In this article, the authors investigate multi/hyperspectral image compression strategies when data and conditions change. In particular the authors compare the classical multi-two-dimensional (2D wavelets and 2D SPIHT) and hybrid (3D wavelets and 2D SPIHT) strategies to full 3D, for which the authors propose a new implementation based on anisotropic 3D wavelets followed by a 3D SPIHT encoder. All strategies are combined with a spectral principal component analysis decorrelation stage to optimize performance. The comparison of the proposed strategy with the other is made with regard to variations in bitrate, spatial, and spectral dimensions of the images. For consistent evaluation, the authors also propose a larger evaluation framework than the conventionally used PSNR, including nine metrics divided into four families. The authors also study the effect of compression by tiles and discuss the time and memory consumption difference between the three compression strategies. Good results are obtained for the proposed method and the benchmark shows the weaknesses and strengths of each strategy.