Multispectral imaging

A handful of broad colours, some of them invisible to the eye, is enough to map how green and vigorous plants are.

The basics

A multispectral sensor measures a handful of broad colours at once, usually between four and a dozen, and some of them lie outside what the eye can see. A fruit grader does something similar when it checks that an apple is red and also measures its brightness in the infrared to judge ripeness. Healthy leaves absorb most red light and reflect a great deal of near-infrared light, and dry soil changes far less between the two. Comparing those two bands pixel by pixel therefore turns a satellite image into a map of how green and vigorous the plants are. That single trick underlies most crop and forest monitoring from space.

Why it matters

Most crop and forest monitoring from space starts here. The same few bands are measured on every pass, so a field or a forest can be followed through the season and from year to year. A simple comparison of two bands turns an image into a map of plant vigour, or of open water, that anyone can compare from one week to the next.

A healthy leaf sends back little red light and a great deal of near-infrared light. Bare soil changes far less between the two, so comparing the two bands picks out growing plants. The heights are illustrative.

True colour, false colour and band combinations

A screen builds every colour from red, green and blue light, so a satellite image is displayed by choosing three of its bands and sending one to each. The result is called an RGB composite. Sending the red, green and blue bands to red, green and blue gives true colour, also called natural colour, close to what the eye would see from above. Any other choice is false colour. The most common is colour infrared, which shows the near-infrared band as red, the red band as green and the green band as blue. Healthy leaves reflect a great deal of near-infrared light, so growing vegetation appears bright red, towns blue-grey and water dark. A combination that includes shortwave infrared, such as shortwave infrared as red, near infrared as green and red as blue, shows living vegetation green and burnt ground dark red. Combinations like it are used to map burn scars, to judge the moisture of soil and plants, and in geology to tell rocks and minerals apart.

The colours are a display choice, not what the eye would see. The same pixels can be shown in many ways, so a caption or legend should say which band went to which colour. A map of an index, such as the vegetation index in the in-depth part of this page, is different again. It shows one computed value through a colour scale rather than three bands.

In depth

Each material has a characteristic spectral reflectance. Figure 1 sketches three. Vegetation absorbs blue and red light in chlorophyll, reflects a little more green, rises steeply across the red edge between about 680 and 750 nanometres, and holds a high plateau in the near infrared that comes from the internal structure of leaves. Water absorption cuts reflectance near 1.4 and 1.9 micrometres. Soil rises gradually with wavelength. Clear water reflects little anywhere and almost nothing beyond the red [1, 2].

(a) Reflectance of three surfaces

(b) Sentinel-2 bands, nominal positions and widths

Figure 1. Schematic reflectance of vegetation, bare soil and water, with the nominal positions and widths of the 13 Sentinel-2 bands. The curves place the features at the wavelengths given in the literature, and their heights are illustrative and not measured values.

The Sentinel-2 instrument is a convenient example of the design. It has 13 bands, four at 10 m pixel size, six at 20 m and three at 60 m, with three narrow bands on the red edge [3]. The red and near-infrared pair at 665 and 833 nanometres drives the best known index, the normalised difference vegetation index [2].

\mathrm{NDVI} = \frac{\rho_{\mathrm{NIR}} - \rho_{\mathrm{red}}}{\rho_{\mathrm{NIR}} + \rho_{\mathrm{red}}}

(1)

where $\mathrm{NDVI}$ is the normalised difference vegetation index, without unit; and $\rho_{\mathrm{NIR}}$ and $\rho_{\mathrm{red}}$ are the reflectances in the near-infrared and red bands, each a fraction between 0 and 1 without unit.

The ratio runs from $-1$ to $+1$ . Dense green vegetation gives high positive values, bare soil sits near zero, and water is negative. Because the numerator and denominator both scale with illumination, much of the effect of lighting cancels. Where vegetation is sparse the soil shows through, and the soil-adjusted index adds a correction factor $L$ , usually 0.5 [4].

\mathrm{SAVI} = \frac{(1+L)\,(\rho_{\mathrm{NIR}} - \rho_{\mathrm{red}})}{\rho_{\mathrm{NIR}} + \rho_{\mathrm{red}} + L}

(2)

where $\mathrm{SAVI}$ is the soil-adjusted vegetation index, without unit; $\rho_{\mathrm{NIR}}$ and $\rho_{\mathrm{red}}$ are the near-infrared and red reflectances of equation 1; and $L$ is the soil brightness correction factor, without unit, from 0 for dense vegetation, where the index equals the NDVI, to 1 for very sparse cover.

Water is picked out by the same logic using green and near-infrared bands. The index of McFeeters is positive over open water and negative over land [5]. Xu replaced the near-infrared band with the shortwave infrared to suppress built-up areas [6]. An index with the same abbreviation but different bands, by Gao, measures the water content of vegetation, so check the formula behind a name, not only the name [7].

\mathrm{NDWI} = \frac{\rho_{\mathrm{green}} - \rho_{\mathrm{NIR}}}{\rho_{\mathrm{green}} + \rho_{\mathrm{NIR}}}

(3)

where $\mathrm{NDWI}$ is the normalised difference water index of McFeeters, without unit; and $\rho_{\mathrm{green}}$ and $\rho_{\mathrm{NIR}}$ are the reflectances in the green and near-infrared bands, each a fraction between 0 and 1 without unit.

References

Rees, W. G. (2012) Physical Principles of Remote Sensing. Cambridge University Press. https://doi.org/10.1017/CBO9781139017411
Tucker, C. J. (1979) Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment, 8(2), 127-150. https://doi.org/10.1016/0034-4257(79)90013-0
Drusch, M., Del Bello, U., Carlier, S. et al. (2012) Sentinel-2: ESA's Optical High-Resolution Mission for GMES Operational Services. Remote Sensing of Environment, 120, 25-36. https://doi.org/10.1016/j.rse.2011.11.026
Huete, A. (1988) A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, 25(3), 295-309. https://doi.org/10.1016/0034-4257(88)90106-X
McFeeters, S. K. (1996) The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17(7), 1425-1432. https://doi.org/10.1080/01431169608948714
Xu, H. (2006) Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing, 27(14), 3025-3033. https://doi.org/10.1080/01431160600589179
Gao, B. C. (1996) NDWI – A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257-266. https://doi.org/10.1016/S0034-4257(96)00067-3