Sequential Gaussian Simulation is a commonly used geostatistical method for populating a grid with a Gaussian random field. The theoretical foundation of this method implies that all previously simulated nodes, referred to as neighbors, should be included in the kriging system of each newly simulated node. This would, however, require solving a large number of linear systems of increasing size as the simulation progresses, which, for computational reasons, is generally not feasible. Traditionally, this problem is addressed by limiting the number of neighbors to the ones closest to the simulated node. This does, however, result in artifacts in the realization. The simulation path, that is, the order in which nodes are visited, is known to influence the location and magnitude of these artifacts. So far, few rigorous studies linking the simulation path to the associated biases are available and, correspondingly, recommendations regarding the choice of the simulation path are largely based on empirical evidence. In this study, a comprehensive analysis of the influence of the path on the simulation errors is presented, based on which guidelines for choosing an optimal path were developed. The most common path types are systematically assessed based on the comparison of the simulation covariance matrices with the covariance of the underlying spatial model. Our analysis indicates that the optimal path is defined as the one minimizing the information lost by the omission of neighbors. Classification into clustering paths, that is, paths simulating consecutively close nodes, and declustering paths, that is, paths simulating consecutively distant nodes, was found to be an efficient way of determining path performance. Common examples of the latter are multi-grid, mid-point, and quasi-random paths, while the former include row-by-row and spiral paths. Indeed, clustering paths tend to inadequately approximate covariances at intermediate and large lag distances, because their neighborhood is only composed of nearby nodes. On the other hand, declustering paths minimize the correlation among nodes, thus ensuring that the neighbors are more diverse, and that only weakly correlated neighbors are omitted.
This is a preview of subscription content, log in to check access.
This study has been supported by a Grant from the Swiss National Research Foundation.
Abdu H, Robinson DA, Seyfried M, Jones SB (2008) Geophysical imaging of watershed subsurface patterns and prediction of soil texture and water holding capacity. Water Resour Res 44(4):W00D18. doi:10.1029/2008WR007043CrossRefGoogle Scholar
Daly C (2005) Higher order models using entropy, Markov random fields and sequential simulation. In: Leuangthong O, Deutsch CV (eds) Geostatistics Banff 2004. Quantitative Geology and Geostatistics, vol 14. Springer, DordrechtGoogle Scholar
Lee SY, Carle SF, Fogg GE (2007) Geologic heterogeneity and a comparison of two geostatistical models: sequential Gaussian and transition probability-based geostatistical simulation. Adv Water Resour 30(9):1914–1932. doi:10.1016/j.advwatres.2007.03.005CrossRefGoogle Scholar
Lin YP, Chang TK, Teng TP (2001) Characterization of soil lead by comparing sequential Gaussian simulation, simulated annealing simulation and kriging methods. Environ Geol 41(1–2):189–199. doi:10.1007/s002540100382CrossRefGoogle Scholar
McLennan J (2002) The effect of the simulation path in sequential gaussian simulation. Technical Representative, University of AlbertaGoogle Scholar
Srinivasan BV, Duraiswami R, Murtugudde R (2008) Efficient kriging for real-time spatio-temporal interpolation Linear kriging. In: 20th conference on probablility and statistics in atmospheric sciences, pp 228–235Google Scholar
Verly GW (1993) Sequential Gaussian cosimulation: a simulation method integrating several types of information. In: Soares A (ed) Geostatistics Tròia ’92. Kluwer Academic Publishers, Dordrecht, pp 543–554. doi:10.1007/978-94-011-1739-5_42CrossRefGoogle Scholar
Zhao Y, Xu X, Huang B, Sun W, Shao X, Shi X, Ruan X (2007) Using robust kriging and sequential Gaussian simulation to delineate the copper- and lead-contaminated areas of a rapidly industrialized city in Yangtze River Delta, China. Environ Geol 52(7):1423–1433. doi:10.1007/s00254-007-0667-0CrossRefGoogle Scholar