AMTAtmospheric Measurement TechniquesAMTAtmos. Meas. Tech.1867-8548Copernicus PublicationsGöttingen, Germany10.5194/amt-9-5423-2016The operational methane retrieval algorithm for TROPOMIHuHailih.hu@sron.nlHasekampOttoButzAndréhttps://orcid.org/0000-0003-0593-1608GalliAndréhttps://orcid.org/0000-0003-2425-3793LandgrafJochenAan de BrughJoostBorsdorffTobiashttps://orcid.org/0000-0002-4421-0187ScheepmakerRemcoAbenIlseSRON Netherlands Institute for Space Research, Utrecht, the NetherlandsDeutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Institut für Physik der Atmosphäre, Oberpfaffenhofen, GermanyPhysics Institute, University of Bern, Bern, SwitzerlandHaili Hu (h.hu@sron.nl)9November20169115423544031March201613April201624October201626October2016This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://amt.copernicus.org/articles/9/5423/2016/amt-9-5423-2016.htmlThe full text article is available as a PDF file from https://amt.copernicus.org/articles/9/5423/2016/amt-9-5423-2016.pdf
This work presents the operational methane retrieval
algorithm for the Sentinel 5 Precursor (S5P) satellite and its performance
tested on realistic ensembles of simulated measurements. The target product
is the column-averaged dry air volume mixing ratio of methane (XCH4), which will be retrieved simultaneously with scattering properties
of the atmosphere. The algorithm attempts to fit spectra observed by the
shortwave and near-infrared channels of the TROPOspheric Monitoring Instrument (TROPOMI) spectrometer aboard S5P.
The sensitivity of the retrieval performance to atmospheric scattering
properties, atmospheric input data and instrument calibration errors is
evaluated. In addition, we investigate the effect of inhomogeneous slit illumination
on the instrument spectral response function. Finally, we discuss the cloud
filters to be used operationally and as backup.
We show that the required accuracy and precision of < 1 % for the
XCH4 product are met for clear-sky measurements over land
surfaces and after appropriate filtering of difficult scenes. The algorithm
is very stable, having a convergence rate of 99 %. The forward model error is
less than 1 % for about 95 % of the valid retrievals. Model errors in the
input profile of water do not influence the retrieval outcome noticeably. The
methane product is expected to meet the requirements if errors in input
profiles of pressure and temperature remain below 0.3 % and 2 K,
respectively. We further find that, of all instrument calibration errors
investigated here, our retrievals are the most sensitive to an error in the
instrument spectral response function of the shortwave infrared channel.
Introduction
Methane (CH4) is the most important anthropogenic greenhouse gas after
carbon dioxide (CO2). While it occurs in smaller concentrations, it
has a higher global warming potential per molecule than CO2. An accurate
understanding of CH4 sources and sinks is essential for a
reliable prediction of climate change. Space-based measurements can provide
continuous and global monitoring of CH4, leading to much-needed
improved constraints on the surface fluxes.
Several current and future satellite missions measure CH4
abundances in the Earth's atmosphere. The SCanning Imaging Absorption
spectroMeter for Atmospheric CartograpHY (SCIAMACHY) aboard ENVISAT () was the first space-based instrument to measure
atmospheric CH4 with sensitivity down to the Earth's surface.
were the first to use these measurements to constrain
CH4 surface fluxes. After loss of contact with ENVISAT in 2012,
the Greenhouse gases Observing SATellite (GOSAT) is currently the only
satellite measuring atmospheric CH4. While GOSAT
has a higher sensitivity and spatial resolution than SCIAMACHY, it has a
fairly low spatial sampling. In 2017, the TROPOspheric Monitoring Instrument
(TROPOMI) will be launched aboard the Sentinel 5 Precursor (S5P) satellite,
and it will provide CH4 measurements as one of its key products
with unprecedented high precision, spatial resolution and global daily
coverage.
The common goal of the above-mentioned missions is to provide atmospheric
CH4 concentrations with sufficient accuracy and spatiotemporal
coverage to allow the assessment of CH4 sources through inverse
modelling. The observation strategy relies on measuring spectra of sunlight,
backscattered by the Earth's surface and atmosphere, in the shortwave
infrared (SWIR) spectral range. Absorption features of CH4
molecules allow for retrieval of its atmospheric concentration with high
sensitivity down to the Earth's surface where the main CH4
sources are located. The applicability of such measurements for estimating
source strengths, however, strongly depends on the precision and accuracy
achieved. Residual systematic biases must be well below 1 % to facilitate
inverse modelling .
Scattering by aerosols and cirrus is one of the major challenges for
retrievals of CH4 from space-based SWIR observations. While
contamination by optically thick clouds can be filtered out reliably,
optically thin scatterers are much harder to detect, yet they still modify the
light path of the observed backscattered sunlight. This can lead to
underestimation or overestimation of the true CH4 column if not
appropriately accounted for. The net light path effect strongly depends on
the amount, size and height distribution of the scatterers as well as on the
reflectance of the underlying surface .
Therefore, retrieval strategies rely on inferring the target gas
concentration either simultaneously with atmospheric scattering properties or
with a light path proxy.
introduced the “proxy” approach for CH4
retrieval from SCIAMACHY measurements around 1600 nm, by using the
simultaneously retrieved CO2 column as a light path proxy. The
proxy approach relies on the assumptions that scattering effects cancel in
the ratio of the CH4 column and the CO2 column, and
that a prior estimate of the CO2 column is sufficiently accurate to
recalculate the CH4 column from the measured CH4/ CO2
ratio. In this case scattering is ignored in the forward modelling. Further
applications of the proxy approach for CH4 retrieval from
SCIAMACHY are described by and
. For GOSAT, the proxy approach has been successfully
applied by and .
Alternatively, scattering-induced light path modification can be taken into
account by simultaneously inferring the atmospheric CH4
concentration and physical scattering properties of the atmosphere. Such
“physics-based” methods have been developed for space-based CO2
and CH4 measurements from SCIAMACHY, GOSAT and the Orbiting
Carbon Observatory (OCO); see e.g. .
The physics-based methods make use of the oxygen (O2) A-band in the
near-infrared (NIR) around 760 nm and absorption bands of the target absorber
in the SWIR spectral range. The advantage of physics-based methods for
CH4 retrieval compared to proxy methods is that they do not
depend on prior information on the CO2 column. On the other hand,
the physics-based algorithms are more complex and computationally expensive.
In addition, they may be limited by the information content of the measurement with
respect to aerosol properties and related forward model errors in the
description of aerosols. A detailed comparison between the two methods for
GOSAT is provided by .
TROPOMI has four spectral channels in the ultraviolet (UV), visible (VIS),
near-infrared (NIR) and shortwave infrared (SWIR), with spectral ranges of
270–320, 310–495, 675–775 and 2305–2385 nm, respectively. We use
TROPOMI measurements in the NIR and SWIR for CH4 retrievals. This
spectral range does not allow for a light-path-proxy approach, and thus the
effect of aerosols and cirrus needs to be accounted for by using a physics-based
method as described above. The goal of this paper is to present the
CH4 retrieval algorithm for TROPOMI and investigate its
sensitivity to algorithm assumptions, atmospheric input data and instrument
calibration errors and filtering criteria. To this end, we simulated
realistic TROPOMI measurements for aerosol- and cirrus-loaded atmospheres
under clear-sky and cloudy conditions.
The outline of this paper is as follows. We start with the methodology in
Sect. , giving an overview of the instrument, the
retrieval algorithm and the methane data product. Then, a detailed error
sensitivity study is presented in Sect. , based on
methane retrievals on a clear-sky global ensemble of simulated spectra. In
Sect. , a study of cloud filtering is performed on a
TROPOMI orbit of simulated spectra that cover a realistic range of cloud
parameters. Conclusions are presented in Sect. .
Methodology
The S5P satellite has a designed 7-year lifetime and will fly in a
sun-synchronous orbit at 824 km altitude. Its single payload, TROPOMI, is a
push-broom imaging spectrometer with a wide swath of 2600 km and a ground
pixel of 7 × 7 km2 in exact nadir. Approximately, TROPOMI observes a
full swath per second, which results in ∼ 216 spectra per second. The
instrument comprises two spectrometer modules, the first containing the UV,
VIS and NIR spectral channels, and the second dedicated to the SWIR channel.
We use the NIR and SWIR channels with spectral resolutions of 0.38 and 0.25
nm and spectral sampling ratios of 2.8 and 2.5, respectively
. Since the NIR and SWIR detectors are incorporated in
different instrument modules, the NIR spectra will be co-registered with the
SWIR spectra before performing CH4 retrievals. Examples of
simulated TROPOMI NIR and SWIR spectra are shown in Fig. .
