Proximal sensing applied to soil texture prediction and mapping in Brazil

Proximal sensors techniques, such as portable X-ray fluorescence (pXRF) spectrometry and magnetic susceptibility (MS), are becoming increasingly popular for predicting soil properties worldwide. However, there are few studies investigating the effectiveness of combining these proximal sensors for prediction and mapping soil texture in tropical soils. This work evaluated the feasibility of combining such sensors for the prediction and mapping of soil texture (sand, silt, and clay contents) through random forest algorithm in an area with varying parent materials, soil classes and land uses. A total of 236 soil samples were collected from A and B horizons, following a regular-grid design with 200 m distance between samples. All samples were scanned with pXRF and susceptibilimeter. Models for A and B horizons separately and combined were built using 70% of the samples and validated with the remaining 30% of the samples. The models with the lowest RMSE values were chosen for soil mapping and further validation. The predictions produced acceptable accuracy in modeling and mapping clay and sand fractions, but were less effective to directly predict silt fraction, although it can be easily calculated through: silt = 100 - sand – clay. MS, Fe, K2O, and SiO2, properties related to soil parent material, were the most important variables for the predictions. The best models achieved an R2 for sand, silt and clay of 0.79, 0.44 and 0.71, respectively. These results represent alternative methods for reducing costs and accelerating the assessment of soil texture spatial variability, supporting agronomic and environmental decision makings.