Soil parent material prediction for Brazil via proximal soil sensing
2020
Mancini, Marcelo | Silva, Sérgio Henrique Godinho | Teixeira, Anita Fernanda dos Santos | Guilherme, Luiz Roberto Guimarães | Curi, Nilton
Parent material (PM) is key in the thorough understanding of soils. However, the complexity of PM distributions and the difficulty of reaching PM in deep soils prevent its detailed assessment. Proximal sensors, such as the portable X-ray fluorescence spectrometer (pXRF), might ease this process. This work attempts to prove the potential of pXRF to predict different PMs from analyses of soil samples. The study encompassed five Brazilian states representing 1,541,309.409 km², from where 310 soil samples of various soil classes derived from 12 different PMs were collected and analyzed by PXRF. Support Vector Machine (SVM) and Random Forest (RF) algorithms were used for modeling. Modeling comprised three datasets: one containing all data (310 samples), a dataset with younger soils (151 samples) and one with older soils, conceptually less influenced by their PM (159 samples), to understand how soil-PM chemical proximity affects prediction performance, assessed via overall accuracy and Kappa coefficient. Data distribution showed pXRF can discriminate PM types via their resulting soils, regardless of the degree of weathering. Prediction results were prominent: RF and SVM achieved roughly 0.9 Kappa and overall accuracy predicting all data. For the remaining datasets, SVM achieved 0.96 Kappa and RF nearly 0.92 for younger soils, and 0.87 and 0.9, respectively, for older soils, confirming that PMs of younger soils are slightly easier to predict, but even soils heavily altered by pedogenetic processes can be accurately predicted. Results confirm the pXRF potential to predict PM from soil data, which might help in soil mapping and its consequent activities in tropical conditions.
Afficher plus [+] Moins [-]Mots clés AGROVOC
Informations bibliographiques
Cette notice bibliographique a été fournie par National Agricultural Library
Découvrez la collection de ce fournisseur de données dans AGRIS