Recent ozone trends in the Chinese free troposphere: role of the local emission reductions and meteorology

25 Free tropospheric ozone (O3) trends in the Central East China (CEC) and export regions are investigated for 2008-2017 using the IASI O3 observations and the LMDZ-OR-INCA model simulations, including the most recent Chinese emission inventory. The observed and modeled trends in the CEC region are -0.07 ± 0.02 DU/yr and -0.08 ± 0.02 DU/yr respectively for the lower free troposphere (3-6km column), and -0.05 ± 0.02 DU/yr and -0.06 ± 0.02 DU/yr respectively for the upper free troposphere (6-9km column). The statistical p-value is smaller to 0.01 for all the derived trends. A good agreement between the observations 30 and the model is also observed in the region including Korea and Japan and corresponding to the region of pollution export from China. Based on sensitivity studies conducted with the model, we evaluate at 60% and 52% the contribution of the Chinese anthropogenic emissions to the trend in the lower and upper free troposphere, respectively. The second main contribution to the trend is the meteorological variability (34% and 50% respectively). These results suggest that the reduction of NOx anthropogenic emissions that occurred since 2013 in China lead to a decrease in ozone in the Chinese free troposphere, 35 contrary to the increase in ozone at the surface. We designed some tests to compare the trends derived by the IASI observations and the model to independent measurements such as IAGOS or other satellite measurements (OMI/MLS). These comparisons https://doi.org/10.5194/acp-2021-476 Preprint. Discussion started: 5 July 2021 c © Author(s) 2021. CC BY 4.0 License.

do not confirm the O3 decrease and stress the difficulty to analyze short-term trends using multiple datasets with various sampling and the risk to overinterpret the results.

