Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/ doi:10.5194/amt-10-119-2017 Author(s) 2017. CC Attribution 3.0 License.
Nicolas Theys1, Isabelle De Smedt1, Huan Yu1, Thomas Danckaert1, Jeroen van Gent1, Christoph Hrmann2, Thomas Wagner2, Pascal Hedelt3, Heiko Bauer3, Fabian Romahn3, Mattia Pedergnana3, Diego Loyola3, and Michel Van Roozendael1
1Royal Belgian Institute for Space Aeronomy (BIRA-IASB), Brussels, Belgium
2Max Planck Institute for Chemistry (MPIC), Hahn-Meitner-Weg 1, 55128 Mainz, Germany
3Institut fr Methodik der Fernerkundung (IMF), Deutsches Zentrum fr Luft und Raumfahrt (DLR), Oberpfaffenhofen, Germany
Correspondence to: N. Theys ([email protected])
Received: 21 September 2016 Published in Atmos. Meas. Tech. Discuss.: 22 September 2016 Revised: 19 November 2016 Accepted: 12 December 2016 Published: 9 January 2017
Abstract. The TROPOspheric Monitoring Instrument (TROPOMI) onboard the Copernicus Sentinel-5 Precursor (S-5P) platform will measure ultraviolet earthshine radiances at high spectral and improved spatial resolution (pixel size of 7 km [notdef] 3.5 km at nadir) compared to its predecessors
OMI and GOME-2. This paper presents the sulfur dioxide (SO2) vertical column retrieval algorithm implemented in the S-5P operational processor UPAS (Universal Processor for UV/VIS Atmospheric Spectrometers) and comprehensively describes its various retrieval steps. The spectral tting is performed using the differential optical absorption spectroscopy (DOAS) method including multiple tting windows to cope with the large range of atmospheric SO2 columns encountered. It is followed by a slant column background correction scheme to reduce possible biases or across-track-dependent artifacts in the data. The SO2 vertical columns are obtained by applying air mass factors (AMFs) calculated for a set of representative a priori proles and accounting for various parameters inuencing the retrieval sensitivity to SO2. Finally, the algorithm includes an error analysis module which is fully described here. We also discuss verication results (as part of the algorithm development) and future validation needs of the TROPOMI SO2 algorithm.
1 Introduction
Sulfur dioxide enters the Earths atmosphere via both natural and anthropogenic processes. Through the formation of sul-fate aerosols and sulfuric acid, it plays an important role on the chemistry at local and global scales and its impact ranges from short-term pollution to climate forcing. While about one-third of the global sulfur emissions originate from natural sources (volcanoes and biogenic dimethyl sulde), the main contributor to the total budget is from anthropogenic emissions mainly from the combustion of fossil fuels (coal and oil) and from smelting. Over the last decades, a host of satellite-based UVvisible instruments have been used for the monitoring of anthropogenic and volcanic SO2 emissions. Total vertical column density (VCD) of SO2 has been retrieved with the sensors TOMS (Krueger, 1983), GOME (Eisinger and Burrows, 1998; Thomas et al., 2005; Khokar et al., 2005), SCIAMACHY (Afe et al., 2004), OMI (Krotkov et al., 2006; Yang et al., 2007, 2010; Li et al., 2013; Theys et al., 2015), GOME-2 (Richter et al., 2009; Bobrowski et al., 2010; Nowlan et al., 2011; Rix et al., 2012; Hrmann et al., 2013) and OMPS (Yang et al., 2013). In particular, the Ozone Monitoring Instrument (OMI) has largely demonstrated the value of satellite UVvisible remote sensing (1) in monitoring volcanic plumes in near-real time (Brenot et al., 2014) and changes in volcanic degassing at the global scale (Carn et al., 2016, and references therein) and (2) in detecting and quantifying large anthropogenic SO2 emissions, weak or
Published by Copernicus Publications on behalf of the European Geosciences Union.
Sulfur dioxide retrievals from TROPOMI onboard Sentinel-5 Precursor: algorithm theoretical basis
120 N. Theys et al.: S-5P SO2 algorithm theoretical basis
given in Sect. 6. Additional information on data product and auxiliary data, as well as a list of acronyms, is provided in the Appendix.
2 TROPOMI SO2 algorithm
2.1 Product requirements
While UV measurements are highly sensitive to SO2 at high altitudes (upper tropospherelower stratosphere), the sensitivity to SO2 concentration in the boundary layer is intrinsically limited from space due to the combined effect of scattering (Rayleigh and Mie) and ozone absorption that hamper the penetration of solar radiation into the lowest atmospheric layers. Furthermore, the SO2 absorption signature suffers from the interference with the ozone absorption spectrum.
The retrieval precision (or random uncertainty) is driven by the signal-to-noise ratio of the recorded spectra and by the retrieval wavelength interval used, the accuracy (or systematic uncertainty) is limited by the knowledge on the auxiliary parameters needed in the different retrieval steps. Among these are the treatment of other chemical interfering species, clouds and aerosol, the representation of vertical proles (gas, temperature, pressure), and uncertainties on data from external sources (e.g., surface reectance).
Requirements on the accuracy and precision for the data products derived from the TROPOMI measurements are specied in the GMES Sentinels 4 and 5 and 5p Mission Requirements Document MRD (Langen et al., 2011), the Report of The Review Of User Requirements for Sentinels-4/5 (Bovensmann et al., 2011) and the Science Requirements Document for TROPOMI (van Weele et al., 2008). These requirements derive from the Composition of the Atmosphere: Progress to Applications in the user CommunITY (CAPACITY) study (Kelder et al., 2005) and have been ne-tuned by the Composition of the Atmospheric Mission concEpts and SentineL Observation Techniques (CAMELOT; Levelt et al., 2009) and Original and New TRopospheric composition and Air Quality measurements (ONTRAQ; Zweers et al., 2010) studies. The CAPACITY study has dened three main themes: the ozone layer (A), air quality (B), and climate (C), with further division into sub-themes. Requirements for SO2 have been specied for a number of these sub-themes. In the following paragraphs, we discuss these requirements and the expected performances of the SO2 retrieval algorithm (summary is given in Table 1).
2.1.1 Theme A3 ozone layer assessment
This theme addresses the importance of measurements in the case of enhanced SO2 concentrations in the stratosphere due to severe volcanic events. The long-term presence (up to several months) of SO2 in the stratosphere contributes to the stratospheric aerosol loading and hence affects the cli-
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
Figure 1. Map of averaged SO2 columns from OMI clear-sky pixels for the 20052009 period.
unreported emission sources worldwide (Theys et al., 2015; Fioletov et al., 2016; McLinden et al., 2016) as well as investigating their long-term changes (Krotkov et al., 2016; van der A et al., 2016; He et al., 2016). An example map of OMI SO2 columns (Theys et al., 2015) averaged over the 20052009 period is shown in Fig. 1, illustrating typical anthropogenic emission hotspots (China, eastern Europe, India and the Middle East) and signals from volcanic activity (e.g., from the volcanoes in DR Congo).
The 7-year-lifetime Sentinel-5p sensor TROPOMI (Veefkind et al., 2012) will y on a polar low-Earth orbit with a wide swath of 2600 km. The TROPOMI instrument is a push-broom imaging spectrometer similar in concept to OMI. It has eight spectral bands covering UV to SWIR wavelengths. The SO2 retrieval algorithm exploits measurements from band 3 (310405 nm), with typical spectral resolution of 0.54 nm, signal-to-noise ratio of about 1000 and pixel size as good as 7 km [notdef] 3.5 km.
TROPOMI will continue and improve the measurement time series of OMI SO2 and other UV sensors. Owing to similar performance to OMI in terms of signal-to-noise ratio and unprecedented spatial resolution, TROPOMI will arguably discern very ne details in the SO2 distribution and will be able to detect point sources with annual SO2 emissions of about 10 kT yr1 or lower (using oversampling techniques).
This paper gives a thorough description of the operational TROPOMI SO2 algorithm and reects the S-5P SO2 L2 Algorithm Theoretical Basis Document v1.0. In Sect. 2, we rst present the product requirements and briey discuss the expected product performance in terms of precision and accuracy. It is then followed by the SO2 column retrieval algorithm description. An error analysis of the retrieval method is presented in Sect. 3. Results from algorithm verication exercise using an independent retrieval scheme is given in Sect. 4. The possibilities for future validation of the retrieved SO2 data product can be found in Sect. 5. Conclusions are
N. Theys et al.: S-5P SO2 algorithm theoretical basis 121
Table 1. Requirements on SO2 vertical column products as derived from the MRD. Left- and right-hand numbers in ranges denote accuracy and precision, respectively.
Horizontal Required Achievable Theme resolution uncertainty uncertainty (table in
(km) MRTD)
Enhanced stratospheric 50200 30 % for Met for VCD > 0.5 DU A3 column VCD > 0.5 DU VCD > 0.5 DUTropospheric 520 3060 % or 50 %/ B1, B2, B3 column 1.3 [notdef] 10
15 molecules cm2 36 [notdef] 10
16 molecules cm2
(least stringent)
Total column 520 3060 % or 50 %/ B1, B2, B31.3 [notdef] 10
15 molecules cm2 36 [notdef] 10
16 molecules cm2
(least stringent)
mate and the stratospheric ozone budget. For such scenarios, the requirements state that the stratospheric vertical column should be monitored with a total uncertainty of 30 %. Although powerful volcanic events generally produce large amounts of SO2, monitoring such a plume over extended periods of time also requires the detection of the plume after it has diluted during the weeks after the eruption.
From an error analysis of the proposed SO2 algorithm (Sect. 3), we have assessed the major sources of uncertainty in the retrieved SO2 column. One of the main contributors to the total uncertainty is instrumental noise. This source of error alone limits the precision to vertical columns above about 0.25 DU (1 DU = 2.69 [notdef] 1016 molec cm2). For SO2
in the stratosphere, the summing-up of the various uncertainties (Sect. 3) is believed to be around the required uncertainty of 30 % for diluted SO2 plumes, provided that the vertical column is larger than 0.5 DU. Explosive volcanic eruptions capable of injecting SO2 into the stratosphere regularly show stratospheric SO2 columns of a few DU to several hundreds of DU or more, as was the case, for example, for the eruptions of Mt. Kasatochi (Yang et al., 2010) and Sarychev Peak (Carn et al., 2011). For very large SO2 concentrations, the dynamical use of different tting windows (see Sect. 2.2) enables the 30 % uncertainty level to be reached (see Sect. 4).
2.1.2 Theme B air quality
This theme includes three sub-themes:
B1 Protocol monitoring: this involves the monitoring of abundances and concentrations of atmospheric constituents, driven by several agreements, such as the Gothenburg protocol, National Emission Ceilings, and EU Air Quality regulations.
B2 Near-real-time (NRT) data requirements: this comprises the relatively fast ( 30 min) prediction and determina
tion of surface concentrations in relation to health and safety warnings.
B3 Assessment: this sub-theme aims at answering several air-quality-related scientic questions, such as the effect on air quality of spatial and temporal variations in oxidizing capacity and long-range transport of atmospheric constituents.
A more detailed description of the air quality sub-themes can be found in Langen et al. (2011).
The user requirements on SO2 products are equal for all three sub-themes. For the total vertical column and the tropospheric vertical column of SO2, the user requirements state an absolute maximum uncertainty of1.3 [notdef] 1015 molecules cm2 or 0.05 DU. This number derives
from the ESA CAPACITY study, where the number was expressed as 0.4 ppbv for a 1.5 km thick boundary layer reaching up to 850 hPa. From the uncertainty due to instrument noise only, it is clear that the 0.05 DU requirement cannot be met on a single-measurement basis. This limitation was already found in the ESA CAMELOT study (Levelt et al., 2009).
For anthropogenic SO2 typically conned in the planetary boundary layer (PBL), calculations performed within the CAMELOT study showed that the smallest vertical column that can be detected in the PBL is of about 13 DU (for a signal-to-noise ratio (S/N) of 1000). Although pollution hotspots can be better identied by spatial or temporal averaging, several uncertainties (e.g., due to varying surface albedo or SO2 vertical prole shape) are not averaging out and directly limit the product accuracy to about 50 % or more. Though the difference between the MRD requirements and the expected TROPOMI performance is rather large, one could argue that the required threshold should not be a strict criterion in all circumstances. The user requirement of 0.05 DU represents the maximum uncertainty to distinguish (anthropogenic) pollution sources from background concentrations. Bovensmann et al. (2011) reviewed the MRD user requirements and motivated a relaxation of certain user requirements for specic conditions. For measurements in the PBL, the document proposes a relative
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
122 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Figure 2. Flow Diagram of the TROPOMI DOAS retrieval algorithm for SO2.
requirement of 3060 % in order to discriminate between enhanced (> 1.5 ppbv), moderate (0.51.5 ppbv), and background concentrations (< 0.5 ppbv). It is expected that it will be possible to discriminate these three levels by averaging (spatiotemporally) TROPOMI data.
For volcanic SO2 plumes in the free troposphere, a better measurement sensitivity is expected for TROPOMI. The expected precision is about 0.5 DU on the vertical column. The accuracy on the SO2 vertical column will be strongly limited by the SO2 plume height and the cloud conditions. As these parameters are highly variable in practice, it is difcult to ascertain the product accuracy for these conditions.
2.2 Algorithm description
The rst algorithm to retrieve SO2 columns from space-borne
UV measurements was developed based on a few wavelength pairs (for TOMS) and has been subsequently applied and rened for OMI measurements (e.g., Krotkov et al., 2006; Yang et al., 2007, and references therein). Current algorithms exploit back-scattered radiance measurements in a wide spectral range using a direct tting approach (Yang et al., 2010; Nowlan et al., 2011), a principal component analysis (PCA) method (Li et al., 2013) or (some form of) differential optical absorption spectroscopy (DOAS; Platt and Stutz, 2008); see, e.g., Richter et al. (2009), Hrmann et al. (2013), or Theys et al. (2015).
Direct tting schemes in which on-the-y radiative transfer simulations are made for all concerned wavelengths and resulting simulated spectra are adjusted to the spectral observations, are in principle the most accurate. They are able to cope with very large SO2 columns (such as those occurring
during explosive volcanic eruptions), i.e., conditions typically leading to a strongly nonlinear relation between the SO2 signal and the VCD. However, the main disadvantage of direct tting algorithms with respect to DOAS (or PCA) is that they are computationally expensive and are out of reach for TROPOMI operational near-real-time processing, for which the level 1b data ow is expected to be massive and deliver around 1.5 million spectral measurements per orbit ( 15 orbits daily) for band 3 (with a corresponding data
size of 6 GB). To reach the product accuracy and processing performance requirements, the approach adopted here applies DOAS in three different tting windows (within the 310390 nm spectral range) that are still sensitive enough to SO2 but less affected by nonlinear effects (Bobrowski et al., 2010; Hrmann et al., 2013).
Figure 2 shows the full ow diagram of the SO2 retrieval algorithm including the dependencies on auxiliary data and other L2 products. The algorithm and its application to OMI data are also described in Theys et al. (2015), although there are differences in some settings. The baseline operation ow of the scheme is based on a DOAS retrieval algorithm and is identical to that implemented in the retrieval algorithm for HCHO (also developed by BIRA-IASB; see De Smedt et al., 2016). The main output parameters of the algorithm are SO2 vertical column density, slant column density, air mass factor, averaging kernels (AKs) and error estimates. Here, we will rst briey discuss the principle of the DOAS VCD retrieval before discussing the individual steps of the process in more detail.
First, the radiance and irradiance data are read from an S-5P L1b le, along with geolocation data such as pixel coor-
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 123
dinates and observation geometry (sun and viewing angles).
At this stage cloud cover information (see Table A3 in Appendix A) is also obtained from the S-5P cloud L2 data, as required for the calculation of the AMF, later in the scheme.Then relevant absorption cross section data, as well as characteristics of the instrument (e.g., slit functions), are used as input for the determination of SO2 slant column density. As a baseline, the slant column t is done in a sensitive window from 312 to 326 nm. For pixels with a strong SO2 signal, results from alternative windows where the SO2 absorption is weaker can be used instead. An empirical offset correction (dependent on the tting window used) is then applied to the SCD. The latter correction accounts for systematic biases in the SCDs. Following the SCD determination, the AMF is estimated based on a pre-calculated weighting functions (or box AMFs) look-up table (LUT). This look-up table is generated using the LInearized Discrete Ordinate Radiative Transfer (LIDORT) code (Spurr, 2008) and has several entries: cloud cover data, topographic information, observation geometry, surface albedo, effective wavelength (representative of the tting window used), total ozone column and the shape of the vertical SO2 prole. The algorithm also includes an error calculation and retrieval characterization module (Sect. 3) that computes the averaging kernels (Eskes and Boersma, 2003), which characterize the vertical sensitivity of the measurement and which are required for comparison with other types of data (Veefkind et al., 2012).
The nal SO2 vertical column is obtained by
Nv =
Ns Nbacks
M , (1)
where the main quantities are the vertical column (Nv), the slant column density (Ns) and the values used for the background correction (Nbacks). M is the air mass factor.
