Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
Contents lists available at ScienceDirect
Journal of Quantitative Spectroscopy &
Radiative Transfer
journal homepage: www.elsevier.com/locate/jqsrt
A climatology of visible surface reflectance spectra
Peter Zoogman a,n, Xiong Liu a, Kelly Chance a, Qingsong Sun b, Crystal Schaaf b,
Tobias Mahr c, Thomas Wagner c
a
b
c
Harvard-Smithsonian Center for Astrophysics, Cambridge, MA, United States
University of Massachusetts Boston, Boston, MA, United States
Max Planck Institute for Chemistry, Mainz, Germany
a r t i c l e i n f o
abstract
Article history:
Received 26 January 2016
Received in revised form
4 April 2016
Accepted 4 April 2016
Available online 13 April 2016
We present a high spectral resolution climatology of visible surface reflectance as a
function of wavelength for use in satellite measurements of ozone and other atmospheric
species. The Tropospheric Emissions: Monitoring of Pollution (TEMPO) instrument is
planned to measure backscattered solar radiation in the 290–740 nm range, including the
ultraviolet and visible Chappuis ozone bands. Observation in the weak Chappuis band
takes advantage of the relative transparency of the atmosphere in the visible to achieve
sensitivity to near-surface ozone. However, due to the weakness of the ozone absorption
features this measurement is more sensitive to errors in visible surface reflectance, which
is highly variable. We utilize reflectance measurements of individual plant, man-made,
and other surface types to calculate the primary modes of variability of visible surface
reflectance at a high spectral resolution, comparable to that of TEMPO (0.6 nm). Using the
Moderate-resolution Imaging Spectroradiometer (MODIS) Bidirection Reflectance Distribution Function (BRDF)/albedo product and our derived primary modes we construct a
high spatial resolution climatology of wavelength-dependent surface reflectance over all
viewing scenes and geometries. The Global Ozone Monitoring Experiment–2 (GOME-2)
Lambertian Equivalent Reflectance (LER) product provides complementary information
over water and snow scenes. Preliminary results using this approach in multispectral
ultraviolet þ visible ozone retrievals from the GOME-2 instrument show significant
improvement to the fitting residuals over vegetated scenes.
& 2016 Elsevier Ltd. All rights reserved.
Keywords:
Surface reflectance
Ozone
Satellite
Retrieval
Trace gas
Albedo
1. Introduction
Tropospheric ozone is a principal component of urban
smog and a major contributor to anthropogenic global
warming. Ozone over the current Environmental Protection Agency (EPA) air quality standard of 75 ppbv affects
53 million people in the United States [33], more than any
other air pollutant. Tropospheric ozone has the thirdlargest short-term radiative forcing among anthropogenic
n
Correspondence to: 60 Garden Street, Cambridge, MA 02138, United
States. Tel.: þ 1 9176129834.
E-mail address: [email protected] (P. Zoogman).
http://dx.doi.org/10.1016/j.jqsrt.2016.04.003
0022-4073/& 2016 Elsevier Ltd. All rights reserved.
greenhouse gases [14]. Tropospheric ozone is formed by
the reaction of nitrogen oxide radicals (NOþ NO2 ¼NOx)
and volatile organic compounds (VOCs) in the presence of
sunlight. However, no current measurements of ozone and
its precursors combine the spatial density, temporal frequency, and vertical resolution to observe and attribute
harmful air quality. Surface and ozonesonde measurements are too spatially sparse to capture ozone distributions [10]. Satellite measurements of ozone and related
atmospheric gases have become an increasingly powerful
tool over the past two decades [22], but are limited by
poor sensitivity to near-surface ozone and return times of
at best one day [9].
40
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
The Tropospheric Emissions: Monitoring of POllution
(TEMPO) satellite instrument holds much promise in
addressing the current shortcomings in the air quality
observing system [5]. TEMPO will be launched in 2018–
2020 and will be stationed in geostationary orbit, providing hourly measurements at 4 8 km2 spatial resolution
over the US, southern Canada, and northern Mexico. This
mission is a member of a global geostationary constellation of air quality focused satellites, which also includes
Sentinel-4 over Europe [13] and GEMS over East Asia
[1,15]. TEMPO will observe in ultraviolet (UV) and visible
spectral regions (290–490 nm and 540–740 nm) to measure ozone and ozone precursors including NO2, HCHO,
and glyoxal. Observation in the weak Chappuis band takes
advantage of the relative transparency of the atmosphere
in the visible to achieve sensitivity to near-surface ozone
[25,30,6]. It has been shown that combining the UV and
visible spectral regions at a high temporal resolution is
necessary for TEMPO to achieve its air quality mission
objectives [39].
