Geologic-related data are basically point data, which are distributed sparsely and irregularly with limited number as compared to the size of target area. Regional imaging of the Earth interior with high accuracy such as geologic structure, physical and chemical properties, ore grade, configuration of petroleum and geothermal reservoirs are required in Earth science and engineering fields. Accordingly, spatial modeling for reconstructing correctly the true distribution from the irregularly-spaced point data has been more and more important. This paper classifies spatial modeling techniques into spline-based, geostatistics, and artificial neural network-based types and reviews shortly their principles and recent examples of developments and applications. The examples focus on improvement of estimation accuracy using supplementary information, applicability extension to vector data such as fracture and category data indispensable to lithofacies distribution, multiscale modeling and decomposition into different scales, application of covariance function to Bayesian-based inverse problem, spatio-temporal multivariate modeling, and extension from two-point to multiple-point statistics. The techniques suitable to these examples are summarized and their usefulnesses are demonstrated through new Earth scientific and engineering findings obtained by spatial modeling. Essential issues for the next step are spatial modeling by a combination of physical and chemical laws, geologic process, and geostatistics; application to largescale dataset and global area and; deepening of consideration of multiscale and heterogeneous structures. In addition, preliminary results and trial ideas concerning these issues are presented.
View full abstract