Smithsonian Astrophysical Observatory Ozone Mapping and Profiler Suite (SAO OMPS) formaldehyde retrieval

Abstract. This paper presents our new formaldehyde (H2CO) retrievals, obtained from spectra recorded by the nadir instrument of the Ozone Mapping and Profiler Suite (OMPS) flown on board NASA's Suomi National Polar-orbiting Partnership (SUOMI-NPP) satellite. Our algorithm is similar to the one currently in place for the production of NASA's Ozone Monitoring Instrument (OMI) operational H2CO product. We are now able to produce a set of long-term data from two different instruments that share a similar concept and a similar retrieval approach. The ongoing overlap period between OMI and OMPS offers a perfect opportunity to study the consistency between both data sets. The different spatial and spectral resolution of the instruments is a source of discrepancy in the retrievals despite the similarity of the physic assumptions of the algorithm. We have concluded that the reduced spectral resolution of OMPS in comparison with OMI is not a significant obstacle in obtaining good-quality retrievals. Indeed, the improved signal-to-noise ratio of OMPS with respect to OMI helps to reduce the noise of the retrievals performed using OMPS spectra. However, the size of OMPS spatial pixels imposes a limitation in the capability to distinguish particular features of H2CO that are discernible with OMI. With root mean square (RMS) residuals  ∼ 5 × 10−4 for individual pixels we estimate the detection limit to be about 7.5 × 1015 molecules cm−2. Total vertical column density (VCD) errors for individual pixels range between 40 % for pixels with high concentrations to 100 % or more for pixels with concentrations at or below the detection limit. We compare different OMI products (SAO OMI v3.0.2 and BIRA OMI v14) with our OMPS product using 1 year of data, between September 2012 and September 2013. The seasonality of the retrieved slant columns is captured similarly by all products but there are discrepancies in the values of the VCDs. The mean biases among the two OMI products and our OMPS product are 23 % between OMI SAO and OMPS SAO and 28 % between OMI BIRA and OMPS SAO for eight selected regions.


Introduction
H 2 CO is ubiquitous in the Earth's troposphere. Background levels in remote regions with concentrations below 1 ppb are due to methane (CH 4 ) oxidation. Over the continents, hotspots of enhanced concentrations are found due to oxidation of short-lived non-methane volatile organic compounds (NMVOCs) of anthropogenic, biogenic, and pyrogenic origin (secondary sources) or directly emitted by wild fires or industrial activities (primary sources). Concentrations of H 2 CO over these hotspots can reach boundary layer levels of 5 ppb or higher (Houweling et al., 1998;Kanakidou et al., 2005;Lowe and Schmidt, 1983;Parrish et al., 2012).
The importance of H 2 CO as an atmospheric trace gas resides in its influence in the chemical pathways of tropospheric ozone (O 3 ) and in the relationship with the concentration of hydroxyl radicals (OH), the main tropospheric oxidant (Anderson et al., 1996). Due to its high reactivity it has a short tropospheric lifetime of a few hours (Brune et al., 1999), making it a useful proxy for NMVOC emissions in satellite observations and for the estimation of topdown emission inventories of isoprene (Barkley et al., 2008;Fu et al., 2007;Marais et al., 2012;Stavrakou et al., 2009aStavrakou et al., , b, 2014Zhu et al., 2014). High concentrations of H 2 CO can be dangerous to human health, causing eye, nose, and throat irritation. It is also known to be a human carcinogen. Fortunately, such high concentrations are not found in the open air (Liteplo et al., 2002) since they are consequence of cigarette smoking, pressed-wood products or fuel-burning appliances.
With UV nadir instruments we can sample concentrations in the troposphere, where most of the atmospheric H 2 CO is located. The first global H 2 CO measurements using UV radiation were reported by Chance et al. (2000), using Global Ozone Monitoring Experiment (GOME) spectra. Since this seminal work, measurements of tropospheric H 2 CO have been obtained using data recorded by the SCanning Imaging Absorption spectrometer for Atmospheric CHartography (SCIAMACHY), GOME-2 instruments and Ozone Monitoring Instrument (OMI) (De Smedt et al., 2008González Abad et al., 2015;Hewson et al., 2015;Kurosu et al., 2004;Palmer et al., 2001;Vrekoussis et al., 2010;Wittrock et al., 2006), and now Smithsonian Astrophysical Observatory Ozone Mapping and Profiler Suite (SAO OMPS; this work and Li et al., 2015). Having overlapping retrievals from OMI and OMPS opens the possibility of building a long-term data set of H 2 CO using similar retrieval algorithms and having a period of cross-calibration to allow us to extract valuable conclusions about the performance of both instruments that will be of great value for future low-Earth orbit (LEO) and geostationary missions such as the Tropospheric Emissions: Monitoring of Pollution (TEMPO; Chance et al., 2013).