Due to the weakness of ozone absorption features in
the visible [3] the combined UVþvisible measurement is
sensitive to the visible surface reflectance, which is highly
variable not only spatially and temporally but also as a
function of wavelength. The surface reflectance spectrum
describes the ratio of reflected radiation to incident solar
radiation at every wavelength. The reflectance and its
spectral dependence vary based on the physical and biological properties as well as the solar illumination and
viewing geometry for each satellite measurement. The
spatial variation of surface reflectance at discrete wavelength bands has been characterized by many land imaging satellite instruments, including Moderate-resolution
Imaging Spectroradiometer [MODIS, [28]], Multi-angle
Imaging SpectroRadiometer [MISR, [23]], and Polarization
and Directionality of the Earth's Reflectances [POLDER,
[16]]. In addition, the capability of these instruments to
observe the same surface scene from many angles has
allowed surface anisotropy to be calculated. These studies
represent a valuable resource to inform current and future
multispectral ozone retrievals.
Current operational satellite retrievals of atmospheric
trace gases in the UV and visible spectral regions do not
account for the variability of surface reflectance spectrum
and its anisotropy. This can lead to large errors in measured trace gas amounts [26,38]. Russell et al. [27] utilized
the MODIS albedo product to develop a new NO2 retrieval
from the OMI satellite instrument that showed better
agreement with aircraft measurements than the operational OMI retrieval, which uses a much more spatially
coarse albedo. Similarly, McLinden et al. [24] used the
MODIS albedo product for improved OMI retrievals of NO2
and SO2 over Canadian oil sands. However, due to the
narrow bands used in the retrievals of these gases a full
surface reflectance spectrum was not needed. Conversely,
multispectral ozone retrievals require continuous surface
reflectance spectra. A key advancement of this study is the
development of surface reflectance spectra as a function of
wavelength targeted at the visible region by combining the
spatially dense (but spectrally discrete) information from
satellites with continuous surface spectra that capture the
potential variability of surface reflectance.
To calculate the primary modes of variability of visible
surface reflectance at a spectral resolution comparable to
than that of TEMPO (0.6 nm) we utilize reflectance measurements of individual plant, man-made, and other surface types. Using the MODIS Bidirectional Reflectance
Distribution Function (BRDF)/albedo gap-filled product
and our derived primary modes we construct a high spatial resolution climatology of wavelength-dependent surface reflectance over all viewing scenes and geometries.
We incorporate the Global Ozone Monitoring Experiment2 (GOME-2) Lambertian Equivalent Reflectance (LER)
product [32] to extend coverage to snow/ice and water
scenes. We implement the use of surface albedo EOFs in
combination with the MODIS BRDF climatology for fitting
GOME-2 visible radiances in our GOME-2 ozone profile
retrieval algorithm to test the efficacy of our approach.
Fig. 1. Land reflectance spectra and EOFs used in this study. The left panel shows the individual surface reflectance spectra for the land cover types that
compose our land cover database. Vegetation spectra (green) share a peak in the green (500–600 nm) and high reflectance in the near-infrared (NIR). The
right panel shows the mean and four leading EOFs of the set land cover spectra. The inset numbers are the amount of variance explained by each EOF. We
find that the first four EOFs capture over 99% of the spectral variation of the land surface reflectance. (For interpretation of the references to color in this
figure legend, the reader is referred to the web version of this article.)
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
2. Visible surface reflectance variability
To capture the range of different surface reflectance
spectra we use spectra from three different reflectance
databases: University of Heidelberg and Max Planck
Institute for Chemistry (UH/MPIC) vegetation, Jet Propulsion Laboratory, Advanced Spaceborne Thermal Emission
Reflection Radiometer (JPL ASTER), and United States
Geological Survey (USGS). The land surface in the conterminous US is primarily covered in crops and natural
vegetation (approximately 83%, NLCD 2011) and thus
having a state-of-the-science database for vegetation
reflectance is of particular importance. We use a new high
spectral resolution database of vegetation reflectance
spectra taken at UH/MPIC by Mahr et al. [21] that covers
the visible to NIR (360–900 nm). They used a spectrometer
with a CCD detector to measure reflectance of samples
taken from 93 plants and trees. A unique feature of this
dataset is its spectral resolution of 0.3 nm, which is finer
than that of TEMPO (0.6 nm). Mahr et al. investigated
selected samples at even higher spectral resolution
(0.15 nm) but did not find any additional structures at that
scale. To extend our spectra beyond vegetation we utilize
the ASTER spectral database (version 2.0, [2]) and the
USGS Digital Spectral Library (version splib06a, [7]). We
use spectra from these databases that describe surface
types expected in the TEMPO field of regard: soils, vegetation, and man-made materials. By combining spectra
from three different databases we include the range of
variation in surface reflectance in our spectra.
Fig. 1 (left panel) shows the reflectance spectra from
the UC/MPIC, ASTER, and USGS databases used in this
work. There are spectral patterns that stand out among
different surface types. Vegetation spectra (green lines)
have a peak in the green (500–600 nm) and a strong peak
in the NIR, called the red edge. Soil spectra (blue lines)
have increasing reflectance with wavelength; man-made
materials (red lines) also have increasing reflectance with
wavelength and can be brighter. The category of manmade materials includes 37 samples, such as concretes and
roofing types, which are the dominant contributors to
urban land cover. The spectral databases we use cover the
variability from all the land cover types expected for the
future geostationary satellites.