2.2.1 Slant column retrieval
The backscattered radiance spectrum recorded by the space instrument differs from the solar spectrum because of the interactions of the photons with the Earths atmosphere and surface reection. Hence, the reectance spectra contains spectral features that can be related to the various absorbing species and their amounts in the atmosphere. The DOAS method aims at the separation of the highly structured trace gas absorption spectra and broadband spectral structures. The technique relies on a number of assumptions that can be summarized as follows:
a. The spectral analysis and atmospheric radiative transfer computations are treated separately by considering one averaged atmospheric light path of the photons traveling from the sun to the instrument.
b. The absorption cross sections are not strongly dependent on pressure and temperature. Additionally, the averaged light path should be weakly dependent on the
wavelength for the tting window used which enables dening an effective absorption (slant) column density. It should be noted that this is not strictly valid for the SO2 DOAS retrieval because of strong absorption by ozone and in some cases SO2 itself (for large SO2 amounts).
c. Spectrally smoothed structures due broadband absorption, scattering and reection processes can be well reproduced by a low-order polynomial as a function of wavelength.
Photons collected by the satellite instrument may have followed very different light paths through the atmosphere depending on their scattering history. However, a single effective light path is assumed, which represents an average of the complex paths of all reected and scattered solar photons reaching the instrument within the spectral interval used for the retrieval. This simplication is valid if the effective light path is reasonably constant over the considered wavelength range. The spectral analysis can be described by the following equation:
ln I ( )
[notdef]0E0 ( ) =
Xj
j ( )Nsj +
Xp
cp p, (2)
where I ( ) is the observed backscattered earthshine radiance (W m2 nm1 sr1), E0 is the solar irradiance (W m2 nm1) and [notdef]0 = cos 0. The rst term on the right-
hand side indicates all relevant absorbing species with absorption cross sections j (cm2 molec1). Integration of the number densities of these species along the effective light path gives the slant column density Nsj (molec cm2). Equation (2) can be solved by least-squares tting techniques (Platt and Stutz, 2008) for the slant column values. The nal term in Eq. (2) is the polynomial representing broadband absorption and (Rayleigh and Mie) scattering structures in the observed spectrum and also accounts for possible errors such as uncorrected instrument degradation effects, uncertainties in the radiometric calibration or possible residual (smooth) polarization response effects not accounted for in the level 01 processing.
Apart from the cross sections for the trace gases of interest, additional t parameters need to be introduced to account for the effect of several physical phenomena on the t result.For SO2 tting, these are the lling-in of Fraunhofer lines (Ring effect) and the need for an intensity offset correction.In the above, we have assumed that for the ensemble of ob-served photons a single effective light path can be assumed over the adopted wavelength tting interval. For the observation of (generally small) SO2 concentrations at large solar zenith angles (SZAs) this is not necessarily the case. For such long light paths, the large contribution of O3 absorption may lead to negative SO2 retrievals. This may be mitigated by taking the wavelength dependence of the O3 SCD over the
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
124 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Table 2. DOAS settings used to retrieved SO2 slant columns.
Fitting intervals 1 and 2 312326 nm (w1), 325335 nm (w2)
Cross sections SO2 : 203K (Bogumil et al., 2003)
O3 : 228 and 243 K with Io correction (Brion et al., 1998)
Pseudo-O3 cross sections ( O3, 2
O3; Pukte et al., 2010)
Ring effect: two eigenvectors (Vountas et al., 1998) generated for 20 and 87 solar zenith angles using LIDORT-RRS (Spurr et al., 2008)
Polynomial Fifth orderFitting interval 3 360390 nm (w3)
Cross sections SO2: Hermans et al. (2009) extrapolated at 203K
NO2 : 220 K (Vandaele et al., 1998)
O2-O2: Greenblatt et al.,1990
Ring effect: single spectrum (Chance and Spurr, 1997)
Polynomial Fourth orderIntensity offset correction Linear offsetSpectrum shift and stretch FittedSpectral spikes removal procedure Richter et al. (2011)
Reference spectrum Baseline: daily solar irradianceForeseen update: daily averaged earthshine spectrum in Pacic region (10 S10 N, 160 E120 W); separate spectrum for each detector row.
NRT: averaged spectra of the last available day; ofine: averaged spectra of the current day
Figure 3. Absorption cross sections of SO2 and O3. The blue, yellow and green boxes delimit the three SO2 tting windows 312
326, 325335 and 360390 nm, respectively.
tting window into account, as will be described in the next section.
The different parts of the DOAS retrieval are detailed in the next subsections and Table 2 gives a summary of settings used to invert SO2 slant columns. Note that, in Eq. (2), the daily solar irradiance is used as a baseline for the reference spectrum. As a better option, it is generally preferred to use daily averaged radiances, selected for each across-track position, in the equatorial Pacic. In the NRT algorithm, the last valid day can be used to derive the reference spectra,
while in the ofine version of the algorithm, the current day should be used. Based on OMI experience, it would allow, for example, for better handling of instrumental artifacts and degradation of the recorded spectra for each detector. At the time of writing, it is planned to test this option during the S-5P commissioning phase.
Wavelength tting windows
DOAS measurements are in principle applicable to all gases having suitable narrow absorption bands in the UV, visible, or near-IR regions. However, the generally low concentrations of these compounds in the atmosphere, and the limited signal-to-noise ratio of the spectrometers, restrict the number of trace gases that can be detected. Many spectral regions contain several interfering absorbers and correlations between absorber cross sections can sometimes lead to systematic biases in the retrieved slant columns. In general, the correlation between cross sections decreases if the wavelength interval is extended, but then the assumption of a single effective light path dened for the entire wavelength interval may not be fully satised, leading to systematic mist effects that may also introduce biases in the retrieved slant columns (e.g., Pukte et al., 2010) . To optimize DOAS retrieval settings, a trade-off has to be found between these effects. In the UVvisible spectral region, the cross-section spectrum of SO2 has its strongest bands in the 280320 nm range (Fig. 3).
For the short wavelengths in this range, the SO2 signal, however, suffers from a strong increase in Rayleigh scattering and ozone absorption. In practice, this leads to a very small SO2 signal in the satellite spectra compared to ozone absorp-
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 125
tion, especially for tropospheric SO2. Consequently, SO2 is traditionally retrieved (for GOME, SCIAMACHY, GOME-2, OMI) using sensitive windows in the 310326 nm range.Note that even in this range the SO2 absorption can be 3 orders of magnitude lower than that of ozone.
The TROPOMI SO2 algorithm is using a multiple-window approach:
312326 nm: classical tting window, ideal for small columns. This window is used as baseline. If nonlinear effects due to high SO2 amounts are encountered, one of the two following windows will be used instead.
325335 nm: in this window, differential SO2 spectral features are 1 order of magnitude smaller than in the classical window. It allows the retrieval of moderate SO2 columns, an approach similar to the one described by Hrmann et al. (2013).
360390 nm: SO2 absorption bands are 23 orders of magnitude weaker than in the classical window and are best suited for the retrieval of extremely high SO2 columns (Bobrowski et al., 2010)
Note that in the 325335 and 360390 nm windows the Rayleigh scattering and ozone absorption are less important than in the baseline 312326 nm window (see also Fig. 3).
Specically, in the rst two intervals, absorption cross sections of O3 at 228 and 243 K are included in the t and, to better cope with the strong (nonlinear) ozone absorption at short wavelengths, the retrieval also includes two pseudo-cross sections following the approach of Pukte et al. (2010): O3 and 2O3 calculated from the O3 cross-section spectrum at 228 K. The correction for the Ring effect is based on the technique outlined by Vountas et al. (1998). This technique involves a PCA of a set of Ring spectra, calculated for a range of solar zenith angles. The rst two of the resulting eigenvectors appear to accurately describe the Ring spectra, with the rst eigenvector representing the lling-in of Fraunhofer lines and the second mostly representing the lling-in of gas absorption features. In the retrieval algorithm, these vectors are determined by orthogonalizing two Ring spectra, calculated by LIDORT-RRS (Spurr et al., 2008), a version of LIDORT accounting for rotational Raman scattering, for a low SZA (20 ) and a high SZA (87 ), respectively.
Wavelength calibration and convolution to TROPOMI resolution
The quality of a DOAS t critically depends on the accuracy of the alignment between the earthshine radiance spectrum, the reference spectrum and the cross sections. Although the level 1b will contain a spectral assignment, an additional spectral calibration is part of the SO2 algorithm. Moreover, the DOAS spectral analysis also includes the t of shift and stretch of radiance spectra because the TROPOMI spectral registration will differ from one ground pixel to another, e.g.,
due to thermal variations over the orbit as well as due to inhomogeneous lling of the slit in ight direction.
The wavelength registration of the reference spectrum can be ne-tuned by means of a calibration procedure making use of the solar Fraunhofer lines. To this end, a reference solar atlas, Es, accurate in absolute vacuum wavelength to better than 0.001 nm (Chance and Kurucz, 2010) is degraded at the resolution of the instrument, through convolution by the TROPOMI instrumental slit function.
Using a nonlinear least-squares approach, the shift ([Delta1]i)
between the reference solar atlas and the TROPOMI irradiance is determined in a set of equally spaced sub-intervals covering a spectral range large enough to encompass all relevant tting intervals. The shift is derived according to the following equation:
E0 ( ) = Es( [Delta1]i), (3)
where Es is the solar spectrum convolved at the resolution of the instrument and [Delta1]i is the shift in sub-interval i. A polynomial is then tted through the individual points in order to reconstruct an accurate wavelength calibration [Delta1]( ) for the complete analysis interval. Note that this approach allows one to compensate for stretch and shift errors in the original wavelength assignment.
In the case of TROPOMI, the procedure is complicated by the fact that such calibrations must be performed (and stored) for each separate spectral eld on the CCD detector array.Indeed, due to the imperfect characteristics of the imaging optics, each row of the TROPOMI instrument must be considered a separate spectrometer for analysis purposes.
In a subsequent step of the processing, the absorption cross sections of the different trace gases must be convolved with the instrumental slit function. The baseline approach is to use slit functions determined as part of the TROPOMI key data.Slit functions are delivered for each binned spectrum and as a function of wavelength. Note that an additional feature of the prototype algorithm allows for an effective slit function of known line shape to be dynamically tted (e.g., asymmetric Gaussian). This can be used for verication and monitoring purpose during commissioning and later on during the mission.
More specically, wavelength calibrations are made for each TROPOMI orbit as follows:
1. The TROPOMI irradiances (one for each row of the CCD) are calibrated in wavelength over the 310390 nm wavelength range, using 10 sub-windows.
2. The earthshine radiances and the absorption cross sections are interpolated (cubic spline interpolation) on the calibrated wavelength grid, prior to the DOAS analysis.
3. During spectral tting, shift and stretch parameters are further derived to align radiance and irradiance spectra.
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
126 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Table 3. Criteria for selecting alternative tting windows.
Window number w1 w2 w3
Wavelength range 312326 nm 325335 nm 360390 nm Derived slant column S1 S2 S3 Application Baseline for S1 > 15 DU S2 > 250 DU every pixel and S2 > S1 and S3 > S2
The reference wavelength grid used in the DOAS procedure is the (optimized) grid of the TROPOMI solar irradiance.
Spike removal algorithm
A method to remove individual hot pixels or detector pixels affected by the South Atlantic Anomaly has been presented for NO2 retrievals in Richter et al. (2011). Often only a few individual detector pixels are affected, and in these cases, it is possible to identify and remove the noisy points from the t. However, as the amplitude of the distortion is usually only of the order of a few percent or less, it cannot always be found in the highly structured spectra themselves. Higher sensitivity for spikes can be achieved by analyzing the residual of the t where the contribution of the Fraunhofer lines, scattering, and absorption is already removed.
When the residual for a single detector pixel exceeds the average residual of all detector pixels by a chosen threshold ratio (the tolerance factor), the pixel is excluded from the analysis, in an iterative process. This procedure is repeated until no further outliers are identied, or until the maximum number of iterations is reached (here xed to 3). This is especially important to handle the degradation of 2-D detector arrays such as OMI or TROPOMI. However, this improvement of the algorithm has a non-negligible impact on the time of processing. At the time of writing, the exact values for the tolerance factor and maximum number of iterations of the spike removal procedure are difcult to ascertain and will only be known during operations. To assess the impact on the processing time, test retrievals have been done on OMI spectra using a tolerance factor of 5 and a limit of three iterations (this could be relaxed), leading to an increase in processing time by a factor of 1.5.
Fitting window selection
The implementation of the multiple-tting-window retrieval requires selection criteria for the transition from one window to another. These criteria are based on the measured SO2 slant columns. As a baseline, the SO2 SCD in the 312
326 nm window will be retrieved for each satellite pixel. When the resulting value exceeds a certain criterion, the slant column retrieval is taken from an alternative window. As part of the algorithm development and during the verication exercise (Sect. 4), closed-loop retrievals have been performed
Figure 4. OMI SO2 vertical columns (DU) averaged for the year 2007 (top) with and (bottom) without background correction. Only clear-sky pixels (cloud fraction lower than 30 %) have been kept. AMFs calculated from SO2 proles from the IMAGES global model are applied to the slant columns (Theys et al., 2015).
and application of the algorithm to real data from the GOME-2 and OMI instruments lead to threshold values and criteria as given in Table 3.
2.2.2 Offset correction
When applying the algorithm to OMI and GOME-2 data, across-track/viewing-angle-dependent residuals of SO2 were found over clean areas and negative SO2 SCDs are found at high SZA which need to be corrected (note that this is a common problem of most algorithms to retrieve SO2 from space
UV sensors). A background correction scheme was found mostly necessary for the SO2 slant columns retrieved in the baseline tting window. The adopted correction scheme depends on across-track position and measured O3 slant column as described below.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 127
The correction is based on a parameterization of the background values that are then subtracted from the measurements. The scheme rst removes pixels with high SZA (> 70 ) or SCDs larger than 1.5 DU (measurements with presumably real SO2) and then calculates the offset correction by averaging the SO2 data on an ozone slant column grid (bins of 75 DU). This is done independently for each across-track position and hemisphere, and the correction makes use of measurements averaged over a time period of 2 weeks preceding the measurement of interest (to improve the statistics and minimize the impact of a possible extended volcanic SO2 plume on the averaged values).
It should be noted that the O3 slant column is dependent on the wavelength when applying the approach of Pukte et al. (2010):
SCD( ) = SCDT 1 + SCDT 2 + SCD + s( )SCDs. (4)
SCDT 1 and SCDT 2 are the retrieved ozone slant columns corresponding to the ozone cross sections at two temperatures (T 1, T 2) included in the t. SCD and SCDs are the retrieved parameters for the two pseudo-cross sections [notdef] s
and 2s (s being the O3 cross section at T 1). In order to apply the background correction, the O3 slant column expression (Eq. 4) is evaluated at 313 nm (read below).
An example of the effect of the background correction is shown in Fig. 4 for OMI. One can see that after correction (top panel) the retrievals show smooth/unstriped results and values close to zero outside the polluted areas. In some regions (in particular at high latitudes), residual columns can be found, but are generally lower than 0.2 DU.
For the two additional tting windows, residual SO2 levels are relatively small in comparison to the column amounts expected to be retrieved in these windows. However, simplied background corrections are also applied to the alternative windows: the offset corrections use parameterizations of the background slant columns based on latitude (bins of 5 ), cross-track position and time (2-week moving averages as for the baseline window). To avoid contamination by strong volcanic eruptions, only the pixels are kept with SCD less than 50 and 250 DU for the tting windows 325335 and 360 390 nm, respectively.
It should be noted that the background corrections do not imply saving 2 weeks of SO2 L2 data in intermediate products, but only the averaged values ([Sigma1]i=1, N SCDi/N) over
the predened working grids (note: the numerators [Sigma1]i=1, N
SCDi and denominators N are stored separately).
This background correction is well suited for the case of a 2-D-detector array such as TROPOMI, for which across-track striping can possibly arise due to imperfect cross-calibration and different dead/hot pixel masks for the CCD detector regions. This instrumental effect can also be found for scanning spectrometers, but since these instruments only have one single detector, such errors do not appear as stripes.These different retrieval artifacts can be compensated for (up
to a certain extent) using background corrections which depend on the across-track position. All of these corrections are also meant to handle the time-dependent degradation of the instrument. Note that experiences with OMI show that the most efcient method to avoid across-track stripes in the retrievals is to use row-dependent mean radiances as control spectrum in the DOAS t.
2.2.3 Air mass factors
The DOAS method assumes that the retrieved slant column (after appropriate background correction) can be converted into a vertical column using a single air mass factor M (representative of the tting interval):
M =
NsNv , (5)
which is determined by radiative transfer calculations with LIDORT version 3.3 (Spurr, 2008). The AMF calculation is based on the formulation of Palmer et al. (2001):
M =
[integraldisplay]
m[prime] (p) [notdef] s (p)dp, (6)
with m[prime] = m(p)/Ctemp(p), where m(p) is the so-called
weighting function (WF) or pressure-dependent air mass factor, Ctemp is a temperature correction (see Sect. 2.2.3.7) and s is the SO2 normalized a priori mixing ratio prole, as a function of pressure (p).
The AMF calculation assumes Lambertian reectors for the ground and the clouds and makes use of pre-calculated WF LUTs at 313, 326 and 375 nm (depending on the tting window used). Calculating the AMF at these three wavelengths was found to give the best results using closed-loop retrievals (see Auxiliary material of Theys et al., 2015).The WF depends on observation geometry (solar zenith angle: SZA; line-of-sight angle: LOS; relative azimuth angle: RAA), total ozone column (TO3), scene albedo (alb), surface pressure (ps), cloud top pressure (pcloud) and effective cloud fraction (feff).