Section 2 includes the description of the SAO OMPS H 2 CO retrievals. We describe the characteristics of the instrument, the spectral fitting, the calculation of vertical column densities (VCDs) by using air mass factors (AMFs), and the reference sector normalization of the VCDs. Section 3 presents the error analysis for the VCDs. Section 4 describes the comparison between NASA and Belgian Institute for Space Aeronomy (BIRA) OMI retrievals and the OMPS SAO product. The conclusions are in Sect. 5.

The OMPS Nadir Mapper (OMPS-NM instrument
The OMPS-NM is one of three sensors that compose the OMPS suite of instruments. OMPS-NM was launched on board the Suomi National Polar-orbiting Partnership (Suomi-NPP) satellite on 28 October 2011. Suomi-NPP orbits the Earth at an altitude of 824 km on a polar Sun-synchronous orbit with an inclination of 98.7 • . Its Equator crossing time in the ascending node is 13:30 local time (LT) which places it close to the afternoon satellites of NASA's A-Train constellation, including Aura. Aura hosts the UV visible spectrometer OMI (Levelt et al., 2006), thus offering a great opportunity for the comparison of the retrievals obtained from both instruments.
The OMPS-NM sensor combines a single grating and a 340 × 740 pixel charge-coupled device (CCD) array detector. It measures UV radiation covering the spectral range between 300 and 380 nm with a pixel sampling of 0.42 nm and a full width half maximum of 1 nm. It has an instantaneous cross-track field of view of approximately 110 • , corresponding to 2800 km at the Earth's surface, providing global daily coverage. For this study we use the OMPS-NM Earth science mode, where measurements are combined in 35 cross-track macropixels given a spatial size of the nadir pixel of 50 km (cross-track) by 50 km (along-track), with an integration period for each measurement of 7.6 s (Dittman et al., 2002;Seftor et al., 2014). The CCD readout is split in the center so that measurements of the central pixel are split, resulting in 36 cross-track pixel positions.

SAO OMPS H 2 CO retrieval
To obtain H 2 CO VCDs our algorithm follows a two-step approach. First we perform a direct spectral fit of the measured radiance to obtain slant column densities (SCDs) using the basic optical absorption spectroscopy approach (BOAS; Chance, 1998). The second step converts SCDs to VCDs using look-up tables of precomputed scattering weights to calculate AMFs. The SAO OMPS H 2 CO algorithm is similar to the SAO OMI H 2 CO retrieval described by González . Modifications are reduced to the minimum required to deal with particular aspects of the OMPS-NM instrument to produce a long-term data set that is as consistent as possible. A more detailed description of the OMI algorithm is in Sect. 2 of González . Here we will briefly describe the characteristics of the OMPS SAO H 2 CO retrieval.

Spectral fitting
We have performed a series of studies to select the most suitable fitting window and fitting parameters for OMPS H 2 CO. Our main interest was to reduce the correlation between H 2 CO columns and the rest of the fitting parameters, especially BrO. We optimized the fitting window at 327.7 to 356.0 nm, different to that used in several consistent GOME, GOME-2, SCIAMACHY, and OMI retrievals, 328.5 to 346.0 nm, which does not include the H 2 CO absorption band located around 353 nm (De Smedt et al., 2008. Figure 1 shows the correlation among H 2 CO SCDs and the rest of the fitted gas SCDs for orbit 3538, an overpass of the Americas on 3 July 2012. Even though the correlation between H 2 CO and BrO shows two minimal bands between 325 and 326 and 330 and 331 nm, two different considerations prevent the use of them. The first band of low H 2 CO-BrO correlations, below 326.5 nm, is discarded due to the increasing strength of O 3 cross sections below 328 nm. The second band is discarded by the increasing correlation between the cross section used to account for the Ring effect above 328 nm for fitting windows ending above 356 nm and the increase in the correlations between O 3 and H 2 CO above 330.7 nm.