Simulated TROPOMI spectra in the NIR and SWIR channels for a
scenario from the global ensemble; see Sect. . The
scenario, located in a rural area in Utah (USA), is observed at nadir and
SZA =50∘.
The retrieval algorithm
The S5P operational CH4 retrieval algorithm is based on RemoTeC,
which was orginally developed for CO2 and CH4
retrievals from OCO and GOSAT obervations . A first
performance study on TROPOMI measurements has been done by ,
where a detailed description of the algorithm can be found. Here, we
summarise the essentials and elaborate on the modifications made since then.
The forward model
The algorithm aims to infer the atmospheric state vector x from
spectral measurements y in the NIR (757–774 nm) and SWIR
(2305–2385 nm) ranges. This requires a forward model F that can
accurately compute the measurement given the atmospheric
state:
y=F(x)+ey+eF,
where ey is the measurement noise error, and eF
is the forward model error. The forward model incorporates the linearised
radiative transfer model LINTRAN . LINTRAN simulates the
radiance at the top of the atmosphere ITOA on a fine internal
spectral grid. The model needs a solar irradiance spectrum on the
internal spectral grid as input, which is inferred from the daily solar measurements
of TROPOMI using the deconvolution approach by and
. The simulated radiance measurement is obtained by a
spectral convolution of ITOA with the instrument spectral
response function (ISRF).
In addition to accuracy, an important requirement of the algorithm is
computational speed. To be able to process TROPOMI's huge number of
measurements, the CPU time per retrieval has to be on the order of seconds.
Our algorithm achieves this by, among other things, using the linear k-method
for multiple scattering . Single scattering is calculated
line by line as it is computationally less expensive. To our knowledge, our
algorithm is among the fastest in use for CH4 full-physics
retrievals, with an average CPU time of 7–10 s per retrieval
.
For a given model atmosphere, the forward model simulates spectra of
backscattered sunlight, taking absorption and scattering by
molecules and particles into account. The model atmosphere is defined for 36
pressure-equidistant vertical layers, with the top at 0.1 hPa and bottom at
the surface pressure psurf. We calculate psurf by
interpolating the meteorological surface pressure, from the European Centre
for Medium-Range Weather Forecasts (ECMWF), on the surface elevation, from a
digital elevation map (). The absorbing trace
gases of interest are O2 in the NIR band, and CH4, H2O and CO in
the SWIR band. The first guess layer subcolumns of these gases are calculated
from input profiles of CH4 and CO (from the global chemical
transport model TM5; ), and temperature, humidity and
pressure profiles (from ECMWF forecast data). Molecular absorption features
are calculated using appropriate spectroscopic databases
. Here, we evaluate the
absorption cross sections on a 72-layer pressure-equidistant grid to account
for the strong temperature dependence and pressure dependence of the cross sections.
Molecular scattering properties are given by Rayleigh theory. Particulate
absorption and scattering are computed with Mie theory using tabulated
aerosol properties by .
In our algorithm, the aerosol type is characterised by the refractive index
and size distribution. The complex refractive index is fixed at n=1.4-0.01i
in the O2 A-band and n=1.47-0.008i in the SWIR band, which are averaged
values derived from the ECHAM5-HAM model . The size
distribution is described by a power-law function with size parameter
α (e.g. ):
n(r)=A,ifr≤r1.A(r/r1)-α,ifr1<r≤r2.0,ifr>r2,
where r1=0.1µm, r2=10µm, r is the particle
radius and A is a normalisation constant. The amount of aerosol and its
vertical distribution is provided by the vertically integrated column number
density Naer and a normalised Gaussian height distribution
h(zk) with zk the height of layer k:
h(zk)=Bexp-4ln2(zk-zaer)2w2,
where w=2000 m, zaer is the central height and B is a
normalisation constant. Thus, in layer k with thickness Δzk, the
layer subcolumn nk is given by
nk=Naerh(zk)Δzk. suggest that the exact choice of the fixed-value parameters,
such as the refractive indices or the width of the height distribution, does
not affect the retrievals significantly. To support this claim, we have
studied the effect of different values for the fixed-value aerosol parameters
on our retrievals, and we arrive at the same conclusion; see
Appendix .
We have modified the retrieval forward model with respect to
to account for chlorophyll fluorescence emission in a simplified manner as
described by . In short, scattering of the
fluorescence emission is ignored, and solely the spectral shape and absorption
features by oxygen (O2) are modelled. This allows fluorescence to be
treated as a simple additive term to the radiance before convolution with the
ISRF. The surface emission at the top of the atmosphere (TOA) is then
modelled as
Fs(λ)TOA=Fs,755surf(1-s(λ-755))e-τO2(λ)/μ,
where τO2(λ) is the vertical optical thickness of
oxygen, μ is the cosine of the viewing zenith angle and λ is the
wavelength in nanometres. Fs,755surf and s represent
the chlorophyll emission at 755 nm and its spectral slope, respectively.
The state vector
The state vector x consists of 25 elements; see
Table . The target is a 12-layer pressure-equidistant
vertical profile of CH4 partial column number densities. The 12
retrieval layers are related to 36-layer model atmosphere by the shared
interfaces; i.e. each retrieval layer is divided in three sublayers for the
forward model calculations. Retrieved ancillary parameters are the total
column number densities of H2O and CO, three scattering
parameters Naer, α and zaer (related to
amount, size and height), the surface albedo (constant-term and first-order
spectral dependence
Note that the algorithm has the option to fit
higher-order spectral dependence of the surface albedo. It will be
investigated whether this is needed with real data.
and in the NIR and SWIR band),
two terms to account for a spectral shift of the measurement (NIR and SWIR
band) and two terms to account for chlorophyll fluorescence emission,
Fs,755surf and s (see Eq. ). An
overview of the state vector elements and their a priori values is given in
Table . We note that for the operational methane
retrievals, the a priori value for Fs,755surf will come
from a dedicated fluorescence retrieval scheme based on the Fraunhofer lines.
In this study however, we simply took an a priori
Fs,755surf=0 photons s-1 m-2 nm-1.
Further, we would like to mention that our retrieved CO column should be regarded
as a diagnostic for the CH4 retrieval. The official TROPOMI CO
product has a dedicated retrieval algorithm described by
.
State vector elements and their a priori values or references to
source.
State vector elementA priori value (units)CH4 subcolumns in 12 vertical layersTM5 (cm-2)CO total columnTM5 (cm-2)H2O total columnderived from ECMWF specific humidity (cm-2)Aerosol column Naerderived from aerosol optical thickness = 0.1 at 760 nm (cm-2)Aerosol size parameter α3.5 (–)Aerosol height parameter zaer5000 (m)Lambertian surface albedo in the NIR bandmaximum of measured reflectance in the NIR band (–)First-order spectral dependence surface albedo in the NIR band0 (–)Lambertian surface albedo in the SWIR bandmaximum of measured reflectance in the SWIR band (–)First-order spectral dependence surface albedo in the SWIR band0 (–)Spectral shift NIR0 (nm)Spectral shift SWIR0 (nm)Fluorescence emission at 755 nm Fs,755surf0 (photons s-1 m-2 nm-1)Fluorescence spectral slope s0 (–)The inversion procedure
The state vector is found by inverting Eq. (), where the
inverse method is based on a Philips–Tikhonov regularisation scheme
. Regularisation is required because the
inverse problem is ill-posed; i.e. the measurements y typically
contain insufficient information to retrieve all state vector elements
independently. Philips–Tikhonov regularisation aims to reduce
contributions from measurement noise to the retrieved state vector while
retaining valuable information. Because the forward model F(x)
is non-linear in x, the inversion is performed iteratively by a
step-size-controlled Gauss–Newton scheme, where at each iteration step the
forward model is linearised.
The inverse algorithm finds x by minimising the cost function that is
the sum of the least-squares cost function and a side constraint weighted by
the regularisation parameter γ according to
x^=minx‖Sy-1/2F(x)-y||2+γ‖W(x-xa)||2,
where Sy is the diagonal measurement error covariance matrix,
which contains the noise estimate. xa is an a priori state
vector, and W is a diagonal weighting matrix that renders the side
constraint dimensionless and ensures that only the CH4 parameters
and the scattering parameters contribute to its norm: Wjj=1/xa,j for the CH4 column number densities and
the three aerosol parameters, and Wjj=10-7/xa,j for all other state vector elements. The
latter are thus retrieved in a least-squares sense. For determining γ,
the L-curve criterion is applied in the baseline
algorithm. However, γ may be fine-tuned later using real observations.
