Lossless compression of 3D ultraspectral sounder data using linear prediction with lattice coding

Ultraspectral sounder data is known for its huge size and sensitivity to noise in retrieving geophysical parameters. Lossless compression of it is thus desired for archival and transfer purposes. In this paper, we propose a lossless compression scheme which consists of a linear prediction, a lattice encoder, and an entropy encoder. The linear prediction serves as a whitening tool to produce a normally distributed residual. The lattice encoder can use low bit rates to approximate a Gaussian source. The entropy encoder is then used for encoding the final residual whose variance is reduced by the lattice encoder. The compression on the standard ultraspectral test dataset set of 10 AIRS granules shows that the proposed scheme outperforms DPVQ, JPEG2000 and the common compression utility BZIP2.