Skip to main content

Colourful Colours 1

Hyper, not multi

For a long time, digital spectral systems of remote sensing attached mostly to satellites had not more than 5-10 bands. Thus, they can store certain ranges of the electromagnetic spectrum separated from the others, e.g. red, green, blue, and infrared radiation. The main problem in multispectral remote sensing is the wavelength width of the bands. The radiation with a wavelength of 0.49 micrometre is stored in the same band as the radiation with a wavelength of 0,57 micrometre and is, therefore, treated as equal.

 

But every material on the surface of the Earth has a certain spectral fingerprint. Roughly, similar material has similar spectral characteristics. But if we want to differentiate between plant species , it is important to detect even the smallest spectral difference between them. Conventional multispectral systems cannot detect whether a plant absorbs more light at 0.49 micrometre than at 0.57 micrometre. Those remote sensing systems average the reflection values of the bandwidth (fig.)

 

Kanäle eines typischen Mulitspektralsensors


Bands of a characteristic multispectral sensor. This example shows a sensor detecting blue, red and green light as well as near and mid-infrared.

 

To remove the vagueness of multispectral remote sensing, spectral systems have been developed recently, holding up to 250 bands. They cover almost the same range of the electromagnetic spectrum but have a smaller bandwidth. Thus, they can detect differences in the reflection characteristics on a scale of 1-5 micrometre.

 

Kanäle eines typischen Hyperspektralsensors


Bands of a characteristic hyperspectral sensor. This example shows a sensor detecting the complete spectrum from blue light to mid-infrared in 128 bands.

 

Due to their enormous number of bands, these spectral systems are called hyperspectral (in contrast to multispectral). The difference between multi- and hyperspectral data becomes apparent if the single images are layered. Hyperspectral systems produce an almost gapless data cube whereas for multispectral systems the term "cube" does not seem appropriate.

 

The swipe below shows two band images of the Rhein meadows in Bonn taken by a hyperspectral sensor. Both images are in the green range of visible light. The first image has a range of 0.49 to 0.50, the second of 0.56 to 0.57. While multispectral systems record the green light in one single band (i.e. 0.49-0.57), hyperspectral systems differentiate between different green ranges. The meadows in Bonn consist of green vegetation. It can be seen that it is reflecting less in the first range of the green spectrum (darker image) than in the last range (lighter image). These differences are not visible in multispectral images.

 

 

Two images taken in the green range of light (start range and end range). There is a sharp distinction between those images, visible only to the hyperspectral sensor. (Images by courtesy of HyVista Corp. (HyMap))


 

 


Conclusion:

The spectral resolution of remote sensing systems has increased highly during the last years. Now, we do not only speak of multispectral, but of hyperspectral systems. Taking all bands together, the reflection characteristics of a landscape can be depicted almost continually in a curve. This was not possible before because one had to work with mean values. Hyperspectral systems even show the slightest differences of the spectral curve immediately.