Estimation of soil penetration resistance with standardized moisture using modeling by artificial neural networks

•We evaluated the models accuracy in prediction of soil penetration resistance (SPR).•We use artificial neural networks (ANN) modeling and linear and nonlinear regressions.•SPR with standardized moisture can be estimated with high accuracy.•ANN has higher accuracy in SPR prediction than linear and nonlinear regressions.•Soil samples from surface layer (0.00–0.10 m) are not recommended for SPR prediction. One of the most used evaluations to monitor soil compaction is based on soil penetration resistance (SPR). However, since SPR is influenced by soil moisture, this evaluation performed in the field may often lead to incorrect interpretations. This study aimed to evaluate the accuracy of models in the estimation of soil penetration resistance with standardized moisture (SPRlab) based on soil penetration resistance measured in the field (SPRfield) and on soil moisture (U) and indicate the best soil layer and best model for that. Samplings were carried out in the years 2016 (72 points – 24 in each layer) and 2017 (270 points – 90 in each layer) in three soil layers (0.00–0.10 m, 0.10–0.20 m and 0.20–0.30 m). Samples collected in 2017 were used to calibrate the models and samples collected in 2016 were used to validate them. The models used were obtained by multiple linear and nonlinear regressions and artificial neural networks (ANNs). Models were calibrated with all sampled layers and stratified per layer. In the latter case, the samples were separated into two parts, one with the surface layer (0.00–0.10 m) and another with subsurface layers (0.10–0.20 m and 0.20–0.30 m). SPRlab can be estimated with high accuracy from SPRfield and U measured in the field. We recommend the use of ANN models (MLP or RBF) and soil samples collected from the 0.10–0.30 m layer for the monitoring of soil penetration resistance.