Modeling soil moisture–reflectance

Spectral information on soil is not easily available as vegetation and farm works prevent direct observation of soil responses. However, there is an increasing need to include soil reflectance values in spectral unmixing algorithms or in classification approaches. In most cases, the impact of soil moisture on the reflectance is unknown and therefore ignored. The objective of this study was to model reflectance changes due to soil moisture in a real field situation using multiresolution airborne Spot data. As the direct observation of soils is only possible in the absence of vegetation, the effective remote sensing of soil moisture is limited to a few days each year. In such a favorable time window, modeling the soil moisture–reflectance relationships was found possible. The proposed exponential model was not valid when all soil categories were considered together. However, when fitted to each category, the RMS error on moisture estimates ranged from 2.0% to 3.5% except for silty soils with crusting problems (6%). Results also indicated that, when the soils have similar colors (i.e. same hue), soil categories can be partly grouped and the model can be simplified, using the same intercept coefficients. This study has potential application for the definition of a more generalized model of the soil reflectance. It shows that the impact of soil moisture on reflectance could be higher than differences in reflectance due to the soil categories.