An efficient and accurate algorithm for generating spatially-correlated random fields (original) (raw)

This paper presents a new computer algorithm for generating spatially-correlated random ÿelds. Such ÿelds are often encountered in hydrology and hydrogeology and in the earth sciences and used as inputs for Monte Carlo simulations. The algorithm is designed by using a multilevel grid strategy and combining the matrix decomposition (MD) method and the screening sequential simulation (SSS) method. The idea originates from the facts: (i) the MD method accounts for all possible nodal correlation values and hence the accuracy of the method is high, but it can be extremely computationally intensive for ÿne meshes with large number of nodes, and (ii) the SSS method is more e cient because a small search neighbourhood for conditioning can be used due to the screening e ect of measurements, however, for large separation distances, correlation values from the SSS method is signiÿcantly inferior to the values obtained by the MD method. Numerical examples are presented to demonstrate the new method. It is shown that the presented method is much more e cient than the MD method and more accurate than the SSS method. The new algorithm is also versatile: it can directly simulate ÿelds of irregular geometry without additional e orts.