The CH4 data product
Although we retrieve methane in n=12 sublayers, there is virtually no
profile information in the measurement. The degree of freedom of signal of
the retrieved methane profile is about 1. Therefore, the methane data product
is given as a column-averaged dry air mixing ratio XCH4. This quantity is
obtained from the methane entries of the retrieved state vector
xi
through
XCH4=∑i=1nxi/Vair,dry,
where Vair,dry is the dry air column (calculated from
meteorological input surface pressure and the water vapour profile).
To interpret the retrieved XCH4 correctly, one also needs the
column averaging kernel Acol that describes the
sensitivity of the retrieved CH4 column to changes to the true
methane profile (see , for details):
Acol,i=∂∑i=1nxi∂xtrue,i.
As an example, in Fig. we show the column averaging
kernel corresponding to the methane retrieval performed on the simulated
spectra of Fig. . We see that the column averaging kernel
is around 1 in the lower atmosphere. From Eq. () it is clear
that the closer Acol is to 1, the more the retrieved
column represents the true column (see also Eq. ). This
illustrates that the retrieval of methane columns from the SWIR has a nearly
ideal sensitivity to methane in the troposphere and the tropospheric boundary
layer.
For validation and interpretation purposes it is important to realise that
the retrieved XCH4 is related to the true methane profile
xtrue and the a priori profile xa as
XCH4=∑i=1nAcol,ixtrue,i+(1-Acol,i)xa,i/Vair,dry+ΔXCH4,F+ΔXCH4,y,
where ΔXCH4,F is the bias caused by forward model
errors, and ΔXCH4,y is the retrieval noise due to
measurement noise. The standard deviation of the retrieval noise, i.e. the
precision σXCH4, follows from the error covariance matrix
Sx, which describes the effect of measurement noise on the
retrieved parameters (see e.g. , Sect. 4.3):
σXCH4=∑i=1n∑j=1nSx,i,jVair,dry.
Together with XCH4 and Acol, the precision
σXCH4 is given in the main data product. Note that for
simulations, the bias ΔXCH4,F can be calculated
from Eq. (), since all other terms are either known from the
simulations or calculated by the retrieval algorithm. In
Sect. , we evaluate ΔXCH4,F, ΔXCH4,y and their
sensitivity to input errors.
Column averaging kernel for a typical methane retrieval. The
retrieval was performed using the simulated NIR and SWIR spectra in
Fig. .
Data filtering
Our algorithm has been designed to be efficient and accurate under certain
assumptions; e.g. the atmosphere is cloud-free and plane-parallel with
optically thin scatterers. Therefore, it is essential to filter out those
scenes for which these assumptions break down to ensure data quality and
reach the required accuracy. Moreover, if filtering can be done a priori, for
example using cloud detection, we can considerably reduce the number of
performed retrievals and hence computation time.
A priori filtering
For operational cloud filtering, measurements from the Visible Infrared
Imaging Radiometer Suite (VIIRS) aboard the Suomi-NPP satellite will be used.
S5P is foreseen to operate in loose formation with Suomi-NPP, meaning that
both missions observe the same ground scene with a time delay of about
5 min. In case VIIRS data are not available, we have developed a backup cloud
filter using H2O retrievals from the weak and strong absorption band,
assuming a non-scattering atmosphere. Here we make use of the fact that
clouds and aerosols will modify the optical path length in the two bands
differently due to the different absorption strengths
. Thus, there will be a difference between
the H2O column retrieved from the strong band and the H2O column
retrieved from the weak band in the presence of clouds.
Based on tests with simulations, we chose 2329–2334 nm as the weak
absorption band and 2367–2377 nm as the strong absorption band (see also
). In Fig. , we show the
absorption features of H2O in the SWIR range and highlighted the proposed
weak and strong absorption band. Because the weak-band and the strong-band
retrievals are differently affected by scattering, the ratio
H2Oweak-H2OstrongH2Ostrong is strongly
correlated with cloud contamination. Scenes in which this ratio exceeds a
certain threshold can be flagged as cloudy and filtered out. Based on a
realistic ensemble of synthetic measurements (see
Sect. ), we find that a threshold of 0.08 filters out
most cloudy scenes and keeps most of the clear-sky scenes. However, a direct
cloud filter based on VIIRS data is shown to be superior to this indirect
two-band cloud approach. Therefore, the two-band cloud filter will only be
used as backup when VIIRS data are unavailable.
Simulated reflectance spectrum showing the absorption features of
H2O in the SWIR spectral range for the same scenario as in
Fig. . The red shaded area represents the weak absorption
band, while the blue area represents the strong absorption band of H2O.
The difference between the water column retrieved from the weak and strong
band is used for the backup cloud filter.
Further, before performing any CH4 retrievals, we filter out
cases with solar zenith angle (SZA) larger than 70∘, and viewing
zenith angle (VZA) larger than 50∘. These thresholds have been
derived from simulations and shall be fine-tuned after launch using real
observations.
A posteriori filtering
identified an a posteriori filter based on retrieved
scattering parameters: SWIR aerosol optical thickness τswir at
2350 nm, size parameter α and height parameter zaer.
Following this work, we filter out retrievals with
τswir⋅zaerα>120m.
This indicates that the retrieval algorithm has difficulties with scenes that
have optically thick scattering layers of large particles at high altitude.
Additionally, we filter out cases with retrieved albedo in the SWIR band smaller than
0.02. More a posteriori filters (e.g. based on goodness of fit) will be
determined after launch using real observations.
Sensitivity studies
Here, we evaluate the sensitivity of the retrieved XCH4 forward
model error to instrument errors and auxiliary input to the retrieval
algorithm (e.g. meteorological data). To put the errors in perspective, the S5P
product accuracy of XCH4 was envisioned to be within 2 % and the
product precision to be within 0.6 % . Accuracy is
defined as the mean deviation from the truth, and precision is the variation
due to random processes such as instrument noise. More recently, the
requirements have been slightly reformulated as 1 % bias and 1 % precision
. From the 1 % bias, 0.8 % is reserved for forward
model errors and 0.6 % for instrument-related errors.
Synthetic measurements: the global ensemble
We performed detailed sensitivity studies of the CH4 algorithm on
a global ensemble of simulated spectra consisting of land-only, clear-sky
scenes on a 2.79∘×2.8125∘
latitude × longitude grid. This ensemble is, to a large extent,
identical to the one used by . It contains realistic aerosol-
and cirrus-loaded scenes for 4 days, one per season. The treatment of
aerosols and cirrus in the simulations is far more complex than in the
retrieval forward model, where only one effective aerosol type is considered;
see Sect. . For the simulations, the aerosol physical
properties and vertical distributions are derived from the global aerosol
model ECHAM5-HAM for five different chemical species and on
a superposition of seven log-normal size distributions. The aerosol optical
thickness is derived from MODIS observations . Furthermore,
the simulations contains cirrus with optical thickness and vertical
distribution based on CALIOP measurements . Finally, the
surface albedo in the NIR is taken from the MODIS land albedo product in the
841–876 nm channel. For the albedo in the SWIR, the SCIAMACHY surface
albedo product at 2350 nm is used . For simplicity, we
assume a constant surface albedo within the NIR and SWIR band in our
simulations. We expect no difficulties to fit a more realistic spectral
dependent surface albedo based on our experience with real GOSAT data. We
refer to Figs. 2 and 3 of for the geographical distribution
of the total optical thickness and surface albedo, respectively, used in the
simulations. The measurements are simulated for the nadir viewing direction
and a solar zenith angle that is representative of TROPOMI with an overpass
time of 13:30 local time. We have 8633 simulated measurements in total.
While only investigated the scattering-induced error, we
increased the inconsistency between the simulation forward model and the
retrieval forward model model. We have attempted to include the most
important contributions to the forward model error, except for errors due to
the underlying spectroscopic database, which have been investigated
elsewhere;
see and . The simulations were computed
using a line-by-line radiative transfer model, whereas the retrieval method
uses the linear k-method. In addition, the simulations have a higher vertical and
spectral resolution than used in the retrieval. Furthermore, we have added
chlorophyll fluorescence emission. In the simulations, fluorescence is modelled
to have a double Gaussian spectral shape , which is
different from the linear spectral shape assumed in the retrieval forward
model (see Eq. ). As in the retrieval scheme, we neglect
scattering of fluorescence emission for simplicity. The fluorescence at the
TOA then becomes
Fs(λ)=Fs,755∑i=1,2Aie-(λ-λi)2σi2e-τO2(λ)/μ.
