Visible-near-infrared reflectance spectra reveal clues about the mineralogy and composition of planetary surfaces. In most cases, light interacts with a mixture of various components on planetary surfaces. While radiative transfer models have been developed to simulate mixture in particulate surfaces, the extent to which spectral features represent the bulk composition of rock remains unclear. In this study, we present the analysis of a new dataset with hyperspectral imaging of rock slabs at different resolutions to characterize the scaling of spectral data. We address the problem of spectral mixing with high-resolution imaging of Martian meteorite NWA480 at a microscopic scale. The linear combination of each pixel provide an approximate to the rock's bulk spectra. However, we find the impurities, fractures and zoning create wide spectral variability in the mineral crystals. In addition, the widespread opaque minerals may also darken the spectra, resulting in non-unique interpretation of the bulk spectra. These complexities in the natural samples prevent an accurate mineral abundance retrieval using unmixing algorithms with select spectral endmembers. We suggest applying recently developed algorithms on unmixing with spectral variability with the consideration of the petrologic context would be necessary to improve the quantitative analysis of planetary surfaces.