Vehicle re-identification (Re-ID) across disparate cameras is essential for intelligent surveillance. While single-spectral methods have advanced, multi-spectral imaging presents unique challenges, particularly related to missing spectral data. This paper introduces a Decoupled Subspaces approach for robust multi-spectral vehicle Re-ID, effectively decomposing features into common and independent components to address high dimensionality and spectral variations. Extensive experiments on the RGBN300, RGBNT100, and MSVR310 datasets demonstrate strong performance, with Rank-1 accuracies of 90.4%, 90.6%, and 46.6%, respectively. Our model exhibits resilience to missing spectral data, maintaining performance despite incomplete information and highlighting its generalization and robustness. These findings establish the Decoupled Subspaces approach as a meaningful contribution to multi-spectral vehicle Re-ID.