2021 Volume 77 Issue 2 Pages I_485-I_494
The spatial distribution of geological properties is often modeled by dividing it into a trend component and a random component. The random component is important for understanding the characteristics of local variability in the spatial distribution. Using Gaussian process regression with superposition of multiple stochastic fields, the trend and random components are separated for two types of 1D and 3D measured data in cone penetration tests, and the autocorrelation function is investigated for the random component. Various autocorrelation functions, such as Gaussian, Markovian, Binary noise, Whittle-Matérn, and Gaussian-Markovian arithmetic mean models, were compared using the information criterion AIC and BIC. As a result of model selection, Whittle-Matérn and the arithmetic mean model were selected as the autocorrelation function for random components. Examples of estimating the trend and random components at arbitrary locations using the selected models are presented.