Jean-Marc Gilliot

Spatial retrieval of soil reflectance from SPOT multispectral data using the empirical line method

The use of empirical relationships to calibrate remotely sensed data has been reported by many authors, but few studies have been devoted to the specific retrieval of soil reflectance and most studies have used two calibration targets only, using few ( = 5) validation targets to assess the error in the empirical relationships. Absolute corrections based on empirical line methods have seldom been applied to satellite data, particularly Satellite pour lObservation de la Terre (SPOT) images. In this study, which deals with bare cultivated soils, the empirical line method was used to retrieve soil reflectance from three programmed SPOT satellite images with 20 m and 10 m resolution based on independent sets of 8 calibration and 15 validation field targets of bare soil. The empirical line method was applied to orthorectified images with digital number (DN) values interpolated both with the cubic convolution and the nearest neighbour interpolation mode. Root mean squared errors (RMSEs) between corrected and measured values at calibration sites were lower than 0.56% for the two visible bands and ranged between 0.61% and 1.97% for the near‐infrared band, whereas at all validation sites they were mostly lower than 2.5% except for most near‐infrared bands. Such accurate results have been facilitated both by the flat topographical conditions of the Beauce region (Western Parisian Basin, France) and the homogeneous sky conditions in a small study area (2500 ha). Differences in illumination conditions in both orbital and field cases did not appear to affect the results dramatically, which suggest that even oblique viewing images can be used for such correction. Errors could also be due to changes in soil moisture and roughness related to soil management practices, or slight vegetation growth over the end of the study period. Due to spatial error, the 20‐m resolution with cubic convolution mode led to a much lower performance than the 10‐m images with either interpolation modes. The achieved results are generalizable to European sedimentary basins with loess deposits, where similar soils and topographic conditions are to be found over thousands of km2.