibility of the combination of bands with morphological variables in discriminating channel unit types. Indices have been developed in environmental studies even though most of these indices are in combination with other bands, and there is the possibility to explore combining band 5 with other bands or with channel morphological variables.