Soil surface infiltration capacity classification based on the bi-directional reflectance distribution function sampled by aerial photographs. The case of vineyards in a Mediterranean area

Spatially distributed hydrological modelling is required to understand and predict erosion, flooding and pollution risks that affect the vine cultivated Mediterranean environment. Previous field studies have demonstrated the dominant influence of soil surface features on overland flow and they therefore constitute an essential input to the hydrological model. In this paper we propose a remote sensing based method to map vineyard soil surface features with a spatial and temporal resolution appropriate for integration into the model. Our goal was to classify each soil surface portion in accordance with a pre-established, field measured infiltration capacity based typology. The radiometric characteristics of the classes of this typology were measured in the field and their Bi-directional Reflectance Distribution Function (BRDF) was modelled. Vineyard sunlit soil surface pixels were automatically extracted from high spatial resolution scanned aerial colour photographs Wassenaar et al., 2001 and Wassenaar et al., 2002. These pixels are radiometrically classified by comparison of their reflectance with BRDF-based reflectance predictions of each soil surface type for the specific illumination and viewing geometry of the pixel.The results show that most hydrological soil surface classes have distinct bi-directional radiometric properties. For one given geometric configuration however, the predicted reflectance ranges of some classes can considerably overlap (tilled soils and stone layers for example), while others can always unambiguously be identified (bare soil crusts, surfaces covered for more than 50% by weed or litter).We conclude that our fuzzy classification approach and the simple radiometric information used, allow us to identify the majority of the hydrological surface types. The method can easily be transposed in time and space. Its performance quite strongly depends on the radiometric and geometric accuracy of the input data. Significant improvements in soil surface classification precision are expected from considering spatial context information and monitoring the soil surface evolution.