IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India, 2002 
  
spectral resolution scanners, such as the multispectral scanners 
(e.g. Landsat- TM/ ETM+, JERS-OPS etc.). This allows 
identification of broad mineral groups like iron-bearing, 
hydroxyl bearing, carbonate bearing etc. Landsat TM data has 
been extensively used for such applications world-wide. 
  
  
  
  
  
  
  
  
s ut 
wu IW 
zl- ES 
« LO > « 
e Zz: z *xllNz 
eje o FAMES 
m. a C n a 
el- e QlifN a 
w .. > li < 
@ |. Ne 
2 i 
0:5 1-0 1-5 2-0 2:5 
(X um) —> 
VIS NIR SWIR 
je vje ie *| 
  
EMITTANCE 
Processing of Hyperspectral remote sensing data is quite 
different from that of multispectral data. There are hundreds of 
channels and the data may be of 12-bit or 16-bit type; therefore 
special processing strategies and high computational facilities 
are required. The rectification of hyperspectral data involves 
first pre-processing (which aims at converting raw radiance data 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
nap ESTee N 
E lui. EN u 
oO! T TENE 
1E LQ 100 ol AN I 
x ju ll o FEN 
«| '3 EIN G 
13 Cim ENE 
=> lui ll. KENT 
meer. Es 
o il CE Tul»? 
Er «el oe NM 
: Lil : 
8:0 9.0 10:0 no 12:0 
(A um) — 
THERMAL IR 
re | 
  
Figure 1. Absorption bands in the optical region, which enable remote sensing mapping of 
mineral assemblages and rocks. Bands in the VIS-NIR-SWIR correspond to low 
reflectance, and those in the TIR to low emittance 
  
Laboratory data show that changes in the chemical composition 
of minerals are characterized in terms of subtle changes in their 
spectral characters. It has been found that these spectral features 
have in general a width of 10-40 nm. Hence, spectral sampling 
at a 10 nm interval is generally considered suitable for obtaining 
information on chemistry of minerals. 
2.1.2 Hyperspectral Sensing: Hyperspectral sensing is used to 
collect image data in a large number (nearly 100 — 200) of 
narrow spectral bands, which allows generation of almost 
continuous spectrum at each pixel (Fig. 2). Sensors used for this 
type of study include the various aerial hyperspectral sensors 
(AVIRIS, GERIS, DIAS, HYDICE, PROBE etc.) and the new 
spaceborne Hyperion. The image data after adequate 
rectification and calibration are compared to field /laboratory / 
library spectra. 
This allows identification of minerals and mapping their relative 
abundances. For this purpose, a large amount of data on spectra 
of various types of objects (minerals, rocks, plants, trees, 
organic substances etc.) has been generated and stored in 
spectral libraries. 
     
   
GROUND SCENE 
into spectrally and spatially rectified at-sensor data) and then 
radiance-to-reflectance transformation. 
After rectification, the hyperspectral sensor data take the form 
of reflectance image data in numerous contiguous bands. The 
large amount of image data has to be processed for a positive 
discrimination and meaningful interpretation. The general 
approach involves characterization of the absorption features, 
comparison to ground truth (spectral libraries) and analysis (e.g. 
Mustard and Sunshine 1999). Fig. 3 shows an example of 
comparison of simplified AVIRIS spectra and the 
corresponding laboratory spectra (van der Meer, 1999). 
2.1.3 Spectral Unmixing: In most cases, a pixel is.composed of 
mixed objects, i.e. often there are many spectrally diverse 
objects present within a pixel. The collective response of all the 
end members present in different proportions is recorded at the 
remote sensor. the aim is to decipher these constituents and 
their relative proportions with-in a pixel. 
In the context of hyperspectral sensing in the SOR, three types 
of physical mixtures are identified: (a) linear mixture (b) non- 
DN (i,j) 
PIXEL SPECTRAL 
CURVE 
    
   
INTENSITY 
  
M SPECTRAL RANGE A2 
Figure 2. Concept of hyperspectral sensing 
470