To capture the variability of surface reflectance contained in our composite dataset we perform an Empirical
Orthogonal Function (EOF) decomposition, also known as
a Principal Component Analysis. Each EOF represents a
distinct mode of variability as a function of wavelength
found in the dataset. Fig. 1 (right panel) shows the mean
spectrum and the four leading EOFs for our spectral
dataset as well as the percent of the spectral variation
described by each EOF. While the enforced orthogonality
of the EOFs precludes straightforward interpretation, we
see features that conform to expectations. The 1st EOF (red
line) has the features of a vegetation spectrum while the
2nd (green line) has characteristics from soils and manmade spectra along with a negative component of vegetation (due to orthogonality). The first 4 EOFs are the
primary modes of variability and capture over 99% of the
spectral variation of the land surface reflectance. The
41
relative number of the different spectra types influence
the ordering and magnitude of variability captured by EOF
but should not significantly affect the shape of the EOFs we
find. As the EOFs describe the variability about the mean
spectrum, we can describe the reflectance spectra of any
land satellite footprint as a linear combination of these
4 EOFs and the mean spectrum.
3. Surface reflectance from satellites
The surface reflectance spectrum in a given scene
observed by a satellite instrument will be a mixture of the
surface types described. To capture the spatial variability
of surface types and their effect on surface reflectance we
utilize the MODIS BRDF gap-filled product [28,29]. MODIS
provides surface reflectance and its dependence on viewing geometry at four wavelength bands relevant to visible
ozone retrievals, centered on: 469, 555, 645, and 859 nm.
Observed reflectances are corrected for atmospheric and
aerosol effects [35]. The MODIS product uses a RossThickLiSparse Reciprocal model for the anisotropy of the surface
reflectance [19], where the total reflectance is parameterized as a linear combination of three components:
isotropic (Lambertian); volumetric reflectance (leaf/vegetation scattering); and gap-driven geometric reflectance
(characteristic of terrain with shadows, e.g. forest or
buildings). MODIS provides parameters fiso, fvol, fgeo, to
compute the reflectance R for a given wavelength λ and
viewing geometry:
Rðλ; θ; ϑ; φÞ ¼ f iso ðλÞ þ f vol ðλÞK vol ðθ; ϑ; φÞ þf geo ðλÞK geo ðθ; ϑ; φÞ
ð1Þ
where the viewing geometry is described by the solar
zenith angle (SZA, θ), viewing zenith angle (VZA, ϑ), and
relative solar azimuth angle (SAA, φ). Kvol, Kgeo are the
contribution from the reflectance parameters given by the
RossThick-LiSparse model based on the viewing geometry.
The V005 BRDF parameters are provided at 30″ ( 1 km)
horizontal resolution at an 8-day temporal frequency by
inverting cloud-free and atmospherically corrected surface
reflectance observations within a grid cell acquired by
MODIS on board the Terra and Aqua satellites. Each 8-day
product is computed based on daily sampling over a 16day period. We calculated a climatology (2002–2011) of
these 8-day periods in order to best capture the average
seasonal variation. The ability to compute the reflectance
at each viewing geometry is critically important to finding
the correct shape of the surface reflectance spectrum, as
the geometrical dependence of the reflectance can vary
with wavelength [12]. Due to the occurrence of multiple
scattering in typical atmospheric observations, surface
reflection includes contributions from all directions rather
than only from the observation geometry. The BRDF model
is often integrated with the radiative transfer model to
calculate the overall surface reflection. MODIS also provides black-sky albedo (BSA) and white-sky albedo (WSA),
which are respectively the direct and diffuse components
of the total albedo. The overall albedo for a given scene can
be approximated as the weighted average of these two
42
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
values, called the blue-sky albedo. The weighting is
determined by the fraction of diffuse downwelling irradiance, which we calculate using the VLIDORT radiative
transfer model (Vector LInearized Discrete Ordinate
Radiative Transfer, [31]). Either the BRDF or the blue-sky
albedo can be used to describe the reflectance properties
of land surfaces using MODIS.
To extend surface reflectance to snow and water scenes,
which are not included in the gap-filled MODIS product,
we use GOME-2 Lambertian Equivalent Reflectance (LER).
This provides global reflectance over all surfaces but at
coarser temporal and spatial resolutions (0.5° 0.5°
monthly) and without any dependence on viewing geometry. GOME-2 LER has higher spectral resolution than
MODIS reflectance products (9 channels instead of
4 channels in the 400–900 nm range). The monthly values
are averaged from observations over the 2007–2013 period. Similar to the MODIS algorithm, the GOME-2 product
removes the modeled atmospheric contribution from the
observed top of the atmosphere radiance, but here the
surface is assumed to have uniform reflectance in all
directions (Lambertian). Snow and water scenes are more
spatially uniform than land scenes, so coarse resolution of
the GOME-2 LER introduces less error. MODIS observations
include coastal regions and are used to distinguish the
boundaries between and land and water scenes that may
not be resolved by the coarser resolution GOME-2. Snow
scenes act as Lambertian reflectors over the range of
viewing geometries planned for TEMPO (o70° SZA;
[20,8]). While ocean scenes can have a strong anisotropic
reflectance effect (glint), this anisotropy is similar over the
visible range [11]. This means that there is no significant
spectral dependence on viewing geometry in visible
wavelengths, i.e. glint affects all visible wavelengths
equally. Thus, ocean glint should not affect the shape of
the reflectance spectrum and so using a Lambertian
satellite product over the open ocean will not negatively
affect the ozone profile retrieval. This exploits the fact that
for trace gas retrievals the shape of the reflectance spectrum, not the magnitude, is the relevant information.
