Multispectral sensors capture images in a discrete number of relatively broad spectral bands (typically 4-12), each covering 20-200 nanometers. Each band targets a specific physical property: blue for water penetration, green for vegetation vigor, red for chlorophyll absorption, near-infrared for vegetation structure, shortwave infrared for moisture and minerals. The result is a multi-layer data cube where each pixel has a reflectance value in every band, enabling band combinations and ratios that discriminate surface materials far beyond what a single band or photograph can achieve.
From optical remote sensing fundamentals, you know that surface materials reflect sunlight differently across wavelengths. Multispectral imaging operationalizes this by sampling the reflected spectrum at strategically chosen wavelength bands selected to maximize discrimination of important surface features.
The design philosophy is targeted sampling. Each band exists for a reason. Landsat 8's OLI has 9 bands: coastal/aerosol for atmospheric studies, blue for water penetration, green for peak vegetation reflectance, red for chlorophyll absorption, NIR for vegetation structure, two SWIR bands for moisture and minerals, a panchromatic band for sharpening, and a cirrus band for thin cloud detection.
The analytical power comes from band math. NDVI uses (NIR - Red)/(NIR + Red) to quantify vegetation density while minimizing illumination effects. Similar normalized differences target water (NDWI), snow (NDSI), and built-up areas (NDBI). False-color composites display non-visible bands as visible colors, making invisible patterns immediately apparent.
The trade-off of multispectral imaging is that broad bands average over many spectral features, potentially mixing distinct absorption signatures. Hyperspectral imaging addresses this but at the cost of data volume and complexity. For most applications, multispectral imaging provides the right balance.