Spectral signatures and vegetation indices

Chapter 5 of Remote sensing.

Spectral reflectance signatures

The reflectance of a material across a range of wavelengths is its spectral reflectance signature. In principle a material can be recognised and mapped from its signature, if the sensor measures enough bands to tell it from other materials’ signatures. That is the basis of multispectral remote sensing.

• Clear water reflects little at all wavelengths, most in blue, less and less towards the infrared, and almost nothing in the near infrared. It looks dark blue.
• Turbid water, with suspended sediment, reflects more in red (about 0.65 µm) and looks brownish.
• Bare soil reflects more and more from blue to short-wave infrared. How much depends on its composition, colour, structure and moisture: wet soil is darker than dry.
• Vegetation has a signature of its own, which makes it easy to tell from other surfaces:
  • low in blue and red, where chlorophyll absorbs the light it uses for photosynthesis;
  • a peak in green, which makes plants look green;
  • a steep rise from red to near infrared around 0.7 µm, the red edge;
  • high in the near infrared, reflected by the cell structure inside the leaves; the arrangement of leaves in the canopy affects it too, which helps tell vegetation types apart;
  • in the short-wave infrared, lower and varying with the water content of the leaves, with absorption bands of water at about 1.45, 1.95 and 2.5 µm. Between those bands, reflectance rises as leaves dry out, which is why SWIR bands help detect drought stress and burnt areas.

The signature also depends on the plant’s condition. Stressed or senescent vegetation reflects less in the near infrared and more in red: the red edge flattens.

Vegetation indices

Bands can be combined to bring out vegetation. Because vegetation reflects much in the near infrared and little in red, the ratio of the two is high for vegetation and low for everything else: the ratio vegetation index,

RVI = NIR / Red

A ratio has no upper limit. A fixed range, from −1 to 1, is often more useful, and is reached with a normalised difference of two bands, (V1 − V2) / (V1 + V2). With near infrared and red it is the normalised difference vegetation index:

NDVI = (NIR − Red) / (NIR + Red)

For Sentinel-2, NDVI = (B8 − B4) / (B8 + B4). Typical values:

Surface	NDVI
Water	below 0
Bare soil, rock, built surfaces	about 0–0.2
Sparse or dry vegetation	about 0.2–0.5
Dense, green vegetation	about 0.6–0.9

Because NDVI is a ratio of reflectances, a common scale factor cancels out: the result is the same whether the bands are stored as reflectance (0–1) or as reflectance × 10,000. An offset does not cancel; it must be removed first (see how reflectance is stored).

What NDVI does not show

• Saturation. Above an NDVI of about 0.8 the index hardly changes as vegetation gets denser. In June, forest and a closed cereal crop have nearly the same NDVI, although a colour-infrared image tells them apart (Vegetation greenness at Ledreborg (Example)).
• Soil background. Where vegetation is sparse, the colour and moisture of the soil beneath change the value.
• Mixed pixels. A 10 m pixel on a road with grass verges, or on a pond edged by trees, averages them. Small or narrow features take the value of their surroundings.
• Date. NDVI follows the growing season. A field is bare in April and green in June, and a harvested field is bare again in August. Values are only comparable between images from the same time of year, and preferably the same sensor.

Other normalised differences

The same form with other bands gives other indices, for example the normalised difference water index (green and NIR, for open water) and the normalised difference moisture index (NIR and SWIR, for the water content of vegetation).

The NDVI layer can also be combined with other bands in a colour composite, to help distinguish types of vegetation.