Remote sensing of soil surface characteristics from a multiscale classification approach

Due to their significant influence on infiltration/runoff partitioning and soil erosion, hydrological soil surface characteristics (SSC) have to be known in a spatially distributed manner. This paper proposes a new approach based on radiometric and spatial remotely sensed information, for the mapping of hydrological SSC classes according to a predefined typology based on infiltration rates. Traditional remote sensing approaches usually focus on single SSC attributes only, where the latter correspond to key structural properties such as micro topography, surface crusting and soil cover. The originality of the method proposed here is considering the composite nature of SSC classes, by combining the aforementioned single attributes. This method makes use of a multiscale image segmentation that allows extracting image objects at two spatial scales of interest. At the fine scale, each Homogeneous Unit (HU) is assigned to an SSC attribute. At the coarse scale, SSC classes are identified by combining HU of SSC attributes assigned at the fine scale. The method was applied on airborne images collected over a Mediterranean vineyard by a small Pixy drone, and validated using intensive ground-based observations. The results showed acceptable performances with an overall accuracy ranging from 63 to 84%, depending on SSC classes and surface conditions. However, unsystematic confusions still remained between SSC classes which significantly differ in terms of hydrological behaviours. Improvements are expected considering richer spectral information, and ancillary information about SSC evolution in the case of intensive temporal monitoring.