SEMI-BLIND SOURCE SEPARATION FOR ESTIMATION OF CLAY CONTENT OVER SEMI-VEGETATED AREAS, FROM VNIR/SWIR HYPERSPECTRAL AIRBORNE DATA
Keywords: Hyperspectral remote sensing, Semi-Blind source separation, Non-Negative Matrix Factorization, partial least squares regression, clay content, semi-vegetated pixels
Abstract. The applicability of Visible, Near-Infrared and Short Wave Infrared (VNIR/SWIR) hyperspectral imagery for soil property mapping decreases when surfaces are partially covered by vegetation. The objective of this research was to develop and evaluate a methodology based on the “double-extraction” technique, for clay content estimation over semi-vegetated surfaces using VNIR/SWIR hyperspectral airborne data. The “double-extraction” technique initially proposed by Ouerghemmi et al. (2011) consists of 1) an extraction of a soil reflectance spectrum ssoil from semi-vegetated spectra using a Blind Source Separation technique, and 2) an extraction of clay content from the soil reflectance spectrum ssoil, using a multivariate regression method. In this paper, the Source Separation approach is Semi-Blind thanks to the integration of field knowledge in Source Separation model. And the multivariate regression method is a partial least squares regression (PLSR) model. This study employed VNIR/SWIR HyMap airborne data acquired in a French Mediterranean region over an area of 24 km2.
Our results showed that our methodology based on the “double-extraction” technique is accurate for clay content estimation when applied to pixels under a specific Cellulose Absorption Index threshold. Finally the clay content can be estimated over around 70% of the semi-vegetated pixels of our study area, which may offer an extension of soil properties mapping, at the moment restricted to bare soils.