Multi-Temporal vs. Hyper-Spectral Imaging for Future Land Imaging at 30 m

We propose to determine the information content of multi-temporal land imaging in discrete Landsat-like spectral bands at 30 m with a 360 km swath width and compare this to the information content of hyper-spectral land imaging at 60 m with a swath width of 145 km.  We will analyze 30 m visible and near infrared cloud-free data collected every two weeks for the entire continuous lower 48-sates in 2011 and 2012. 

The extremely successful Landsat series of satellites have collected invaluable imagery of the Earth’s surface since Landsat-1 was launched in 1972.  Since 1982 with Landsat-4’s thematic mapper instrument, 30 m multispectral imagery have been collected in discrete visible, near-infrared, and short wave infrared bands complemented by thermal imagery at coarser resolutions.  Landsat-8, launched in 2013, and Landsat-7, launched in 1999 and since 2003 suffering from a lack of scan line corrections, are the sources of current US land imaging data.  JPL and their associates have proposed the replacing the Landsat 30 m discrete multispectral visible, near-infrared, and short wave infrared imaging with hyper-spectral imagers, patterned after HyspIRI, a JPL instrument.   

The argument hyper-spectral imager enthusiasts make for replacing a discrete band Landsat-type instrument is there is more information in hyper-spectral data, because you have so many more spectral bands.  JPL’s hyper-spectral HyspIRI instrument, scheduled for launch in 2016, has a 60 m spatial resolution, 212 spectral bands, and a 145 km swath width.  This argument never considers information theory and the fact that there is a very high correlation between adjacent spectral intervals in the visible, near infrared, and short-wave infrared regions.  This has been investigated with hyper-spectral data by Tucker and Maxwell (1976) and Tucker (1978) who found extremely high correlations between adjacent 5 nm spectral intervals in the visible and near-infrared spectral regions.  These results have been further extended by Tucker and Sellers (1986).

The “hyper-spectral conundrum” results from the trade off between the number of spectral bands, spatial resolution, radiometric accuracy, and swath width or revisit frequency.  It is difficult ir not impossible for a hyper-spectral instrument with hundreds of bands to have a 30 m spatial resolution and a short revisit frequency. 

Tucker, C.J. and E.L. Maxwell, 1976.   Sensor Design for Monitoring Vegetation Canopies. Photogrammetric Engineering and Remote Sensing 42(11):1399-1410.

Tucker, C. J. 1978.  A Comparison of Satellite Sensor Bands for Vegetation Monitoring. Photogrammetric Engineering and Remote Sensing 44(11):1169-1180.

Tucker. C.J. and P.J. Sellers, 1986. Satellite remote sensing of primary production. International Journal of Remote Sensing 7:1395-1416.