Spectral Bands

Sunlight and the energy the Earth emits span a wide range of wavelengths, and most surfaces reflect or radiate that energy differently at each one. A remote sensing instrument turns this into information by measuring several separate slices of the spectrum at once. Each slice is a spectral band, and the set of bands a sensor carries decides what it can and cannot tell apart. Understanding bands — what a band actually measures, which parts of the spectrum are most useful, and how measurements in different bands combine — is what lets you move from a picture to a physical interpretation. These ideas hold regardless of the software you later use to open the data, and they build directly on the Digital Imagery concept, where each pixel already holds one number per band.

From wavelengths to bands

Electromagnetic energy is described by its wavelength, and the full spectrum runs continuously from short, high-energy wavelengths through the visible range and on into the infrared and beyond. A sensor does not record this continuum in full. Instead, each of its bands integrates the energy arriving within a defined wavelength interval and reports it as a single number. A band therefore has a centre wavelength and a bandwidth rather than a single exact wavelength, and its sensitivity is described by a spectral response function — a curve that tails off toward the edges instead of stopping at a hard cutoff. Two sensors with bands of the same name can still respond slightly differently, which is why calibrated values and documented band definitions matter when you compare data from different instruments.

The number, placement, and width of a sensor’s bands are deliberate design choices. A few broad bands capture a lot of energy and support fine spatial detail but blur spectral distinctions; many narrow bands resolve subtle differences in how a surface reflects but collect less light per band. This is the difference between broadly multispectral and finely hyperspectral instruments, and it is a trade-off, not a ranking: the right band design depends on what you are trying to observe.

The main regions of the spectrum

A handful of spectral regions do most of the work in optical remote sensing, and each reveals something different. The visible region, roughly 0.4 to 0.7 micrometres, is the reflected sunlight our eyes perceive as colour; it responds to pigments, water, and the general brightness of a surface. The near-infrared (NIR) region, just beyond red at roughly 0.7 to 1.3 micrometres, is invisible to us but strongly reflected by healthy vegetation and strongly absorbed by water, which makes it one of the most informative bands for separating living plants from bare ground and open water. The shortwave-infrared (SWIR) region, from roughly 1.3 to 3 micrometres, is sensitive to moisture in leaves and soil and to the composition of rocks and minerals, and it helps distinguish snow from cloud and map burned areas.

These three regions all measure reflected solar energy, so they depend on daylight. The thermal-infrared region, near 8 to 14 micrometres, is different in kind: it measures energy the surface itself emits as heat rather than reflects, so it relates to temperature and can be sensed by day or night. Because the physical quantity behind a thermal band is emitted radiation, not reflected sunlight, it is interpreted with a different logic from the optical bands, even though a sensor may collect all of them together. The exact boundaries between these regions are conventions and vary a little between sensors; what stays true is the distinct physical behaviour each region captures.

Band combinations and visual interpretation

A single band is a greyscale image — one measurement per pixel. Colour appears on a screen only when three bands are assigned to the display’s red, green, and blue channels to form a composite. Mapping the visible red, green, and blue bands to their matching channels produces a natural-colour image that looks much as the scene would to the eye. But the assignment is a choice, not a fixed rule, and mapping invisible bands into visible channels — a false-colour composite — makes otherwise hidden structure legible. A common example routes the near-infrared band into the red channel, so vigorous vegetation, which reflects NIR strongly, stands out in bright red. Composites do not add new measurements; they arrange existing band values so a human interpreter can see contrasts that a single visible image would hide. How those values are then stretched, ramped, and scaled so the composite reads honestly — and stays comparable across dates — is the subject of the Visualization concept.

Spectral signatures of surfaces

Because materials reflect and emit energy differently across wavelengths, each surface type has a characteristic pattern of reflectance across the bands — its spectral signature. Recognising these signatures is what turns band values into identification. Healthy vegetation absorbs visible light for photosynthesis, especially in the blue and red, reflects a little green, and then reflects sharply and strongly in the near-infrared, producing a steep rise known as the red edge; its SWIR reflectance dips where leaf water absorbs. Water reflects weakly across the visible, more in the blue than the red, and absorbs almost everything in the NIR and SWIR, so clear water appears very dark in those bands. Bare soil typically reflects more as wavelength increases from the visible into the NIR and SWIR, with the exact shape shifting according to moisture, organic content, and iron minerals. Snow is brilliantly reflective across the visible yet drops off steeply in the SWIR, and that contrast is precisely what separates bright snow from equally bright cloud. Built and paved surfaces tend to be spectrally flat and moderately bright, and because a single pixel often covers several materials at once, urban areas commonly show mixed signatures rather than one clean curve. Reading these differences across bands is the foundation for Classification, which formalises the step from signature to labelled surface type.

Band ratios and indices as derived measurements

Once you know that surfaces differ across bands, you can design a measurement that isolates a property of interest. A band ratio or spectral index combines two or more bands arithmetically so that the contrast you care about is amplified while nuisance variation is suppressed. The most familiar is the normalised difference vegetation index, NDVI, computed as the difference between the near-infrared and red bands divided by their sum. It works because healthy vegetation is high in NIR and low in red, so the normalised difference is large for vigorous plants and small or negative for water and bare ground. Dividing by the sum matters as much as the difference: it largely cancels effects that scale both bands together, such as changing illumination and terrain shadow, leaving a value that is comparable from place to place and, over time, from date to date. The same template — a normalised difference tuned to a known signature — underlies indices for surface water, moisture, snow, and burn severity.

An index is a derived measurement, not a raw observation, so its reliability depends on the inputs. Because a ratio compares bands directly, the bands should be calibrated reflectance rather than raw digital numbers if the result is to mean the same thing across scenes, which is why the Digital Imagery concept precedes this one. Used well, indices compress a rich spectral signature into a single interpretable layer and become the raw material for Change Detection and time-series analysis. Bands are the measurements; signatures are the patterns they reveal; indices are the purpose-built quantities you build on top — and together they are what makes it possible to observe the surface, not just to picture it.