Abstract

The estimation of topsoil properties such as texture and organic carbon (SOC), from satellite imaging spectroscopy is hampered by the variation of soil moisture at the time of data acquisition. In this work, a soil moisture estimation model is applied to PRISMA data, to select spectra acquired under dry conditions, from data collected in different dates in three study sites in Italy. The dataset is compared to one obtained by averaging the spectra acquired at different dates on the same elementary sampling units. The estimation of soil properties is performed using machine learning regression algorithms (MLRA). The selection of relatively dry spectra provided better results for the estimation of clay, for which a GPR model gave a relative root mean squared error (RRMSE) of 26.52. However, for sand and SOC the mean spectra from the different PRISMA acquisition dates provided better results.