Spatial quantification of soil total carbon, in Djerid arid area, by merging Visible-Near infrared laboratory data to ASTER image data
Résumé
Spatial quantification of soil properties is required to monitor soil resources. In Djerid arid area (Tunisia), our aim was to map total carbon (totC) over 580 ha bare soils covered by an ASTER image. We set up Partial Least Squares Regression-kriging approach. Indeed, 144 soil samples were collected in nodes of a 200 m square grid. We calibrated a PLSR model, based on the 144 spectra extracted from the 9 Visible-Near infrared ASTER image bands, previously radiometrically corrected with the 144 soil spectra acquired in laboratory between 400 and 2500 nm according to the empirical line method. Calibrated model generates a first prediction map. Residual kriging has improved PLSR prediction accuracy. Indeed, we obtain respectively (R-2 = 0.78 and RMSE = 0.16%) for PLSR model against (R-2 = 0.53 and RMSE = 0.52%) for PLSR-kriging. Furthermore the spatial distribution of these quantifications has a physical significance. Our results open interesting perspectives for soil properties mapping on large territories.