02985naa a2200217 a 450000100080000000500110000800800410001910000170006024501430007726000090022052022370022965300260246665300370249265300130252970000170254270000170255970000170257670000230259370000140261677301370263011279542018-11-21       2017    bl uuuu     u00u1 u    #d1 aSILVA, E. B.  aVIS-NIR SPECTROSCOPY FOR PREDICTING SAND, SILT AND CLAY CONTENTS USING LEGACY SOIL SAMPLES OF SANTA CATARINA STATE.h[electronic resource]  c2017  aThe quantitative distribution of particle size is one of the most important soil  properties. Diffuse reflectance spectroscopy in the  visible and near infrared  (Vis-NIR)  region  have been employed to characterize and quantify soil properties in a fast, non-destructive and  low cost manner. The objective of this  study  was to  develop a calibration model with  laboratory-based soil  Vis-NIR spectra for  sand, silt and clay content using multivariate  analysis techniques.  Materials and Methods  -  A set of 1534 legacy soil samples  representative of the crops from 260 municipalities (80% of total) of Santa Catarina state were  used.  Sand, silt and clay content (in percentage) were determined  in the laboratory  by the  pipette method. Soil spectra was obtained using the FieldSpec 3 (ASD) sensor with a range of  370 to 2500 nm (Vis-NIR). Four multivariate techniques (Partial least-squares - PLS, Support  vector machine  -  SVM,  Random  forest  -  RF,  Gaussian process regression  -  GPR) and  eight pre-processing transformations (including raw spectra,  smoothing,  derivatives, normalization  and non-linear transformations) of spectral data were compared.  The coefficient of  determination (R 2 ) and the root mean square error (RMSE) were used to evaluate  the models. All analyses and modeling were performed using the AlradSpectra package in R. Results and  Discussion - On average, the predictive performance of the multivariate techniques decreased  in the following order:  PLS&gtRF&gtGPR&gtSVM  for sand; RF&gtPLS&gtGPR&gtSVM  for silt;  and  PLS&gtRF&gtGPR&gtSVM  for clay.  The combinations of pre-processing transformation and  multivariate technique that  achieved  the best  predictive result  was  Savitzky-Golay first  derivative/RF (SGD1-RF) for sand (R 2 v=0.67, RMSEv=11.56%); SGD1-RF for silt (R 2 v=0.43, RMSEv=8.51%);  and  Normalization/SVM  for clay  (R 2 v=0.76, RMSEv=8.68%).  Conclusions - This study offers the possibility to perform different modeling strategies using legacy data in  order to predict sand, silt and clay content, by  Vis-NIR  spectroscopy technique.  On the other  hand,  silt  was not predictable with the same effectiveness and more studies should be  conducted.  achemometrics modeling  aDiffuse reflectance spectroscopy  aspectral1 aGIASSON, É.1 aDOTTO, A. C.1 aCATEN, A. T.1 aDEMATTÊ, J. A. M.1 aVEIGA, M.  tIn: CONGRESSO BRASILEIRO DE CIÊNCIA DO SOLO, 36., 2017, Belém. Resumos... Viçosa: Sociedade Brasileira de Ciência do Solo, 2017.