Another factor we have considered to select the fitting interval is the minimization of fitting uncertainties by increasing the number of spectral points considered. Extending the fitting window above 345 nm allows us to include the H 2 CO absorption band centered at 353 nm. Including this band also implies including the wing of the O 2 −O 2 band centered at 360 nm. To minimize the correlation between H 2 CO and O 2 −O 2 we decided to start the fitting window at 327.7 nm, and to include the H 2 CO absorption band at 353 nm we chose a finishing wavelength of 356 nm.
After selecting the fitting window we perform the direct fit of the measured radiance to the semi-empirical model of the top-of-atmosphere radiance described by Eq. (1). Prior to building the model we wavelength calibrate the radiance by cross-correlating it with a high-resolution solar spectrum (Chance and Kurucz, 2010).
The radiance, I , is calculated starting from a mean radiance reference, I 0 . We construct I 0 using earthshine radiances over the remote Pacific Ocean, where H 2 CO concentrations are assumed to be at background levels. These radiances are as close as possible to 165 • W and were measured within 1 day of the orbit we are retrieving. With the set of selected radiances over the remote Pacific Ocean we calculate a mean radiance reference for each cross-track position. While computing the mean radiance reference I 0 we only consider spectra recorded between 30 • S and 30 • N. The reason behind using the mean radiance reference instead of the solar irradiance spectrum is to avoid cross-track stripes in the retrieval, a common problem in many sensors that use 2-D CCD array detectors (Veihelmann and Kleipool, 2006). The caveat of this solution is that effectively we are retrieving the differential SCDs ( SCDs) between the observation and the radiance reference spectra. To account for this fact we apply a correction based on a reference sector normalization, which is described in higher detail in Sect. 2.4 (Khokhar et al., 2005).
In modeling the radiance we also include the effects of the Beer-Lambert law absorption (e − j α j X j ) with contributions from H 2 CO (Chance and Orphal, 2011), O 3 at two temperatures 228 and 295 K (Malicet et al., 1995), BrO (Wilmouth et al., 1999), NO 2 (Vandaele et al., 1998), the O 2 −O 2 collision complex (Thalman and Volkamer, 2013), and the Ring effect (Chance and Spurr, 1997). We also include a correction term to account for instrument undersampling of the spectra ( i α i X i ) (Chance, 1998;Chance et al., 2005). The absorption cross sections are convolved with the OMPS-NM slit functions. The functions were measured prior to launch using a tunable laser to scan the band passes of selected band centers at representative pixel locations on the focal plane; the resulting functions were then iterated to calculate a bandpass for each CCD pixel. The laser was tuned over a wavelength range of ±2 pixels for each of the OMPS-NM sensor band centers. Since the wavelength step size is coarse, approximately 0.4 pixels, five adjacent pixel responses were interpolated to a fine grid and combined to form a single observed band-pass function. In Eq. (1) we are using two closure third-order polynomials ( n α n X n + m α m X m ) to account for low-frequency features of the spectra such as Rayleigh scattering and the effects of aerosols. In Eq. (1), α represents the fitting parameter while X represents the undersampling correction, the cross sections, or the different orders of the polynomials. The characteristics of the spectral fitting are summarized in Table 1. To retrieve the SCDs we use a nonlinear least-squares inversion method implemented in the ELSUNC software (Lindström and Wedin, 1988) to minimize the cost function χ 2 defined as the weighted square difference between the measured and modeled radiance. Figure 2 shows the result of the fitting for three pixels in orbit 3538 over the Americas. These three pixels have been chosen to illustrate how with just a visual inspection, when the amount of H 2 CO is big enough, we can clearly distinguish the H 2 CO signal in the fitting (two top panels with retrieved H 2 CO concentrations of 1.8 × 10 16 and 0.8 × 10 16 , respectively), while in the last case, as we approach the detection limit, it is difficult to tell visually whether there is measurable H 2 CO and we must rely on statistics to obtain the information. Typical retrieved SCDs range between −0.5×10 15 and 2×10 16 molecules cm −2 for background levels and hotspots with fitting uncertainties of about 4.0×10 15 .  Figure 3 shows the SCDs, the fitting uncertainty, the solar zenith angle (SZA), and the relative fitting uncertainty for this orbit. The effect of the South Atlantic Anomaly (SAA) in the quality of the retrieval is evident in the SCDs and the fitting uncertainties (swath lines 20-120). Note the increase in the fitting uncertainty as the SZAs increase. Excluding the SAA, the fitting uncertainty remains around 5×10 15 molecules cm −2 or less for SZAs below 70 • . To estimate the detection limit we explore the relationship between the root mean square (RMS), H 2 CO cross sections, and its optical depth using this expression, SCD × σ max > RMS, where σ max is the maximum H 2 CO cross section in the fitting window (Schönhardt et al., 2008). With typical RMS values of ∼ 5 × 10 −4 the detection limit is estimated at ∼ 7.5 × 10 15 molecules cm −2 . As the background concentration due to CH 4 oxidation in remote regions is 2 to 7 × 10 15 molecules cm −2 , the retrieval is unable to clearly resolve background concentrations. This situation is evident when we consider the standard deviation over the remote Pacific Ocean, which is around 4 × 10 15 molecules cm −2 , similar to the concentrations expected in remote regions.