Examples of SO2 weighting functions are displayed in Fig. 5 (as a function of height for illustration purpose) and show the typical variations in the measurement sensitivity as a function of height, wavelength and surface albedo.
The generation of the WF LUT has been done for a large range of physical parameters, listed in Table 4. In practice, the WF for each pixel is computed by linear interpolation of the WF LUT at the a priori prole pressure grid and using the auxiliary data sets described in the following subsections. Linear interpolations are performed along the cosine of solar and viewing angles, relative azimuth angle and surface albedo, while a nearest-neighbor interpolation is performed in surface pressure. In particular, the grid of surface pressure
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
128 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Table 4. Physical parameters that dene the WF look-up table.
Parameter Number of grid points
Grid values Symbol
Atmospheric pressure (hPa) 64 1056.77, 1044.17,1031.72, 1019.41, 1007.26, 995.25, 983.38, 971.66, 960.07, 948.62, 937.31, 926.14, 915.09, 904.18, 887.87, 866.35, 845.39, 824.87, 804.88, 785.15, 765.68, 746.70, 728.18, 710.12, 692.31, 674.73, 657.60, 640.90, 624.63, 608.58, 592.75, 577.34, 562.32, 547.70, 522.83, 488.67, 456.36, 425.80, 396.93, 369.66, 343.94, 319.68, 296.84, 275.34, 245.99, 210.49, 179.89, 153.74, 131.40, 104.80, 76.59, 55.98, 40.98, 30.08, 18.73, 8.86,4.31, 2.18, 1.14, 0.51, 0.14, 0.03, 0.01, 0.001
pl
zl
Altitude corresponding to the atmospheric pressure, using a US standard atmosphere (km)
64 0.35, 0.25, 0.15, 0.05, 0.05, 0.15, 0.25, 0.35,
0.45, 0.55, 0.65, 0.75, 0.85, 0.95, 1.10, 1.30, 1.50, 1.70,1.90, 2.10, 2.30, 2.50, 2.70, 2.90, 3.10, 3.30, 3.50, 3.70,3.90, 4.10, 4.30, 4.50, 4.70, 4.90, 5.25, 5.75, 6.25, 6.75,7.25, 7.75, 8.25, 8.75, 9.25, 9.75, 10.50, 11.50, 12.50,13.50, 14.50, 16.00, 18.00, 20.00, 22.00, 24.00, 27.50,32.50, 37.50, 42.50, 47.50, 55.00, 65.00, 75.00, 85.00,95.00
Solar zenith angle ( ) 17 0, 10, 20, 30, 40, 45, 50, 55, 60, 65, 70, 72, 74, 76, 78,80, 85
0
Line-of-sight angle ( ) 10 0, 10, 20, 30, 40, 50, 60, 65, 70, 75
Relative azimuth angle ( )
5 0, 45, 90, 135, 180 '
Total ozone column (DU) 4 205, 295, 385, 505 TO3
Surface albedo 14 0, 0.01, 0.025, 0.05, 0.075, 0.1, 0.15, 0.2, 0.25, 0.3 0.4,0.6, 0.8, 1.0
As
Surface/cloud top pressure (hPa)
17 1063.10, 1037.90, 1013.30, 989.28, 965.83, 920.58,
876.98, 834.99, 795.01, 701.21, 616.60, 540.48, 411.05, 308.00, 226.99, 165.79, 121.11
ps
column values from 200 to 500 DU for a set of typical ozone proles. The total ozone column is directly available from the operational processing of the S-5P total ozone column product.
Surface albedo
The albedo value is very important for PBL anthropogenic SO2 but less critical for volcanic SO2 when it is higher in the atmosphere. For the surface albedo dimension, we use the climatological monthly minimum Lambertian equivalent reector (minLER) data from Kleipool et al. (2008) at 328 nm for w1 and w2, and 376 m for w3. This database is based on OMI measurements and has a spatial resolution of 0.5 [notdef] 0.5 .
Note that other surface reectance databases with improved spatial resolution (more appropriate for TROPOMI) will likely become available and these data sets will be considered for next algorithmic versions.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
AMF wavelength 3 313, 326, 375
is very thin near the ground in order to minimize interpolation errors caused by the generally low albedo of ground surfaces. Furthermore, the LUT and model pressures are scaled to the respective surface pressures in order to avoid extrapolations outside the LUT range.
Observation geometry
The LUT covers the full range of values for solar zenith angles, line-of-sight angles and relative azimuth angles that can be encountered in the TROPOMI measurements. The observation geometry is readily present in the L1b data for each satellite pixel.
Total ozone column
The measurement sensitivity at 313 nm is dependent on the total ozone absorption. The LUT covers a range of ozone
N. Theys et al.: S-5P SO2 algorithm theoretical basis 129
Figure 5. SO2 box AMFs at 313, 326 and 375nm for albedo of (a) 0.06 and (b) 0.8. SZA: 40 ; LOS: 10 ; RAA: 0 ; surface height: 0 km.
Clouds
The AMF calculations for TROPOMI partly cloudy scenes use the cloud parameters (cloud fraction fc, cloud albedo
Ac, cloud top pressure ctp) supplied by the nominal S-5P cloud algorithm OCRA/ROCINN in its Clouds as Reecting Boundaries (CRB) implementation (Loyola et al., 2016).The cloud surface is considered to be a Lambertian reecting surface and the treatment of clouds is achieved through the independent pixel approximation (IPA; Martin et al., 2002), which considers an inhomogeneous satellite pixel as being composed (as for the radiance intensity) of two independent homogeneous scenes, one completely clear and the other completely cloudy. The weighting function is expressed as
m(p) = [Phi1]mcloud (p) + (1 [Phi1])mclear (p), (7)
where [Phi1] is the intensity-weighted cloud fraction or cloud radiance fraction:
[Phi1] =
feffIcloud
feffIcloud + (1 feff)Iclear
. (8)
The sufxes clear and cloudy refer to the WF and intensity calculation corresponding to a fully clear or cloudy pixel, respectively. The WF LUT is therefore accompanied by an intensity LUT with the same input grids. Both LUTs have been generated for a range of cloud cover fractions and cloud top pressures.
Note that the variations in the cloud albedo are directly related to the cloud optical thickness. Strictly speaking, in a Lambertian (reective) cloud model approach, only thick clouds can be represented. An effective cloud fraction corresponding to an effective cloud albedo of 0.8 (feff
= fc Ac0.8)
can be dened in order to transform optically thin clouds into equivalent optically thick clouds of reduced extent. Note that
in some cases (thick clouds with Ac > 0.8) the effective cloud fraction can be larger than one and the algorithm assumes feff = 1. In such altitude-dependent air mass factor calcula
tions, a single cloud top pressure is assumed within a given viewing scene. For low effective cloud fractions (feff lower than 10 %), the current cloud top pressure output is highly unstable and it is therefore reasonable to consider the observation a clear-sky pixel (i.e., the cloud fraction is set to 0 in Eq. 8) in order to avoid unnecessary error propagation through the retrievals, which can be as high as 100 %. Moreover, it has been shown recently by Wang et al. (2016) using multi-axis DOAS (MAX-DOAS) observations to validate satellite data that, in the case of elevated aerosol loadings in the PBL (typically leading to apparent feff up to 10 %), it is recommended to apply clear-sky AMFs rather than total AMFs (based on cloud parameters) that presumably correct implicitly for the aerosol effect on the measurement sensitivity.
It should be noted that the formulation of the pressure-dependent air mass factor for a partly cloudy pixel implicitly includes a correction for the SO2 column lying below the cloud and therefore not seen by the satellite, the so-called ghost column. Indeed, the total AMF calculation as expressed by Eqs. (6) and (7) assumes the same shape factor and implies an integration of the a priori prole from the top of the atmosphere to the ground, for each fraction of the scene. The ghost column information is thus coming from the a priori prole shapes. For this reason, only observations with moderate cloud fractions (feff lower than 30 %)
are used, unless it can be assumed that the cloud cover is mostly situated below the SO2 layer, i.e., a typical situation for volcanic plumes injected into the upper troposphere or lower stratosphere.
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
pixel. The TM5 SO2 prole is shifted to start at ps and scaled so that volume mixing ratios are preserved (see Zhou et al., 2009).
Temperature correction
The SO2 absorption cross sections of Bogumil et al. (2003) show a clear temperature dependence which has an impact on the retrieved SO2 SCDs depending on the tting window used. However, only one temperature (203 K) is used for the DOAS t, therefore a temperature correction needs to be applied: SCD = Ctemp.SCD. While the SO2 algorithm pro
vides vertical column results for a set of a priori proles, applying this correction to the slant column is not simple and as a workaround it is preferred to apply the correction directly to the AMFs (or box AMFs to be precise) while keeping the (retrieved) SCD unchanged: AMF = AMF/Ctemp. This for
mulation implicitly assumes that the AMF is not strongly affected by temperature, which is a reasonable approximation (optically thin atmosphere). The correction to be applied requires a temperature prole for each pixel (which is obtained from the TM5 model):
Ctemp = 1/[notdef]1 .(T [notdef]K[notdef] 203)[notdef], (10)
where equals 0.002, 0.0038 and 0 for the tting windows 312326, 325335 and 360390 nm, respectively. The parameter has been determined empirically by tting Eq. (10) through a set of data points (Fig. 6), for each tting window.Each value in Fig. 6 is the slope of the tting line between the SO2 differential cross sections at 203 K vs. the cross section at a given temperature. In the tting window 360 390 nm, no temperature correction is applied ( = 0) because
the cross sections are quite uncertain. Moreover, the 360 390 nm wavelength range is meant for extreme cases (strong volcanic eruptions) for SO2 plumes in the lower stratosphere, where a temperature of 203 K is a good baseline.
Aerosols
The presence of aerosol in the observed scene (likely when observing anthropogenic pollution or volcanic events) may affect the quality of the SO2 retrieval (e.g., Yang et al., 2010). No explicit treatment of aerosols (absorbing or not) is foreseen in the algorithm as there is no general and easy way to treat the aerosols effect on the retrieval. At processing time, the aerosol parameters (e.g., extinction prole or single-scattering albedo) are unknown. However, the information on the S-5P UV absorbing aerosol index (AAI) by Zweers (2016) will be included in the L2 SO2 les as it gives information to the users on the presence of aerosols for both anthropogenic and volcanic SO2. Nevertheless, the
AAI data should be used/interpreted with care. In an ofine future version of the SO2 product, absorbing aerosols might be included in the forward model, if reliable information on
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
130 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Surface height
The surface height (zs) is determined for each pixel by interpolating the values of a high-resolution digital elevation map, GMTED2010 (Danielson and Gesch, 2011).
Prole shapes
It is generally not possible to know at the time of observation what the SO2 vertical prole is and whether the observed
SO2 is of volcanic origin or from pollution (or both). Therefore, the algorithm computes four vertical columns for different hypothetical SO2 proles.
Three box proles of 1 km thickness, located in the boundary layer, upper troposphere and lower stratosphere, are used.The rst box prole stands for typical conditions of well mixed SO2 (from volcanic or anthropogenic emissions) in the boundary layer, while the upper-troposphere and lower-stratosphere box proles are representative of volcanic SO2 plumes from effusive and explosive eruptions, respectively.
In order to have more realistic SO2 proles for polluted scenes, daily forecasts calculated with the global TM5 chemical transport model (Huijnen et al., 2010) will also be used. TM5 will be operated with a spatial resolution of 1 [notdef] 1 in latitude and longitude, and with 34 sigma
pressure levels up to 0.1 hPa in the vertical direction. TM5 will use 3 h meteorological elds from the European Centre for Medium-Range Weather Forecast (ECMWF) operational model (ERA-Interim reanalysis data for reprocessing, and the operational archive for real time applications and forecasts). These elds include global distributions of wind, temperature, surface pressure, humidity, (liquid and ice) water content, and precipitation. A more detailed description of the TM5 model is given at http://tm.knmi.nl/
Web End =http://tm.knmi.nl/ and by van Geffen et al. (2016).
For the calculation of the air mass factors, the proles are linearly interpolated in space and time, at the pixel center and S-5P local overpass time, through a model time step of 30 min. For NRT processing, the daily forecast of the TM5 model (located at KNMI) will be ingested by the UPAS operational processor.
To reduce the errors associated to topography and the lower spatial resolution of the model compared to the TROPOMI 7 km [notdef] 3.5 km spatial resolution, the a priori pro
les need to be rescaled to effective surface elevation of the satellite pixel. The TM5 surface pressure is converted by applying the hypsometric equation and the assumption that temperature changes linearly with height (Zhou et al., 2009):
ps = pTM5(
TTM5
(TTM5 + [Gamma1](zTM5 zs))
)
g
R[Gamma1] (9)
where pTM5 and TTM5 are the TM5 surface pressure and temperature, [Gamma1] = 6.5 K km1 the lapse rate, zTM5 the TM5 ter
rain height, and zs surface elevation for the satellite ground
N. Theys et al.: S-5P SO2 algorithm theoretical basis 131
Figure 6. Effect of temperature (relative to 203 K) on SO2 retrieved SCD for tting windows 312326 (left) and 325335 nm (right). The red lines show the adopted formulation of Ctemp (Eq. 10). Note that, for the 312326 nm window, the result at 273K has been discarded from the t as it is seems rather inconsistent with the dependence at other temperatures.
absorbing aerosol can be obtained from the AAI and the S-5P aerosol height product (Sanders and de Haan, 2016).
3 Error analysis
3.1 Introduction
The total uncertainty (accuracy and precision) on the SO2 columns produced by the algorithm presented in Sect. 2 is composed of many sources of error (see also, e.g., Lee et al., 2009). Several of them are related to the instrument, such as uncertainties due to noise or knowledge of the slit function. These instrumental errors propagate into the uncertainty on the slant column. Other types of error can be considered model errors and are related to the representation of the physics in the algorithm. Examples of model errors are uncertainties on the trace gas absorption cross sections and the treatment of clouds. Model errors can affect the slant column results or the air mass factors.
The total retrieval uncertainty on the SO2 vertical columns can be derived by error propagation, starting from Eq. (1) and if one assumes uncorrelated retrieval steps (Boersma et al., 2004; De Smedt et al., 2008):
2Nv = [parenleftBig]
The error analysis is complemented by the total column averaging kernel (AK) as described in Eskes and Boersma (2003):
AK(p) =
m[prime](p)
M (12)
(m[prime] is the weighting function, Eq. 6), which is often used to characterize the sensitivity of the retrieved column to a change in the true prole.
3.2 Error components
The following sections describe and characterize 20 error contributions to the total SO2 vertical column uncertainty.
These different error components and corresponding typical values are summarized in Tables 5 and 6. Note that, at the time of writing, the precise effect of several S-5P-specic error sources are unknown and will be estimated during operations.
A difculty in the error formulation presented above comes from the fact that it assumes the different error sources/steps of the algorithm to be independent and uncor-related, which is not strictly valid. For example, the background correction is designed to overcome systematic features/deciencies of the DOAS slant column tting, and these two steps cannot be considered independent. Hence, summing up all the corresponding error estimates would lead to overestimated error bars. Therefore, several error sources will be discussed in the following subsections without giving actual values at this point. Their impact is included and described in later subsections.
Another important point to note is that one should also (be able to) discriminate systematic and random components of
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
NS
M
2+
Nback S M
2+
NS NbackS
[parenrightbig]
M M2
!2, (11)
where Ns and backNs are the errors on the slant column NS and on the background correction NbackS, respectively.
132 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Table 5. Systematic and random error components contributing to the total uncertainty on the SO2 slant column.
# Error source Type* Parameter uncertainty Typical uncertainty on SO2 SCD
1 SO2 absorption S 6 % (window 1) 6 % cross section 6 % (window 2)
unknown (window 3)
2 SO2 and O3 absorption S & R Errors 9 & 10
3 Other atmospheric S & R Error 9 absorption or interference
4 Radiance shot noise R S/N = 8001000 0.30.5 DU (window 1)
5 DU (window 2)
60 DU (window 3)
5 DOAS settings S 1 nm, polynomial order < 11 % (window 1) < 6 % (window 2) < 8 % (window 3)
6 Wavelength and S Wavelength calibration Wavelength calibration and spectral shifts radiometric can be corrected by thecalibration algorithm to less than 5 % effecton the slant column
Radiometric calibration Intensity offset correction in Additive errors should principle treats (small) remain below 2 % radiometric calibration errors
7 Spectral response TBD TROPOMI-specic function Expected uncertainty: 10 %
8 Other spectral Strongly dependent features on interfering signal
9 Background S & R 0.2 DU correction
R: random; S: systematic.
about 20 % + 0.2 DU of the background-corrected slant col
umn (Ns,syst = 0.2 [notdef] (Ns Nbacks)+ 0.2 DU).
For the random component of the slant column errors, the error on the slant columns provided by the DOAS t is considered (hereafter referred to as SCDE) as it is assumed to be dominated by and representative of the different random sources of error.
Error source 1: SO2 cross section
Systematic errors on slant columns due to SO2 cross-section uncertainties are estimated to be around 6 % (Vandaele et al., 2009) in window 1 (312326 nm) and window 2 (325 335 nm) and unknown in window 3 (360390 nm). In addition, the effect of the temperature on the SO2 cross sections has to be considered as well. Refer to Sect. 3.2.2 for a discussion of this source of error.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
a given error source V:
2V =
2V(rand)
n + 2V(syst), (13)
here n is the number of pixels considered. However, they are hard to separate in practice. Therefore, each of the 20 error contributions are (tentatively) classied as either random or systematic errors, depending on their tendencies to average out in space/time or not.
