Abstract
A methodology for estimation of spatial distribution known as Geostatistics or Kriging has been used widely in various fields. The formulation to estimate trend component is one of important topics in Geostatistics. This study proposes a method that estimates trend component based on concept of sparse modelling, and random component based on Kriging, namely conventional least square method. Through application of the proposed method to simple hypothetical one dimensional data, we point out a problem that conventional LASSO type formulation produces bias to the estimated trend. A method to correct the bias is also proposed and demonstrated. Applying the proposed method to actual data obtained by cone penetration test in river dike, we estimate trend and random component of two dimensional spatial distribution of SPT value.