Slant column to vertical column calculation
The proportionality between VCDs and SCDs can be expressed via AMFs (Eq. 2). The AMFs contain information, some of it a priori, about the state of the atmosphere (gas con- Figure 3. SCD, fitting uncertainty, SZA and relative fitting uncertainty for orbit 3538. The effect of the SAA is clearly visible in the two top panels (swath lines 20 to 120). In the top right panel, which shows the fitting uncertainty, it is possible to appreciate the effect of splitting the central cross-track pixel  centrations, clouds, and aerosols), the characteristics of the spatial pixel footprint (surface reflectance and terrain height), and the geometry of the retrieval.
To compute AMFs we follow the theoretical approach described by Palmer et al. (2001). The main idea is that for weak trace gas absorption such as that of H 2 CO, we can describe the AMFs as the combination of a shape factor containing information about the vertical distribution of H 2 CO concentrations and the scattering weights (w(z)) describing how the radiation is transmitted throughout different layers of the atmosphere. w(z) is a function of the geometry of the observation (SZA, viewing zenith angle (VZA), and relative azimuth angle (RAA)), the altitude, the aerosol load, the surface properties (reflectance and altitude), and the characteristics of the clouds present in the scene, if any (cloud fraction and cloud pressure). In our algorithm cloudy pixels are treated by considering the independent pixel approximation (Martin et al., 2002). In this approximation the pixel w(z) is the combination of the cloud-free scattering weight (w clear ) and the cloudy scattering weight (w cloud ) as shown in Eq. (3) for a given altitude. w clear is a function of the geometry of the problem, surface albedo (a s ), surface elevation (h s ), and the altitude while w cloud depends on the cloud albedo (a c ), defined here as 0.8, the cloud centroid pressure (c p ), and the altitude. is the radiative cloud fraction.
We extract the a priori H 2 CO information from the same monthly GEOS-Chem  climatology we use in our SAO OMI H 2 CO retrieval. w(z) are precomputed using VLIDORT (Spurr, 2006) and saved in look-up tables. These tables consider the w(z) dependency with viewing geometry (SZA, VZA, and RAA), surface altitude, surface reflectance, and cloud properties. Information about the pixel surface reflectance is obtained from the Total Ozone Mapping Instrument (TOMS) climatology (Changwoo Ahn, personal communication, 2015). This surface reflectance climatology is the same one that is used by Vasilkov et al. (2014) to retrieve cloud information ( and c p ). Given the small variation of the AMF within the fitting interval (usually less than 7 %), we calculate it at one wavelength (λ), 340 nm, which we consider representative of their mean value within the fitting window. A detailed description of AMF calculations can be found in González Abad et al. (2015).

Reference sector normalization of VCDs
We are retrieving SCDs due to the fact that we are using a radiance reference as a starting point in the modeling of radiances. To obtain SCDs starting from these SCDs we calculate a normalization to modeled values. We work out the difference between the retrieval over the Pacific Ocean and the GEOS-Chem climatology. The process, inspired by earlier work by Khokhar et al. (2005) andDe Smedt et al. (2008), is described in detail by González .
Here we mention the essential steps and ideas behind it.