For the parameters A1, A2, λ1, λ2, σ1 and
σ2, we use the same values as in . We only
included fluorescence emission for scenes in the global ensemble with
albedoNIR/ albedoSWIR>5 as a rough selection
criterion for regions with vegetation.
After convolving the simulated TOA radiance with the ISRF, the spectra are
superimposed with instrument noise from the TROPOMI noise model
. For the NIR, the noise consists solely of shot noise, while
for the SWIR, the noise is composed of both shot noise and a
signal-independent term. The corresponding continuum signal-to-noise ratios
are 500 in the NIR and 100 in the SWIR for a reference scene with surface
albedo As=0.05, viewing zenith angle VZA=0∘ and
solar zenith angle SZA=70∘.
The baseline performance of the operational CH4 algorithm is
tested on the simulated global ensemble described above. In
Fig. , we show a world map of the bias ΔXCH4,F before and after applying the a posteriori
filters based on retrieved scattering parameters and albedo. Note that the
error due to measurement noise ΔXCH4,y has been
subtracted from the total XCH4 error. This is a random error and
is evaluated separately in Sect. In Fig. , the
cumulative probability distribution of the absolute XCH4 retrieval
error is shown (blue line). We get a convergence rate of 99 %, and 53 % of
the converged retrievals pass the filters. Finally, 94 % of the valid
retrievals have an absolute error < 1 %. In Fig. , we also
plotted the cumulative probability distribution in case fluorescence emission
is not fitted (red line). Then, we have a convergence rate of 95 and 92 %
of the valid retrievals have an error < 1 %. Thus our retrieval
results are improved by fitting fluorescence, which is why we have included
this in the baseline.
We note that most of the rejected retrievals are found in the desert and dust
regions in northern Africa and central Asia in April and July where our
simulations capture dust storm events. For real TROPOMI measurements though,
we will less likely filter out large regions because TROPOMI has higher
spatial resolution than our simulations. Furthermore, our simulations only
represent 1 day per season with a mean atmospheric state. Therefore our
results merely indicate that certain regions will have fewer days with
successful retrievals in certain seasons, not that entire regions will
be rejected.
Residual forward model XCH4 errors for the baseline retrieval
method for the clear-sky global TROPOMI measurement ensemble
before (a) and after (b) a posteriori filtering. The
ensemble covers scenes for a day in January (JAN), April (APR), July (JUL)
and October (OCT) as described in
Sect. .
Cumulative probability distribution of the absolute XCH4
forward model error for the baseline retrieval (blue line) and for the case
that fluorescence emission is not fitted (red line).
Sensitivity to atmospheric input data
We investigated the effect of imprecise atmospheric input from TM5 and ECMWF
on the CH4 retrievals. The results are summarised in
Fig. .
A priori CH4 profile
For the baseline performance test in Sect. we used
the true profile of methane xtrue as the a priori profile for the
retrievals. The bias ΔXCH4,F as defined in
Eq. () should not depend on the choice of the a priori
profile because this effect is accounted for by the averaging kernel. To test
this, we take the zonal mean as the a priori profile, where we averaged over all
longitudes within a 2.79∘ latitude bin, with a column deviation up
to ±2 %. To illustrate the effect on the global ensemble, we evaluate
the root mean square (rms) of the XCH4 bias (i.e. the total
XCH4 error minus the contribution due to noise) of all retrievals
that pass the a posteriori filters. Figure (first panel,
blue line) shows that the a priori CH4 does not influence the
retrieval accuracy, in terms of the rms of the XCH4 bias, nor the
stability, in terms of the convergence rate or number of valid retrievals.
A priori H2O profile
To investigate the sensitivity to errors on the assumed H2O profile, we
follow the same procedure as in Sect. . The error on the
prior H2O profile is established in the same way as for CH4, i.e. by
taking a normalised zonal mean profile per latitude bin. Note that for
H2O, there is an additional (minor) influence on the retrieval of
XCH4 through the dry air column. The H2O column error is
varied up to ±10 %. Figure (left panel, red line)
shows that the prior H2O profile has negligible influence on the rms of
the XCH4 bias, increasing it with < 0.01 %. There is a
small effect on the convergence rate, reducing it from 99 to 97 % and
leading to a reduction of valid retrievals from 53 to 50 %. We note that
taking a zonal mean profile for H2O represents a worst case in terms of
accuracy for the specific humidity of ECMWF.
Pressure
An erroneous pressure affects the retrieval of XCH4 in two ways:
first of all, through the pressure dependence of the cross sections and,
secondly, through the retrieved air column that is used to convert the
CH4 total column to the dry air mixing ratio, XCH4. The latter
will introduce a retrieval error of the same magnitude as the pressure error.
To evaluate the net effect of a pressure error, the prior pressure profile is
perturbed with a scaling factor up to ±0.3 %, corresponding to
|ΔPsurf|≈ 3 hPa. We expect a better accuracy from
the ECMWF surface pressure together with the digital elevation map
. Figure
(right panel, blue line) shows that the increase in the rms of the
XCH4 bias is < 0.15 %. There is no visible effect on the
stability of the algorithm.
Temperature
An error in the temperature will propagate to the XCH4 retrievals
though the temperature dependence of the cross sections. To investigate this
effect, the temperature profile is offset up to ±2 K.
Figure (right panel, blue line) shows that the increase
in the rms of the XCH4 bias is < 0.15 %. There is a small effect on
the stability of the algorithm, reducing the convergence rate to 97 % and
the number of valid retrievals to 47 %.
Influence of errors in atmospheric input on the accuracy (panels 1
and 2) and stability (panels 3 and 4) of XCH4 retrievals. The
upper and lower x axes refer to perturbations. Accuracy refers to the
XCH4 bias, and stability refers to the fraction of converged and
valid retrievals. The profiles of methane and water have been perturbed in
panels 1 and 3; the pressure and temperature profiles have been perturbed in
panels 2 and 4.
Sensitivity to instrument errors
We investigated the effect of different possible instrument and calibration
errors on the CH4 retrievals. The results are summarised in
Table . Below we discuss each effect separately.
Signal-to-noise ratio
The simulated spectra include instrument noise as described in
Sect. . The precision is given by the standard
deviation of the retrieval noise σXCH4; see
Eq. (). The world map in Fig. shows the
precision relative to the retrieved XCH4. Typically the precision is
better than the accuracy. The signal-to-noise ratio only becomes a limiting
factor for scenarios with snow-covered ground and large SZA, which is why we
filter for SZA < 70∘ and albedo > 0.02 to keep this error
relatively small.
Relative precision of XCH4 due to the instrument noise
for the a posteriori filtered data set for the clear-sky ensemble in
Fig. .
Instrument spectral response function
The synthetic measurements were created by convolving the underlying
line-by-line spectra with a Gaussian ISRF with a full width half maximum
(FWHM) of 0.38 nm in the NIR band and 0.25 nm in the SWIR band. The mission
requirements state that the ISRF shall be known within 1 % of its maximum
. This is achieved approximately by varying the FWHM by
1 %. Table give the results for retrievals
with an assumed error in the FWHM. We find that the retrievals are mostly
sensitive to the accuracy of the ISRF in the SWIR band, leading to an
increase of the rms XCH4 bias from 0.56 % (baseline) to
0.67 %. We note that of all the instrument errors investigated here, the
ISRF gives the largest error contribution to the methane retrievals. Thus, our
study indicates that accurate calibration of the ISRF should have high
priority.
Spectral calibration
According to the instrument requirements, the centre wavelengths of spectral
channels are known within 2 pm (picometres). For our global ensemble, a spectral
shift of 2 pm has negligible effect on the error characteristics. Here, we
evaluate the effect of a wavelength shift of 1/20 of the spectral sampling
distance, i.e. 10 and 5 pm for the NIR and SWIR band, respectively. The
reference retrieval fits a spectral shift. To test this fitting option, the
synthetic spectra were shifted with a constant wavelength:
λk,meas=λk+Δλ,
where λk is the real wavelength, λk,meas is the
measured wavelength at pixel k and Δλ the spectral shift. In
Table , results are given for the case that a
spectral shift is not fitted and is fitted (between brackets). When fitted,
the performance is as good as for the reference retrieval, i.e. simulations
without an error in the spectral position. This is as expected and indicates
that the spectral shift fitting is robust.