3.2.1 Errors on the slant column
Error sources that contribute to the total uncertainty on the slant column originate both from instrument characteristics and uncertainties/limitations on the representation of the physics in the DOAS slant column tting algorithm. For the systematic errors on the slant column, the numbers provided in Table 5 have been determined based on sensitivity tests (using the QDOAS software).
With all effects summed in quadrature, the various contributions are estimated to account for a systematic error of
N. Theys et al.: S-5P SO2 algorithm theoretical basis 133
Table 6. Systematic and random error components contributing to the total uncertainty on the SO2 air mass factor.
# Error Type1 Parameter Typical uncertainty on the AMF uncertainty
10 AMF wavelength S 10 % dependence
11 Model atmosphere S O3 prole 510 %
P , T proles small
12 Forward model S < 5 % < 5 %
13 Surface albedo2 S 0.02 15 % (PBL) 5 % (FT)1 % (LS)
14 Cloud fraction2 R 0.05 5 % (PBL) 15 % (FT) 1 % (LS)
15 Cloud top pressure2 R 50 hPa 50 % (PBL) 50 % (FT)1 % (LS)
16 Cloud correction R < 5 % on yearly averaged data
17 Cloud model TBD
18 SO2 prole shape S anthropogenic SO2
2035 %
volcanic SO2 large (low albedo), < 50 %(high albedo)
19 Aerosol S & R Anthropogenic SO2 15 %
(Nowlan et al., 2011).
Volcanic SO2 (aerosols: ash/sulfate):
20 %(Yang et al., 2010)
20 Temperature R 5 %
correction
1 R: random; S: systematic. 2 Effect on the AMF estimated from Fig. 6.
Error source 2: O3 and SO2 absorption
Nonlinear effects due to O3 absorption are to a large extent accounted for using the Taylor expansion of the O3 optical depth (Pukte et al., 2010). Remaining systematic biases are then removed using the background correction; hence, residual systematic features are believed to be small (please read also the discussion on errors 9 and 10). The random component of the slant column error contributes to SCDE.
Nonlinear effects due to SO2 absorption itself (mostly for volcanic plumes) are largely handled by the triple windows retrievals, but as will be discussed in Sect. 4 the transition between the different tting windows is a compromise and there are cases where saturation can still lead to rather
large uncertainties. However, those are difcult to assess on a pixel-to-pixel basis.
Error source 3: other atmospheric absorption/interferences
In some geographical regions, several systematic features in the slant columns remain after the background correction procedure (see discussion on error 9: background correction error) and are attributed to spectral interferences not fully accounted for in the DOAS analysis, such as incomplete treatment of the Ring effect. This effect also has a random component and contributes to the retrieved SCD error (SCDE).
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
134 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Error source 4: radiance shot noise
It has a major contribution to the SCDE and it can be estimated from typical S/N values of S-5P in UV band 3 (800 1000, according to Veefkind et al., 2012). This translates to typical SCD random errors of about 0.30.5, 5 and 60 DU for window 1, 2 and 3, respectively. Note that real measurements are needed to consolidate these numbers.
Error source 5: DOAS settings
Tests on the effect of changing the lower and upper limits of the tting windows by 1 nm and the order of the closure polynomial (4 instead of 5) have been performed. Based on a selection of orbits for the Kasatochi eruption (wide range of measured SCDs), the corresponding SCD errors are less than 11, 6 and 8 % for window 1, 2 and 3, respectively.
Error source 6: wavelength and radiometric calibration
Tests on the effect of uncertainties in the wavelength calibration have been performed in the ESA CAMELOT study. The numbers are for a shift of 1/20th of the spectral sampling in the solar spectrum and 1/100th of the spectral sampling in the earthshine spectrum. The shift can be corrected for, but interpolation errors can still lead to a remaining uncertainty of a few percent.
Regarding radiometric calibration, the retrieval result is in principle insensitive to at (spectrally constant) offsets on the measured radiance because the algorithm includes an intensity offset correction. From the ESA ONTRAQ study it was found that additive error signals should remain within 2 % of the measured spectrum.
Error source 7: spectral response function
Uncertainties in the S-5P instrumental slit functions can lead to systematic errors on the retrieved SO2 slant columns (to be determined).
Error source 8: other spectral features
Unknown or untreated instrumental characteristics such as stray light and polarization sensitivity can introduce spectral features that may lead to bias in the retrieved slant column data. To a certain extent these can be prevented by the DOAS polynomial and the intensity offset correction settings, as long as the perturbing signals are a smooth function of wavelength. Conversely, high-frequency spectral structures can have potentially a large impact on SO2 retrievals depending on their amplitude and whether they interfere with SO2 absorption structures. At the time of writing, it is hard to evaluate these measurement errors. Once the instrument will be operating, such measurement errors will be characterized and correct for, either as part of L1b processor or in the form of pseudo-absorption cross sections in the DOAS analysis.
In the ONTRAQ study, testing sinusoidal perturbation signals showed that the effect of spectral features on the retrieval result depends strongly on the frequency of the signal. Additive signals with an amplitude of 0.05 % of the measurement affect the retrieved SO2 slant column by up to 30 %. The effect scales more or less linearly with the signal amplitude.
Error source 9: background/destriping correction
This error source is mostly systematic and important for anthropogenic SO2 or for monitoring degassing volcanoes.
Based on OMI and GOME-2 test retrievals, the uncertainty on the background correction is estimated to be < 0.2 DU. This value accounts for limitations of the background correction and is compatible with residual slant columns values typically found (after correction) in some clean areas (e.g., above the Sahara), or for a possible contamination by volcanic SO2, after a strong eruption.
3.2.2 Errors on the air mass factor
The error estimates on the AMF are listed in Table 6 and are based on simulations and closed-loop tests using the radiative transfer code LIDORT. One can identify two sources of errors on the AMF. First, the adopted LUT approach has limitations in reproducing the radiative transfer in the atmosphere (forward model errors). Secondly, the error on the AMF depends on input parameter uncertainties. This contribution can be broken down into a squared sum of terms (Boersma et al., 2004):
2M = [parenleftbigg]
@M
@alb [notdef] alb[parenrightbigg]2 + [parenleftbigg]
@M
@ctp [notdef] ctp[parenrightbigg]2
+
[parenleftbigg]
@M
@feff [notdef] feff[parenrightbigg]2 + [parenleftbigg]
2, (14)
where alb, ctp, feff, s are typical uncertainties on the albedo, cloud top pressure, cloud fraction and prole shape, respectively.
The contribution of each parameter to the total air mass factor error depends on the observation conditions. The air mass factor sensitivities ( @M
@parameter ), i.e., the air mass factor derivatives with respect to the different input parameters, can be derived for any particular condition of observation using the altitude-dependent AMF LUT, created with LIDORTv3.3, and using the a priori prole shapes. In practice, a LUT of AMF sensitivities has been created using reduced grids from the AMF LUT and a parameterization of the prole shapes based on the prole shape height.
Error source 10: AMF wavelength dependence
Because of strong atmospheric absorbers (mostly ozone) and scattering processes, the SO2 AMF shows a wavelength dependence. We have conducted sensitivity tests to determine
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
@M
@s [notdef] s
N. Theys et al.: S-5P SO2 algorithm theoretical basis 135
Error source 12: radiative transfer model
This error source is believed to be small, less than 5 % (Hendrick et al., 2006; Wagner et al., 2007).
Error source 13: surface albedo
A typical uncertainty on the albedo is 0.02 (Kleipool et al., 2008). This translates to an error on the air mass factor after multiplication by the slope of the air mass factor as a function of the albedo (Eq. 14) and can be evaluated for each satellite pixel. As an illustration, Fig. 8 shows the expected dependence of the AMF with albedo and also with the cloud conditions. From Fig. 8a, one concludes that the retrievals of SO2 in the BL are much more sensitive to the exact albedo value than for SO2 higher up in the atmosphere, for this particular example.
More substantial errors can be introduced if the real albedo differs considerably from what is expected, for example in the case of the sudden snowfall or ice cover. The snow/ice cover ag in the L2 le will therefore be useful for such cases.
Error source 14: cloud fraction
An uncertainty on the cloud fraction of 0.05 is considered.
The corresponding AMF error can be estimated through Eq. (14; see Fig. 8b) or by analytic derivation from Eqs. (6)(8).
Error source 15: cloud top pressure
An uncertainty on the cloud top height of 0.5 km ( 50 hPa)
is assumed. The corresponding AMF error can be estimated through Eq. (14). Figure 8c illustrates the typical behavior of signal amplication/shielding for a cloud below/above the SO2 layer. One can see that the error (slope) dramatically increases when the cloud is at a height similar to the SO2 bulk altitude.
Error source 16: cloud correction
Sensitivity tests showed that applying the independent pixel approximation or assuming cloud-free pixels makes a difference of only 5 % on yearly averaged data (for anthropogenic BL SO2 VC with cloud fractions less than 40 %).
Error source 17: cloud model
Cloud as layer (CAL) is the baseline of the S-5P cloud algorithm, but a Lambertian equivalent reector (LER) implementation will be used for NO2, SO2 and HCHO retrievals.
The error due to the choice of the cloud model will be evaluated during the operational phase.
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
Figure 7. Retrieved SO2 slant columns versus simulated SCDs at a wavelength of 313 nm from synthetic spectra (SZA: 30, 70 ) in the spectral range 312326 nm and for SO2 layers in the boundary layer, upper troposphere and lower stratosphere. The different points correspond to different values for the line-of-sight angle (0, 45 ), surface albedo (0.06, 0.8), surface height (0, 5 km) and total ozone column (350, 500 DU). SO2 vertical columns as input of the
RT simulations are a maximum of 25 DU.
the optimal wavelengths for AMF calculations representative of each of the three tting windows. To do so, synthetic radiances and SO2 SCDs have been generated using LIDORT for typical observations scenarios and at spectral resolution and sampling compatible with S-5P. The spectra have been analyzed by DOAS and the retrieved SCDs have been compared to the calculated SCDs at different wavelengths. It appears from this exercise that 313, 326 and 375 nm provide the best results, for window 1, 2 and 3, respectively. Figure 7 shows an illustration of these sensitivity tests in the baseline window; an excellent correlation and slope close to 1 is found for the scatter plot of retrieved versus simulated slant columns using an effective wavelength of 313 nm for the AMF. Overall, for low solar zenith angles, the deviations from the truth are less than 5 % in most cases, except for boundary layer (BL) SO2 at a 1 DU column level and for low-albedo scenes (deviations up to 20 %). For high solar zenith angles deviations are less than 10 % in most cases, except for BL SO2 at a 1 DU column level and for low-albedo scenes (underestimation by up to a factor of 2).
Error source 11: model atmosphere
This error relates to uncertainties in the atmospheric proles used as input of LIDORT for the weighting function lookup-table calculations.
Although the effect of O3 absorption on the AMF is treated in the algorithm, the O3 proles used as input of LIDORT are not fully representative of the real proles and typical errors (including error due to interpolation) of 510 % can occur.
A test has been performed by replacing the US standard atmosphere pressure and temperature proles by high latitude winter proles and the impact on the results is found to be small.
136 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Figure 8. Air mass factors at 313 nm for SO2 in the boundary layer (BL: 01 km), free troposphere and lower stratosphere (FT, LS: Gaussian proles with maximum height at 6,15 km; FWHM: 1 km). Calculations are for SZA = 40 , Los = 10 , RAA = 0 and surface height = 0 km.
AMFs are displayed as a function of the (a) albedo for clear-sky conditions, (b) cloud fraction for albedo = 0.06, cloud albedo = 0.8 and
cloud top height = 2 km and (c) cloud top height for albedo = 0.06, cloud albedo = 0.8 and cloud fraction = 0.3.
Error source 18: prole shape
A major source of systematic uncertainty for most SO2 scenes is the shape of the vertical SO2 distribution. The corresponding AMF error can be estimated through Eq. (14) and estimation of uncertainty on the prole shape. Note that vertical columns are provided with their averaging kernels, so that column data might be improved for particular locations by using more accurate SO2 prole shapes based on input from models or observations.
For anthropogenic SO2 under clear-sky conditions, sensitivity tests using a box prole from 0 to 1 [notdef] 0.5 km
above ground level, or using the different proles from the
CAMELOT study (Levelt et al., 2009), give differences in AMFs in the range of 2035 %. Note that for particular conditions SO2 may also be uplifted above the top of the boundary layer and sometimes reach upper-tropospheric levels (e.g., Clarisse et al., 2011). SO2 weighting functions displayed in Fig. 5 show that the measurement sensitivity is then increased by up to a factor of 3 and therefore constitutes a major source of error.
In the SO2 algorithm, the uncertainty on the prole shape is estimated using one parameter describing the shape of the TM5 prole: the prole height, i.e., the altitude (pressure) below which resides 75 % of the integrated SO2 prole. @M@s is approached by @M@sh , where sh is half of the prole height.
Relatively small variations in this parameter have a strong impact on the total air mass factors for low-albedo scenes, because altitude-resolved air mass factors decrease strongly in the lower troposphere, where the SO2 proles peak (see, e.g., Fig. 5).
For volcanic SO2, the effect of the prole shape uncertainty depends on the surface or cloud albedo. For low-albedo scenes (Fig. 5a), if no external information on the SO2 plume height is available, it is a major source of error at all wavelengths. Vertical columns may vary by up to a factor of 5. For high-albedo scenes (Fig. 5b), the error is less than 50 %. It should be noted that these conditions are often encountered for strong eruptions injecting SO2 well above the cloud deck (high reectivity). Further uncertainty on the retrieved SO2 column may arise if the vertical distribution shows dis-
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 137
tinct layers at different altitudes, due to the different nature of successive phases of the eruption.
In the SO2 algorithm, three 1 km thick box proles are used in the AMF calculations, mostly to represent typical volcanic SO2 proles. The error due to the prole shape uncertainty is estimated by varying the box center levels by 100 hPa.
Error source 19: aerosols
The effect of aerosols on the air mass factors are not explicitly considered in the SO2 retrieval algorithm. To some extent, however, the effect of the non-absorbing part of the aerosol extinction is implicitly included in the cloud correction (Boersma et al., 2004). Indeed, in the presence of aerosols, the cloud detection algorithm is expected to overestimate the cloud fraction, resulting partly in a compensation effect for cases where aerosols and clouds are at similar heights. Absorbing aerosols have a different effect on the air mass factors and can lead to signicant errors for high aerosol optical depths (AODs). In the TROPOMI SO2 product, the absorbing aerosol index eld can be used to identify observations with elevated absorbing aerosols.
Generally speaking, the effect of aerosols on AMF is highly variable and strongly depends on aerosol properties (AOD, height and size distribution, single-scattering albedo, scattering phase function, etc.). Typical AMFs uncertainties due to aerosols found in the literature are given in Table 6. As aerosols affect cloud fraction, cloud top height and to some extent the albedo database used, correlations between uncertainties on these parameters are to be expected.
Error source 20: temperature correction
The DOAS scheme uses an SO2 cross section at only one temperature (Bogumil et al., 2003, at 203 K), which is in general not representative of the effective temperature corresponding to the SO2 vertical prole. This effect is in principle accounted for by the temperature correction (which is applied in practice to the AMFs; see Sect. 2.2.3.7) but with a certain associated error of 5 %.
4 Verication
The SO2 retrieval algorithm presented in Sect. 2, and hereafter referred as prototype algorithm, has been applied to OMI and GOME-2 spectra. The results have been extensively veried and validated against different satellite and ground-based data sets (e.g., Theys et al., 2015; Fioletov et al., 2016;Wang et al., 2016). Here we report on further scientic verication activities that took place during the ESA S-5P L2WG project.
In addition to the prototype algorithm, a scientic algorithm (referred as verication algorithm) has been developed in parallel. Both algorithms have been applied to syn-
thetic and real (OMI) spectra and results were compared. In this study, we only present and discuss a selection of results (for OMI).
4.1 Verication algorithm
The S-5P TROPOMI verication algorithm was developed in close cooperation between the Max Planck Institute for Chemistry (MPIC) in Mainz (Germany) and the Institut fr Methodik und Fernerkundung as part of the Deutsches Institut fr Luft- und Raumfahrt Oberpfaffenhofen (DLR-IMF). Like the prototype algorithm (PA), the verication algorithm (VA) uses a multiple tting window DOAS approach to avoid nonlinear effects during the SCD retrieval in the case of high SO2 concentrations in volcanic plumes. However, especially the alternatively used tting windows differ strongly from the ones used for the PA and are entirely located in the lower UV range:
312.1324 nm (standard retrieval SR): similar to baseline PA tting window, ideal for small columns
318.6335.1 nm (medium retrieval MR): this tting window is essentially located in between the rst and second tting window of the PA and was mainly introduced to guarantee a smoother transition between the baseline window and the one used for high SO2 concentrations. The differential SO2 spectral features are still about 1 order of magnitude smaller than in the baseline window.
323.1335.1 nm (alternative retrieval AR): similar to the intermediate tting window of the PA. This tting window is used in the case of high SO2 concentrations.
Although it is expected that volcanic events with extreme SO2 absorption are still affected by nonlinear absorption in this window, the wavelength range is sufcient for most volcanic events.