Using the same radiance reference we perform two retrievals. One retrieval is for the orbit of interest, for example a satellite flight over Africa and Europe, and the second one is the retrieval of the orbit we used to compute the radiance reference which is always over the remote Pacific Ocean. Using the second retrieval and the monthly GEOS-Chem climatology we work out a latitudinally dependent correction which accounts for the difference between the SCDs and the model SCDs. This correction is different for each radiance reference, so it changes daily, and is applied only to orbits retrieved with that particular radiance reference. We then assume that this correction is longitudinally constant so that we can apply the correction to results from the orbitof-interest retrieval. Before applying the correction we take into account the particularities of each scene by using the information contained in the AMFs. Equations (4) and (5) summarize the basic operations involved in the process.
We work out the amplitude of the correction at 500 latitude grid points extending from 90 • N to 90 • S. To obtain the value for each grid point we consider the median value of all j pixels of the radiance reference orbit whose central latitudes are in between any given couple of contiguous grid points. After interpolating the correction to the latitude value of each pixel (i) in the orbit of interest we apply the information contained in the particular AMF to that pixel to obtain the reference sector corrected VCD. Figure 4 shows VCDs (defined as SCD/AMF), VCDs, VCDs fitting uncertainty and RMS for orbit 6202 a January 2013 overpass of Africa and Europe. Enhanced concentrations in the middle of the orbit correspond to hotspots over Central Africa.

Error analysis
For an individual pixel VCD the associated error (ε VCD ) is given by Eq. (6), which assumes no correlation between terms. Given the small contributions arising from correlations, this is a good approximation for the description of the VCD error (Boersma et al., 2004).
In Eq. (6), ε SCD ramdon is the random component of the spectral fitting process. It can be estimated by considering the residuals of the fitting process according to Eq. (7), which takes into consideration the diagonal term of the covariance matrix for H 2 CO (C j ) and the RMS fitting residual weighted by the degrees of freedom (m is the number of spectral points and n the number of fitted variables). This method to estimate ε SCD ramdon also includes contributions from systematic errors that consistently show up in the fitting residuals. To evaluate them we have computed an average residual ("common mode") for each cross-track position and included it in the semi-empirical calculation of the radiances (Eq. 1) as an extra term. The difference in the fitting uncertainty between retrievals performed with and without the common mode is about 25 %. We attribute this difference to systematic errors due to instrumental effects such as characterization of the instrumental line shape, wavelength calibration, and stray light. As mentioned in Sect. 2.2.1, ε SCD ramdon values range between 10 % over H 2 CO hotspots to 100 % or more for pixels with low concentrations.
We have evaluated the systematic component of SCD error (ε SCD sys ) considering only the error associated with the uncertainties of the different cross sections included in the semi-empirical model of the radiances (Eq. 1). Table 2 summarizes the uncertainties associated to each molecule as reported in the literature. Considering that the uncertainties for different cross sections are not correlated we have found ε SCD sys to be about 7 %.
To estimate the error associated with AMF calculations we have performed a sensitivity analysis by perturbing one at the time the input parameters of AMF calculations. The parameters considered are cloud parameters, surface reflectance, terrain height, and GEOS-Chem climatological profiles as well as the wavelength dependency of w(s) within the fitting window. The amplitude of each perturbation is reflecting the uncertainties reported in the literature for each one of the parameters. A shortcoming of this method is that it assumes that errors due to different parameters are not correlated.
We already mentioned that the dependency with λ of w(z) is weak, around 7 %; by choosing a λ near the center of the fitting window, 340 nm, we estimate that the error due to the variation of w(z) in the fitting window is about 4 %. In the UV spectral range, most surfaces reflect between 1 and 10 % of the incoming light. The errors of the TOMS surface climatology are estimated to be around 40 %, which translates to an error of 6 % in the AMF for a typical surface with an albedo of 0.03. Brighter surfaces such as deserts with albedos of 0.1 will have an AMF error of 14 %, while for a snowy  pixel with an albedo of 0.4 the AMF error will be about 26 %. A change in the cloud pressure from 800 to 900 hPa will imply a change in the AMF from 0.6 to 1.2. The cloud product we are using is derived from analysis of the rotational Raman scattering around 350 nm (Vasilkov et al., 2014). Errors of the cloud centroid pressure are estimated to be about 50 hPa, which translates to variations in the AMF values of 23 %. A change in the cloud fraction of 0.1 for a cloud pressure of 800 hPa will change the AMF by 25 %. To finish the sensitivity analysis of the AMF calculation we tested the impact of the a priori profiles of H 2 CO. Since it is the shape of the profile what affects the AMF calculation we have applied the reported uncertainty only to layers showing the highest volume mixing ratios, defined as those that contribute to the 90 % percentile concentrations. Those uncertainties are defined as the GEOS-Chem biases with respect to in situ measurements. GEOS-Chem biases vary with location around the world. For example, GEOS-Chem simulations over the USA have biases of around 10 % (Millet et al., 2006), which translate to AMF uncertainties up to 16 %, while over tropical regions Barkley et al. (2011) reported biases up to 25 %, which translate to 40 % AMFs uncertainties.  Considering the contributions from all these factors we estimate ε AMF to be about 38 %, for an optimistic case with small GEOS-Chem profile uncertainty and cloud parameters being the most significant source of uncertainty, to 50 % when surface reflectance and GEOS-Chem profile uncertainties are at their maximum, becoming significant sources of AMF uncertainty that contribute as much as the uncertainty in the cloud parameters.