4. Land surface reflectance
To construct the surface reflectance spectrum of a given
land scene we combine the satellite-derived reflectance at
discrete wavelengths with the EOFs capturing the variation in land reflectance. A continuous spectrum is needed
for the multispectral ozone retrievals but the satellite data
are only for discrete wavelength bands; hence the EOFs are
needed. To determine the contribution of each EOF to the
surface spectrum we perform a linear fit of the EOFs to the
satellite-derived surface reflectances. This is done by solving for x, the vector of relative contributions of each EOF
in the equation
Ax ¼ b
ð2Þ
where the columns of A are the values of each EOF at the
satellite measurement wavelengths and b contains the
satellite-derived reflectances at those wavelengths with
the mean land surface reflectance removed. In the case of
using MODIS and the four leading EOFs (Fig. 1b) A is a
square matrix and Eq. (2) can be solved exactly. If the
number of satellite wavelengths and EOFs does not match
(e.g. using GOME-2 LER or changing the number of EOFs
fitted) the equation does not have an exact solution. We
then solve for x that minimizes the sum-of-squares error
x ¼ ðAt AÞ 1 At b
ð3Þ
In either case we reconstruct our best estimate of the
surface reflectance spectrum as a linear combination of the
mean surface reflectance and the EOFs fitted, with the
coefficients given by the corresponding values of x. The
calculation of x is fast since ðAt AÞ 1 At can be precalculated once the discrete wavelengths are selected.
This algorithm has been implemented both for BRDF and
albedo products.
This method is similar to that employed by Vidot and
Borbas [36] to construct the Radiative Transfer for the
Television infrared observation satellite Operational Vertical sounder (RTTOV) reflectance database. However, our
product differs by: (a) limiting the wavelength window for
the EOF calculation and fitting to 400–900 nm compared
to 400 nm to 2.5 mm and (b) inclusion of a high-resolution
vegetation spectra database. This leads to EOFs and land
Fig. 2. Continuous land surface reflectance spectra from fitting land Empirical Orthogonal Functions (EOFs) to MODIS reflectances (x's) for a proposed
TEMPO viewing geometry. The seasonal variability of the reflectance spectra is shown over a forested scene in New York (75°W, 41°N) and over an urban
scene (Boston, MA; 71°W, 42°N).
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
spectra that are more accurate for the visible spectral
range relevant for TEMPO ozone and other trace gas
retrievals; RTTOV was developed principally for NIR
applications. The wavelength window is limited on the UV
side of the domain to 400 nm due to limitation of both the
spectral databases and satellite BRDF data. EOF decomposition is found to perform better at capturing the
variability of surface reflectances than using average
spectra for each land cover classification.
Fig. 2 shows examples of fitting the EOFs to MODIS
BRDF data for two scenes for dates spanning a seasonal
cycle. The MODIS reflectances (x's) are calculated using the
BRDF kernels using a proposed TEMPO viewing geometry
(geostationary orbit over 100° W). The left panel shows
reflectance spectra for a forested scene in New York State
(75°W, 41°N), where the surface reflectance strongly
resembles the vegetation reflectances from Fig. 1a. There is
seasonal variation, but the vegetation signal is dominant
throughout the year; the scene is a combination of evergreen and deciduous vegetation. The right panel shows
reflectance spectra for the city Boston, where the surface
reflectance spectra is a combination of vegetation and
man-made materials. Boston is shown for comparable
viewing geometry between the two scenes. The seasonal
variation for Boston is damped compared to the forested
scene due to the lower vegetation fraction. These examples
demonstrate how the combination of satellite reflectances
and land reflectance EOFs can be used to account for the
temporal and spatial variability of the surface reflectance
spectrum. These examples also highlight the usefulness of
the MODIS channel centered at 859 nm for the shape of
the visible surface reflectance, even though it is not in the
visible itself: this channel helps constrain the vegetation
fraction of the scene and thus the contribution to the
overall spectrum of the vegetation reflectance.
Although TEMPO will be observing from a fixed point
relative to the surface of the Earth, the anisotropy of the
surface reflectance (or BRDF) must be accounted for due to
the differences in viewing angle from scene-to-scene, the
changing position of the incident solar radiation relative to
both the surface and TEMPO, and the various degrees of
multiple scattering. Fig. 3 shows the effect of the sun's
position on surface reflectance at 600 nm, the peak of
ozone absorption in the Chappuis band, for three scenes.
