Riparian Vegetation Mapping for Hydraulic Roughness Estimation Using Very High Resolution Remote Sensing Data Fusion

For detailed hydraulic modeling, accurate spatial information of riparian vegetation patterns needs to be derived in automatic fashion. We propose a supervised classification for heterogeneous riparian corridors with a low number of spectrally separate classes using data fusion of a Quickbird image and LIDAR data. The approach considers nine land cover classes including three woody riparian species, brush, cultivated areas, grassland, urban infrastructures, bare soil and water. The classical stacked vector approach is adopted for data fusion, while the nonparametric weighted feature-extraction method and the pixel-oriented maximum likelihood algorithm are used for feature-reduction and classification purposes, respectively. We test the approach over a 14-km stretch of the Sieve River (Tuscany Region, Italy). A one-dimensional river modeling is applied over the study reach comparing the results of a classification-derived hydraulic roughness map and a traditional ground-based approach. Despite the complex study reach, the classification method produced encouraging accuracies (OKS=0.77) and represents a useful tool to delineate application domains of flow resistance models suited to different hydrodynamic patterns (e.g., stiff/flexible vegetation). Hydraulic modeling results showed that the remotely derived floodplain roughness parameterization captures the equivalent Manning coefficient over 20 test cross sections with uncertainty distributions described by low mean and standard deviation values.