Finally we consider ε Corr , the error associated with the GEOS-Chem climatology used to perform the reference sector correction. As mentioned above this climatology is estimated to have a 10 % error, which we have assumed to be directly carried over to the final H 2 CO VCDs. This assumption intentionally provides a high error estimate.
To summarize, users of this product should assume that a typical pixel VCD will have an uncertainty ranging between 40 and 100 % for hotspots and background level concentra-tion pixels, respectively. This conclusion can be appreciated better in Fig. 5. Here we show absolute and relative error density probability functions (DPFs) for all orbits available in the months of January and July. Four million pixels are used in the calculation of January's DPFs while 4 million and a quarter are used for July's DPFs. The first conclusion is that the shape of the distribution is similar for both months but there are some differences. The relative error DPF peaks at 38 % in January and 36 % in July and both have a minimum relative error cutoff around 15 %. January errors are more spread than July errors. For example, in January 46 % of the pixels have relative errors below 60 and 73 % errors below 100 %, while in July these numbers increase to 54 and 79 %. This behavior is not surprising when we consider the higher concentrations observed in July vs. the concentrations observed in January: 58 % of the pixels in January have retrieved concentrations above 5 × 10 15 molecules cm −2 while in July this percentage increases to 68 %.

Comparison between OMI and OMPS H 2 CO retrievals
We have compared our retrieval with two OMI retrievals using monthly mean averaged VCDs at a spatial resolution of 0.25 • × 0.25 • for the time period of August 2012 to August 2013. We have only considered pixels with SZA smaller than 70 • and radiative cloud fraction below 40 %. We have explored the impact of using other radiative cloud fraction to filter pixels with results from the OMPS and OMI SAO retrievals. The results for different cloud fractions are consistent between them with the exception of winter months in southeastern USA and Europe. This situation is most likely due to the reduced total columns for those months, the viewing conditions with higher SZAs, and the reduced planetary boundary layer. One question that remains to be answered for future studies combining both OMI and OMPS products is which radiative fractions threshold should be used with each product to best combine them.
Two of the most significant advantages of comparing OMPS with OMI are the close overpass time for both instruments, around 13:30 LT, and the similar concepts of the instruments. Both designs use 2-D CCD detectors. A significant disadvantage is their difference in the spatial pixel size. Figure 6 shows an almost coincident OMPS and OMI Africa overpass to better illustrate the differences in the pixel size. OMI has a nadir pixel size of 24 × 13 km while OMPS, in the configuration we are using, has a nadir pixel size of 50 × 50 km. The apparent reduction of noise in OMPS retrievals in comparison with the OMI retrievals shown in Fig. 6 is a direct consequence of OMPS lower spatial and spectral resolution, which translate to signal-to-noise ratio (SNR) increases. Figure 7 shows the eight regions we have selected for comparison. The limits are explicitly shown in Table 3. The three retrievals are able to capture the seasonality of H 2 CO columns, showing similar patterns, as seen in Fig. 8 where we have plotted the time series of monthly means for the different regions. The error bars are the standard deviations of the pixels included in the calculation of monthly means. Temporal correlations between OMI BIRA and OMPS SAO retrievals (see Table 4 for more details) vary between a low of 0.25 over the Pacific Ocean and a maximum of 0.96 over the Amazon basin. When comparing OMI SAO and OMPS retrievals these correlations improve with a minimum of 0.72 over the Pacific Ocean and a maximum of 0.99 over the Amazon basin.   Fig. 6. Error bars represent the standard deviation of the mean value with in the region. Green lines and data points represent OMPS SAO retrievals, red is used for OMI SAO, and black for OMI BIRA. The shapes of seasonal cycles are reproduced similarly by the three retrievals despite having significant offsets, which are further analyzed in Fig. 8.
