No abstract

No abstract

This paper presents the multiple linear regression (MLR) models developed from electrical resistivity and seismic refraction surveys for quick prediction of subsurface soil’s shear strength parameters. A total of four parameters have been considered with electrical resistivity and seismic refraction velocity as the independent variables: and soil cohesion and internal friction angle as the dependent variables. In order to mitigate the effects of nonlinearity of resistivity and velocity, both datasets were initially log-transformed to conform with the fundamental assumptions of regression analysis. Two models were therefore built based on the strong multiple linear relationships between explanatory and response variables, with coefficient of determination (R2), 0.777, p-values, < 0.050, Durbin-Watson value, 1.787 and multicollinearity, 1.185. The obtained models’ coefficients were transferred and used for the estimation of 2D models soil cohesion and internal angle of friction for validation. Thereafter, the developed models demonstrated good performance, having subjected to accuracy assessment with results at < 5%, and < 10% for the root mean square error (RMSE) and weighted mean absolute percentage error (MAPE) respectively. Therefore, the new developed soil’s shear strength MLR models have provided continual description of soil properties in two-dimensional form, enhancing the subsurface information for site investigations as compared, to one-dimensional information from the invasive method.

This paper presents multiple linear regression (MLR) soil shear strength models developed from electrical resistivity and seismic refraction tomography data. The MLR technique is used to estimate the value of dependent variables of soil shear strength based on the value of two independent variables, namely, resistivity and velocity. These parameters were regressed using regression statistics technique for generating MLR model. The results of MLR model, which is based on the estimation of model dependent parameters (Log10 resistivity and Log10 velocity), calculated for p-value, are less than 0.05 and VIF value less than 10 for cohesion and friction angle models. This result shows that there is a statistically significant relationship between cohesion and friction angle with geophysical parameters (independent variables). The estimation accuracy of the MLR models is also conducted for verification, and the result shows that RMSE value for predicted cohesion and predicted friction angle is 0.77 kN/m2 and 1.73° which is close to zero. Meanwhile, MAPE value was found to be 4.57 % and 7.61 %, indicating highly accurate estimation for the MLR models of predicted cohesion and predicted friction angle. Based on the application of near surface, the study area was successfully classified into two regions, namely, medium and hard clayey sand. Thus, it is concluded that MLR method is suitable in estimating the subsurface characterization that covered more regions compared to the traditional method (laboratory test).