Assessing the statistical features of a remote sensing data set is a fundamental task in a lot of data analysis.For example, in the image clustering method, we often need to estimate the statistical similarity measure [1]; In many classification algorithms, we need spatial statistics-based expressions to create the decision boundary between various classes [2]; In researching of end member detection, we also need to consider the spatial distribution of end members[3],[4]using the statistical characteristics of the data set. For big data, we often estimate a vector of model parameters given a training data set. Such statistical feature estimation provides us far more information than a simple inquiry and can be used to improve human interpretation of inferential outputs, do bias correction, perform hypothesis testing, make more efficient use of available resources, perform active learning, and optimize feature selection, among many more potential uses. There are a large number of studies that focus on the statistical features of big data sets[5]–[11].However,in many remote sensing applications related to big data sets, we often do not directly assess their statistical features. In order to manifest some of the statistical character is tics of the data sets, we often represent them based on some transforms. How to represent big data sets is one of the fundamental problems in researching big data,as most data processing tasks rely on an appropriate data representation. For many image processing tasks, the wavelet transform [12] of the data is the preferred transform. For remote sensing big data, multi-resolution representation by wave let transform is more and more important for many algorithms such as image segmentation[13],image de-noising [14], image restoration [15], image fusion [16], change detection [17], feature extraction [18], and image interpretation.Therefore,the estimation of statistical features of big data in the wave let transform domain is one of the most important problems.