Furthermore, the VA selection criteria for the transition from one window to another are not just based on xed SO2 SCD thresholds. The algorithm allows for a slow and smooth transition between different t ranges by linearly decreasing the weight of the former tting window and at the same time increasing the weight of the following tting window:
1. for SO2 SCD 4 [notdef] 1017 molec cm2 ( 15 DU):
SO2 SCD = SR;
2. for 4 [notdef] 1017 molec cm2 < SO2 SCD < 9 [notdef] 1017 molec
cm2:
SO2 SCD = SR [notdef]
1
SR9 [notdef] 1017 moleccm2 [bracketrightbigg]
+MR [notdef]
SR
[bracketleftbigg][parenleftBigg]9 [notdef] 1017 moleccm2 [bracketrightbigg];
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
138 N. Theys et al.: S-5P SO2 algorithm theoretical basis
3. for SO2 SCD 9 [notdef] 1017 molec cm2 ( 33 DU): SO2 SCD = MR;
4. for 9 [notdef] 1017 molec cm2 < SO2 SCD < 4.6 [notdef] 1018
molec cm2:
SO2 SCD = MR [notdef]
1
MR4.6 [notdef] 1018 moleccm2 [bracketrightbigg]
+AR [notdef]
the a priori settings. Main deviations between prototype and verication algorithm are therefore expected to be caused by the usage of different t windows (determining their sensitivity and t error) and especially the corresponding transition criteria.
Figure 9 shows the resulting maps of the SO2 VCD for the VA (upper panels) and PA (lower panels) for the three selected test cases. As can be seen, both algorithms result in similar SO2 VCDs; however, a closer look reveals some differences, such as the maximum VCDs, which are not necessarily appearing at the same locations. For the Anatahan case for instance, the maximum VCD is seen closer to the volcano at the eastern end of the plume for the PA, while it appears to be further downwind for the VA. This effect can be explained by the corresponding t windows used for both algorithms, which may result in deviating SO2 VCDs, especially for SO2 scenarios where the best choice is difcult to assess. This is illustrated in Fig. 10, showing scatter plots of VA versus PA SO2 VCDs for the three test cases (Anatahan, Norilsk and
Kasatochi) color-coded differently depending on the tting window used for VA (left) and PA (right), respectively. While the PA uses strictly separated results from the individual t windows, the VA allows a smooth transition whenever resulting SO2 SCDs are found to be located in between subsequent t ranges.
For all three test cases, it appears that the PA is less affected by data scattering for low SO2 or SO2 free measurements than the VA. For the shortest UV t windows, both algorithms mainly agree, but VA VCDs tend to be higher by 1015 % than the PA VCDs for the Anatahan and Kasatochi measurements but interestingly not for the Norilsk case. For SO2 VCDs around 7 DU the PA seems to be slightly affected by saturation effects in 312326 nm window, while VA already makes use of a combined SR/MR SCD. For larger SO2
VCDs (> 10 DU), data sets from both algorithms show an increased scattering, essentially resulting from the more intensive use of tting windows at longer wavelengths (for which the SO2 absorption is weaker). While it is difcult to conclude which algorithm is closer to the actual SO2 VCDs, the combined t windows of the VA probably are better suited (in some SO2 column ranges) for such scenarios as the
SO2 cross section is generally stronger for lower wavelength (< 325 nm) when compared to the intermediate t window of the PA.
For extremely high SO2 loadings, i.e., for the Kasatochi plume on 8 August 2008, the DOAS retrievals from PA and VA require all three t windows to prevent systematic underestimation of the resulting SO2 SCDs due to nonlinear absorption caused by very high SO2 concentrations within the volcanic plume. Figure 9 (right panel) shows that the SO2 distribution is similar for both algorithms, including the location of the maximum SO2 VCD.
From Fig. 10 (lowest panel), it can be seen that the VA shows higher values for SO2 VCDs < 100 DU, for all three t windows. For very high SO2 VCDs, it seems that the ver-
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
AR
[bracketleftbigg][parenleftBigg]4.6 [notdef] 1018 moleccm2 [bracketrightbigg];5. for SO2 SCD 4.6 [notdef] 1018 molec cm2 ( 171 DU): SO2 SCD = AR.
To convert the nal SO2 SCDs into vertical column densities, a single-wavelength AMF for each of the three tting windows (SO2 SR, MR and AR) is calculated using the LI
DORT LRRS v2.3 (Spurr et al., 2008). The AMF depends on the viewing angles and illumination, surface and cloud conditions as well as on the O3 total column, which is taken from the O3 total column retrieval. A cloudy and clear-sky AMF is calculated using temperature-dependent cross sections for SO2 (Bogumil et al., 2003) and O3 (Brion et al.,
1984): AMF( ) =
ln
I+SO2 ISO2
[parenrightbigg]
SO2 with (I+SO2) and (ISO2) being simulated earthshine spectra with and without including SO2 as a trace gas, respectively. Both AMFs are combined using the cloud fraction information. Like the PA, the VA is calculated for different a priori SO2 proles (center of mass at 2.5, 6 and 15 km) and a temperature correction is applied (see Sect. 2.2.3.7). In contrast to the PA the VA uses Gaussian-shaped SO2 proles with a FWHM of 2.5km rather than box proles as in the PA. This choice, however, has only a minor inuence on the AMF.
For further details on the VA, the reader is referred to the S-5P Science Verication Report (available at https://earth.esa.int/web/sentinel/user-guides/sentinel-5p-tropomi/document-library/-/asset_publisher/w9Mnd6VPjXlc/content/sentinel-5p-tropomi-science-verification-report
Web End =https:// https://earth.esa.int/web/sentinel/user-guides/sentinel-5p-tropomi/document-library/-/asset_publisher/w9Mnd6VPjXlc/content/sentinel-5p-tropomi-science-verification-report
Web End =earth.esa.int/web/sentinel/user-guides/sentinel-5p-tropomi/ https://earth.esa.int/web/sentinel/user-guides/sentinel-5p-tropomi/document-library/-/asset_publisher/w9Mnd6VPjXlc/content/sentinel-5p-tropomi-science-verification-report
Web End =document-library/-/asset_publisher/w9Mnd6VPjXlc/ https://earth.esa.int/web/sentinel/user-guides/sentinel-5p-tropomi/document-library/-/asset_publisher/w9Mnd6VPjXlc/content/sentinel-5p-tropomi-science-verification-report
Web End =content/sentinel-5p-tropomi-science-verication-report ) for more detailed description and results.
4.2 Verication results
For the intercomparison, the prototype algorithm and verication algorithm were applied to OMI data for three different SO2 emission scenarios: moderate volcanic SO2 VCDs on 1
May 2005, caused by the eruption of the Anatahan volcano, elevated anthropogenic SO2 VCDs, on 1 May 2005, from the
Norilsk copper smelter (Russia), and strongly enhanced SO2 VCDs, on 8 August 2008, after the massive eruption of Mt. Kasatochi.
In the following, both algorithms use the same assumption of an SO2 plume located at 15 km altitude for the AMF calculation. Even if this choice is not realistic for some of the presented scenarios, it minimizes the inuence of differences in
N. Theys et al.: S-5P SO2 algorithm theoretical basis 139
Figure 9. OMI SO2 VCD (expressed in DU) for the verication (upper panels) and prototype algorithms (lower panels) for the three selected scenarios: during the Anatahan eruption (left), over the Norilsk copper smelter area (center) and for the volcanic eruption of Kasatochi (right).
Note that, for each case, the color bar has been scaled to the maximum SO2 VCD from both algorithms.
ication algorithm is already slightly affected by an underestimation of the SO2 VCD caused by nonlinear radiative transfer effects in the SO2 AR t window, while the PA retrievals in the 360390 nm t range are insensitive to saturation effects. We note, however, that the Kasatochi plume also contained signicant amounts of volcanic ash and we cannot rule out a possible retrieval effect of volcanic ash on the observed differences between PA and VA SO2 results.
Finally ,we have also investigated other cases with extreme concentrations of SO2, and contrasting results were found compared to the Kasatochi case. For example, on 4 September 2014, PA retrieved up to 260 DU of SO2 during the Icelandic Brarbunga ssure eruption, while VA only found 150 DU (not shown). Compared to Kasatochi, we note that this specic scenario is very different as for the plume height (the SO2 plume was typically in the lowermost troposphere
3 km a.s.l.) and it is likely to play a role in the discrepancy between PA and VA results.
In summary, we found that the largest differences between prototype and verication algorithms are due to the tting window transitions and differences of measurement sensitivity of the tting windows used (all subject differently to nonlinear effects). Verication results have shown that the prototype algorithm produces reasonable results for all the expected scenarios, from modest to extreme SO2 columns, and are therefore adequate for treating the TROPOMI data. In a future processor update, the method could, however, be rened.
5 Validation of TROPOMI SO2 product
In this section, we give a brief summary of possibilities (and limitations) to validate the TROPOMI SO2 product with independent measurements.
Generally speaking, the validation of a satellite SO2 column product is a challenge for several reasons, on top of which is the representativeness of the correlative data when compared to the satellite retrievals. Another reason comes from the wide range of SO2 columns in the atmosphere that vary from about 1 DU level for anthropogenic SO2 and low-level volcanic degassing to 101000 DU for medium to extreme volcanic explosive eruptions.
The space-borne measurement of anthropogenic SO2 is difcult because of the low column amount and reduced measurement sensitivity close to the surface. The SO2 signal is covered by the competing O3 absorption and the column accuracy is directly affected by the quality of the background correction applied. Among the many parameters of the SO2 retrieval algorithm that affect the results, the SO2 vertical prole shape is of utmost importance for any comparison with correlative data. The SO2 column product accuracy is also directly impacted by the surface albedo used as input for the AMF calculation, the cloud correction/ltering and aerosols. In principle, all these effects will have to be addressed in future validation efforts.
The measurement of volcanic SO2 is facilitated by SO2 columns often larger than for anthropogenic SO2. However,
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
140 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Figure 10. OMI SO2 VCD (DU) scatter plots for PA (x axis) and VA (y axis) for the three test cases: the Anatahan eruption, Norislk anthropogenic emissions and the Kasatochi eruption (from top to bottom). The different t windows used for both algorithms are color-coded: VA on left panels (blue: SR; purple: SR/MR; green: MR; orange: MR/AR; red: AR), PA on right panels (blue: 312326 nm; green: 325335 nm; red: 360390 nm). For the three scenarios, the prototype and verication algorithms agree fairly well with r2 0.9.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
but this requires techniques to convert satellite SO2 vertical column into mass uxes (see, e.g., Theys et al., 2013, and references therein; Beirle et al., 2014). Similarly, fast-sampling UV cameras are becoming increasingly used to measure and invert SO2 uxes and are also relevant to validate TROPOMI SO2 data over volcanoes or anthropogenic point sources (e.g., power plants). It should be noted, however, that ground-based remote sensing instruments operating nearby SO2 point sources are sensitive to newly emitted SO2 plumes, while a satellite sensor like TROPOMI will measure aged plumes that have been signicantly depleted in SO2. While in some cases it is possible to compensate for this effect by estimating the SO2 lifetime, e.g., directly from the space measurements (Beirle et al., 2014), the general situation is that the SO2 loss rate is highly variable (especially in volcanic environments), and this can lead to strong discrepancies when comparing satellite and ground-based SO2 uxes.
In addition to optical devices, there are also in situ instruments measuring surface SO2 mixing ratios. This type of instrument can only validate surface concentrations, and additional information on the SO2 vertical prole (e.g., from model data) is required to make the link with the satellite retrieved column. However, in situ instruments are being operated for pollution monitoring in populated areas, and allow for extended and long-term comparisons with satellite data (see, e.g., Nowlan et al., 2011).
5.2 Aircraft and mobile measurements
Airborne and mobile instruments provide valuable and complementary data for satellite validation.
In the case of volcanic explosive eruptions, satisfactory validation results can be obtained by comparing satellite and xed ground DOAS measurements of drifting SO2 plumes, as shown by Spinei et al. (2010), but the comparison generally suffers from the small number of coincidences. Dedicated aircraft campaign ights (e.g., Schumann et al., 2011) can in principle improve the situation. Their trajectory can be planned with relative ease to cross sustained eruptive plumes. However, localized high SO2 concentrations may be carried away too quickly to be captured by aircraft or have diluted below the threshold limit for satellite detection before an aircraft can respond. An important database of SO2 aircraft measurements is provided by the CARIBIC/IAGOS project, which exploits automated scientic instruments operating long-distance commercial ights. Measurements of volcanic SO2 during the eruptions of Mt. Kasatochi and Eyjafjallajkull and comparison with satellite data have been reported by Heue et al. (2010, 2011).
An attempt to validate satellite SO2 measurements using a mobile DOAS instrument for a fast moving (stratospheric) volcanic SO2 plume was presented by Carn and
Lopez (2011). Although the agreement between both data sets was found reasonable, the comparison was complicated
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
N. Theys et al.: S-5P SO2 algorithm theoretical basis 141
the total SO2 column is strongly dependent on the height of the SO2 plume, which is highly variable and usually unknown. For most volcanoes, there is no ground-based equipment to measure SO2 during an appreciable eruption and even if it is the case, the data are generally difcult to use for validation. For strong eruptions, volcanic plumes are transported over long distances and can be measured by ground-based and aircraft devices, but generally there are only a handful of data sets available and the number of coincidences is rather small.
For both anthropogenic and volcanic SO2 measurements, the vertical distribution of SO2 is a key parameter limiting the product accuracy. If reliable (external) information on the SO2 prole (or prole shape) is available, it is recommended to recalculate the SO2 vertical columns by using this piece of information and the column averaging kernels that can be found in the TROPOMI SO2 L2 les.
5.1 Ground-based measurements
When considering the application of ground-based instruments for the validation of satellite SO2 observations, several types of instruments are to be considered.
Brewer instruments have the advantage to operate as part of a network (http://www.woudc.org
Web End =http://www.woudc.org ), but the retrieved SO2 columns are generally found inaccurate for the validation of anthropogenic SO2. Yet in some cases they might be used for coincidences with volcanic clouds, typically for SO2 VCDs larger than 510 DU. Multi-axis DOAS (MAX-DOAS) or direct-sun DOAS measurements (e.g., from Pandora instruments) can be used to validate satellite SO2 columns from anthropogenic emissions (e.g., Theys et al., 2015; Jin et al., 2016; Wang et al., 2016), but caution must be exercised in the interpretation of the results because realistic SO2 prole shapes must be used by the satellite retrieval scheme.While direct-sun DOAS retrievals are independent of the SO2 prole shape, MAX-DOAS observations carry information on the SO2 vertical distribution, but it is not obvious that the technique is directly applicable to the validation of satellite SO2 retrievals, because the technique is not able to retrieve the full SO2 prole. Another important limitation comes from the fact that ground-based DOAS and satellite instruments have very different elds of view and are therefore probing different air masses. This can cause a large discrepancy between ground-based and satellite measurements in the case of strong horizontal gradients of the SO2 column eld. DOAS instruments scanning through volcanic plumes are now routinely measuring volcanic SO2 emissions, as part of the Network for Observation of Volcanic and Atmospheric Change (NOVAC; Galle et al., 2010), for an increasing number of degassing volcanoes. Ongoing research focusses on calculating SO2 uxes from those measurements and accounting for nontrivial radiative transfer effects (e.g., light dilution; see Kern et al., 2009). NOVAC ux data could be used for comparison with TROPOMI SO2 data,
142 N. Theys et al.: S-5P SO2 algorithm theoretical basis
tical way to assess the quality of the retrievals. To overcome sampling issues mentioned above, intercomparison of SO2 masses integrated over the measured volcanic plume is often performed. For TROPOMI, current satellite instruments will be an important source of data for cross-comparisons.Although non-exhaustive, the following is a list of satellite sensors that could be used: OMI, OMPS, GOME-2 and IASI (MetOp-A, -B, and the forthcoming -C), AIRS, CrIS, VIIRS and MODIS. As mentioned above, the intercomparison of satellite SO2 products is difcult and in this respect the plume altitude is a key factor of the satellite SO2 data accuracy. Comparison of TROPOMI and other satellite SO2 products will benet not only from the advent of scientic algorithms for the retrieval of SO2 plume heights but also from the use of volcanic plume height observations using space lidar instruments (e.g., CALIOP and the future EarthCare mission).
For both anthropogenic SO2 and volcanic degassing SO2, the satellite UV sensors OMI, GOME-2 and OMPS can be compared to TROPOMI SO2 data by averaging data over certain polluted regions. This procedure will give valuable information on the data quality, but, in some cases, the comparison will suffer from differences in spatial resolution. A more robust and in-depth comparison would be to use different TROPOMI SO2 data sets generated by different retrieval algorithms and investigate the differences in the various retrieval steps (spectral tting, corrections, radiative transfer simulations, error analysis).
6 Conclusions
Based on the heritage from GOME, SCIAMACHY, GOME-2 and OMI, a DOAS retrieval algorithm has been developed for the operational retrieval of SO2 vertical columns from
TROPOMI level 1b measurements in the UV spectral range.
Here we describe its main features.
In addition to the traditionally used tting window of 312 326 nm, the new algorithm allows for the selection of two additional tting windows (325335 and 360390 nm), reducing the risk of saturation and ensuring accurate SO2 column retrieval even for extreme SO2 concentrations as observed for major volcanic events. The spectral tting procedure also includes an advanced wavelength calibration scheme and a spectral spike removal algorithm.
