IEEE Geoscience and Remote Sensing Magazine - June 2019 - 40

L Lamb
noadj(m) =

[A(m) + B(m)] t t(m)
+ L up(m).
1 - t t(m) S(m)

polishing method, consisting of a linear transform derived
from spectrally smooth pixels in the scene, was the empirical flat field optimal reflectance transformation (known as
EFFORT) [53]-[55], which proved to be very valuable for
mineral classification. Related polishing methods were later
developed for atmospheric compensation codes, such as
ACORN, FLAASH, and ATCOR. An example of polishing
using FLAASH's method is shown in Figure 8. User selection
of the degree of smoothing may help guard against removing true features that are widespread in the scene.

(7)

This approximation results in some inaccuracy in scenes
with both haze and high-contrast clutter.
HIGH-ACCURACY ATMOSPHERIC CORRECTION
FOR HYPERSPECTRAL DATA
The HATCH [50], [51] model, introduced circa 2001, uses
the same radiance equation [i.e., (7)] as ATREM but introduces some improvements. These include the incorporation of a correlated-k method to derive the atmospheric
transmission from the high-resolution transmission molecular absorption database (HITRAN) 2000 [52] spectral line database, a fast RT equation solver, an improvement to the three-band ratioing technique for water-vapor
amount retrieval (i.e., a spectral smoothness test), and an
automated spectral recalibration method. We can see the
results of these improvements, illustrated in Figure 7, on
1999 AVIRIS data over Boulder, Colorado. Figure 7 shows
the spectral plots of soil measured with an Analytical
Spectral Devices, Inc. spectrometer versus HATCH and
ATREM surface reflectances. The biggest improvement results from the spectral recalibration.

ATMOSPHERIC CORRECTION NOW
The ACORN model [56] is a commercial compensation
product that uses the RT code MODTRAN. It was originally
available through Analytical Imaging and Geophysics but is
now maintained by the developer, ImSpec; the latest version
is ACORN 6 [57]. Its basic at-sensor radiance model follows
that of ATREM and HATCH and does not account for adjacency averaging. ACORN uses MODTRAN-derived look-up
tables to determine gas absorption and scattering from aerosols and molecules. As in ATREM and HATCH, it estimates
water vapor on a pixel-by-pixel basis, keying in on the 940and 1,140-nm water-absorption bands. An advanced feature
of ACORN, which most other codes lack, is separate estimation of atmospheric water vapor from liquid water in surface
materials, such as vegetation, making the results for those
materials more accurate. Visibility is estimated with nonlinear least squares fitting between the image data and modeled radiances. ACORN also has options for wavelength
recalibration, removal of noisy channels, and a type of
spectral smoothing. As with ATREM and HATCH, ACORN
outputs a map of column water-vapor estimates and a scaled
surface-reflectance data cube. Performance assessments of
ACORN have been made by using
five different data sets collected with
the hyperspectral digital imagery collection experiment (HYDICE) HSI
sensor [58].

SPECTRAL POLISHING
It is essentially impossible for a first-principles physics-based
method to model the atmosphere and sensor properties with
complete accuracy. Therefore, a postprocessing procedure
called polishing is often used to remove atmospheric residuals and other structured spectral artifacts from the retrieved
reflectance spectra. The challenge is to smooth the spectra
without disturbing real absorption features. The original

0.6

0.5

HATCH
With Spectral
Calibration

Reflectance

0.4
HATCH
0.3

0.2

ATREM

0.1

Surface
Measurement
0.5

1.5
Wavelength (µm)

2.5

FIGURE 7. A comparison of a ground truth surface measurement with HATCH and ATREM-derived

surface reflectances. The upper spectra are plotted with a 15% offset for clarity [51].
40

FAST LINE-OF-SIGHT
ATMOSPHERIC ANALYSIS
OF SPECTRAL HYPERCUBES
A commonly used atmospheric compensation model for both hyperspectral and multispectral imagery, developed in the late 1990s, is FLAASH
[5], [25], [59], [60]. A commercial version is available in the Atmospheric
Correction Module for Harris Corp.'s
ENVI software. As with ACORN,
FLAASH also takes advantage of the RT
code MODTRAN. FLAASH and MODTRAN are well integrated, because they
are both maintained by Spectral Sciences, Inc. [60]. FLAASH is based on
the model presented in (2), and it

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

JUNE 2019

IEEE Geoscience and Remote Sensing Magazine - June 2019

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