IEEE Geoscience and Remote Sensing Magazine - June 2019 - 45

(red boxes). Figure 13(b) and (c) shows the retrieved reflectance for the white and black panels with ATCOR,
FLAASH, QUAC, and the ELM. The ELM was based on
these same white and black calibration panels, so it
comes as no surprise that the ELM results (green line) almost exactly match ground truth (black line). The results
from physics-based approaches like ATCOR (red line) and
FLAASH (blue line) are very similar and close to ground
truth. The QUAC results (magenta line) are somewhat
off in the visible region because this scene does not quite
meet the assumptions of material diversity required by
QUAC (see the "Quick Atmospheric Correction" section).
However, beyond 900 nm, the QUAC results are quite
close to the others.
Figure 14 shows reflectance retrievals for the red and
blue panels from the SHARE 2012 imagery. Again, we can
see that ATCOR (red line) and FLAASH (blue line) are almost identical and are very close to ground truth. The ELM
(green line) is even closer to ground truth, which is more
than likely due to the fact that the scene was spatially uniform in atmospheric characteristics and adjacency scattering. Therefore, there was not much atmospheric variation
from the panels (where the ELM coefficients were determined) to the red and blue panels, which were 350 m away.
As before, QUAC did a very good job in the SWIR and nearinfrared regions.

0.6

SHARE2012, Vegetation
ATCOR

Reflectance

0.5

FLAASH

In Figure 15, we take a closer look at the retrieved vegetation results, even though there is no ground truth to compare against. Here, we have
separated the compensation
approac hes: Fig ure 15(a)
AS IT STANDS, THE FIELD
shows the physics-based apOF HYPERSPECTRAL
proach, and Fig ure 15(b)
ATMOSPHERIC
shows the scene-based apCOMPENSATION IS
proach, so we can see how
FAIRLY MATURE.
close ATCOR and FLA ASH
really are for this scene. Additionally, we can see how
closely QUAC matches ELM, which performed very well
with this data set.
LITTORAL- OR COASTAL-BASED IMAGERY
A recent article [78] examined the atmospheric compensation of a coastal scene, comparing TAFKAA, FLAASH,
and QUAC. Four regions of a Compact Airborne Spectrographic Imager (CASI) image, shown in Figure 16, were
identified and studied: 1) an ocean area, 2) a vegetation
region, 3) a track area, and 4) a white panel. The intent
was to compare a scene-based approach, such as QUAC,
against physics-based methods for a scene that is largely
water. This required the application of a "mud" or shallow
water filter to weed out highly structured spectra from the
QUAC endmembers. Unfortunately, no ground truth reflectances were available.
The results are shown in Figure 17. QUAC has the advantage of eliminating negative reflectances found with
TAFKAA and FLAASH, because it used the empirical DOS
method. In Figure 17(a), we see that FLAASH produces higher

0.4
0.3
0.2
0.1
0
0.4

0.6

0.9

1.4
1.9
Wavelength (µm)
(a)

SHARE2012, Vegetation

1

2

QUAC
ELM

0.5
Reflectance

2.4

0.4

4

3

0.3
0.2
0.1
0
0.4

0.9

1.4
1.9
Wavelength (µm)
(b)

2.4

FIGURE 15. The retrieved reflectance of vegetation from the
SHARE 2012 imagery. (a) An illustration of the physics-based approaches. (b) A comparison of scene-based methods.
JUNE 2019

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

FIGURE 16. A CASI HSI image of Sabina Point in Shoalwater Bay,
Queensland, Australia. Highlighted are ROIs, including 1) an ocean
area, 2) vegetation, 3) a track area, and 4) a white panel.

45
IEEE Geoscience and Remote Sensing Magazine - June 2019

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