IEEE Geoscience and Remote Sensing Magazine - June 2019 - 59

with an acceptable level of under- and oversaturated pixels. The default 30-95-percentile stretch is a good starting
point, as it keeps the 30% of pixels with the shallowest features in the dark, as it were, to avoid displaying noisy and
indistinct features in the area. However, if the shallower
features are to be displayed as well, the stretch values can
be manually overwritten by values that are determined
based on, e.g., the histogram of the depth layer after step 1.
For a multiflight-line data set that has undergone absolute atmospheric correction to the reflectance values, it
is advantageous to keep the depth stretch values fixed because this provides consistent intensity values across flight
lines. In contrast, data sets with relative atmospheric correction or multitemporal data sets often offer more consistent results with relative (percentile-based) stretch values
because they partially compensate for relative depth differences between flight lines or acquisition dates.
CASE STUDIES
To illustrate the application and scalability of the two tools,
we demonstrate them on two data sets of different scale.
In the first case, we apply the Minimum Wavelength Mapper to a laboratory imaging spectrometer data set of a single
rock specimen to show the internal spatial patterns of hydrothermal alteration along veinlets. In the second case,
we use data from an airborne hyperspectral sensor to map
alteration patterns in the landscape on the scale of an entire alteration system. The toolboxes and data sets used for
this tutorial are available for download at https://www.itc
.nl/grs, so the steps can be reproduced by interested parties.
APPLICATION TO PROXIMAL SENSING
INTRODUCTION
This proximal-sensing study shows how the Wavelength
Mapper can be used to assess the mineralogical diversity
of a mineralized granitic rock sample. The work illustrates
the type of mineralogical information that can be extracted
from SWIR hyperspectral imagery and discusses both the
strengths of the method and its limitations in this particular example.
The mineralized granitic rock sample used in this study
(Figure 6) was obtained from a porphyry copper mine and
had an ore-grade content of 0.95% copper [50]. The granitic
rock had a fine-grained, phaneritic igneous microstructure
and was crosscut by multiple veins of different sizes and
directions. The granitic wall rock contained visually recognizable crystals of quartz, plagioclase, and K-feldspar,
while the veins consisted predominantly of quartz but also
contained tourmaline and chalcopyrite (Table 3, column 3,
and Figure 6). The chalcopyrite was the source of the copper in the rock. X-ray diffraction analysis [50] confirmed
the visually recognizable mineralogical composition and
also indicated that the rock contained muscovite. This suggested that the rock was derived from a phyllic alteration
zone within the porphyry copper system [51], [52].
JUNE 2019

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

IMAGE ACQUISITION AND PROCESSING
A SWIR hyperspectral image of the granitic rock was acquired with a Spectral Imaging Ltd. SisuCHEMA imager
between 940 and 2,540 nm with a spectral resolution of
6 nm and a spatial resolution of 0.21 mm per pixel [34].
The image was acquired from a cut and smooth surface
that was dry and dust free. The image was then corrected
to reflectance using white reference and dark current measurements. The image was spectrally subset to a wavelength
range between 1,000 and 2,450 nm to remove the noisy
bands at the start and end of the spectral range.
HypPy software [47] was used to apply the Wavelength
Mapper to the SWIR range using 2,100-2,400 nm as the
spectral subset for step 1 and 2,200-2,215 nm for step 2. Additionally, step 1 of the Wavelength Mapper was also calculated for the range of 1,850-2,100 nm. This provided a depth
estimate of the water feature near 1,900 nm (Depth1900),
which was further used, together with depth estimates of
the Al-OH feature from the 2,100-2,400-nm spectral range
(Depth2200), to calculate the illite-muscovite crystallinity
values (Crystallinity = Depth2200 / Depth1900).
RESULTS
The reflectance spectra of the hyperspectral image pixels
that correspond to the locations indicated in Figure 6 are
shown in Figure 7. These spectra show a variety of brightness levels, ranging from approximately 0.1 to 0.5. Most of
the spectra contain multiple clear absorption features. The
minerals interpreted from the spectra are illite, muscovite,
kaolinite, tourmaline, and chalcopyrite (Table 3, column 4).
The chalcopyrite spectrum does not contain diagnostic
features. Chalcopyrite was inferred by visual comparison
of the spectrum with the appearance of the mineral in the
rock at that location. All of the reflectance spectra, except

C
D
B

E
F

A
G
2 cm
FIGURE 6. A color photo of a granitic rock sample. The circles indicate the locations of the pixel spectra in Figure 7 and link to their
spectral parameters and interpretation in Table 3.

https://www.itc.nl/grs https://www.itc.nl/grs

IEEE Geoscience and Remote Sensing Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2019