332 
2.2. Processing techniques 
There are several methods available for converting the data collected by the airborne sensor to 
reflectance. These include the empirical line method (Conel et al., 1987), logarithmic residuals (Green 
and Craig, 1985), Internal Average Relative Reflectance (IARR)(Kruse et al., 1985), flat field correction 
(Roberts et al., 1986) and radiative transfer modelling (Gao et al., 1993). This study will be confined to 
IARR and radiative transfer modelling which do not require ground spectral information for the processing 
step. 
2.2.1. Radiative Transfer Modelling. Involves the application of a radiative transfer code (LOWTRAN, 
MODTRAN, 5S) to model the effects of changes in solar irradiance with wavelength, changes in 
transmission due to atmospheric absorptions and the scattering processes in the visible and near infrared. 
The model requires accurate inputs for the flight parameters and atmospheric conditions. This method 
theoretically provides an accuracy exceeding other methods in recovering the ground surface reflectance. 
However, when applied to the 1990 AVIRIS data, consistent errors were discernible in all pixels which are 
related to inaccuracies in the solar irradiance data (Green and Gao, 1993). These inaccuracies are present 
in both the LOWTRAN and MODTRAN codes and have large effects in the mineralogically important 
2.0pm to 2.5pm wavelength region. For this reason this technique was not used further. 
2.2.2. Internal Average Relative Reflectance. The method uses a division of the average spectral curve 
in the scene into each individual pixel spectral curve. The average curve should be dominated by 
atmospheric absorptions (which have multiplicative effects on the spectra), common calibration effects 
and atmospheric scattering (an additive effect essentially confined to the 0.4 pm to 1.0pm wavelength 
region). This method can deal with the atmospheric absorptions and calibration effects, but can not deal 
with additive effects such as atmospheric scattering. To avoid the problems with atmospheric scattering, 
the processing was confined to the 2.03pm to 2.37pm wavelength range, where most of the characteristic 
absorption features of the economically important minerals are clearly defined. 
The method is dependent on the type and proportions of the components in the scene, 
therefore artifacts of the processing are possible and the result is only approximate. However, given a 
heterogeneous scene it is possible to recover the major spectral absorption features and thus the identity of 
the surface mineralogy. Given the problems of the radiative transfer model it was decided to use this 
technique as the main form of processing for the 2.03pm to 2.37pm wavelength region. 
23. Spectral libraries 
The output from the processing is compared in a various ways to well characterised spectral libraries of 
pure minerals. The aim is to directly identify the surface mineralogy from the spectral response by finding 
the best mineral fit from the library. In many cases this is not possible due to mixed spectral responses of 
multi-component surfaces. However, for mineral exploration purposes where there is often a zoned system 
of spectrally dominant minerals a simple method of comparison ignoring surface mixtures in the first 
instance has proved to be very effective. 
There are several mineral spectral libraries available. In this study both the JPL Mineral 
library (Grove et al., 1992) and the PIMA based library created as part of the IGCP Project 264 will be 
used. The libraries show a consistency in the position of the major absorption features to within 3nm to 
7nm in most cases. 
3. IMAGING SPECTROMETERS - DATA ANALYSIS 
There are many different methods of analysis available. Those listed below are generally in use or 
currently being investigated. They are, 
Parameterisation. 
Cross-correlation. 
Neural networks. 
3.1. Parameterisation 
This technique was one of the earliest developed and has been used by many workers (Grove et al., 1992; 
Kruse et al., 1993). The method usually involves a feature extraction step to isolate the diagnostic 
absorption features from the “background” continuum spectral curve.