After the slant column retrieval, the next step is a background correction, which is empirically based on the O3 slant column (for the baseline tting window) and across-track position and accounts for possible across-track dependencies and instrumental degradation.
The SO2 slant columns are then converted into vertical columns by means of air mass factor calculations. The latter is based on weighting function look-up tables with dependencies on the viewing geometry, clouds, surface pressure, albedo and ozone and is applied to pre-dened box proles
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
Figure 11. Comparison of SO2 SCDs between prototype algorithm and operational processor for the OMI test data of 8 August 2008.
by the relatively fast displacement of the volcanic cloud with respect to the ground spectrometer and clear heterogeneity on scales smaller than a satellite pixel. For degassing volcanoes or new ssure eruptions, mobile DOAS traverse measurements under the plume offer unique opportunities to derive volcanic SO2 uxes that could be used to validate satellite measurements.
For polluted regions, measurements of anthropogenic SO2 by airborne nadir-looking DOAS sensors are able to produce high-spatial-resolution mapping of the SO2 column eld (e.g., during the AROMAT campaigns, http://uv-vis.aeronomie.be/aromat/
Web End =http:// http://uv-vis.aeronomie.be/aromat/
Web End =uv-vis.aeronomie.be/aromat/ http://uv-vis.aeronomie.be/aromat/
Web End = ) that could be used to validate TROPOMI SO2 product or give information on horizontal gradients of the SO2 eld (e.g., in combination with coincident mobile DOAS measurements) that would be particularly useful when comparing satellite and MAX-DOAS data (see discussion in Sect. 5.1). Equally important are also limb-DOAS or in situ instruments to provide information on the vertical distribution of SO2 which is crucial for satellite validation (e.g., Krotkov et al., 2008).
5.3 Satellite measurements
Intercomparison of satellite SO2 measurements generally provides a convenient and easy way to evaluate at a glance the quality of a satellite product, for instance by comparing SO2 maps. Often, it also provides improved statistics and geographical representativeness, but it poses a number of problems because, when different satellite sensors are compared, they have also different overpass times, swaths, spatial resolutions and measurement sensitivities to SO2.
For volcanic SO2, satellite measurements often provide the only data available for the rst hours to days after an eruption event and satellite intercomparison is thus the only prac-
N. Theys et al.: S-5P SO2 algorithm theoretical basis 143
and TM5 CTM forecast proles. In addition, the algorithm computes DOAS-type averaging kernels and a full error analysis of the retrieved columns.
In this paper we have also presented verication results using an independent algorithm for selected OMI scenes with enhanced SO2 columns. Overall, the prototype algorithm agrees well with the verication algorithm, demonstrating its ability in retrieving accurately medium to very high SO2 columns. We have discussed the advantages and limitations of both prototype and verication algorithms.
Based on the experience with GOME-2 and OMI, the TROPOMI SO2 algorithm is expected to have a comparable level of accuracy. Due to its high signal-to-noise ratio, TROPOMI will be capable of at least achieving comparable retrieval precision as its predecessors but at a much ner spatial resolution of 7 km [notdef] 3.5 km at best. For single measure
ments, the user requirements for tropospheric SO2 concentrations will not be met, but improved monitoring of strong pollution and volcanic events will be possible by spatial and temporal averaging the increased number of observations of TROPOMI. Nevertheless, it will require signicant validation work and here we have discussed some of the inherent challenges for both volcanic and anthropogenic SO2 retrievals. Correlative measurements from ground-based, aircraft/mobile, and satellite instruments will be needed over different regions and various emission scenarios to assess and characterize the quality of TROPOMI SO2 retrievals.
The baseline algorithm presented here, including all its modules (slant column retrieval, background correction, air mass factor calculation and error analysis), has been fully implemented in the S-5P operational processor UPAS by the DLR team. Figure 11 illustrates the status of the implementation for one day of OMI test data, as an example for the slant columns retrievals. A nearly perfect agreement is found between SCD results over 4 orders of magnitude. A similar match between prototype algorithm and operational processor is found for all other retrieval modules.
For more information on the TROPOMI SO2 L2 data les, the reader is referred to the S-5P SO2 Product User Manual (Pedergnana et al., 2016).
7 Data availability
The TROPOMI SO2 retrieval algorithm has been tested on OMI L1 and L2 operational data, publicly available from the NASA Goddard Earth Sciences (GES) Data and Information Services Center (http://disc. sci.gsfc.nasa.gov/Aura/OMI/omso2.shtml
Web End =http://disc.sci.gsfc.nasa.gov/ http://disc. sci.gsfc.nasa.gov/Aura/OMI/omso2.shtml
Web End =Aura/OMI/omso2.shtml ). The static auxiliary datasets used as input of the TROPOMI SO2 retrieval algorithm are publicly available. The links to the data sets are in the references included in Table A2. Other underlying research data are available upon request from Nicolas Theys ([email protected]).
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
144 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Appendix A: Feasibility, information on data product and ancillary data
High-level data product description
In addition to the main product results, such as SO2 slant column, vertical column and air mass factor, the level 2 data les will contain several additional parameters and diagnostic information. Table A1 gives a minimum set of data elds that will be present in the level 2 data. A one-orbit SO2 column level 2 le will be about 640 MB. More details about the operational level 2 product based on the NetCDF data format and the CF metadata convention are provided in the SO2 Product User Model (Pedergnana et al., 2016).
It should be noted that the averaging kernels are given only for the a priori proles from the TM5 CTM (to save space). The averaging kernels for the box proles can be estimated by scaling the provided averaging kernel (corresponding to TM5 proles): AKbox(p) = AK(p). Following the AK for
mulation of Eskes and Boersma (2004), the scaling factor is given simply by AMF ratios: AMFTM5 / AMFbox.
Auxiliary information
The algorithm relies on several external data sets. These can be either static or dynamic. An overview is given in Tables A2 and A3.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 145
Table A1. List of output elds in the TROPOMI SO2 products. nAlong [notdef] nAcross corresponds to the number of pixels in an orbit along
track and across track, respectively; n.u.: no unit.
Name/data Symbol Unit Description Data type Number of entries per observation
Date n.u. Date and time of the measurement YYMMDDHHMMSS.MS
characters nAlong
Latitudes lat degree Latitudes of the four pixel corners + center oat 5 [notdef] nAlong [notdef] nAcross
Longitudes long degree Longitudes of the four pixel corners + cen
ter
oat 5 [notdef] nAlong [notdef] nAcross
SZA 0 degree Solar zenith angle oat nAlong [notdef] nAcross
VZA degree Viewing zenith angle oat nAlong [notdef] nAcross
RAA ' degree Relative azimuth angle oat nAlong [notdef] nAcross
SCD Ns mol m2 SO2 slant column density oat nAlong [notdef] nAcross
SCDcorr Nc
s mol m2 SO2 slant column density background corrected
oat nAlong [notdef] nAcross
oat 4 [notdef] nAlong [notdef] nAcross
Wdow ag Wag n.u. Flag for the tting window used (1, 2, 3) integer nAlong [notdef] nAcross
AMF M n.u. Air mass factor (4 values) oat 4 [notdef] nAlong [notdef] nAcross
Cloud-free AMF Mclear n.u. Cloud-free air mass factor (4 values) oat 4 [notdef] nAlong [notdef] nAcross
Cloudy AMF Mcloud n.u. Fully cloudy air mass factor (4 values) oat 4 [notdef] nAlong [notdef] nAcross
CF fc n.u. Cloud fraction oat nAlong [notdef] nAcross
CRF [Phi1] n.u. Cloud radiance fraction oat nAlong [notdef] nAcross
CP pcloud Pa Cloud top pressure oat nAlong [notdef] nAcross
CH zcloud m Cloud top height oat nAlong [notdef] nAcross
CA Acloud n.u. Cloud top albedo oat nAlong [notdef] nAcross
Albedo As n.u. Surface albedo oat nAlong [notdef] nAcross
Aerosol index AAI n.u. Absorbing aerosol index oat nAlong [notdef] nAcross
Chi-squared chi2 n.u. Chi-squared of the t oat nAlong [notdef] nAcross
VCD error Nv mol m2 Total error on the vertical column (individual measurement)
oat 4 [notdef] nAlong [notdef] nAcross
SCD random error Ns_rand mol m2 Random error on the slant column oat nAlong [notdef] nAcross
SCD systematic error
Ns_syst mol m2 Systematic error on the slant column oat nAlong [notdef] nAcross
AMF random error Mrand n.u. Random error on the air mass factor (4 values)
Msyst n.u. Systematic error on the air mass factor (4 values)
Averaging kernel AK n.u. Total column averaging kernel (for a priori prole from CTM)
scaling box n.u. Factors to apply to the averaging kernel function to obtain the corresponding averaging kernels for the three box proles
SO2 prole na n.u. A priori prole from CTM (volume mixing ratio)
oat 34 [notdef] nAlong [notdef] nAcross
Surface altitude zs m Digital elevation map oat nAlong [notdef] nAcross
Surface pressure ps Pa Effective surface pressure of the satellite pixel
oat nAlong [notdef] nAcross
oat 24
Ai n.u. oat 24
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
VCD Nv mol m2 SO2 vertical column density(4 values)
oat 4 [notdef] nAlong [notdef] nAcross
AMF systematic error
oat 4 [notdef] nAlong [notdef] nAcross
oat 34 [notdef] nAlong [notdef] nAcross
Averaging kernel scalings for box proles
oat 3 [notdef] nAlong [notdef] nAcross
TM5 level coefcient a
Ai Pa TM5 pressure level coefcients that effectively dene the mid-layer levels (from ECMWF)
TM5 level coefcient b
146 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Table A2. Static auxiliary data for the S-5P SO2 algorithm.
Name/data Symbol Unit Source Pre-process Comments needs
Absorption cross sections
SO2 SO2 cm2 molec1 Bogumil et al. (2003), 203, 223, 243, 293 K Hermans et al. (2009), all temperatures
Convolution at the instrumental spectral resolution using the provided slit function
Ozone o3218 o3243 cm2 molec1 Brion et al. (1998);218 and 243 K
BrO BrO cm2 molec1 Fleischmann et al. (2004), 223 K
NO2 NO2 cm2 molec1 Vandaele et al. (1998),220 K
O4 (O2-O2) O4 cm5 molec2 Greenblatt et al. (1990)
High-resolution reference solar spectrum
Es W m2 nm1 Chance and Kurucz(2010)
Calculated in an ozone containing atmosphere for low and high SZA, using LIDORT_RRS (Spurr et al., 2008) and a standard atmosphere (CAMELOT European Pollution atmospheric prole)
Ring effect ringev1ringev2 cm2 molec1 Two Ring cross sections generated internally
A high-resolution reference solar spectrum andthe instrumentslit function are needed to generate the data set
Nonlinear O3 absorption effect
o3l o3sq nm cm2 molec1
cm4 molec2
Two pseudo-cross sections generated internally
The O3 cross section at 218 K is needed
Calculated from the Taylor expansion of the wavelength andthe O3 optical depth (Pukte et al., 2010)
Instrument slit function
SF n.u. Slit function by wavelength/detector
Values between 300 and 400 nm
Surface albedo As n.u. OMI-based monthly minimum LER (update of Kleipool et al., 2008)
Digital elevation map
zs
m
GMTED2010 (Danielson et al., 2011)
Average over the ground pixel area.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 147
Table A2. Continued.
Name/data Symbol Unit Source Pre-process Comments needs
SO2 prole na n.u. One-kilometer-thick box proles, with three different peak altitudes, representing different altitude regimes:boundary layer: from the surface altitude to 1 km above it;free troposphere: centered around 7 km altitude;lower stratosphere: centered around 15 km altitudeDaily SO2 proles forecast from
TM5
TM5 proles from the last available day if the TM5 proles of the current day are not available
Look-up table of pressure-resolved AMFs
m n.u. Calculated internally with the LIDORTv3.3 RTM (Spurr, 2008)
For the different tting windows (312326, 325335, 360390 nm), the assumed vertical column is 5, 100, 500 DU, respectively
Temperature correction parameters
K1 Bogumil et al. (2003)
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
148 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Table A3. Dynamic auxiliary data for the S-5P SO2 algorithm.
Name/data Symbol Unit Source Pre-process Backup if not needs available
S-5P level 1B I mol s1 m2 nm1 sr1 S-5P L1b product No retrieval Earth radiance
S-5P level 1B E0 mol s1 m2 nm1 S-5P L1b product Wavelength recalibrated Use previous sun irradiance using a high-resolution measurement reference solar spectrum
S-5P cloud fc n.u. fraction S-5P operational cloud product
S-5P cloud top pcloud Pa based on a Lambertian cloud No retrieval pressure model (Loyola et al., 2016)
S-5P cloud top Acloud n.u. UPAS processor albedo
SO2 prole na n.u. Daily forecast from TM5 Use TM5 CTM
CTM run at KNMI. prole from last available day
Temperature prole T K Daily forecast from TM5 Use TM5 CTM prole CTM run at KNMI prole from last available day
S-5P absorbing AAI n.u. S-5P operational AAI product Missing aerosol index (Zweers, 2016) information ag
Used for agging
KNMI processor
Snow/ice ag n.u. Near-real-time global Ice Use snow/and Snow Extent (NISE) ice climatology data from NASA
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 149
Table A4. Acronyms and abbreviations.
AAI Absorbing aerosol index
AK Averaging kernelAMF Air mass factorAOD Aerosol optical depthAR Alternative retrievalBrO Bromine monoxideCAL Cloud as layerCAMELOT Composition of the Atmospheric Mission concEpts and SentineL Observation Techniques CAPACITY Composition of the Atmosphere: Progress to Applications in the user CommunITYCCD Charge-coupled deviceCRB Clouds as Reecting BoundariesCTM Chemical transport modelDOAS Differential optical absorption spectroscopyDU Dobson unit (1 DU = 2.6867 [notdef] 10
16 molecules cm2)
ECMWF European Centre for Medium-Range Weather Forecast
ESA European Space Agency
FT Free troposphereFWHM Full width at half maximumGMES Global Monitoring for Environment and Security GOME-2 Global Ozone Monitoring Experiment2HCHO FormaldehydeIPA Independent pixel approximationIR InfraredL2WG Level-2 Working GroupLER Lambertian equivalent reectorLIDORT LInearized Discrete Ordinate Radiative Transfer LOS Line-of-sight angleLS Lower stratosphereLUT Look-up tableMAX-DOAS Multi-axis DOASMR Medium retrievalNO2 Nitrogen dioxide
NOVAC Network for Observation of Volcanic and Atmospheric Change
NRT Near-real timeOCRA Optical Cloud Recognition AlgorithmO3 Ozone
OMI Ozone Monitoring Instrument
OMPS Ozone Mapping Proler SuitePA Prototype algorithm(P)BL Planetary boundary layer
PCA Principal component analysisROCINN Retrieval Of Cloud Information using Neural NetworksRRS Rotational Raman scatteringRTM Radiative transfer modelRAA Relative azimuth angleS-5P Sentinel-5 PrecursorSCIAMACHY SCanning Imaging Absorption spectroMeter for Atmospheric ChartograpHY SCD Slant column densitySCDE Slant column density errorS/N Signal-to-noise ratioSO2 Sulfur dioxide
SR Standard retrieval
SWIR Shortwave infraredSZA Solar zenith angleTOMS Total Ozone Mapping SpectrometerTROPOMI Tropospheric Monitoring InstrumentUPAS Universal Processor for UV/VIS Atmospheric Spectrometers UV UltravioletVA Verication algorithmVC(D) Vertical column densityWF Weighting function
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
150 N. Theys et al.: S-5P SO2 algorithm theoretical basis
Acknowledgements. This work has been performed in the frame of the TROPOMI project. We acknowledge nancial support from ESA S-5P, Belgium Prodex TRACE-S5P projects, and the Bayerisches Staatsministerium fr Wirtschaft und Medien, Energie und Technologie (grant 07 03/893 73/ 5 /2013).
Edited by: J. KimReviewed by: two anonymous referees
References
Afe, O. T., Richter, A., Sierk, B., Wittrock, F., and Burrows, J.P.: BrO emissions from volcanoes: a survey using GOME and SCIAMACHY measurements, Geophys. Res. Lett., 31, L24113, doi:http://dx.doi.org/10.1029/ 2004GL020994
Web End =10.1029/ 2004GL020994 , 2004.
Beirle, S., Hrmann, C., Penning de Vries, M., Drner, S., Kern,C., and Wagner, T.: Estimating the volcanic emission rate and atmospheric lifetime of SO2 from space: a case study for Kilauea volcano, Hawaii, Atmos. Chem. Phys., 14, 83098322, doi:http://dx.doi.org/10.5194/acp-14-8309-2014
Web End =10.5194/acp-14-8309-2014 http://dx.doi.org/10.5194/acp-14-8309-2014
Web End = , 2014.
Bobrowski, N., Kern, C., Platt, U., Hrmann, C., and Wagner, T.: Novel SO2 spectral evaluation scheme using the 360
390 nm wavelength range, Atmos. Meas. Tech., 3, 879891, doi:http://dx.doi.org/10.5194/amt-3-879-2010
Web End =10.5194/amt-3-879-2010 http://dx.doi.org/10.5194/amt-3-879-2010
Web End = , 2010.
