Multiple Machine Learning Algorithms and Pedoenvironmental Attributes Applied in Geophysical Surveys

Geophysical techniques applied to soils and machine learning algorithms have been used to understand the dynamics of the pedosphere and model soil attributes, but studies exploring the spatial prediction of geophysical data are still scarce. In this research, we aimed to model and map geophysical and soil attributes using parent material and terrain attributes with different machine learning algorithms. In addition, we tested nested leave-one-out cross validation (nested-LOOCV) methodology to deal with datasets with a small number of samples. We collected soil physico-chemical and geophysical data (gamma-ray emission from uranium, thorium and potassium; magnetic susceptibility and, apparent electric conductivity) using three sensors: a gamma-ray spectrometer, a susceptibilimeter and conductivimeter. The performances of the best algorithms based on the root mean square error (RMSE) and mean absolute error (MAE) for each predictor were higher compared to the use of a mean value for the entire area (NULL Model). The most important variables were the parent material and the digital elevation model for most predictions. The nested-LOOCV methodology proved to be adequate for small samples. Machine learning techniques are potential tools for the model and mapping geophysical data when only field data with proximal sensors are available