Disclaimer

Acknowledgements

V. Choice of Spectral Bands

There is a rich heritage of knowledge from over 20 years of electro-optical remote sensing research. Draw upon this and use it as a starting point when approaching a data set. Fundamental relationships between specific spectral bands and plants and soils are well documented. Look at the relationships between red and NIR and various plant characteristics. Then try NDVI. When specifying spectral bands for analysis, channels with band centers close to the band centers of Landsat ETM+ are a good choice:

Terra satellite MODIS or the EO-1 satellite Hyperion instrument spectral band configurations are also reasonable choices.

Note the potential pitfalls of specifying narrow spectral bandwidths below 500 nm and above 1000 nm as explained in the section on Silicon Detector Sensitivity. Another factor to consider is the bandwidth of the system. The AISA, like many low-cost airborne scanner systems are "band-width limited", meaning that only so much data can be handled at a time. This forces users to balance the need for many spectral bands with the need for high spatial resolution ground instantaneous field of view (GIFOV). As the GIFOV increases, the number of available spectral bands will decrease.

HYPErspectral or Multispectral?

Despite over 20 years of electro-optical multispectral and hyperspectral image research, NDVI is still the most widely used approach for analysis. Hyperspectral exploitation has lagged, presumably because of the lack of widespread data availability to the remote sensing research community, and because of a lack of pressing scientific justification. One can better understand this by addressing why the HIRES instrument was deleted from the NASA EOS/Terra satellite. Thus, spectral vegetation index (SVI) approaches are still very appropriate.

Land cover classification accuracies have been shown to improve from the use of hyperspectral imagery. Recent work on the value of improved sensitivity of narrow spectral bandwidths to vegetation characteristics when using SVI analysis is further evidence of the potential of hyperspectral imaging. Red Edge analysis also relies on hyperspectral data as does spectral unmixing.