Optionally a wavelength-dependent shift, Δλsqueeze,
can be fitted. To test this fitting option, the synthetic wavelength grid was
“squeezed”:
λk,meas=λk+Δλsqueeze×(λk-λmid)/(λend-λmid),
where λend and λmid are the wavelengths
at the end and middle of the band, respectively.
Table shows the performance of retrievals with
this assumed error on the measured wavelength grid. Since a spectral squeeze
is a second-order effect and has a much smaller impact on the
XCH4 retrievals than a spectral shift, it is not fitted in the
baseline. However, our results show that, if needed, the option to fit a
spectral squeeze can be used reliably.
Effect of instrument calibration errors on convergence rate,
fraction of valid retrievals after filtering and rms values of
XCH4 bias and precision for the global ensemble. Note that all
sensitivities include the baseline forward model error, caused mainly by
aerosol and cirrus scattering. The terms between brackets are for the cases
where the relevant quantity is also retrieved. For each instrument
calibration error, multiple simulation runs were performed with all
combinations of errors in NIR and SWIR channels. The results shown here
correspond to the runs with poorest performance in terms of the rms
error.
Methane bias due to heterogenous slit illumination for spatially
varying surface reflection over a marsh scene at Siberia close to the river
Ob at latitude 62.8∘ N and longitude 72.1∘ E. Measurement
simulations are performed with the instrument model by
for an instantaneous field of view of 3.4 km across the slit and 7.0 km
along the slit.
Radiometric offset: additive factor
The effect of an unknown systematic offset in the Earth radiance is
investigated. The offsets in the NIR and SWIR bands are independently varied
with ±0.1 % of the continuum. Table
shows the effect of a radiometric offset to the XCH4 retrievals.
We note that a radiometric offset in the SWIR band causes a larger
XCH4 error than an offset in the NIR band. The latter is partly
compensated by the retrieved fluorescence.
Radiometric gain: multiplicative factor
The absolute radiometric accuracy of the measurement of the Earth spectral
radiance shall be better than 2 % according to the system requirements.
To investigate the effect of such an error, the synthetic spectra were
multiplied by a scaling factor G. Table
shows that there is negligible effect of an error of 2 % in radiometric
gain. This error is largely compensated by the retrieved surface albedo. It
follows that interaction between surface albedo and aerosols has a negligible
impact for gain errors < 2 %.
Heterogenous slit illumination
For the two-dimensional TROPOMI push-broom spectrometer, light in across-slit
dimension is dispersed by the instrument grating in order to spectrally
resolve the received signal. The along-slit direction is aligned across
flight direction to achieve the desired spatial resolution. For a homogeneous
illumination of the instrument slit, the spectral instrument response is
characterised extensively during the pre-flight calibration of the TROPOMI
instrument, and it is used as baseline to simulate the TROPOMI radiometric
measurement in our retrieval. In space, however, the instrument slit will be
illuminated inhomogeneously due to ground scene heterogeneities on scales
smaller than the instrument's field of view. Inhomogeneous illumination
across the slit leads to a distortion of the ISRF as described by
, and , and it can
affect the retrieval accuracy of the TROPOMI methane product. Small-scale
heterogeneities of the ground scene are generally caused by spatial
variations of surface reflection and by broken clouds. Because of our strict
cloud filtering, spatial variations in surface reflection are the only cause
of methane retrieval biases due to inhomogeneous slit illumination. To
evaluate the effect of surface scene heterogeneity on our XCH4
product, we employ the instrument model described by
for both the NIR and SWIR bands of our retrieval. Furthermore, we use a high
spatial resolution MODIS albedo map for the 50 × 50 km2 marsh
region in central Siberia with structures in the surface reflection due to
ponds, shown in Fig. . Depending on the scene
heterogeneity in the flight direction, the XCH4 error shows an
oscillation structure with a maximum amplitude ≤ 0.4 %, a standard
deviation of 0.12 % and a mean error of -0.01 %. For a particular
temporal and spatial sampling of the scene, a pseudo-random scatter is
introduced to the XCH4 product. This means that overall the
effect can be considered small. One may consider this error as a limitation
when interpreting very localised sources in surroundings of heterogenous
surface reflection, but for most applications some averaging either in time
or space will be done, which reduces this error.
Cloud filtering
The global ensemble from as used in
Sect. cannot be used to evaluate the cloud filters,
because it consists purely of cloud-free scenes. Therefore, the cloud filters
are tested using a new ensemble of synthetic measurements, which also include
scenes with water clouds.
Synthetic measurements: the TROPOMI test orbit
Simulations of a TROPOMI orbit. Panel (a) shows MODIS cloud
fraction resampled on the orbit's ground pixels. Panel (b) gives the
XCH4 bias of all processed pixels that converged.
Panel (c) and (d) give XCH4 bias of the valid
retrievals after cloud filtering with MODIS data and the backup cloud filter,
respectively. Here, we also applied the a posteriori filter based on
retrieved scattering parameters and albedo.
Error statistics of XCH4 retrievals from the L1B orbit
using different cloud filters in comparison to the global clear-sky orbit.
The performance of the MODIS filter is expected to be comparable with the
operational cloud filter using VIIRS data.
Retrievals with XCH4error< 1 %XCH4error< 0.5 %rms of XCH4 biasrms of XCH4 precisionGlobal clear-sky ensemble94 %78 %0.56 %0.43 %L1B orbit with MODIS filter96 %80 %0.56 %0.31 %L1B orbit with backup cloud filter94 %79 %0.71 %0.27 %
To test the performance of the proposed backup cloud filter, we have
constructed synthetic TROPOMI L1B radiance spectra for an entire orbit that
passes over Africa. We used realistic viewing geometries from a TROPOMI orbit
simulator provided by the Royal Netherlands Meteorological Institute (KNMI).
The meteorological data are from ECMWF. We used CO and CH4 model
profiles from TM5 . The surface elevation comes from a
digital elevation map constructed by KNMI based on USGS
and NASA data . The aerosol and cirrus properties and surface
albedos are taken from the global ensemble, which means that these are the
same for all TROPOMI ground pixels within a
∼ 3∘× 3∘ latitude × longitude box.
While the global ensemble used for the sensitivity studies is land-only and
clear-sky, the test orbit contains cloudy scenes over land and ocean. Cloud
fraction, cloud optical thickness and cloud top pressure are obtained from
MODIS Aqua measurements at 5×5 km2. Here, the cloud top height is
derived from the cloud top pressure and the surface pressure, also provided
by MODIS. The cloud properties are collocated in time and space to the
TROPOMI orbit and used in the measurement simulation. For fractional clouds,
the independent-pixel approximation is used to combine the cloudy and
clear-sky parts of the scene. For reference, the cloud fraction used in the
simulations is shown in Fig. , upper left. We note that
28 % of the TROPOMI ground pixels are fully cloud-free for this test
orbit. Globally, one would expect, on average, 20 % cloud-free pixels
. Resampling of the auxiliary data on the TROPOMI
ground pixels has been performed using the Multi-Instrument Preprocessor
(MIPrep) developed at SRON.
Performance of cloud filters
First, we show the performance of the XCH4 retrievals on the test
orbit when no a posteriori data filtering is applied, only a priori filtering
of ocean pixels and SZA > 70∘ and VZA > 50∘; see
Fig. , upper right. It is clear that cloud-contaminated
measurements lead to large XCH4 errors (> 2 %).
Assuming that we have cloud data from VIIRS, we would then be able to filter
out the cloudy pixels almost perfectly. To illustrate the effect, we filtered
out pixels with cloud fraction > 0.02 in Fig. , lower
left. Note that in this case we have also applied a posteriori filtering
based on retrieved scattering parameters and albedo. One is then left with
valid retrievals of ∼ 3 % of all simulations in the test orbit.
In comparison, the performance of the backup cloud filter based on the
difference between the H2O column retrieved from strong and weak bands
(see Sect. ) is shown in Fig. , lower
right. The backup cloud filter removes most cloudy pixels, but some remain.
In Table and Fig. , the
statistics of XCH4 retrievals on the orbit are summarised. After
cloud filtering with MODIS data (representative of operational VIIRS data),
the results for the test orbit are comparable to the clear-sky global
ensemble. However, the backup cloud filter is less effective. The rms of the
XCH4 bias is then 0.71 % instead of the 0.56 % that is
expected for the operational VIIRS cloud mask.
Cumulative probability distribution of the absolute
XCH4 bias for the simulated level 1b orbit. Cloud-contaminated
measurements are filtered using MODIS data (blue line) or the backup cloud
filter (red line).
