For the reduction of the computation, in the reference [34], the HMM model is simplified by exploiting the inherent self-similarity of real world images.AlsointroducedisaBayesianuniversalHMM that fixes a few of the parameters no training is required. In reference [35], the authors propose a new ‘‘upwarddownward’’ algorithm, in which a Viterbi-like algorithm for global restoration of the hidden state tree is introduced. Almost all the methods that model the statistical characteristics of wavelet coefficients are only applicable to a small data set. Most of them are related to de-noising, texture analysis, segmentation, classification, and retrieval algorithms. For big data sets, such as space-temporal remote sensing data sets, we also need to model their wavelet coefficients to find the changing trends, discover the intrinsic mechanisms, and represent the rules of their evolution process. In this paper, we use the GMM to denote the statistical properties of wave let coefficients of are mote sensing big data set. Our contribution is to estimate the model parameters of a big data set using different aspects or dimensions such as time, spectral bands, scales, and textures.