IEEE Geoscience and Remote Sensing Magazine - June 2019 - 57

THE MINIMUM WAVELENGTH MAPPER TOOL
The Minimum Wavelength Mapper (Wavelength Mapper
for short) is an approach to map the wavelength position
of the deepest absorption features in spectral images [24],
[25]. The method combines the position and depth information of the deepest absorption features to give a per-pixel
overview map of the dominating mineral occurrences in an
area. It is unsupervised and repeatable, and the results can
be easily understood by geologic experts with limited image processing background. In contrast to IDL DISPEC, this
technique does not analyze asymmetry and focuses mainly
on the dominating spectral feature. Its big advantage, however, is that it was developed for spectral images and can
highlight spatial patterns in the data.
The method builds on earlier approaches of extracting
the wavelength position-e.g., Commonwealth Scientific
and Industrial Research Organisation [59] (unpublished
work), [46]-by fusing the wavelength position of the features with its respective feature depth. This allows the analyst to highlight in a single image a proxy for mineralogy
and a proxy for its mineral abundance, respectively, which
accentuates the dominating mineral patterns and suppresses spectral noise in the results by deemphasizing shallow
features that are less relevant.
The Wavelength Mapper is part of the Hyperspectral
Python (HypPy) [47] toolbox developed by the Faculty of
Geo-Information Science and Earth Observation, University of Twente (UT-ITC), Enschede, The Netherlands. It is
an open source toolbox written in Python 3 that is free
to download from the UT-ITC website (https://www.itc
.nl/grs).
THE METHOD
The Wavelength Mapper works on a pixel-by-pixel
basis and consists of two distinct steps. In step 1, the
JUNE 2019

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

continuum of the pixel spectrum is removed, and the
deepest absorption features in the spectrum are extracted with a parabolic fit through three adjacent spectral
bands (Figure 4). This parabolic interpolation allows the
detection of subtle spectral shifts that are smaller than
the spacing of the spectral bands. The interpolated position and depth of the deepest feature are stored in the
process, which is repeated for the second and third local
minima of the spectra (their use falls outside the scope
of this tutorial). The output of step 1 is two bands that
contain 1) the interpolated wavelength position of the
minimum reflectance per pixel and 2) the interpolated
depth of the absorption feature.
In step 2, the wavelength position and depth information of the deepest features are fused with a hue-saturationvalue transformation [48], [49] into a single output image,
where the hue is representative of different absorption
feature wavelengths and the intensity (value) captures the
variation in feature depth. Step 2 is typically done on only
the deepest feature but can be extended to the second and
third features as well.
THE SETTINGS
The Wavelength Mapper does not require complex user
input but allows the adjustment of a few control parameters that steer the process. We will briefly explain these
settings and useful default values for particular situations. The case studies in the latter part of this tutorial
further illustrate some of these settings and their respective results.

Reflectance (Continuum Removed)

alunite, positive for dickite; negative for halloysite, positive
for kaolinite).
Spectra can have compound features (double, triplet,
or, in general, smaller features superimposed on a larger
one). When a spectrum of a pure material has a compound
feature, it is often this attribute that is diagnostic. When
a spectrum of a mixture has compound features, it is frequently the unique absorption peaks that are diagnostic.
The maximum value approach measures the multiple features in a compound feature as a single feature and misses
information on the individual absorption features. By contrast, the inflection point approach can characterize only
the shape and area of each single feature and cannot measure the compound feature as a unity. IDL DISPEC thus provides complete information about individual spectral features only when features are in isolation or when the nature
of the materials (pure or a mixture) is known. The feature
position (center wavelength) will, however, always be reported correctly, which is information that can be propagated to the application of the second tool that we present.

Interpolated
Depth of
Absorption
Feature

0.98
0.96
0.94
0.92
0.9
0.88 Interpolated
Wavelength of
0.86
Minimum Reflectance
2.16

Measured
Wavelength of
Minimum Reflectance

2.18
2.2
2.22
Wavelength (µm)

2.24

FIGURE 4. An illustration of the parabolic interpolation used to extract the interpolated minimum wavelength position and depth of
the feature. The open circles represent the spectral band positions
and their respective continuum-removed reflectance values. The
dashed lines highlight the positions of the lowest band value and
the lowest interpolated band value, respectively. The interpolation
method allows wavelength shifts that are smaller than the spectral
distance between bands to be captured. (Used with permission
from [25].)

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