Multivariate distribution models of soil electrical resistivity

Abstract

Soil electrical resistivity is an important parameter in geotechnical engineering. In this paper, a multivariate distribution model is used to predict soil electrical resistivity. The Box-Cox method was used to transform the soil parameters, which were then analysed by a correlation matrix. Multivariate distribution models with different input parameters were established, verified and analysed, and compared with conventional empirical models. The results show that the predictive accuracy of the multivariate distribution models was improved significantly by increasing the number of parameters. The correlation coefficient (R2) of the calibration data increased from 0.05 to 0.88, and that of the validation data increased from 0.06 to 0.83. The model with the best predictive accuracy was the RE−{G, F, Sr} model (R2 = 0.87, E(ε) = 1.22, COV(ε) = 0.98). The predictive accuracy of the multivariate distribution model was obviously higher than that of the conventional empirical model. The multivariate distribution model is an effective and simple way to predict soil electrical resistivity. The proposed model exhibits good performance using limited electrical resistivity data from frozen soils.