Pattern-based simulation is a growing area of the multi-point geostatistical simulation. The successfulness of any pattern-based simulation depends on how efficiently a pattern is searched from the pattern database. A discrete wavelet-based algorithm is proposed in this paper; this algorithm assists in reducing the dimensionality of the pattern database and preserves maximum pattern information. Wavelets by construction preserve maximum information through few coefficients termed scaling coefficients. The scaling coefficients of the wavelet decomposed patterns are used for clustering the pattern database. The k-means clustering algorithm is then applied to classify patterns. The proposed algorithm is validated by simulating conditionally and unconditionally sets of categorical and continuous data and analyzing results. The comparative evaluation with filtersim shows that the proposed wavelet-based simulation algorithm performs better in all cases.
Copyright 2012 International Geostatistics Congress
