Geostatistical Inverse Problem: A Modified Technique for Characterizing Heterogeneous Fields

Abstract

This paper describes a modification of the self-calibrating method for generating equally likely realizations (conditional simulations) of the transmissivity field, that honour measurements of transmissivity and dependent variables (heads, concentrations, etc.). Soft data (e. g. geophysics) can also be included in the conditioning procedure as a external drift. Moreover, spatial variability patterns of the “real” field (as observed through field or lab experiments) are respected. The results of the algorithm are compared with those obtained by the most commonly used methods in groundwater, such as zonation and pilot points (conditional estimation methods). The performance of these geostatistical inverse approaches was compared on a synthetic data set, where the outcome is based on qualitative (resemblance between the obtained transmissivity fields and the ‘real’ one) and quantitative criteria (goodness of fit between computed and measured heads). Results show that the inclusion of head data in the conditioning procedure provides a better solution than the one obtained including only transmissivity data. Final comparison (simulations/estimations conditioned to both type of data) shows similar results. The choice of the best method depends on whether the modeller seeks small-scale variability (conditional simulation methods) or large-scale trends (conditional estimation methods).