IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002 
   
IMPACT OF ON-BOARD HYPERSPECTRAL DATA COMPRESSION ON MINERAL 
MAPPING PRODUCTS 
K. Staenz ? , R. Hitchcock P , S. Qian ? , and R.A. Neville ? 
? Canada Centre for Remote Sensing, Natural Resources Canada, 588 Booth Street, Ottawa, ON, 
Canada K1A 0Y7 (karl.staenz Q ccrs.nrcan.gc.ca) 
P Prologic Systems Ltd., 75 Albert Street, Suite 206, Ottawa, ON, Canada 
(robert.hitchcock G ccrs.nrcan.gc.ca) ©) Canadian Space Agency, 6767 Route de 1’ Aéroport, St. Hubert, QC, Canada J3Y 8Y9 
(Shen-en.Qian G space.gc.ca) 
Commission VII, WG VII/1 
KEY WORDS: Hyper spectral, Geology, Compression, Processing, Classification, Impact Analysis 
ABSTRACT: 
As on-board data compression is an option for future operational hyperspectral satellite systems, its impact on the data products need 
to be investigated. Accordingly, the study presented in this paper investigated the impact of lossy Hierarchical Self-organizing 
Cluster Vector Quantification (HSOCVQ) data compression on the identification and mapping of minerals in environments with 
sparse vegetation cover. For this purpose, an Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) radiance cube acquired over 
the Cuprite mining district area in Nevada on June 12, 1996 was compressed by factors of 10, 20, and 40. The original data and the 
de-compressed data were processed separately, applying atmospheric correction using MODTRAN4.2 and spectra post-processing 
prior to automatic Iterative Error Analysis (IEA) endmember selection, and subsequent constrained spectral linear unmixing to 
produce mineral (endmember) abundance maps. The results indicate that the errors between original radiance data and de- 
compressed data increase with increasing compression ratio. This trend is also true for the derived mineral abundance maps. In 
general, for most of the endmembers, the 10:1 and 20:1 compression ratios produced abundance maps which are spatially similar to 
those extracted from the original data, when only fractions larger than 0.5 are mapped. Only these higher fractions are of interest for 
exploration purposes. One endmember out of 15 was lost using the 40:1 compression ratio and, consequently, this particular 
endmember could not be mapped. 
1.0 INTRODUCTION 
With the launch of spaceborne hyperspectral sensors, data 
transmission becomes an issue due to the high data rate 
required to cope with the large volumes of hyperspectral data. 
This is especially true when moving towards operational 
systems, as compared to technology demonstrators such as 
NASA's Hyperion. In order to overcome this problem, lossy 
data compression can be used to reduce the data volume while 
preserving enough information for the generation of application 
products in various areas, such as forestry, agriculture, 
environment, coastal/inland waters, and geoscience. Suitable 
data compression techniques are those which use Vector 
Quantization (VQ) (Qian et al, 1997 and 2000). These 
techniques are characterised by their near lossless property with 
high compression ratio and relatively simple structure. 
The goal of this study is to apply a VQ compression technique, 
called the “Hierarchical Self-organizing Cluster Vector 
Quantification” (HSOCVQ: Qian et al, 2002), to calibrated 
(radiance) hyperspectral airborne data to investigate the impact 
of this technique on mineral mapping products. For this 
purpose, an Airborne Visible/Infrared Imaging Spectrometer 
(AVIRIS: Green et al., 1998) data set acquired on June 12, 
1996 over a test site near Cuprite, Nevada, U.S.A., was used. 
The products retrieved from data compressed with HSOCVQ 
were compared quantitatively and qualitatively with the 
products extracted from the original (uncompressed) data. The 
ratios used in the compression are 10:1, 20:1, and 40:1. The 
subsequently de-compressed data sets were processed in the 
  
same way as the original data. Major processing steps include 
the removal of atmospheric effects, automatic extraction of 
endmembers, and application of a constrained linear spectral 
unmixing technique to map the minerals. These processing 
steps were carried out using the Imaging Sp ectrometer Data 
Analysis System (ISDAS: Staenz et al., 1998) developed at the 
Canada Centre for Remote Sensing. The paper describes in 
detail the aforementioned processing steps together with the 
extracted results. Special emphasis is given to the analysis of 
the end products (fractions of minerals) but also to the 
comparison of intermediate outputs such as the endmembers. 
2.0 DATA USED 
A Cuprite standard Jet Propulsion Laboratory (JPL) AVIRIS 
data set, collected on June 12, 1996, has been used for this 
study. This sensor acquires imagery at approximately 20-m 
ground resolution from an ER-2 aircraft in 224 spectral bands, 
each about 10 nm wide, in the 400-nm to 2500-nm wavelength 
range. Additional information for identification of endmembers 
includes the USGS (2002) spectral data base. 
The site selected for this study lies within the Cuprite mining 
district of Nevada (37.6 °N and 117.2 ° W). This site has been 
used as a test area for mineral mapping in hyperspectral remote 
sensing for many years (Goetz and Srivastava, 1985; Hook and 
Rast,..1990; Swayze et.31,.1992; Neville et al, 2003). 
Accordingly, this site is very well characterized in terms of 
mineralogy. This and the excellent exposure of alteration 
minerals such as alunite, kaolinite, buddingtonite, and others 
together with limited soil development and sparse vegetation