The VZA for each pixel is fixed (geostationary orbit over
43
100°W) but the circles show the surface reflectance for the
full range of SZAs and SAAs. The SZA is given by the radial
distance and the SAA is given by the angle from the bottom of each plot. The reflectance at 600 nm is computed
by finding the MODIS reflectances for the given geometry
and fitting the land reflectance EOFs. We compare two
scenes at similar viewing angles with different surface
types (left and center panels) and two urban scenes at
greatly different viewing angles (center and right panels).
Generally, scenes are brighter in a back-scattered viewing
geometry, when the sun is on the same side of a scene as
the satellite (bottom half of panels) than in a forwardscattered viewing geometry. We see that the dependence
on the solar position of the reflectance is significantly
different for different land types and viewing angles.
5. Snow and ocean reflectance spectra using GOME-2
To generate surface reflectance spectra for scenes not
captured by the MODIS BRDF product we use the GOME-2
LER product at seven visible reflectance channels. The
limitations of GOME-2 LER (coarse spatial resolution, does
not account for anisotropy) are minimized for snow and
ocean scenes, as discussed in Section 3. As the spectral
variability over these scenes is less well characterized in
the reflectance databases, the additional channels provided by GOME-2 over MODIS are beneficial. Further
investigation of MODIS snow albedo products [37] could
be beneficial for future TEMPO retrievals over snow. Over
land scenes, we can also use the discrete GOME-2 LERs to
generate LER spectrum. Fig. 4 shows the comparison of
land surface reflectance spectra computed by fitting the
calculated land EOFs to MODIS blue-sky albedo or GOME-2
LER. The black line is the mean difference over the TEMPO
field of regard over the year for snow-free land scenes and
the red lines denote the middle 68% (2σ) of the differences
at each wavelength. The fitted spectra show very good
agreement for wavelengths shorter than 700 nm. This
agreement holds particularly true between 500 nm and
700 nm where the ozone absorption in the Chappuis band
is strongest. This gives confidence that using MODIS and
GOME-2 reflectances to retrieve ozone profiles over different scenes will not introduce significant errors. The
positive MODIS bias relative to GOME-2 for wavelengths
Fig. 3. Surface reflectance at 600 nm (the peak of visible ozone absorption) as a function of solar position for 3 scenes viewed from TEMPO geostationary
orbit (100°W). The radial distance is the solar zenith angle (0° at the center to 90° at the edge) and the angular distance from the bottom of the plot is the
relative solar azimuth angle between the sun and the satellite.
44
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
Fig. 4. Difference in fitted land surface reflectance spectra between using
MODIS blue-sky albedo and GOME-2 Lambertian Equivalent Reflectance
(LER ) as the satellite data to fit our land Empirical Orthogonal Functions
(EOFs). The mean (black line) and middle 68% interval (bounded by
dotted red lines) of the differences of land reflectance by fitting to MODIS
minus the reflectance by fitting to GOME-2 are shown over the TEMPO
field of regard over a year for snow-free land scenes. (For interpretation
of the references to color in this figure legend, the reader is referred to
the web version of this article.)
longer than 700 nm is likely due to the GOME-2 LER product not extending into the red edge at as long a wavelength as the MODIS blue-sky albedo (772 nm vs. 859 nm).
We generate continuous spectra for snow/ice and
ocean scenes by adding a mean snow/ice spectrum and a
mean ocean spectrum from the USGS database to our suite
of land EOFs. We then fit this extended set of primary
modes of variability to the GOME-2 LER product over
snow/ice and ocean scenes. Fig. 5 shows an example of this
algorithm for a scene in Canada (110°W, 60°N) that transitions from snow-covered to vegetated during late winter
to early spring. In February the scene is snow-covered: the
reflectance is close to unity and mostly constant with
wavelength, apart from the decrease starting at 750 nm.
These features agree with snow scenes measured from
aircraft [20]. Over the following two months the snow
fraction of the scene decreases: the reflectance becomes
much lower and by April the spectrum primarily resembles vegetation. Fig. 6 shows the mean ocean spectra and
its variation (red lines denote 95% of the spectra) globally
over the yearly cycle. We show the aggregated statistics in
this case since the variability of the ocean spectrum is
much smaller than that of snow or land scenes. The
magnitude and shape of this spectrum agrees well with
airborne measurements of water reflectances by Varotsos
et al. [34]. There may be additional variation in coastal
waters; the significance of this variation on trace gas
retrievals is still to be studied.
6. Improvements on GOME-2 visible spectral fitting
We have implemented the use of surface albedo EOFs
in combination with the MODIS BRDF climatology for fitting GOME-2 visible radiances in our GOME-2 ozone profile retrieval algorithm. The GOME-2 algorithm was adapted from our GOME and OMI ozone profile algorithms
[17,18] and was first described in Cai et al. [4] for ozone
profile retrievals from UV radiances. It has been modified
to perform ozone profile retrievals from joint UV (289–
307 nm in band 1, 315–340 nm in band 2) and visible
spectra. For visible, we use radiances in band 3 (540–
595 nm) and band 4 (602–646 nm, excluding the overlapping region which has degraded quality), which contains most of the visible spectral region that contributes to
ozone sensitivity. Prior to the use of surface EOFs, the
surface albedo spectrum is fitted as polynomials for both
UV and visible regions; this has been shown to perform
well for UV retrievals and is still used for fitting the UV
part. To take advantage of the visible surface reflectance
spectrum developed in this study, we use MODIS BRDF
parameters averaged over the GOME-2 footprint to derive
the blue-sky albedo at the four MODIS wavelengths for the
given observation geometry. The four EOFs in Fig. 1 and the
derived MODIS albedos are combined to initialize the
surface albedo spectrum based on Eq. (2). The three
leading EOFs matching MODIS values are currently further
fitted in the retrievals.