Conclusions
This paper describes the algorithm baseline of the operational methane
retrievals from the S5P measurements. The level 2 product includes the
column-averaged dry air mixing ratio XCH4, the column averaging kernel and
the noise standard deviation. In order to account for the effect of aerosols
and cirrus, the algorithm developed
retrieves the methane column
simultaneously with effective scattering parameters related to particle
amount, size and height distribution. The choice of scattering parameters
reflects the information content of the measurements as closely as possible.
The retrieval algorithm uses the radiance and irradiance measurements in the
SWIR 2305–2385 nm band and additionally in the NIR band between
757 and 774 nm (O2 A-band). The forward model of the retrieval algorithm
uses online radiative transfer calculations, fully including multiple
scattering in an efficient manner. Absorption cross sections of the relevant
atmospheric trace gases and optical properties of aerosols are calculated
from lookup tables. The inversion is performed using Phillips–Tikhonov
regularisation in combination with a reduced-step-size Gauss–Newton iteration
scheme.
To test the developed algorithm we generated two ensembles of simulated
measurements that cover the range of scenes that will likely be encountered
by the S5P instrument: one clear-sky global ensemble and one test orbit
containing cloud-contaminated measurements. Overall, the algorithm developed
performs well in correcting for the effect of aerosols and cirrus clouds on
the retrieved XCH4. For both ensembles, ∼ 80 % of the cases have
an XCH4 error < 0.5 % and ∼ 95 % have an error
< 1 %. To achieve this, it is necessary to apply a priori filtering of
cloud-contaminated scenes and a posteriori filtering based on retrieved
parameters. We illustrated the performance of the proposed backup cloud
filter based on retrievals of H2O from weak and strong absorption bands in
the SWIR under the assumption of a non-scattering atmosphere. It should be
noted that the cloud filter based on S5P measurements itself is less
efficient than the VIIRS cloud mask for water clouds.
Apart from forward model errors induced by aerosols, we also studied effects
of model errors in temperature, pressure and water vapour profiles. We expect
to stay within product requirements for errors in input profiles of water,
pressure and temperature below 10 %, 0.3 % and 2 K, respectively.
Another relevant source of errors to the CH4 data product could
be spectroscopic errors. This has been studied in detail by
and . Note that a study is ongoing to improve the
spectroscopic data for TROPOMI's SWIR spectral range .
Concerning instrument errors, we found that the most critical error source is
an error in the ISRF in the SWIR band. To conclude, we have shown that for a
compliant instrument, our algorithm provides a methane product that meets the
requirements.
Data availability
The underlying research data for the simulated atmospheric profiles and
TROPOMI measurement simulation and retrievals are available upon request from
Haili Hu (h.hu@sron.nl). HITRAN spectroscopic line parameters
are available through HITRAN online
(http://hitran.org), and the line parameters from
are available in their Supplement.
Sensitivity of retrievals to fixed-value aerosol parameters
Effect of fixed-value aerosol parameters on
convergence rate, fraction of valid retrievals after filtering and rms
values of XCH4 bias and precision for the global ensemble. See
text for values used in the baseline run.
Our retrieval algorithm retrieves three effective aerosol parameters, namely
total column, size parameter and central height. Other aerosol parameters are
assumed to be fixed, such as the refractive indices n and width w of the aerosol
layer (i.e. FWHM of Gaussian height distribution). The values in the baseline
retrieval are fixed at nNIR=1.4-0.01i,
nSWIR=1.47-0.008i and w=2 km, respectively. To justify this
assumption, we studied the effect of the fixed-value aerosol parameters on
the methane retrievals for our synthetic global ensemble as described
Sect. . We have performed four retrieval test runs
with refractive indices of 1.37-0.0005i and 1.55-0.02i, representative of the majority
of aerosol types , and with widths of 1 and
3 km. For each test run the other values are kept to the baseline values.
For simplicity, we used the same refractive index for NIR and SWIR in the test runs. The results are summarised in Table from
which we conclude that, within a reasonable range, the exact value of the
fixed aerosol parameters does not affect the methane retrievals significantly.
We do see a slight deterioration of our results when a relatively low
refractive index is used.
Acknowledgements
We thanks M. Sneep (The Royal Netherlands Meteorological Institute, KNMI)
and S. Houweling (SRON) for providing us with auxiliary data needed to
simulate TROPOMI measurements. This research has been funded in part by the
TROPOMI national programme from the Netherlands Space Office (NSO).
André Butz is supported by the Deutsche Forschungsgemeinschaft through the
Emmy Noether Programme, grant BU2599/1-1 (RemoteC). Edited by: J. Kim Reviewed by: two anonymous
referees
ReferencesAben, I., Hasekamp, O., and Hartmann, W.: Uncertainties in the
space-based measurements of CO2 columns due to scattering in the Earth's
atmosphere, J. Quant. Spectrosc. Ra., 104, 450–459,
10.1016/j.jqsrt.2006.09.013, 2007.Bergamaschi, P., Frankenberg, C., Meirink, J. F., Krol, M.,
Dentener, F., Wagner, T., Platt, U., Kaplan, J. O., KöRner, S.,
Heimann, M., Dlugokencky, E. J., and Goede, A.: Satellite chartography
of atmospheric methane from SCIAMACHY on board ENVISAT: 2. Evaluation based
on inverse model simulations, J. Geophys. Res.-Atmos., 112, D02304,
10.1029/2006JD007268, 2007.Bergamaschi, P., Frankenberg, C., Meirink, J. F., Krol, M.,
Villani, M. G., Houweling, S., Dentener, F., Dlugokencky, E. J.,
Miller, J. B., Gatti, L. V., Engel, A., and Levin, I.: Inverse
modeling of global and regional CH4 emissions using SCIAMACHY satellite
retrievals, J. Geophys. Res.-Atmos., 114, D22301,
10.1029/2009JD012287, 2009.
Bovensmann, H., Burrows, J. P., Buchwitz, M., Frerick, J.,
Noël, S., Rozanov, V. V., Chance, K. V., and Goede, A. P. H.:
SCIAMACHY: Mission Objectives and Measurement Modes, J. Atmos. Sci., 56,
127–150, 1999.Butz, A., Hasekamp, O. P., Frankenberg, C., and Aben, I.: Retrievals
of atmospheric CO_2 from simulated space-borne measurements of backscattered
near-infrared sunlight: accounting for aerosol effects, Appl. Opt., 48,
3322, 10.1364/AO.48.003322, 2009.Butz, A., Hasekamp, O. P., Frankenberg, C., Vidot, J., and Aben,
I.: CH4 retrievals from space-based solar backscatter measurements:
Performance evaluation against simulated aerosol and cirrus loaded scenes,
J. Geophys. Res.-Atmos., 115, D24302, 10.1029/2010JD014514, 2010.Butz, A., Galli, A., Hasekamp, O., Landgraf, J., Tol, P., and Aben, I.:
{TROPOMI} aboard Sentinel-5 Precursor: Prospective performance of {CH4}
retrievals for aerosol and cirrus loaded atmospheres, Remote Sens. Environ.,
120, 267–276, 10.1016/j.rse.2011.05.030, 2012.
Caron, J., Sierk, B., Bezy, J., Loescher, A., and Meijer, Y.: The CarnoSat
candidate mission: radiometric and specteral performances over spatially
heterogeneouse scenes, International Conference on Space Optics, ICOS, 7–10 October 2014, Tenerife, Spain,
2014.Checa-Garcia, R., Landgraf, J., Galli, A., Hase, F., Velazco, V. A., Tran,
H., Boudon, V., Alkemade, F., and Butz, A.: Mapping spectroscopic
uncertainties into prospective methane retrieval errors from Sentinel-5 and
its precursor, Atmos. Meas. Tech., 8, 3617–3629,
10.5194/amt-8-3617-2015, 2015.Connor, B. J., Boesch, H., Toon, G., Sen, B., Miller, C., and
Crisp, D.: Orbiting Carbon Observatory: Inverse method and prospective
error analysis, J. Geophys. Res.-Atmos., 113, D05305,
10.1029/2006JD008336, 2008.
