Lossless compression techniques provide efficient compression of hyperspectral satellite data. The present invention combines the advantages of a clustering with linear modeling. A number of visualizations are presented, which help clarify why the approach of the present invention is particularly effective on this dataset. At each stage, the algorithm achieves an efficient grouping of the data points around a relatively small number of lines in a very large dimensional data space. The parametrization of these lines is very efficient, which leads to efficient descriptions of data points. The method of the present invention yields compression ratios that compare favorably with what is currently achievable by other approaches.