The three retrievals, however, observe different amplitudes of seasonal variations. There is also a bias that can be significant for particular months and regions. The difference between the BIRA and SAO retrievals over the Pacific Ocean is induced by the underlying difference of the chemical transport models used for the reference sector correction. As expected, the differences between OMI BIRA and OMPS retrievals are greater than those between OMI SAO and OMPS. The absolute mean offset between the OMPS SAO retrieval and the OMI BIRA is 28 %, while it is 23 % between OMI SAO and OMPS. Part of the bias between both SAO products can be explained by the differences in the AMFs. Despite using the same scattering weight look-up tables, as well as GEOS-Chem climatology, the differences between OMI and OMPS cloud products as well as surface reflectance are the cause for OMI SAO AMFs to be on average 10 % higher than OMPS SAO AMFs. This difference in AMFs can explain 10 % of the quasi-systematic 23 % low bias of OMI SAO with respect to OMPS SAO retrievals.
While these mean absolute offsets give us an idea of how close the retrievals are, from a general perspective it is more interesting to consider the differences for particular regions. The most evident differences between OMI BIRA and OMPS, easy to appreciate in Fig. 9, happen during the winter months over southeastern USA and Europe. For example, in December over southeastern USA the bias between OMI BIRA and OMPS is 75 % while between OMI SAO and OMPS is −16 %. During the same month over Europe the differences are 109 % between OMI BIRA and OMPS and 63 % between OMI SAO and OMPS. Table 5 summarizes the amplitude, maximum, and minimum of the yearly variation for each one of the retrievals and regions. The amplitude observed with OMPS is in general bigger than the amplitude observed with OMI; however, considering the errors associated with each derived amplitude the values obtained with all three retrievals for each region always fall within the error bars. We would like to remind any possible user of these products of the increased difficulty of winter retrievals over Europe due to the viewing geometry (high SZA).
With few exceptions (winter months over southeastern USA, Europe, and eastern China) retrieved H 2 CO VCDs using OMPS are higher than columns retrieved using OMI as shown by the dominant blue color in Fig. 9. This figure shows the bias between OMPS and OMI retrievals as the percentage of OMI minus OMPS over OMPS. It is clear that to have a better idea of the accuracy of the different retrievals, independent validation studies like the one recently published by Zhu et al. (2016) are needed. In this paper H 2 CO satellite products produced from different groups using different instruments are compared with in situ measurements. The products included in this comparison are BIRA OMI, GOME2A and GOME2B, SAO OMI and OMPS, and NASA Goddard OMPS. SAO OMPS retrievals are found to be biased 39 % lower over the southeastern USA between 5 August and 25 September 2013.
The error bars of Fig. 8 contain information about the variation within each region and month expressed as the standard deviation of the VCDs considered for each monthly mean calculation. In general we observe a stronger variability within one region and one month in the BIRA retrieval than in the SAO retrievals. The average VCDs variability for the retrievals considering all regions and months is 2.6 × 10 15 molecules cm −2 for OMI BIRA, 1.9 × 10 15 molecules cm −2 for OMI SAO, and 2.2 × 10 15 molecules cm −2 for the OMPS SAO retrieval. We have included Fig. 10 showing the standard deviation for each retrieval, month, and region.

Conclusions
We have developed a new OMPS H 2 CO retrieval by adapting the OMI SAO H 2 CO retrieval to OMPS-NM. Our twostep retrieval includes the direct fitting of radiances to obtain SCDs, the calculation of AMFs to convert SCDs to VCDs, and the application of a reference sector correction based on a monthly climatology derived from GEOS-Chem simulations to obtain the final VCDs. Despite the reduced spectral resolution of OMPS (1 nm) with respect to other sensors like OMI or GOME-2, the retrievals presented here provide compelling evidence of the suitability of OMPS-NM sensor to perform H 2 CO retrievals. Even more, given the good SNR of OMPS-NM the retrieval has reduced noise compared with OMI retrievals and a detection limit of 7.5 × 10 15 vs. 1 × 10 16 molecules cm −2 in the case of the OMI SAO retrieval. The improved SNR (and detection limit) comes in part from the bigger number of photons captured by larger spatial pixels by OMPS-NM with respect to OMI pixels as well as the reduced OMPS spectral resolution. Typical values of the OMPS H 2 CO VCDs range between 0.5 × 10 16 and 2×10 16 molecules cm −2 with associated errors of around 40 % over pixels with high concentrations that increase to 100 % or even more for pixels with low concentrations.
We have compared the OMPS SAO retrieval with OMI retrievals developed at BIRA and SAO. While all three retrievals similarly capture the general trends of seasonal variation, they show significant differences in the amplitude of the seasonal signals with significant offsets between them as well. As expected, the agreement between the SAO retrievals is better than the agreement between the SAO OMPS and BIRA OMI retrieval. The mean offset for selected regions with strong seasonal variations plus the remote Pacific Ocean is 23 % between the SAO retrievals and 28 % between the SAO OMPS and OMI BIRA retrievals. Obviously, to have an assessment of the absolute accuracy of the retrieval it is necessary to perform validation studies with independent in situ data. These studies are outside of the scope of this paper but will constitute part of ongoing efforts to validate the latest version of the SAO OMI retrieval. Table 5. Minimum (Min.), maximum (Max.), and amplitude (Amp.) of monthly mean time series for each region and retrieval. H 2 CO VCDs are expressed in 10 15 molecules cm −2 . The error associated with each value is the standard deviation of the values considered in the calculation of the monthly mean. In parentheses, the month in which the minimum or the maximum was observed is indicated. Pacific Ocean 0.3 ± 0.1 0.4 ± 0.2 0.2 ± 0.2 0.4 ± 0.1 0.4 ± 0.1 0.0 ± 0.1 0.4 ± 0.1 0.4 ± 0.1 0.0 ± 0.1 (Aug) (Dec) (Oct) (Mar) (Oct) (Mar) Southeastern USA 0.4 ± 0.3 1.0 ± 0.3 0.6 ± 0.4 0.2 ± 0.2 0.8 ± 0.2 0.6 ± 0.2 0.2 ± 0.2 1.3 ± 0.3 1.1 ± 0.4 (Feb) (Jul) (Feb) (Jul) (Feb) (Aug) Amazon basin 0.5 ± 0.3 1.4 ± 0.5 0.9 ± 0.5 0.6 ± 0.2 1.5 ± 0.4 0.9 ± 0.4 0.8 ± 0.2 2.1 ± 0.7 1.3 ± 0.7 . Monthly mean offsets between OMPS SAO retrieval and OMI retrievals from BIRA (left panel) and SAO (right panel). The offsets are expressed as percent of the OMI retrieval minus the OMPS retrieval. As expected the biases between OMPS SAO and OMI BIRA retrievals are bigger than the biases between OMPS SAO and OMI SAO. Southern Africa, Europe, and southeastern USA show the most relevant differences.
Finally we want to highlight the benefit of having similar retrievals from two instruments overlapping in time. The data record for OMI already extends for more than 10 years. OMPS was launched on board the SUOMI-NPP satellite in October 2011 and it is therefore in the early stages of its operational life. The ongoing overlapping period between both instruments offers a great opportunity to construct a consistent extended long-term data set. To fully exploit the sci- entific benefits from this opportunity further studies analyzing the impact of different spectral and spatial resolutions in the retrievals of H 2 CO are necessary. The prospect of other OMPS instruments being launched in future missions of the National Oceanic and Atmospheric Administration (NOAA) such as the Joint Polar Satellite System (JPSS) will create the possibility of a multi-decadal set of consistent measurements of H 2 CO from afternoon LEO platforms. Another instrument that will contribute to the afternoon constellation, TROPOMI/S5P, will be launched in 2016 (Veefkind et al., 2012).

Data availability
SAO OMPS H 2 CO retrievals have been performed using OMPS radiance data publicly accessible at https://ozoneaq. gsfc.nasa.gov/data/omps/ . SAO OMPS H 2 CO can be obtained by contacting the corresponding author. OMI SAO and OMI BIRA H 2 CO retrievals are publicly available following the links included in Sect. 4.