IEEE Geoscience and Remote Sensing Magazine - June 2019 - 39

PHYSICS-BASED METHODS
First-principles physics-based compensation approaches
aim to obtain absolute surface reflectance information
without any prior knowledge of materials in the scene. However, they require data that are accurately calibrated, both
spectrally and radiometrically. First-principles RT codes are
used to model atmospheric effects; examples include MODTRAN [6]-[8], 6S [23], Laser Environmental Effects Definition and Reference [41], and Discrete Anisotropic Radiative
Transfer [42].
The most sophisticated first-principles methods allow
nonuniformity of the atmospheric water vapor and incorporate locally averaged background albedo in the adjacency term. A water-vapor column amount is retrieved locally, typically pixel by pixel, and the result is applied in
the atmospheric compensation. This locally based water
retrieval can compensate for effects not strictly related to
the atmosphere, such as varying terrain elevation. Haze is
most often treated as uniform across the scene because its
optical depth can be accurately retrieved only over dark materials, such as vegetation, water, or shadow, in the absence
of other information.
ATMOSPHERE REMOVAL AND THE ALGORITHM
FORMERLY KNOWN AS ATMOSPHERE REMOVAL
One of the earliest VNIR-SWIR atmospheric compensation approaches, which has its origin in the late 1980s,
is the ATREM program [2]. This technique was originally designed to produce surface reflectance from JPL's
HSI sensor, AVIRIS [29]. It assumes Lambertian surface
behavior and derives the integrated column water vapor
on a pixel basis using the 940- and 1,140-nm water-vapor
bands and a three-channel ratio technique [11]. Given
the solar geometry, observation geometry, and derived

water-vapor values, the transmission of seven gases (H2O,
CO 2, O 3, N2O, CO, CH 4, and O 2) is estimated along with
scattering effects, using the RT code 6S. Early versions of
the code used a previous version of the 6S RT code (i.e.,
5S) and the Malkmus [43] narrow-band spectral model
to derive the transmission, although this was later updated with a line-by-line model [44]. The output of this
algorithm is a per-pixel estimated water-vapor map and a
compensated HSI data cube. An overview of this process
is shown in Figure 6.
ATREM, however, is no longer supported and is not
available to new users [46], with its last publicly released
version being 3.1. A variant of ATREM exists under the
name TAFKAA [47], [48]; it has been used to derive waterleaving reflectances from AVIRIS, from the space-based
Hyperion sensor [49], and from other sensors that measure
ocean color. TAFKAA is actually two codes: one based on
ATREM4 (TAFKAA_6S) and one that uses tabulated RT
calculations (TAFKAA_tabular) from a vector (polarized)
model. Because scattering (particularly Rayleigh) is partially polarized, the vector RT calculation provides the highest
accuracy in the backscatter term. This accuracy is needed
over water, which is so dark that backscattering makes a
proportionally large contribution to the radiance. Another feature of TAFKAA is its accounting for differing view
angles across the scene. TAFKAA also provides a method
designed to retrieve both the aerosol type and the optical
depth over water surfaces. For all of these reasons, TAFKAA
is one of the preferred atmospheric compensation methods
for ocean remote sensing.
A limitation of ATREM and TAFKAA is their neglect of
adjacency averaging by setting t b = t t : that is, the background reflectance is replaced with that of the pixel itself.
This amounts to replacing (2) with

Input:
Sensor Information, Solar
Zenith Angle, Aerosol

Top-ofAtmosphere
Solar Irradiance

At-Sensor
Reflectance

RT Code

Three-Channel Water
Vapor Ratio at
940 µm and 1.14 µm

Input:
Radiance Cube

Water Vapor
Look-Up Table

Average
Integrated
Water Vapor

Pixel-by-Pixel
Total
Transmittance

Output:
Reflectance Cube
FIGURE 6. A general flowchart, inspired by [45], of the ATREM atmospheric compensation algorithm.
JUNE 2019

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

IEEE Geoscience and Remote Sensing Magazine - June 2019

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