Boersma, K. F., Eskes, H. J., and Brinksma, E. J.: Error analysis for tropospheric NO2 retrieval from space, J. Geophys. Res., 109,
D04311, doi:http://dx.doi.org/10.1029/2003JD003962
Web End =10.1029/2003JD003962 http://dx.doi.org/10.1029/2003JD003962
Web End = , 2004.
Bogumil, K., Orphal, J., Homann, T., Voigt, S., Spietz, P., Fleischmann, O., Vogel, A., Hartmann, M., Bovensmann, H., Frerick, J., and Burrows, J. P.: Measurements of molecular absorption spectra with the SCIAMACHY Pre-Flight Model: instrument characterization and reference data for atmospheric remote-sensing in the 2302380 nm region, J. Photoch. Photobio. A, 157, 167184, 2003.
Bovensmann, H., Peuch, V.-H., van Weele, M., Erbertseder, T., and
Veihelmann, B.: Report Of The Review Of User Requirements For Sentinels-4/-5, ESA, EOP-SM/2281/BV-bv, issue: 2.1, 2011. Brenot, H., Theys, N., Clarisse, L., van Geffen, J., van Gent, J.,
Van Roozendael, M., van der A, R., Hurtmans, D., Coheur, P.-F., Clerbaux, C., Valks, P., Hedelt, P., Prata, F., Rasson, O., Sievers,K., and Zehner, C.: Support to Aviation Control Service (SACS): an online service for near-real-time satellite monitoring of volcanic plumes, Nat. Hazards Earth Syst. Sci., 14, 10991123, doi:http://dx.doi.org/10.5194/nhess-14-1099-2014
Web End =10.5194/nhess-14-1099-2014 http://dx.doi.org/10.5194/nhess-14-1099-2014
Web End = , 2014.
Brion, J., Daumont, D., and Malicet, J.: New measurements of the absolute absorption cross-sections of ozone at 294 and 223 K in the 310350 nm spectral range, J. Phys., 45, L57L60, 1984. Brion, J., Chakir, A., Charbonnier, J., Daumont, D., Parisse, C., and Malicet, J.: Absorption spectra measurements for the ozone molecule in the 350830 nm region, J. Atmos. Chem., 30, 291 299, doi:http://dx.doi.org/10.1023/A:1006036924364
Web End =10.1023/A:1006036924364 http://dx.doi.org/10.1023/A:1006036924364
Web End = , 1998.
Carn, S. A. and Lopez, T. M.: Opportunistic validation of sulfur dioxide in the Sarychev Peak volcanic eruption cloud, Atmos. Meas. Tech., 4, 17051712, doi:http://dx.doi.org/10.5194/amt-4-1705-2011
Web End =10.5194/amt-4-1705-2011 http://dx.doi.org/10.5194/amt-4-1705-2011
Web End = , 2011.
Carn, S. A., Clarisse, L., and Prata, A. J.: Multi-decadal satellite measurements of global volcanic degassing, J. Volcanol. Geoth. Res., 311, 99134, doi:http://dx.doi.org/10.1016/j.jvolgeores.2016.01.002
Web End =10.1016/j.jvolgeores.2016.01.002 http://dx.doi.org/10.1016/j.jvolgeores.2016.01.002
Web End = , 2016.
Chance, K. and Spurr, R. J.: Ring effect studies: Rayleigh scattering including molecular parameters for rotational Raman scattering, and the Fraunhofer spectrum, Appl. Opt., 36, 52245230, 1997.
Chance, K. and Kurucz, R. L.: An improved high-resolution solar reference spectrum for earths atmosphere measurements in the ultraviolet, visible, and near infrared, J. Quant. Spectrosc. Ra., 111, 12891295, 2010.
Clarisse, L., Fromm, M., Ngadi, Y., Emmons, L., Clerbaux, C., Hurtmans, D., and Coheur, P.-F.: Intercontinental transport of anthropogenic sulfur dioxide and other pollutants; an infrared remote sensing case study, Geophys. Res. Lett., 38, L19806, doi:http://dx.doi.org/10.1029/2011GL048976
Web End =10.1029/2011GL048976 http://dx.doi.org/10.1029/2011GL048976
Web End = , 2011.
Danielson, J. J. and Gesch, D. B.: Global multi-resolution terrain elevation data 2010 (GMTED2010): US Geological Survey Open-File Report 20111073, 26 pp., 2011.
De Smedt, I., Mller, J.-F., Stavrakou, T., van der A, R., Eskes,H., and Van Roozendael, M.: Twelve years of global observations of formaldehyde in the troposphere using GOME and SCIAMACHY sensors, Atmos. Chem. Phys., 8, 49474963, doi:http://dx.doi.org/10.5194/acp-8-4947-2008
Web End =10.5194/acp-8-4947-2008 http://dx.doi.org/10.5194/acp-8-4947-2008
Web End = , 2008.
De Smedt, I., Yu, H., Danckaert, T., Theys, N., van Gent, J., Van Roozendael, M., Richter, A., Hilboll, A., Loyola, D., and Veefkind, P.: Formaldehyde retrievals from TROPOMI on-board Sentinel-5 Precursor: Algorithm Theoretical Basis, Atmos.Meas. Tech., in preparation, 2016.
Eisinger, M. and Burrows, J. P.: Tropospheric sulfur dioxide ob-served by the ERS-2 GOME instrument, Geophys. Res. Lett., 25, 41774180, 1998.
Eskes, H. J. and Boersma, K. F.: Averaging kernels for DOAS total-column satellite retrievals, Atmos. Chem. Phys., 3, 12851291, doi:http://dx.doi.org/10.5194/acp-3-1285-2003
Web End =10.5194/acp-3-1285-2003 http://dx.doi.org/10.5194/acp-3-1285-2003
Web End = , 2003.
Fioletov, V. E., McLinden, C. A., Krotkov, N., Yang, K., Loyola,D. G., Valks, P., Theys, N., Van Roozendael, M., Nowlan, C. R., Chance, K., Liu, X., Lee, C., and Martin, R. V.: Application of OMI, SCIAMACHY, and GOME-2 satellite SO2 retrievals for detection of large emission sources, J. Geophys. Res.-Atmos., 118, 1139911418, doi:http://dx.doi.org/10.1002/jgrd.50826
Web End =10.1002/jgrd.50826 http://dx.doi.org/10.1002/jgrd.50826
Web End = , 2013.
Fioletov, V. E., McLinden, C. A., Krotkov, N., Li, C., Joiner, J., Theys, N., Carn, S., and Moran, M. D.: A global catalogue of large SO2 sources and emissions derived from the Ozone
Monitoring Instrument, Atmos. Chem. Phys., 16, 1149711519, doi:http://dx.doi.org/10.5194/acp-16-11497-2016
Web End =10.5194/acp-16-11497-2016 http://dx.doi.org/10.5194/acp-16-11497-2016
Web End = , 2016.
Galle, B., Johansson, M., Rivera, C., Zhang, Y., Kihlman, M., Kern,C., Lehmann, T., Platt, U., Arellano, S., and Hidalgo, S.: Network for Observation of Volcanic and Atmospheric Change (NOVAC) A global network for volcanic gas monitoring: Network layout and instrument description, J. Geophys. Res., 115, D05304, doi:http://dx.doi.org/10.1029/2009JD011823
Web End =10.1029/2009JD011823 http://dx.doi.org/10.1029/2009JD011823
Web End = , 2010.
Greenblatt, G. D., Orlando, J. J., Burkholder, J. B., and Ravishankara, A. R.: Absorption measurements of oxygen between 330 and 1140 nm, J. Geophys. Res., 95, 1857718582, doi:http://dx.doi.org/10.1029/JD095iD11p18577
Web End =10.1029/JD095iD11p18577 http://dx.doi.org/10.1029/JD095iD11p18577
Web End = , 1990.
He, H., Vinnikov, K. Y., Li, C., Krotkov, N. A., Jongeward, A. R., Li, Z., Stehr, J. W., Hains, J. C. and Dickerson, R. R.: Response of SO2 and particulate air pollution to local and regional emission controls: A case study in Maryland, Earths Future, 4, 94109, doi:http://dx.doi.org/10.1002/2015EF000330
Web End =10.1002/2015EF000330 http://dx.doi.org/10.1002/2015EF000330
Web End = , 2016.
Heue, K.-P., Brenninkmeijer, C. A. M., Wagner, T., Mies, K., Dix,B., Frie, U., Martinsson, B. G., Slemr, F., and van Velthoven,
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 151
P. F. J.: Observations of the 2008 Kasatochi volcanic SO2 plume by CARIBIC aircraft DOAS and the GOME-2 satellite, Atmos.
Chem. Phys., 10, 46994713, doi:http://dx.doi.org/10.5194/acp-10-4699-2010
Web End =10.5194/acp-10-4699-2010 http://dx.doi.org/10.5194/acp-10-4699-2010
Web End = , 2010.
Heue, K.-P., Brenninkmeijer, C. A. M., Baker, A. K., Rauthe-Schch, A., Walter, D., Wagner, T., Hrmann, C., Sihler, H., Dix, B., Frie, U., Platt, U., Martinsson, B. G., van Velthoven,P. F. J., Zahn, A., and Ebinghaus, R.: SO2 and BrO observation in the plume of the Eyjafjallajkull volcano 2010: CARIBIC and GOME-2 retrievals, Atmos. Chem. Phys., 11, 29732989, doi:http://dx.doi.org/10.5194/acp-11-2973-2011
Web End =10.5194/acp-11-2973-2011 http://dx.doi.org/10.5194/acp-11-2973-2011
Web End = , 2011.
Hendrick, F., Van Roozendael, M., Kylling, A., Petritoli, A., Rozanov, A., Sanghavi, S., Schoeld, R., von Friedeburg, C., Wagner, T., Wittrock, F., Fonteyn, D., and De Mazire, M.: Intercomparison exercise between different radiative transfer models used for the interpretation of ground-based zenith-sky and multi-axis DOAS observations, Atmos. Chem. Phys., 6, 93108, doi:http://dx.doi.org/10.5194/acp-6-93-2006
Web End =10.5194/acp-6-93-2006 http://dx.doi.org/10.5194/acp-6-93-2006
Web End = , 2006.
Hermans, C., Vandaele, A. C., and Fally, S.: Fourier transform measurements of SO2 absorption cross sections: I.
Temperature dependence in the 24 00029 000 cm1 (345 420 nm) region, J. Quant. Spectrosc. Ra., 110, 756765, doi:http://dx.doi.org/10.1016/j.jqsrt.2009.01.031
Web End =10.1016/j.jqsrt.2009.01.031 http://dx.doi.org/10.1016/j.jqsrt.2009.01.031
Web End = , 2009.
Hrmann, C., Sihler, H., Bobrowski, N., Beirle, S., Penning de Vries, M., Platt, U., and Wagner, T.: S ystematic investigation of bromine monoxide in volcanic plumes from space by using the GOME-2 instrument, Atmos. Chem. Phys., 13, 47494781, doi:http://dx.doi.org/10.5194/acp-13-4749-2013
Web End =10.5194/acp-13-4749-2013 http://dx.doi.org/10.5194/acp-13-4749-2013
Web End = , 2013.
Huijnen, V., Williams, J., van Weele, M., van Noije, T., Krol, M., Dentener, F., Segers, A., Houweling, S., Peters, W., de Laat, J., Boersma, F., Bergamaschi, P., van Velthoven, P., Le Sager, P., Eskes, H., Alkemade, F., Scheele, R., Ndlec, P., and Ptz, H.-W.: The global chemistry transport model TM5: description and evaluation of the tropospheric chemistry version 3.0, Geosci. Model Dev., 3, 445473, doi:http://dx.doi.org/10.5194/gmd-3-445-2010
Web End =10.5194/gmd-3-445-2010 http://dx.doi.org/10.5194/gmd-3-445-2010
Web End = , 2010.
Jin, J., Ma, J., Lin, W., Zhao, H., Shaiganfar, R., Beirle, S., and Wagner, T.: MAX-DOAS measurements and satellite validation of tropospheric NO2 and SO2 vertical column densities at a rural site of North China, Atmos. Environ., 133, 1225, 2016.
Kelder, H., van Weele, M., Goede, A., Kerridge, B., Reburn, J., Bovensmann, H., Monks, P., Remedios, J., Mager, R., Sassier,H., and Baillon, Y.: Operational Atmospheric Chemistry Monitoring Missions CAPACITY: Composition of the Atmosphere: Progress to Applications in the user CommunITY, Final Report of ESA contract no. 17237/03/NL/GS, 2005.
Kern, C., Deutschmann, T., Vogel, A., Whrbach, M., Wagner, T., and Platt, U.: Radiative transfer corrections for accurate spectroscopic measurements of volcanic gas emissions, B. Volcanol., 72, 233247, 2009.
Khokhar, M. F., Frankenberg, C., Van Roozendael, M., Beirle, S., Khl, S., Richter, A., Platt, U., and Wagner, T.: Satellite Observations of Atmospheric SO2 from Volcanic Eruptions during the
Time Period of 1996 to 2002, J. Adv. Space Res., 36, 879887, doi:http://dx.doi.org/10.1016/j.asr.2005.04.114
Web End =10.1016/j.asr.2005.04.114 http://dx.doi.org/10.1016/j.asr.2005.04.114
Web End = , 2005.
Kleipool, Q. L., Dobber, M. R., de Haan, J. F., and Levelt, P. F.: Earth surface reectance climatology from 3 years of OMI data,J. Geophys. Res., 113, D18308, doi:http://dx.doi.org/10.1029/2008JD010290
Web End =10.1029/2008JD010290 http://dx.doi.org/10.1029/2008JD010290
Web End = , 2008.
Krotkov, N. A., Carn, S. A., Krueger, A. J., Bhartia, P. K., and Yang, K.: Band residual difference algorithm for retrieval of SO2 from the Aura Ozone Monitoring Instrument (OMI), IEEE Trans.
Geosci. Remote Sensing, AURA Special Issue, 44, 12591266, doi:http://dx.doi.org/10.1109/TGRS.2005.861932
Web End =10.1109/TGRS.2005.861932 http://dx.doi.org/10.1109/TGRS.2005.861932
Web End = , 2006.
Krotkov, N., McClure, B., Dickerson, R., Carn, S., Li, C., Bhartia, P.K., Yang, K., Krueger, A., Li, Z., Levelt, P., Chen, H., Wang, P., and Lu, D.: Validation of SO2 retrievals from the Ozone Monitoring Instrument over NE China, J. Geophys. Res., 113, D16S40, doi:http://dx.doi.org/10.1029/2007JD008818
Web End =10.1029/2007JD008818 http://dx.doi.org/10.1029/2007JD008818
Web End = , 2008.
Krotkov, N. A., McLinden, C. A., Li, C., Lamsal, L. N., Celarier,E. A., Marchenko, S. V., Swartz, W. H., Bucsela, E. J., Joiner,J., Duncan, B. N., Boersma, K. F., Veefkind, J. P., Levelt, P. F., Fioletov, V. E., Dickerson, R. R., He, H., Lu, Z., and Streets,D. G.: Aura OMI observations of regional SO2 and NO2 pollution changes from 2005 to 2015, Atmos. Chem. Phys., 16, 4605
4629, doi:http://dx.doi.org/10.5194/acp-16-4605-2016
Web End =10.5194/acp-16-4605-2016 http://dx.doi.org/10.5194/acp-16-4605-2016
Web End = , 2016.
Krueger, A. J.: Sighting of El Chichon sulfur dioxide clouds with the Nimbus 7 total ozone mapping spectrometer, Science, 220, 13771379, 1983.
Langen, J., Meijer, Y., Brinksma, E., Veihelmann, B., and Ingmann,P.: GMES Sentinels 4 and 5 Mission Requirements Document (MRD), ESA, EO-SMA-/1507/JL, issue: 3, 2011.
Lee, C., Martin, R. V., van Donkelaar, A., OByrne, G., Krotkov, N., Richter, A., Huey, L. G., and Holloway, J. S.: Retrieval of vertical columns of sulfur dioxide from SCIAMACHY and OMI: Air mass factor algorithm development, validation, and error analysis, J. Geophys. Res., 114, D22303, doi:http://dx.doi.org/10.1029/2009JD012123
Web End =10.1029/2009JD012123 http://dx.doi.org/10.1029/2009JD012123
Web End = , 2009.
Levelt, P., Veefkind, J., Kerridge, B., Siddans, R., de Leeuw, G., Remedios, J., and Coheur, P.: Observation Techniques and Mission Concepts for Atmospheric Chemistry (CAMELOT), Report, European Space Agency, Noordwijk, the Netherlands, 2009.
Li, C., Joiner, J., Krotkov, N. A., and Bhartia, P. K.: A fast and sensitive new satellite SO2 retrieval algorithm based on principal component analysis: Application to the ozone monitoring instrument, Geophys. Res. Lett., 40, 63146318, doi:http://dx.doi.org/10.1002/2013GL058134
Web End =10.1002/2013GL058134 http://dx.doi.org/10.1002/2013GL058134
Web End = , 2013.
Loyola, D., Garcia, S.G., Lutz, R., and Spurr, R.: S5P Cloud Products ATBD, available at: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =https://sentinel.esa.int/web/sentinel/ https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =technical-guides/sentinel-5p/appendices/referencesandhttp: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =//www.tropomi.eu/documents/level-2-products (last access: 6 January 2017), 2016.
