Insights into geomorphic and vegetation spatial patterns within dynamic river floodplains using soft classification approaches

Abstract:

Lowland rivers in broad alluvial floodplains create one of the most dynamic landscapes, governed by multiple, and commonly nonlinear, interactions among geomorphic, hydrologic, and ecologic processes. Fluvial landforms and land-cover patches composing the floodplains of lowland rivers vary in their shapes and sizes because of variations in vegetation biomass, topography, and soil composition (e.g., of abandoned meanders versus accreting bars) across space. Such floodplain heterogeneity, in turn, influences future river-channel evolution by creating variability in channel-migration rates. In this study, using Landsat 5 Thematic Mapper data and alternative image-classification approaches, we investigate geomorphic and vegetation spatial patterns in a dynamic large tropical river. Specifically, we examine the spatial relations between river-channel planform and fluvial-landform and land-cover patterns across the floodplain. We classify the images using both hard and soft classification algorithms. We characterize the structure of geomorphic landform and vegetation components of the floodplain by computing a range of class-level landscape metrics based on the classified images. Results indicate that comparable classification accuracies are accrued for the inherently hard and (hardened) soft classification images, ranging from 89.8% to 91.8% overall accuracy. However, soft classification images provide unique information regarding spatially-varying similarities and differences in water-column properties of oxbow lakes and the main river channel. Proximity analyses, where buffer zones along the river with distances corresponding to 5, 10, and 20 river-channel widths are constructed, reveal that the average size of forest patches first increase away from the river banks but they become sparse after a distance of 10 channel widths away from the river.