Additional data channels, like gradient magnitude, data value, etc., are compressed in the same order as the positions of the voxels to exploit spatial coherence also. Huffman encoding of differences of successive values and additional zlib compression (for further reduction of uniform areas) is used.