To show the improvements of GOME-2 visible spectral
fitting with the use of surface EOFs, we compare with
retrievals that fit visible surface albedo as a second-order
polynomial (also with 4 parameters). Fig. 7 shows such an
example for a GOME-2 pixel over the South-eastern United
States, which is mostly covered with forest and agricultural vegetation. We find that the fitting residuals are
significantly reduced from 0.83% to 0.14% with the use of
four surface EOFs. Generally, we found significant
improvement over vegetation scenes, but insignificant
difference for soil and desert scenes that do not have much
surface spectral structure in the selected spectral region.
The use of four EOFs for visible surface reflectance in
conjunction with high-resolution MODIS BRDF climatology
is an important step toward using visible to improve
retrieval sensitivity in the lower troposphere. Testing is
ongoing on how to best use the combination of surface
EOFs and satellite BRDF information. However, the above
fitting still requires further optimization to reduce the
systematic calibration differences among bands 2–4 as the
visible fitting residual of 0.14% is still much larger than the
random-noise of the spectra ( 0.05%).
7. Summary
We developed a global climatology of surface reflectance spectra at high spectral resolution in the visible. Our
goal was to produce this climatology for use in satellite
retrievals of ozone and other atmospheric constituents. To
accomplish this we used spectra from three different
reflectance databases to capture the range of different
surface reflectance spectra through an EOF analysis. We
then utilized the MODIS BRDF and GOME-2 LER satellite
products which inform the surface reflectance at discrete
wavelengths for each given scene. By fitting our high
spectral resolutions EOFs to the satellite-derived reflectances we generated a climatology of geometry dependent
surface reflectance spectra for all locations and days of the
year. We implemented the use of surface albedo EOFs in
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
Fig. 5. Continuous surface reflectance spectra from fitting land Empirical
Orthogonal Functions (EOFs) as well as snow/ice and water spectra to
GOME-2 LER (x's, Lambertian Equivalent Reflectance) over a snowy scene
in Canada (110°W, 60°N). The temporal evolution over three months from
snow-covered to snow-free is shown.
Fig. 6. Ocean surface reflectance spectra from modes of surface reflectance variability to GOME-2 Lambertian Equivalent Reflectance (LER). The
mean (black line) and middle 95% of the data (bounded by dotted red
lines) over the tropics and mid-latitude (60°S to 60°N) are shown. (For
interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)
45
combination with the MODIS BRDF climatology for fitting
GOME-2 visible radiances in our GOME-2 ozone profile
retrieval algorithm. This algorithm is the forerunner of
future ozone profile retrievals from the TEMPO mission,
which has measurement of near-surface ozone as a primary mission objective. The surface reflectance information can be provided to the retrieval algorithm either as
one surface spectrum or as BRDF parameters and surface
EOFs; testing is currently ongoing as to the best method
for multispectral ozone retrievals. These data are provided
for each MODIS 8-day period and adjusted for the solar
geometry for each retrieval.
An important component of this work is the inclusion
of a high-resolution vegetation spectra database. This,
along with limiting the wavelength window for the EOF
calculation and fitting to 400–900 nm leads to land spectra
that are as accurate as possible for the visible spectral
range relevant for TEMPO ozone and other trace gas
retrievals. The improvement shown over vegetated scenes
for GOME-2 spectral fitting over vegetation is promising as
the land surface in the conterminous US is primarily covered in crops and natural vegetation. The use of a surface
albedo climatology is intended to support the near-realtime processing planned for TEMPO operational products.
However, we can also exploit high time resolution satellite
reflectance data for specific time periods (as becomes
available) for TEMPO research products that are processed
on a less time-sensitive basis. Also, further optimization of
the ozone profile algorithm is needed since the fitting
residuals remain above acceptable limits even after this
quantification of the surface reflectance spectrum.
The TEMPO geostationary satellite mission represents a
transformative development for remote sensing of ozone
air pollution and related atmospheric species. Measuring
near-surface ozone has presented a significant technical
challenge but will yield great benefits in our understanding of tropospheric chemistry as well as our assessment and attribution of pollution events. The precise
quantification of the visible surface reflectance spectra
that we have performed has proved necessary to enable
these ozone measurements.
Acknowledgments
This work was supported by NASA Grant NNX13AR40G.
References
Fig. 7. Comparison of fitting residuals in the visible for a GOME-2 pixel
over the Southeastern United States (centered at 87.6°W, 33.5°N) on
1 July 2008. We compare fitting four surface albedo Empirical Orthogonal
Functions (EOFs, red line) and fitting a second-order polynomial (purple
line) for the surface albedo spectrum. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
[1] Bak J, Kim JH, Liu X, Chance K, Kim J. Evaluation of ozone profile and
tropospheric ozone retrievals from GEMS and OMI spectra. Atmos
Meas Tech 2013;6:239–49.
[2] Baldridge AM, Hook SJ, Grove CI, Rivera G. The ASTER spectral library
version 2.0. Remote Sens Environ 2009;113:711–5.
[3] Brion J, Chakir A, Charbonnier J, Daumont D, Parisse C, Malicet J.
Absorption spectra measurements for the ozone molecule in the
350–830 nm region. J Atmos Chem 1998;30:291–9.
[4] Cai Z, Liu Y, Liu X, Chance K, Nowlan CR, Lang R, et al. Characterization and correction of Global Ozone Monitoring Experiment
2 ultraviolet measurements and application to ozone profile retrievals. J Geophys Res—Atmos 2012;117:D07305.
46
P. Zoogman et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 180 (2016) 39–46
[5] Chance K, Liu X, Suleiman RM, Flittner DE, Al-Saadi J, Janz SJ. Tropospheric emissions: monitoring of pollution (TEMPO). Earth Observ
Syst 2013;Xviii, 8866:88660D.
[6] Chance K, Burrows J, Perner D, Schneider W. Satellite measurements
of atmospheric ozone profiles, including tropospheric ozone, from
ultraviolet/visible measurements in the nadir geometry: a potential
method to retrieve tropospheric ozone. J Quant Spectrosc Radiat
Transf 1997;57:467–76.
[7] Clark, RN, Swayze, GA, Wise, R, Livo, KE, Hoefen, TM, Kokaly, RF, et al.
USGS Digital Spectral Library splib06a; 2007.
[8] Dumont M, Brissaud O, Picard G, Schmitt B, Gallet J-, Arnaud Y.
High-accuracy measurements of snow Bidirectional Reflectance
Distribution Function at visible and NIR wavelengths – comparison
with modelling results. Atmos Chem Phys 2010;10:2507–20.
[9] Fishman J, Iraci LT, Al-Saadi J, Chance K, Chavez F, Chin M, et al. The
United States' next generation of atmospheric composition and
coastal ecosystem measurements NASA's Geostationary Coastal and
Air Pollution Events (GEO-CAPE) Mission. Bull Am Meteorol Soc
2012;93:1547.
[10] Fishman J, Bowman KW, Burrows JP, Richter A, Chance KV, Edwards
DP, et al. Remote sensing of tropospheric pollution from space. Bull
Am Meteorol Soc 2008;89:805–21.
[11] Gatebe C, King M, Lyapustin A, Arnold G, Redemann J. Airborne
spectral measurements of ocean directional reflectance. J Atmos Sci
2005;62:1072–92.
[12] Gatebe C, King M, Platnick S, Arnold G, Vermote E, Schmid B. Airborne spectral measurements of surface-atmosphere anisotropy for
several surfaces and ecosystems over southern Africa. J Geophys Res
—Atmos 2003;108:8489.
[13] Ingmann P, Veihelmann B, Langen J, Lamarre D, Stark H, CourregesLacoste GB. Requirements for the GMES Atmosphere Service and
ESA's implementation concept: Sentinels-4/ 5 and-5p. Remote
Sens Environ 2012;120:58–69.
[14] IPCC: Climate Change 2013: The Physical Science Basis. Contribution
of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change; 2013.
[15] Kim J. GEMS (Geostationary Enviroment Monitoring Spectrometer)
onboard the GeoKOMPSAT to monitor air quality in high temporal
and spatial resolution over Asia-Pacific region. EGU; 2012.
[16] Leroy M, Deuze J, Breon F, Hautecoeur O, Herman M, Buriez J, et al.
Retrieval of atmospheric properties and surface bidirectional
reflectances over land from POLDER/ADEOS. J Geophys Res—Atmos
1997;102:17023–37.
[17] Liu X, Bhartia PK, Chance K, Spurr RJD, Kurosu TP. Ozone profile
retrievals from the ozone monitoring instrument. Atmos Chem Phys
2010;10:2521–37.
[18] Liu X, Chance K, Sioris C, Spurr R, Kurosu T, Martin R, et al. Ozone
profile and tropospheric ozone retrievals from the global ozone
monitoring experiment: algorithm description and validation. J
Geophys Res—Atmos 2005;110:D20307.
[19] Lucht W, Schaaf C, Strahler A. An algorithm for the retrieval of
albedo from space using semiempirical BRDF models. IEEE Trans
Geosci Remote Sens 2000;38:977–98.
[20] Lyapustin A, Gatebe CK, Kahn R, Brandt R, Redemann J, Russell P,
et al. Analysis of snow bidirectional reflectance from ARCTAS
Spring-2008 Campaign. Atmos Chem Phys 2010;10:4359–75.
[21] Mahr, T, Peper, E, Warnach, S, Pohler, D, Beirle, S, Mies, K, et al.
Spectrally high-resolved satellite remote sensing of vegetation. In:
Poster presentation at the 2013 EUMETSAT meteorological satellite
conference and 19th American meteorological society AMS satellite
meteorology, oceanography, and climatology conference, Vienna;
16–20 September 2013.
[22] Martin RV. Satellite remote sensing of surface air quality. Atmospheric
Environment 2008. http://dx.doi.org/10.1016/j.atmosenv.2008.07.018.
[23] Martonchik J, Diner D, Kahn R, Ackerman T, Verstraete M, Pinty B,
et al. Techniques for the retrieval of aerosol properties over land
and ocean using multiangle imaging. IEEE Trans Geosci Remote Sens
1998;36:1212–27.
[24] McLinden CA, Fioletov V, Boersma KF, Kharol SK, Krotkov N, Lamsal
L, et al. Improved satellite retrievals of NO2 and SO2 over the
Canadian oil sands and comparisons with surface measurements.
Atmos Chem Phys 2014;14:3637–56. http://dx.doi.org/10.5194/
acp-14-3637-2014.
[25] Natraj V, Liu X, Kulawik S, Chance K, Chatfield R, Edwards DP, et al.
Multi-spectral sensitivity studies for the retrieval of tropospheric
and lowermost tropospheric ozone from simulated clear-sky GEOCAPE measurements. Atmos Environ 2011;45:7151–65.
[26] Noguchi K, Richter A, Rozanov V, Rozanov A, Burrows JP, Irie H, et al.
Effect of surface BRDF of various land cover types on geostationary
observations of tropospheric NO2. Atmos Meas Tech 2014;7:
3497–508.
[27] Russell AR, Perring AE, Valin LC, Bucsela EJ, Browne EC, Min K, et al.
A high spatial resolution retrieval of NO2 column densities from
OMI: method and evaluation. Atmos Chem Phys 2011;11:8543–54.
[28] Schaaf C, Gao F, Strahler A, Lucht W, Li X, Tsang T, et al. First
operational BRDF, albedo nadir reflectance products from MODIS.
Remote Sens Environ 2002;83:135–48.
[29] Schaaf CB, Liu J, Gao F, Strahler AH. In: Ramachandran B, Justice C,
Abrams M, editors. MODIS Albedo and Reflectance Anisotropy Products from Aqua and Terra, In Land Remote Sensing and Global
Environmental Change: NASA's Earth Observing System and the
Science of ASTER and MODIS, Remote Sensing and Digital Image
Processing Series, vol. 11. Springer-Cerlag; 2011. p. 873.
[30] Sellitto P, Di Noia A, Del Frate F, Burini A, Casadio S, Solimini D. On
the role of visible radiation in ozone profile retrieval from nadir UV/
VIS satellite measurements: an experiment with neural network
algorithms inverting SCIAMACHY data. J Quant Spectrosc Radiat
Transf 2012;113:1429–36.
[31] Spurr RJD. VLIDORT: a linearized pseudo-spherical vector discrete
ordinate radiative transfer code for forward model and retrieval
studies in multilayer multiple scattering media. J Quant Spectrosc
Radiat Transf 2006;102:316–42.
[32] Tilstra, LG, Tuinder, ONE, Stammes, P. GOME-2 surface LER product –
Algorithm Theoretical Basis Document; 2014.
[33] United States Environmental Protection Agency: Health Risk and
Exposure Assessment for Ozone; 2014.
[34] Varotsos CA, Melnikova IN, Cracknell AP, Tzanis C, Vasilyev AV. New
spectral functions of the near-ground albedo derived from aircraft
diffraction spectrometer observations. Atmos Chem Phys 2014;14:
6953–65.
[35] Vermote E, El Saleous N, Justice C, Kaufman Y, Privette J, Remer L,
et al. Atmospheric correction of visible to middle-infrared EOSMODIS data over land surfaces: background, operational algorithm
and validation. J Geophys Res—Atmos 1997;102:17131–41.
[36] Vidot J, Borbas E. Land surface VIS/NIR BRDF atlas for RTTOV-11:
model and validation against SEVIRI land SAF albedo product. Q J R
Meteorol Soc 2014;140:2186–96.
[37] Wang Z, Schaaf C, Strahler AH, Chopping MJ, Roman MO, Shuai Y,
et al. Evaluation of MODIS albedo product (MCD43A) over grassland, agriculture and forest surface types during dormant and snowcovered periods. Remote Sens Environ 2014;140:60–77.
[38] Zhou Y, Brunner D, Spurr RJD, Boersma KF, Sneep M, Popp C, et al.
Accounting for surface reflectance anisotropy in satellite retrievals
of tropospheric NO2. Atmos Meas Tech 2010;3:1185–203.
[39] Zoogman P, Jacob DJ, Chance K, Zhang L, Le Sager P, Fiore AM, et al.
Ozone air quality measurement requirements for a geostationary
satellite mission. Atmos Environ 2011;45:7143–50.