A set of 86 soil profiles, belonging to 13 national Italian soil typologies, were selected for the spectroscopy analysis. A total of 208 samples from the soil horizons was analysed by using Fourier transform infrared spectrometer (FTIR) with DRIFT detector for MIR region. The samples selected for PLS calibration models were from the soil archive of CRA-ABP (Research Center for Agrobiology and Pedology, Firenze). Ten grams of soil samples <2 mm were grounded by a mill and sieved to <100 μm particle diameter. MIRDRIFT spectra were processed for PLS calibration by Unscrambler X. The PLS calibration models were calculated to predict: sand, clay, total organic carbon, and inorganic carbon. A subset of 132 soil horizon samples, belonging to 57 soil profiles, were used for the calibration models, and a test set of 76 soil horizon samples (29 profiles) were used to validate the models. Different calibration strategies were used to establish the best calibration models for the prediction of the selected soil properties: a) random selection (set generated by a randomized experimental design method); b) selection based on a diversity index; c) selection based on a set of standardization samples. The different strategies, respect to the random selection, seem to have not driven to sensible improvements for prediction of the main properties of this set representing the main typologies of Italian soils. Prediction accuracy using random selection was: for sand, R2 = 0.83; RPD = 2.4; for clay R2 = 0.86; RPD = 2.8; for total organic carbon, R2 = 0.90; RPD = 3.1; and inorganic carbon R2 = 0.93; RPD = 3.1, respectively.
Soil properties prediction of selected Italian soil typologies by means of MID-Infrared diffuse Reflectance spectroscopy
Luigi P D'Acqui;Alessandra Bonetti;
2014
Abstract
A set of 86 soil profiles, belonging to 13 national Italian soil typologies, were selected for the spectroscopy analysis. A total of 208 samples from the soil horizons was analysed by using Fourier transform infrared spectrometer (FTIR) with DRIFT detector for MIR region. The samples selected for PLS calibration models were from the soil archive of CRA-ABP (Research Center for Agrobiology and Pedology, Firenze). Ten grams of soil samples <2 mm were grounded by a mill and sieved to <100 μm particle diameter. MIRDRIFT spectra were processed for PLS calibration by Unscrambler X. The PLS calibration models were calculated to predict: sand, clay, total organic carbon, and inorganic carbon. A subset of 132 soil horizon samples, belonging to 57 soil profiles, were used for the calibration models, and a test set of 76 soil horizon samples (29 profiles) were used to validate the models. Different calibration strategies were used to establish the best calibration models for the prediction of the selected soil properties: a) random selection (set generated by a randomized experimental design method); b) selection based on a diversity index; c) selection based on a set of standardization samples. The different strategies, respect to the random selection, seem to have not driven to sensible improvements for prediction of the main properties of this set representing the main typologies of Italian soils. Prediction accuracy using random selection was: for sand, R2 = 0.83; RPD = 2.4; for clay R2 = 0.86; RPD = 2.8; for total organic carbon, R2 = 0.90; RPD = 3.1; and inorganic carbon R2 = 0.93; RPD = 3.1, respectively.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