Danielson, J. and Gesch, D.: Global multi-resolution terrain elevation
data 2010 (GMTED2010), US Geological Survey Open-File Report, 2011-1073, 26,
2011.Dubovik, O., Holben, B., Eck, T. F., Smirnov, A., Kaufman, Y. J.,
King, M. D., Tanré, D., and Slutsker, I.: Variability of
Absorption and Optical Properties of Key Aerosol Types Observed in Worldwide
Locations, J. Atmosp. Sci., 59, 590–608,
10.1175/1520-0469(2002)059<0590:VOAAOP>2.0.CO;2, 2002.Dubovik, O., Sinyuk, A., Lapyonok, T., Holben, B. N., Mishchenko,
M., Yang, P., Eck, T. F., Volten, H., MuñOz, O., Veihelmann,
B., van der Zande, W. J., Leon, J.-F., Sorokin, M., and Slutsker, I.:
Application of spheroid models to account for aerosol particle nonsphericity
in remote sensing of desert dust, J. Geophys. Res.-Atmos., 111, D11208,
10.1029/2005JD006619, 2006.Farr, T. G., Rosen, P. A., Caro, E., Crippen, R., Duren, R.,
Hensley, S., Kobrick, M., Paller, M., Rodriguez, E., Roth, L.,
Seal, D., Shaffer, S., Shimada, J., Umland, J., Werner, M.,
Oskin, M., Burbank, D., and Alsdorf, D.: The Shuttle Radar Topography
Mission, Rev. Geophys., 45, RG2004, 10.1029/2005RG000183, 2007.Frankenberg, C.: OCO-2 IMAP-DOAS preprocessor algorithm theoretical basis
document, Report, Jet Propulsion Laboratory, California Institute of
Technology, Pasadena, CA, USA, available at:
http://disc.sci.gsfc.nasa.gov/OCO-2/documentation/oco-2-v5/IMAP_OCO2_ATBD_prelaunch.pdf,
2014.Frankenberg, C., Platt, U., and Wagner, T.: Retrieval of CO from SCIAMACHY
onboard ENVISAT: detection of strongly polluted areas and seasonal patterns
in global CO abundances, Atmos. Chem. Phys., 5, 1639–1644,
10.5194/acp-5-1639-2005, 2005.Frankenberg, C., Warneke, T., Butz, A., Aben, I., Hase, F., Spietz, P., and
Brown, L. R.: Pressure broadening in the 2ν3 band of methane and its
implication on atmospheric retrievals, Atmos. Chem. Phys., 8, 5061–5075,
10.5194/acp-8-5061-2008, 2008.Frankenberg, C., O'Dell, C., Guanter, L., and McDuffie, J.: Remote sensing of
near-infrared chlorophyll fluorescence from space in scattering atmospheres:
implications for its retrieval and interferences with atmospheric CO2
retrievals, Atmos. Meas. Tech., 5, 2081–2094, 10.5194/amt-5-2081-2012,
2012.Galli, A., Butz, A., Scheepmaker, R. A., Hasekamp, O., Landgraf, J., Tol, P.,
Wunch, D., Deutscher, N. M., Toon, G. C., Wennberg, P. O., Griffith, D. W.
T., and Aben, I.: CH4, CO, and H2O spectroscopy for the Sentinel-5
Precursor mission: an assessment with the Total Carbon Column Observing
Network measurements, Atmos. Meas. Tech., 5, 1387–1398,
10.5194/amt-5-1387-2012, 2012.Gloudemans, A. M. S., Schrijver, H., Hasekamp, O. P., and Aben, I.: Error
analysis for CO and CH4 total column retrievals from SCIAMACHY
2.3 µm spectra, Atmos. Chem. Phys., 8, 3999–4017,
10.5194/acp-8-3999-2008, 2008.Guanter, L., Alonso, L., Gómez-Chova, L., Meroni, M., Preusker,
R., Fischer, J., and Moreno, J.: Developments for vegetation
fluorescence retrieval from spaceborne high-resolution spectrometry in the
O2-A and O2-B absorption bands, J. Geophys. Res.-Atmos., 115,
D19303, 10.1029/2009JD013716, 2010.Hansen, P.: Rank-Deficient and Discrete Ill-Posed Problems: Numerical Aspects
of Linear Inversion, Mathematical Modeling and Computation, Society for
Industrial and Applied Mathematics, 263 pp., 10.1137/1.9780898719697,
1998.Hasekamp, O., Hu, H., Galli, A., Tol, P., Landgraf, J., and Butz, A.:
Algorithm Theoretical Baseline Document for Sentinel-5 Precursor methane
retrieval, Report, SRON netherlands Institute for Space Research, available
at:
http://www.tropomi.eu/sites/default/files/files/SRON-S5P-LEV2-RP-001_TROPOMI_ATBD_CH4_v1p0p0_20160205.pdf,
2016.Hasekamp, O. P. and Butz, A.: Efficient calculation of intensity and
polarization spectra in vertically inhomogeneous scattering and absorbing
atmospheres, J. Geophys. Res-Atmos., 113, D20309,
10.1029/2008JD010379, 2008.Houweling, S., Krol, M., Bergamaschi, P., Frankenberg, C., Dlugokencky, E.
J., Morino, I., Notholt, J., Sherlock, V., Wunch, D., Beck, V., Gerbig, C.,
Chen, H., Kort, E. A., Röckmann, T., and Aben, I.: A multi-year methane
inversion using SCIAMACHY, accounting for systematic errors using TCCON
measurements, Atmos. Chem. Phys., 14, 3991–4012,
10.5194/acp-14-3991-2014, 2014.Krijger, J. M., van Weele, M., Aben, I., and Frey, R.: Technical Note: The
effect of sensor resolution on the number of cloud-free observations from
space, Atmos. Chem. Phys., 7, 2881–2891, 10.5194/acp-7-2881-2007, 2007.Kuze, A., Suto, H., Nakajima, M., and Hamazaki, T.: Thermal and near
infrared sensor for carbon observation Fourier-transform spectrometer on the
Greenhouse Gases Observing Satellite for greenhouse gases monitoring, Appl.
Opt., 48, 6716, 10.1364/AO.48.006716, 2009.
Landgraf, J., aan de Brugh, J., Scheepmaker, R., Borsdorff, T., Hu, H.,
Houweling, S., Butz, A., Aben, I., and Hasekamp, O.: Carbon monoxide total
column retrievals from TROPOMI shortwave infrared measurements, Atmos. Meas.
Tech., 9, 4955–4975, doi:10.5194/amt-9-4955-2016, 2016.
Langen, J., Meijer, Y., Brinksma, E., Veihelmann, B., and Ingmann, P.: GMES
Sentinels 4 and 5 mission requirements document, Mrd, European Space Agency
(ESA), 2011.
Loos, J., Birk, M., Wagner, G., Didier, M., Kassi, S.,
Vasilchenko, S., Campargue, A., Hase, F., Orphal, J., Perrin, A.,
Tran, H., Coudert, L., Dufour, G., Eremenko, M., Cuesta, J.,
Daumont, L., Rotger, M., Bigazzi, A., and Zehner, C.: Spectroscopic
Database for TROPOMI/Sentinel-5 Precursor, ESA ATMOS 2015 conference proceedings (ESA SP-735), 8–12 June 2015, Heraklion, Greece, 2015.Meirink, J. F., Eskes, H. J., and Goede, A. P. H.: Sensitivity analysis of
methane emissions derived from SCIAMACHY observations through inverse
modelling, Atmos. Chem. Phys., 6, 1275–1292, 10.5194/acp-6-1275-2006,
2006.Mishchenko, M. I., Geogdzhayev, I. V., Cairns, B., Rossow, W. B., and
Lacis, A. A.: Aerosol retrievals over the ocean by use of channels 1 and 2
AVHRR data: sensitivity analysis and preliminary results, Appl. Opt., 38,
7325–7341, 10.1364/AO.38.007325, 1999.Noël, S., Bramstedt, K., Bovensmann, H., Gerilowski, K., Burrows, J. P.,
Standfuss, C., Dufour, E., and Veihelmann, B.: Quantification and mitigation
of the impact of scene inhomogeneity on Sentinel-4 UVN UV-VIS retrievals,
Atmos. Meas. Tech., 5, 1319–1331, 10.5194/amt-5-1319-2012, 2012.O'Dell, C. W., Connor, B., Bösch, H., O'Brien, D., Frankenberg, C.,
Castano, R., Christi, M., Eldering, D., Fisher, B., Gunson, M., McDuffie, J.,
Miller, C. E., Natraj, V., Oyafuso, F., Polonsky, I., Smyth, M., Taylor, T.,
Toon, G. C., Wennberg, P. O., and Wunch, D.: The ACOS CO2 retrieval
algorithm – Part 1: Description and validation against synthetic
observations, Atmos. Meas. Tech., 5, 99–121, 10.5194/amt-5-99-2012,
2012.Parker, R., Boesch, H., Cogan, A., Fraser, A., Feng, L., Palmer,
P. I., Messerschmidt, J., Deutscher, N., Griffith, D. W. T., Notholt,
J., Wennberg, P. O., and Wunch, D.: Methane observations from the
Greenhouse Gases Observing SATellite: Comparison to ground-based TCCON data
and model calculations, Geophys. Res. Lett., 38, L15807,
10.1029/2011GL047871, 2011.Phillips, D. L.: A Technique for the Numerical Solution of Certain Integral
Equations of the First Kind, J. ACM, 9, 84–97, 10.1145/321105.321114,
1962.Remer, L. A., Kaufman, Y. J., Tanré, D., Mattoo, S., Chu,
D. A., Martins, J. V., Li, R.-R., Ichoku, C., Levy, R. C.,
Kleidman, R. G., Eck, T. F., Vermote, E., and Holben, B. N.: The
MODIS Aerosol Algorithm, Products, and Validation, J. Atmos. Sci., 62,
947–973, 10.1175/JAS3385.1, 2005.Reuter, M., Buchwitz, M., Schneising, O., Heymann, J., Bovensmann, H., and
Burrows, J. P.: A method for improved SCIAMACHY CO2 retrieval in the
presence of optically thin clouds, Atmos. Meas. Tech., 3, 209–232,
10.5194/amt-3-209-2010, 2010.