Martin, R. V., Chance, K., Jacob, D. J., Kurosu, T. P., Spurr, R.J. D., Bucsela, E., Gleason, J .F., Palmer, P. I., Bey, I., Fiore, A.M., Li, Q., Yantosca, R. M., and Koelemeijer, R. B. A.: An improved retrieval of tropospheric nitrogen dioxide from GOME,J. Geophys. Res., 107, 4437, doi:http://dx.doi.org/10.1029/2001JD001027
Web End =10.1029/2001JD001027 http://dx.doi.org/10.1029/2001JD001027
Web End = , 2002.
McLinden, C. A., Fioletov, V., Shephard, M. W., Krotkov, N., Li,C., Martin, R. V., Moran, M. D., and Joiner, J.: Space-based detection of missing sulfur dioxide sources of global air pollution, Nat. Geosci., 9, 496500, doi:http://dx.doi.org/10.1038/ngeo2724
Web End =10.1038/ngeo2724 http://dx.doi.org/10.1038/ngeo2724
Web End = , 2016.Nowlan, C. R., Liu, X., Chance, K., Cai, Z., Kurosu, T. P., Lee, C., and Martin, R. V.: Retrievals of sulfur dioxide from the Global Ozone Monitoring Experiment 2 (GOME-2) using an optimal estimation approach: Algorithm and initial validation, J. Geophys.Res., 116, D18301, doi:http://dx.doi.org/10.1029/2011JD015808
Web End =10.1029/2011JD015808 http://dx.doi.org/10.1029/2011JD015808
Web End = , 2011.
Palmer, P. I., Jacob, D. J., Chance, K. V., Martin, R. V., Spurr, R.J. D., Kurosu, T. P., Bey, I., Yantosca, R., and Fiore, A.: Air
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
152 N. Theys et al.: S-5P SO2 algorithm theoretical basis
mass factor formulation for spectroscopic measurements from satellites: Application to formaldehyde retrievals from the Global Ozone Monitoring Experiment, J. Geophys. Res., 106, 14539 14550, doi:http://dx.doi.org/10.1029/2000JD900772
Web End =10.1029/2000JD900772 http://dx.doi.org/10.1029/2000JD900772
Web End = , 2001.
Pedergnana, M., Loyola, D., Apituley, A., Sneep, M., and Veefkind, P.: S5P Level 2 Product User Manual Sulfur Dioxide SO2, available at: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =https://sentinel.esa.int/web/sentinel/
https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =technical-guides/sentinel-5p/appendices/referencesandhttp: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =//www.tropomi.eu/documents/level-2-products (last access: 6 January 2017), 2016.
Platt, U. and Stutz, J.: Differential Optical Absorption Spectroscopy (DOAS), Principle and Applications, ISBN 3-340-21193-4, Springer Verlag, Heidelberg, 2008.
Pukte, J., Khl, S., Deutschmann, T., Platt, U., and Wagner, T.: Extending differential optical absorption spectroscopy for limb measurements in the UV, Atmos. Meas. Tech., 3, 631653, doi:http://dx.doi.org/10.5194/amt-3-631-2010
Web End =10.5194/amt-3-631-2010 http://dx.doi.org/10.5194/amt-3-631-2010
Web End = , 2010.
Richter, A., Wittrock, F., Schnhardt, A., and Burrows, J. P.: Quantifying volcanic SO2 emissions using GOME2 measurements,
Geophys. Res. Abstr., EGU2009-7679, EGU General Assembly 2009, Vienna, Austria, 2009.
Richter, A., Begoin, M., Hilboll, A., and Burrows, J. P.: An improved NO2 retrieval for the GOME-2 satellite instrument, Atmos. Meas. Tech., 4, 11471159, doi:http://dx.doi.org/10.5194/amt-4-1147-2011
Web End =10.5194/amt-4-1147-2011 http://dx.doi.org/10.5194/amt-4-1147-2011
Web End = , 2011.
Rix, M., Valks, P., Hao, N., Loyola, D. G., Schlager, H., Huntrieser,H. H., Flemming, J., Koehler, U., Schumann, U., and Inness,A.: Volcanic SO2, BrO and plume height estimations using GOME-2 satellite measurements during the eruption of Eyjafjallajkull in May 2010, J. Geophys. Res., 117, D00U19, doi:http://dx.doi.org/10.1029/2011JD016718
Web End =10.1029/2011JD016718 http://dx.doi.org/10.1029/2011JD016718
Web End = , 2012.
Sanders, A. and de Haan, J.: S5P ATBD of the Aerosol Layer Height product, available at: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =https://sentinel.esa.int/web/sentinel/ https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =technical-guides/sentinel-5p/appendices/referencesandhttp: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =//www.tropomi.eu/documents/level-2-products (last access: 6 January 2017), 2016.
Schumann, U., Weinzierl, B., Reitebuch, O., Schlager, H., Minikin,A., Forster, C., Baumann, R., Sailer, T., Graf, K., Mannstein, H., Voigt, C., Rahm, S., Simmet, R., Scheibe, M., Lichtenstern, M., Stock, P., Rba, H., Schuble, D., Tafferner, A., Rautenhaus, M., Gerz, T., Ziereis, H., Krautstrunk, M., Mallaun, C., Gayet, J.-F., Lieke, K., Kandler, K., Ebert, M., Weinbruch, S., Stohl, A., Gasteiger, J., Gro, S., Freudenthaler, V., Wiegner, M., Ansmann,A., Tesche, M., Olafsson, H., and Sturm, K.: Airborne observations of the Eyjafjalla volcano ash cloud over Europe during air space closure in April and May 2010, Atmos. Chem. Phys., 11, 22452279, doi:http://dx.doi.org/10.5194/acp-11-2245-2011
Web End =10.5194/acp-11-2245-2011 http://dx.doi.org/10.5194/acp-11-2245-2011
Web End = , 2011.
Spinei, E., Carn, S. A., Krotkov, N. A., Mount, G. H., Yang,K., and Krueger, A. J.: Validation of ozone monitoring instrument SO2 measurements in the Okmok volcanic plume over
Pullman, WA in July 2008, J. Geophys. Res., 115, D00L08, doi:http://dx.doi.org/10.1029/2009JD013492
Web End =10.1029/2009JD013492 http://dx.doi.org/10.1029/2009JD013492
Web End = , 2010.
Spurr, R.: LIDORT, and VLIDORT: Linearized pseudo-spherical scalar and vector discrete ordinate radiative transfer models for use in remote sensing retrieval problems, Light Scattering Reviews, 3, edited by: Kokhanovsky, A., Springer, 2008.
Spurr, R., de Haan, J. F., van Oss, R., and Vasilkov, A.: Discrete Ordinate Radiative Transfer in a Stratied Medium with First Order
Rotational Raman Scattering, J. Quant. Spectros. Ra., 109, 404 425, doi:http://dx.doi.org/10.1016/j.jqsrt.2007.08.011
Web End =10.1016/j.jqsrt.2007.08.011 http://dx.doi.org/10.1016/j.jqsrt.2007.08.011
Web End = , 2008.
Theys, N., Campion, R., Clarisse, L., Brenot, H., van Gent, J., Dils,B., Corradini, S., Merucci, L., Coheur, P.-F., Van Roozendael,M., Hurtmans, D., Clerbaux, C., Tait, S., and Ferrucci, F.: Volcanic SO2 uxes derived from satellite data: a survey using OMI,
GOME-2, IASI and MODIS, Atmos. Chem. Phys., 13, 5945 5968, doi:http://dx.doi.org/10.5194/acp-13-5945-2013
Web End =10.5194/acp-13-5945-2013 http://dx.doi.org/10.5194/acp-13-5945-2013
Web End = , 2013.
Theys, N., De Smedt, I., van Gent, J., Danckaert, T., Wang, T., Hendrick, F., Stavrakou, T., Bauduin, S., Clarisse, L., Li, C., Krotkov, N. A., Yu, H., and Van Roozendael, M.: Sulfur dioxide vertical column DOAS retrievals from the Ozone Monitoring Instrument: Global observations and comparison to ground-based and satellite data, J. Geophys. Res.-Atmos., 120, 2470 2491, doi:http://dx.doi.org/10.1002/2014JD022657
Web End =10.1002/2014JD022657 http://dx.doi.org/10.1002/2014JD022657
Web End = , 2015.
Thomas, W., Erbertseder, T., Ruppert, T. van Roozendael, M., Verdebout, J., Balis, D., Meleti, C., and Zerefos, C.: On the retrieval of volcanic sulfur dioxide emissions from GOME backscatter measurements, J. Atmos. Chem., 50, 295320, doi:http://dx.doi.org/10.1007/s10874-005-5544-1
Web End =10.1007/s10874-005-5544-1 http://dx.doi.org/10.1007/s10874-005-5544-1
Web End = , 2005.
Vandaele, A. C., Hermans, C., Simon, P. C., Carleer, M., Colin,R., Fally, S., Mrienne, M. F., Jenouvrier, A., and Coquart,B.: Measurements of the NO2 absorption cross-section from 42 000 cm1 to 10 000 cm1 (2381000 nm) at 220 K and 294 K, J. Quant. Spectrosc. Ra., 59, 171184, 1998.
Vandaele, A. C., Hermans, C., and Fally, S.: Fourier transform measurements of SO2 absorption cross sections: II.
Temperature dependence in the 29 00044 000 cm1 (227 345 nm) region, J. Quant. Spectrosc. Ra., 110, 21152126, doi:http://dx.doi.org/10.1016/j.jqsrt.2009.05.006
Web End =10.1016/j.jqsrt.2009.05.006 http://dx.doi.org/10.1016/j.jqsrt.2009.05.006
Web End = , 2009.van der A, R. J., Mijling, B., Ding, J., Koukouli, M. E., Liu, F.,
Li, Q., Mao, H., and Theys, N.: Cleaning up the air: Effectiveness of air quality policy for SO2 and NOx emissions in China,
Atmos. Chem. Phys. Discuss., doi:http://dx.doi.org/10.5194/acp-2016-445
Web End =10.5194/acp-2016-445 http://dx.doi.org/10.5194/acp-2016-445
Web End = , in review, 2016.van Geffen, J., van Roozendaal, M., Rix, M., and Valks, P.: Initial
Validation of GOME-2 GDP 4.2 SO2 Total ColumnsORR B, TN-IASB-GOME2-O3MSAF-SO2-01, September, 2008.
van Geffen, J. H. G. M., Boersma, K. F., Eskes, H. J., Maasakkers, J. D., and Veefkind, J. P. : S5P NO2 data products ATBD, available at: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =https://sentinel.esa.int/web/sentinel/
https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =technical-guides/sentinel-5p/appendices/referencesandhttp: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/referencesand http://www.tropomi.eu/documents/level-2-products
Web End =//www.tropomi.eu/documents/level-2-products (last access: 6 January 2017), 2016.van Weele, M., Levelt, P., Aben, I., Veefkind, P., Dobber, M., Eskes, H., Houweling, S., Landgraf, J., and Noordhoek, R.: Science Requirements Document for TROPOMI. Volume 1, KNMI & SRON, RS-TROPOMI-KNMI-017, issue: 2.0, 2008. Veefkind, J. P., Aben, I., McMullan, K., Frster, H., de Vries,J., Otter, G., Claas, J., Eskes, H. J., de Haan, J. F., Kleipool,Q., van Weele, M., Hasekamp, O., Hoogeven, R., Landgraf,J., Snel, R., Tol, P., Ingmann, P., Voors, R., Kruizinga, B., Vink, R., Visser, H., and Levelt, P. F.: 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, 7083, doi:http://dx.doi.org/10.1016/j.rse.2011.09.027
Web End =10.1016/j.rse.2011.09.027 http://dx.doi.org/10.1016/j.rse.2011.09.027
Web End = , 2012.
Atmos. Meas. Tech., 10, 119153, 2017 www.atmos-meas-tech.net/10/119/2017/
N. Theys et al.: S-5P SO2 algorithm theoretical basis 153
Vountas, M., Rozanov, V. V., and Burrows, J. P.: Ring effect: impact of rotational Raman scattering on radiative transfer in earths atmosphere, J. Quant. Spectrosc. Ra., 60, 943961, 1998.
Wagner, T., Burrows, J. P., Deutschmann, T., Dix, B., von Friedeburg, C., Frie, U., Hendrick, F., Heue, K.-P., Irie, H., Iwabuchi,H., Kanaya, Y., Keller, J., McLinden, C. A., Oetjen, H., Palazzi,E., Petritoli, A., Platt, U., Postylyakov, O., Pukite, J., Richter,A., van Roozendael, M., Rozanov, A., Rozanov, V., Sinreich,R., Sanghavi, S., and Wittrock, F.: Comparison of box-air-mass-factors and radiances for Multiple-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) geometries calculated from different UV/visible radiative transfer models, Atmos. Chem. Phys., 7, 18091833, doi:http://dx.doi.org/10.5194/acp-7-1809-2007
Web End =10.5194/acp-7-1809-2007 http://dx.doi.org/10.5194/acp-7-1809-2007
Web End = , 2007.
Wang, Y., Beirle, S., Lampel, J., Koukouli, M., De Smedt, I., Theys,N., Xie, P. H., Van Roozendael, M., and Wagner, T.: Validation of OMI and GOME-2A and GOME-2B tropospheric NO2,
SO2 and HCHO products using MAX-DOAS observations from 2011 to 2014 in Wuxi, China, Atmos. Chem. Phys. Discuss., doi:http://dx.doi.org/10.5194/acp-2016-735
Web End =10.5194/acp-2016-735 http://dx.doi.org/10.5194/acp-2016-735
Web End = , in review, 2016.
Yang, K., Krotkov, N., Krueger, A., Carn, S., Bhartia, P. K., and Levelt,P.: Retrieval of Large Volcanic SO2 columns from the Aura Ozone Monitoring Instrument (OMI): Comparisons and Limitations, J. Geophys. Res., 112, D24S43, doi:http://dx.doi.org/10.1029/2007JD008825
Web End =10.1029/2007JD008825 http://dx.doi.org/10.1029/2007JD008825
Web End = , 2007.
Yang, K., Liu, X., Bhartia, P., Krotkov, N., Carn, S., Hughes, E., Krueger, A., Spurr, R., and Trahan, S.: Direct retrieval of sulfur dioxide amount and altitude from spaceborne hyperspectral UV measurements: Theory and application, J. Geophys. Res., 115, D00L09, doi:http://dx.doi.org/10.1029/2010JD013982
Web End =10.1029/2010JD013982 http://dx.doi.org/10.1029/2010JD013982
Web End = , 2010.
Yang, K., Dickerson, R. R., Carn, S. A., Ge, C., and Wang, J.: First observations of SO2 from the satellite Suomi NPP OMPS: Widespread air pollution events over China, Geophys. Res. Lett., 40, 49574962, doi:http://dx.doi.org/10.1002/grl.50952
Web End =10.1002/grl.50952 http://dx.doi.org/10.1002/grl.50952
Web End = , 2013.
Zhou, Y., Brunner, D., Boersma, K. F., Dirksen, R., and Wang, P.: An improved tropospheric NO2 retrieval for OMI observations in the vicinity of mountainous terrain, Atmos. Meas. Tech., 2, 401416, doi:http://dx.doi.org/10.5194/amt-2-401-2009
Web End =10.5194/amt-2-401-2009 http://dx.doi.org/10.5194/amt-2-401-2009
Web End = , 2009.
Zweers, S.: S5P ATBD for the UV aerosol index, available at: https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/references
Web End =https://sentinel.esa.int/web/sentinel/technical-guides/ https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/appendices/references
Web End =sentinel-5p/appendices/references and http://www.tropomi.eu/documents/level-2-products
Web End =http://www.tropomi. http://www.tropomi.eu/documents/level-2-products
Web End =eu/documents/level-2-products (last access: 6 January 2017), 2016.
Zweers, S., Levelt, P. F., Veefkind, J. P., Eskes, H., de Leeuw, G, Tamminen, J., Coheur, P. F., Prunet, P., and Camy-Peyret, C.: TRAQ Performance Analysis and Requirements Consolidation for the Candidate Earth Explorer Mission TRAQ, Final report, KNMI, RP-ONTRAQ-KNMI-051, issue: 1.0, 2010.
www.atmos-meas-tech.net/10/119/2017/ Atmos. Meas. Tech., 10, 119153, 2017
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Copyright Copernicus GmbH 2017
Abstract
The TROPOspheric Monitoring Instrument (TROPOMI) onboard the Copernicus Sentinel-5 Precursor (S-5P) platform will measure ultraviolet earthshine radiances at high spectral and improved spatial resolution (pixel size of 7km × 3.5km at nadir) compared to its predecessors OMI and GOME-2. This paper presents the sulfur dioxide (SO<sub>2</sub>) vertical column retrieval algorithm implemented in the S-5P operational processor UPAS (Universal Processor for UV/VIS Atmospheric Spectrometers) and comprehensively describes its various retrieval steps. The spectral fitting is performed using the differential optical absorption spectroscopy (DOAS) method including multiple fitting windows to cope with the large range of atmospheric SO<sub>2</sub> columns encountered. It is followed by a slant column background correction scheme to reduce possible biases or across-track-dependent artifacts in the data. The SO<sub>2</sub> vertical columns are obtained by applying air mass factors (AMFs) calculated for a set of representative a priori profiles and accounting for various parameters influencing the retrieval sensitivity to SO<sub>2</sub>. Finally, the algorithm includes an error analysis module which is fully described here. We also discuss verification results (as part of the algorithm development) and future validation needs of the TROPOMI SO<sub>2</sub> algorithm.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer