Impact of Spaceborne Carbon Monoxide Observations from the S-5 P platform on 1 Tropospheric Composition Analyses and Forecasts 2 3

Impact of Spaceborne Carbon Monoxide Observations f rom the S-5P platform on 1 Tropospheric Composition Analyses and Forecasts  2 3 R. Abida, J.-L. Attié, L. El Amraoui , P. Ricaud, W. Lahoz, H. Eskes , A. Segers , L. Curier, J. 4 de Haan, J. Kujanpää , A. O. Nijhuis, J. Tamminen , R. Timmermans , and P. Veefkind 4 5 6 1 CNRM-GAME, Météo-France/CNRS UMR 3589, Toulouse, Fr ance 7 2 Université de Toulouse, Laboratoire d’Aérologie, CN RS UMR 5560, Toulouse, France  8 3 NILU – Norwegian Institute for Air Research, P.O. B ox 100, 2027 Kjeller, Norway  9 4 Royal Netherlands Meteorological Institute (KNMI), P.O. Box 201, 3730 AE De Bilt, The 10 Netherlands 11 5 TNO, Business unit Environment, Health and Safety, P.O. Box 80015, 3508 TA Utrecht, The 12 Netherlands 13 6 Finnish Meteorological Institute, Earth Observation U it, P.O. Box 503, 00101 Helsinki, Finland 14


32
Over the last decade, the capabilities of satellite instruments for sensing the lower troposphere have 33 improved, and opened the way for monitoring and better understanding of atmospheric pollution processes, 34 e.g., tropospheric chemistry (Jacob, 2000), long-range transport (HTAP, 2007), and emissions (e.g. Streets, 35 2013 and references therein). Satellite instruments provide global measurements of many pollutants (e.g., 36 ozone; carbon monoxide, CO; nitrogen dioxide, NO 2 ; and aerosols), including information on their trans-37 boundary transport, and complement in situ measurements from ground-based stations (e.g., the European quality (AQ). The challenge for future space-borne missions will be to assess directly the local scales of 49 transport and/or chemistry for tropospheric pollutants (1 hour or less, 10 km or less) and to facilitate the use 50 of remote sensing information for improving local-and regional-scale (from country-wide to continental 51 scales) AQ analyses and forecasts. Building on this effort, various LEO satellite platforms and/or 52 constellations of GEO satellite platforms will help extend AQ information from continental scales to global 53 scales (e.g., Lahoz et al., 2012, and references therein for LEO/GEO platforms; Barré

86
The goal of the S-5 and S-5P platforms is to provide global daily measurements of atmospheric pollutants 87 Page 5 of 56 (e.g., CO, ozone, NO 2 , SO 2 , BrO, and formaldehyde), climate related trace gases (e.g., methane, CH 4 ) and 88 aerosols, at relatively high spatial resolution (from below 8 km to below 50 km, depending on wavelength).

90
The S-5P is the ESA pre-operational mission required to bridge the gap between the end of the OMI (Ozone The OSSE concept consists of simulating observations and their associated errors from a representation of 157 reality (the "Nature Run" or NR) and providing this information to a data assimilation system to produce 158 estimates of the NR states. Thereafter, one compares these estimates of the NR states from an assimilation

185
We obtain the NR by combining the LOTOS-EUROS CO profiles from the surface to 3.5 km with the TM5 186 CO profiles from 3.5 km to the top of the atmosphere (identified by the TM5 model top at 0.1 hPa). We use 187 spatial interpolation to merge the values near the boundary between the two models at a height of 3.5 km.

188
The model simulations used to construct the NR have a spin-up period of three months. We archive the NR 189 output data on an hourly basis. To enable sounding of the lower atmosphere at finer scales, TROPOMI has an unprecedented spatial 256 resolution of 7x7 km 2 at nadir. This relatively high spatial resolution is necessary for air quality applications 257 at local to regional scales. It will resolve emission sources with 15% of accuracy and 10% precision 258 (

Cloud properties 285
We obtain cloud fields from the high-resolution operational weather forecast archive of the ECMWF. We

Averaging kernel and measurement uncertainty lookup tables 304
Because of the large number of observations that will become available from the S-5P instrument, full 305 radiative transfer calculations for each observation separately are not feasible. We thus choose to build look-306 up tables for a set of geometries based on a radiative transfer code that employs the adding-doubling method 307 in combination with optimal estimation (using the radiative transfer toolbox DISAMAR; de Haan, 2012).  Table 1. We provide the kernels on 21 pressure levels between 1050.0 and 0.1 hPa. We specify the 312 uncertainties for clear-sky and cloudy-sky separately.

314
Each simulation with DISAMAR consists of a forward calculation of the satellite-observed spectrum, 315 followed by a retrieval step based on the optimal estimation method (Rodgers, 2000). We convert instrument 316 Page 13 of 56 noise, listed in Table 1, into uncertainties for the retrieved CO column. We take a-priori trace gas profiles 317 from the CAMELOT study (Levelt et al., 2009). As indicated above, we assume that both the cloud and the 318 surface are Lambertian reflectors. Kujanpää   The S-5P will produce large amounts of data owing to its wide swath and relatively high spatial resolution of 370 about 7x7 km 2 . Thus, a pre-processing step is necessary to reduce the data volume for the data assimilation 371 Page 15 of 56 experiments. For this study, we consider only pixels inside the OSSE simulation domain (Note that retrieval 372 pixels in each single cross-track are essentially instantaneous measurements of CO.). This has the advantage 373 of alleviating the data volume burden. However, a single cross-track over Europe could have more than 374 80,000 valid retrieval pixels. Furthermore, each individual pixel is associated with an averaging kernel vector 375 given at 34 vertical pressure levels, from the surface up to the top of the atmosphere (identified as 0.1 hPa). 376 377 Figure 4 shows an example of averaging kernels at the surface, as well as the averaging kernels 378 representative of retrievals including pixels with different cloud fractions (less than 10%, greater than 30%, 379 and greater than 80%). In addition, we discard data points with standard deviation exceeding 20% of the 380 retrieval or with solar zenith angles larger than 80%. The retrieval over sea is noise-dominated. Because of 381 this, we only consider CO partial columns above cloudy sea scenes with cloud fraction more than 80% and The spatial binning not only reduces considerably the data volume but also results in an improved spatial 399 representativeness of the CO measurements by reducing the random error of each data pixel. 400 401

The Control Run 402
To generate the CR, it is important to use a state-of-the-art modelling system, which simulates the

408
Consequently, by using two independent models the OSSE will simulate more realistically the assimilation 409 of real observations. This allows us to design an OSSE that is not too overoptimistic.

Evaluation of the assimilation run 485
In this section, we evaluate the impact of the assimilation of the S-5P CO total column. First, we evaluate the 486 consistency of the assimilation run by separating the clear-sky pixels from their cloudy counterparts (Sect.

Consistency of the assimilation run 492
We perform two OSSEs. The first one includes all pixels in the OSSE domain, regardless of whether they are 493 cloudy or clear-sky and the second only includes clear-sky pixels. We consider a pixel to be clear when the 494 cloud fraction is less than 10%. Comparison of the ARs from these two OSSEs indicates that the impact of 495 including all pixels is small. The largest differences between the respective ARs in relation to the NR are 4% 496 in regions over North Europe (North Sea and Scandinavia), with the AR for clear-sky pixels closer to the NR 497 (not shown). We can explain these results by the fact the summer generally has low amounts of cloud.

498
Consequently, we only present the results from the OSSE with all pixels.

499
To evaluate the AR, we calculate the χ 2 diagnostic associated with the Observation minus Forecast (OmF) 500 differences (see, e.g., Lahoz et al., 2007a). Here, we normalize the OmF differences by the background error.

501
We also calculate histograms of the Observation minus Analysis (OmA) differences, the observation and the 502 simulation from the CR (observation-minus-control run, hereafter OmC) differences, and the OmF 503 differences. We use the observational error to normalize the differences building the histograms of OmA, 504 OmC and OmF.

Study of increments 543
To understand further the impact on the surface CO field of the simulated S-5P CO total column The shape of the S-5P increments is similar to that of typical SCIAMACHY analysis increments, which also 568 extend through a deep layer and have a maximum at the surface (Tangborn et al., 2009). The fact that both 569 these analysis increments stretch out over a deep layer is owing to similarities in the S-5P and SCIAMACHY 570 averaging kernels -both are close to unity over cloud-free land (see Fig. 5). Note that the situation shown in 571 Fig. 7 is a snapshot and depends on the particular conditions for this time. An average of the increments over 572 the summer period would tend to show a uniform distribution in height. 573 574 Figure 8 shows the fields of surface CO from the CR, and the NR and the AR, averaged over the northern 577 summer period. One can see the general change of CO over land between the CR (top left panel) and the AR 578 (bottom panel). We can ascribe this to the contribution of simulated S-5P total column CO data sampled from 579 the NR. This figure shows several differences between the CR and AR fields that indicate the superior 580 behaviour of the AR in capturing features in the NR. For example, over Eastern Europe and Russia, the AR 581 CO concentration values are closer to those in the NR (with a mean bias between -1.5 and +1.5 ppbv); in 582 particular, the CR shows generally lower values than in the NR (mean bias around -6 ppbv). Nevertheless, 583 over Portugal, where the NR shows the forest fires that occurred over the summer, the AR captures them 584 only slightly better than the CR. We expect this relatively poor performance of the CR regarding fires, as the 585 fires are not included in the CR set-up (see Sect. 2.4). Although the AR, in the operational set-up, captures 586 the CO concentrations emitted by forest fires slightly better than the CR (through assimilation of CO 587 measurements), the relatively poor temporal resolution of the S-5P ultimately limits its performance.

588
However, the most important deficiency is due to the criterion used in the operational set-up in which we 589 activate a data-screening test to discard observations far away from the model (see section 3.2.5). A 590 geostationary satellite, given its relatively high temporal resolution, should be able to capture better the 591 temporal variability of CO from these forest fires (Edwards et al., 2009). 592 593

Statistical metrics 594
Page 23 of 56 In this section, we provide a quantitative assessment of the benefit from S-5P CO total column measurements  Figure 9 presents the zonal and meridional means of the difference between the CR and the AR averaged 623 over the northern summer 2003 (1 June -31 August). We also plot the confidence interval representing the 624 areas where the AR is not significantly different to the CR at the 99% confidence limit (highlighted in the 625 grey colour). These two figures show that there is benefit from the S-5P CO total column data over the first 626 few bottom levels of the troposphere, i.e., the lowermost troposphere. Between the surface and 800 hPa, a 627 negative peak is present in the zonal difference field (over Scandinavia), and in the meridional difference 628 field (over Eastern Europe). Note that the zonal field shows two areas, one with positive values and the other 629 with negative values representing a CR greater than the AR and a CR smaller than the AR, respectively. The 630 positive peak, at a slightly higher level (i.e., lower pressure) than the negative peak, is representative of the 631 Mediterranean Sea, whereas the negative peak is more representative of the land areas (Scandinavia and 632 Eastern Europe). We also calculate the RMSE as well as the reduction rate of the RMSE, RMSERR (Figure 11), both keeping 647 the systematic error (Fig. 11, top), and removing the systematic error (Fig. 11, bottom). We calculate the bias 648 in the AR and CR by subtracting the NR field from each of them, producing an unbiased AR and CR. For the 649 Page 25 of 56 case where we remove the systematic error, we perform the statistics on the unbiased AR and CR. If we 650 examine the RMSE statistics, Fig. 9   In Figure 12, we show the correlation between the CR and the NR, and the correlation between the AR and 665 the NR, at the surface for the three northern summer months (1 June -31 August). The AR is closer than the 666 CR to the NR with the correlation coefficient reaching 0.9 over land. By contrast, the correlation coefficient 667 between the CR and the NR is typically less than 0.5, with very low values over Eastern Europe, where CO 668 sources are sparse.  (25°E-53°N), where the 675 reduction of RMSE (i.e., RMSERR) is much larger than for other regions (Fig. 13, bottom panel). For all 676 three areas, the AR is generally closer to the NR than the CR, showing the impact of the simulated 677 observations. We calculate the biases between the AR and CR vs the NR by computing the difference NR-X,

678
Page 26 of 56 where X is AR or CR, and normalizing by the number of observations over the northern summer period. The

726
We perform a regional-scale Observing System Simulation Experiment (OSSE) over Europe to explore the 727 impact of the LEO satellite mission S-5P carbon monoxide (CO) total column measurements on lowermost 728 tropospheric air pollution analyses, with a focus on CO surface concentrations and the Planetary Boundary 729 Layer (PBL). The PBL varies in depth throughout the year, but is contained within the lowermost 730 troposphere (heights 0-3 km), and typically spans the heights 0-1 km. We focus on northern summer 2003, 731 which experienced a severe heat wave with severe societal impact over Europe. 732 733 Our guiding principle in the set-up of this OSSE study is to avoid overoptimistic results. To achieve this, we 735 address several factors considered likely to contribute to an overoptimistic OSSE. (i) We use different 736 models for the NR and the OSSE experiments. (ii) We check that the differences between the NR and actual 737 measurements of CO are comparable to the CO field differences between the model used for the OSSE and 738 the NR. (iii) We remove the systematic error (calculated as the bias against the NR) in the OSSE outputs (AR 739 and CR) and compare the unbiased results to the NR. (iv) We perform a quantitative evaluation of the OSSE 740 results, including performing statistical significance tests, and self-consistency and chi-squared tests. Based 741 on the specifications of the TROPOMI instrument, we anticipate relatively low CO column uncertainties of 742 around 5% over the European continent. Finally, our approach was to study the performance of S-5P alone 743 without taking into account other existing or future missions (i.e., MOPITT, CrIS or IASI).

745
The OSSE results indicate that simulated S-5P CO total column measurements during northern summer 2003 746 benefit efforts to monitor surface CO. The largest benefit occurs over land in remote regions (Eastern 747 Europe, including Russia) where CO sources are sparse. Over these land areas, and for the case when we 748 remove the systematic error, we obtain a lower RMSE value (by ~10 ppbv) for the AR than for the CR, in 749 both cases vs the NR. Over sea and Scandinavia, we also obtain a lower RMSE (by ~10%) for the AR than 750 for the CR, in both cases vs the NR. Consistent with this behaviour, we find the AR is generally closer to the 751 NR than the CR to the NR, with a correlation coefficient reaching 0.9 over land (NR vs AR). By contrast, the 752 correlation coefficient between the CR and the NR is typically less than 0.5, with very low values over 753 Eastern Europe, where CO sources are sparse. In general, for all the metrics calculated in this paper, there is 754 an overall benefit over land from the S-5P CO total column measurements in the free troposphere, but also at 755 the surface. Significance tests on the CR and AR results indicate that, generally, the differences in their 756 performance are significant at the 99% confidence level. This indicates that the S-5P CO total column 757 measurements provide a significant benefit to monitor surface CO.

759
We further show that, locally, the AR is capable of reproducing the peak in the CO distribution at the surface 760 due to forest fires (albeit, weaker than the NR signal), even if the CR does not have the signature of the fires 761 in its emission inventory. A second OSSE shows that this relatively weak signal of the forest fires in the AR 762 arises from the use of a default criterion to discard CO total column observations too far from model values, Page 29 of 56 a criterion not appropriate to situations resulting in excessive values in the CO concentrations, as is the case 764 for forest fires. This second OSSE shows a much stronger signal in the AR, which is now much closer to the 765 NR than the CR, confirming the benefit of S-5P CO total column measurements and the limitations of using 766 standard operational criteria in this case. 767 768 Further work will involve extending the OSSE approach to other S-5P measurements, such as ozone total 769 column, and NO 2 and formaldehyde tropospheric columns. These studies will complement similar studies on 770 the benefit from Sentinel-4 and -5 measurements. Collectively, these OSSE studies will provide insight into 771 the relative benefits from the Sentinel-4, -5 and -5P platforms for monitoring atmospheric pollution