Rodgers, C.: Inverse Methods for Atmospheres: Theory and Practice, World Scientific Publishing, Singapore,
240 pp., 2000.Rothman, L., Gordon, I., Barbe, A., Benner, D. C., Bernath, P. F., Birk, M.,
Boudon, V., Brown, L. R., Campargue, A., Champion, J.-P., Chance, K.,
Coudert, L. H., Dana, V., Devi, V. M., Fally, S., Flaud, J.-M., Gamache, R.
R., Goldman, A., Jacquemart, D., Kleiner, I., Lacome, N., Lafferty, W. J.,
Mandin, J.-Y., Massie, S. T., Mikhailenko, S. N., Miller, C. E.,
Moazzen-Ahmadi, N., Naumenko, O. V., Nikitin, A. V., Or- phal, J., Perevalov,
V. I., Perrin, A., Predoi-Cross, A., Rins- land, C. P., Rotger, M.,
Šimečková, M., Smith, M. A. H., Sung, K., Tashkun, S. A.,
Tennyson, J., Toth, R. A., Vandaele, A. C., and Vander Auwera, J: The
HITRAN 2008 molecular spectroscopic database, J. Quant. Spectrosc. Ra.,
110, 533–572, 10.1016/j.jqsrt.2009.02.013, available at: http://hitran.org, 2009.Salstein, D., Ponte, R. M., and Cady-Pereira, K.: Uncertainties in
atmospheric surface pressure fields from global analyses, J. Geophys. Res.,
113, D14107, 10.1029/2007JD009531, 2007.Scheepmaker, R. A., Frankenberg, C., Galli, A., Butz, A., Schrijver, H.,
Deutscher, N. M., Wunch, D., Warneke, T., Fally, S., and Aben, I.: Improved
water vapour spectroscopy in the 4174–4300 cm-1 region and its impact
on SCIAMACHY HDO/H2O measurements, Atmos. Meas. Tech., 6, 879–894,
10.5194/amt-6-879-2013, 2013.Scheepmaker, R. A., aan de Brugh, J., Hu, H., Borsdorff, T., Frankenberg, C.,
Risi, C., Hasekamp, O., Aben, I., and Landgraf, J.: HDO and H2O total column
retrievals from TROPOMI shortwave infrared measurements, Atmos. Meas. Tech.,
9, 3921–3937, 10.5194/amt-9-3921-2016, 2016.Schepers, D., Guerlet, S., Butz, A., Landgraf, J., Frankenberg, C.,
Hasekamp, O., Blavier, J.-F., Deutscher, N. M., Griffith, D. W. T.,
Hase, F., Kyro, E., Morino, I., Sherlock, V., Sussmann, R., and
Aben, I.: Methane retrievals from Greenhouse Gases Observing Satellite
(GOSAT) shortwave infrared measurements: Performance comparison of proxy and
physics retrieval algorithms, J. Geophys. Res.-Atmos., 117, D10307,
10.1029/2012JD017549, 2012.Schepers, D., aan de Brugh, J. M. J., Hahne, P., Butz, A.,
Hasekamp, O. P., and Landgraf, J.: LINTRAN v2.0: A linearised vector
radiative transfer model for efficient simulation of satellite-born
nadir-viewing reflection measurements of cloudy atmospheres, J. Quant.
Spectrosc. Ra., 149, 347–359, 10.1016/j.jqsrt.2014.08.019, 2014.Schneising, O., Buchwitz, M., Reuter, M., Heymann, J., Bovensmann, H., and
Burrows, J. P.: Long-term analysis of carbon dioxide and methane
column-averaged mole fractions retrieved from SCIAMACHY, Atmos. Chem. Phys.,
11, 2863–2880, 10.5194/acp-11-2863-2011, 2011.Schrijver, H., Gloudemans, A. M. S., Frankenberg, C., and Aben, I.: Water
vapour total columns from SCIAMACHY spectra in the 2.36 µm window,
Atmos. Meas. Tech., 2, 561–571, 10.5194/amt-2-561-2009, 2009.Stier, P., Feichter, J., Kinne, S., Kloster, S., Vignati, E., Wilson, J.,
Ganzeveld, L., Tegen, I., Werner, M., Balkanski, Y., Schulz, M., Boucher, O.,
Minikin, A., and Petzold, A.: The aerosol-climate model ECHAM5-HAM, Atmos.
Chem. Phys., 5, 1125–1156, 10.5194/acp-5-1125-2005, 2005.Taylor, T. E., O'Dell, C. W., Frankenberg, C., Partain, P. T., Cronk, H. Q.,
Savtchenko, A., Nelson, R. R., Rosenthal, E. J., Chang, A. Y., Fisher, B.,
Osterman, G. B., Pollock, R. H., Crisp, D., Eldering, A., and Gunson, M. R.:
Orbiting Carbon Observatory-2 (OCO-2) cloud screening algorithms: validation
against collocated MODIS and CALIOP data, Atmos. Meas. Tech., 9, 973–989,
10.5194/amt-9-973-2016, 2016.
Tikhonov, A. N.: Solution of incorrectly formulated problems and a method of
regularization, Doklady Akademii Nauk SSSR (Translated in: Soviet
Mathematics, 4, 1035–1038), 151, 501–504, 1963.
Tol, P., Landgraf, J., and Aben, I.: Instrument noise model for the
Sentinel 5 SWIR bands, Report, Netherlands Insitute for Space Research,
SRON, Utrecht, the Netherlands, 2011.Tran, H., Boulet, C., and Hartmann, J.-M.: Line mixing and
collision-induced absorption by oxygen in the A band: Laboratory
measurements, model, and tools for atmospheric spectra computations, J.
Geophys. Res.-Atmos., 111, D15210, 10.1029/2005JD006869, 2006.
van Deelen, R., Hasekamp, O. P., and Landgraf, J.: Accurate modeling of
spectral fine-structure in Earth radiance spectra measured with the Global
Ozone Monitoring Experiment, Appl. Opt., 46, 243–252, 2007.Veefkind, J., Aben, I., McMullan, K., örster, H., de Vries, J.,
Otter, G., Claas, J., Eskes, H., de Haan, J., Kleipool, Q., van
Weele, M., Hasekamp, O., Hoogeveen, R., Landgraf, J., Snel, R.,
Tol, P., Ingmann, P., P., V., Kruizinga, P., Vink, R., Visser, H.,
and Levelt, P.: TROPOMI on the ESA Sentinel-5 Precursor: A GMES mission
for global observations of the atmospheric composition for climate, air
quality and ozone layer applications, Remote Sens. Environ., 120, 70–83,
10.1016/j.rse.2011.09.027, 2012.Wassmann, A., Borsdorff, T., aan de Brugh, J. M. J., Hasekamp, O. P., Aben,
I., and Landgraf, J.: The direct fitting approach for total ozone column
retrievals: a sensitivity study on GOME-2/MetOp-A measurements, Atmos. Meas.
Tech., 8, 4429–4451, 10.5194/amt-8-4429-2015, 2015.Winker, D. M., Hunt, W. H., and McGill, M. J.: Initial performance
assessment of CALIOP, Geophys. Res. Lett., 34, L19803,
10.1029/2007GL030135, 2007.