Regional predictions of eight common soil properties and their spatial structures from hyperspectral Vis-NIR data

Abstract The potential of the visible–near infrared (Vis–NIR; 400–2500 nm) laboratory spectroscopy for the estimation of soil properties has been previously demonstrated in the literature, and the Vis–NIR spatial spectroscopy is expected to provide direct estimates of these properties at the soil surface. The aim of this work was to examine whether Vis–NIR airborne spectroscopy could be used for mapping eight of the most common soil properties, including clay, sand, silt, calcium carbonate (CaCO 3 ), free iron, cation-exchange capacity (CEC), organic carbon and pH, without mispredicting the local values of these properties and their spatial structures. Our study was based on 95 soil samples and a HyMap hyperspectral image available over 192 bare soil fields scattered within a 24.6 km² area. Predictions of soil properties from HyMap spectra were computed for the eight soil properties using partial least squares regression (PLSR). The results showed that 1) four out of the eight soil properties (CaCO 3 , iron, clay and CEC) were suitable for mapping using hyperspectral data, and both accurate local predictions and good representations of spatial structures were observed and 2) the application of prediction models using hyperspectral data over the study area provided statistical characterizations within soilscape variations and variograms that describe in details the short range soil variations. All results were consistent with the previous pedological knowledge of the studied region. This study opens up the possibility of more extensive use of hyperspectral data for digital soil mapping of these successfully predicted soil properties.