Introduction
Tropospheric ozone is a harmful pollutant close to the surface impacting human health and ecosystems (Lelieveld et al., 2015;Monks et al., 2015). Tropospheric ozone is also a short-lived climate forcer with an impact on surface temperature greatest in 5 the upper troposphere lower stratosphere (UTLS) and then contributes to climate change (Riese et al., 2012). The recent Tropospheric Ozone Assessment Report (TOAR) has stated that free tropospheric O3 increased during industrial times and the last decades (Gaudel et al., 2018;Tarasick et al., 2019). At the surface, the trends depend on the considered regions: a decrease is observed during summertime in North America and in Europe, and an increase is observed in Asia (e.g. Gaudel et al., 2018Gaudel et al., , 2020. However, conclusions are more difficult to draw for the recent trends of tropospheric ozone. In addition to the statistical 10 robustness of these trends, Gaudel at al. (2018) point out inconsistencies between satellite trends derived from ultraviolet (UV) sounders, which show mainly positive trends (e.g. Cooper et al., 2014;Ziemke et al., 2019) and infrared (IR) sounders, which shows mainly negative trends (Wespes et al., 2017).
In China and Central East China (CEC), one of the most polluted regions worldwide (e.g. Wang et al., 2017;Fan et al., 2020), stringent pollutant emission controls for NOx, SO2, and primary PM (particulate matter) emissions have been applying during 15 the last decade (Zhang et al., 2019;Zheng et al., 2018). The main objective of these restrictions was to decrease primary and secondary PM concentrations (e.g. Zhai et al., 2019;Zhang et al., 2019). However, these reductions have leaded to a worsening of urban ozone pollution (Li et al., 2020;Liu and Wang, 2020a, b;Lu et al., 2018;Ma et al., 2021), attributed to O3-precursors reductions in the large urban VOC-limited regions and directly to the aerosol reductions, which slow down the aerosol sink of hydroperoxy radicals (RO2) and then increase the ozone production (Li et al., 2019;Ma et al., 2021). Most of the studies are 20 based on surface observations and model simulations.
Satellite observations are more difficult to use to derive information on surface ozone due to their lack of sensitivity to surface concentration. Shen et al. (2019) show a relatively good correlation between OMI and surface measurements, especially in Southern China and state a possibility to infer trends for the subtropical latitudes. This was already partly reported by Hayashida et al. (2015). For individual events, IASI (Dufour et al., 2015) and IASI+GOME2 (Cuesta et al., 2018) products 25 show abilities to inform on pollution events in the North China Plain. The IASI+GOME2 O3 product shows a better ability to reproduce ozone surface concentrations with good comparisons with surface measurements in Japan .
Despite this encouraging partial sensitivity to surface or boundary layer ozone, satellite observations such as IASI are mostly suited to probe free tropospheric ozone. IASI is however able to separate, at least partly, the information from the lower and the upper troposphere with a maximum of sensitivity between 3 and 6 km (Dufour et al., 2010(Dufour et al., , 2012(Dufour et al., , 2015. Based on the IASI 30 observations, Dufour et al. (2018) discuss lower tropospheric O3 trends (surf-6km) over the NCP for the 2008-2016 period and associate driving factors using a multivariate regression model. They show that O3 trend derived from IASI is negative (-0.24 DU/yr or -1.2 %/yr) and explained by large-scale dynamical processes such as El Niño and changes in precursors emissions since 2013. The hypothesis to explain the negative impact of precursors reduction compared to the positive one at the surface is related to the chemical regime turning from VOC-limited at the surface to NOx-limited in altitude. In this study, we question the ability of IASI to derive free tropospheric ozone trends in China by comparison with the state-of-the-art global chemistry- in-situ measurements and compared in section 3. Section 4 presents the observed and simulated O3 trends in the troposphere.
The results are discussed in section 5.   (Clerbaux et al., 2009). Three versions of the instrument are currently operational on the same orbit: one aboard the 5 Metop-A platform since October 2006, one aboard the Metop-B platform since September 2012, and one aboard the Metop-C platform since November 2018. The IASI instruments operate in the thermal infrared between 645 and 2760 cm -1 with an apodized resolution of 0.5 cm -1 . The field of view of the instrument is composed of a 2 × 2 matrix of pixels with a diameter at nadir of 12 km each. IASI scans the atmosphere with a swath width of 2200 km and crosses the equator at two fixed local solar times 9:30 am (descending mode) and 9:30 pm (ascending mode), allowing the monitoring of atmospheric composition twice 10 a day at any location.
Ozone profiles are retrieved from the IASI radiances using the KOPRA radiative transfer model, its inversion tool (KOPRAFIT) and an analytical altitude-dependent regularization method as described in (Eremenko et al., 2008) and (Dufour et al., 2012(Dufour et al., , 2015. In order to avoid the potential impact of versioning of the auxiliary parameters (such as temperature profile, clouds screening, etc) on the ozone retrieval (Van Damme et al., 2017), surface temperature and temperature profiles are 15 retrieved before the ozone retrieval. A data screening procedure is applied to filter cloudy scenes and to insure the data quality (Eremenko et al., 2008;Dufour et al., 2010Dufour et al., , 2012. Three different a priori and constraints (polar < 10 km, midlatitudes -10-14 km, tropical > 14 km) are used depending on the tropopause height, which is based on the 2 PV geopotential height product from the ECMWF (European Center for Medium-range Weather Forecasts). The a priori profiles are compiled from the ozonesonde climatology of McPeters et al. (2007). Compared to the previous version of the ozone product (Dufour et al., 20 2018), water vapor is fitted simultaneously with ozone to account for remaining interferences in the spectral windows used for the retrieval and improve the retrieval in the current version v3.0 of the product. From the retrieved profiles, different ozone partial columns can be calculated. In this study, we consider four partial columns: the lowermost tropospheric (LMT) column from the surface to 3 km (named 0-3km), the lower free tropospheric (LFT) column from 3 km to 6 km (named 3-6km), the upper free tropospheric (UFT) column from 6 km to 9 km (named 6-9km), and the upper tropospheric -lowermost 25 stratospheric (UT-LMS) column from 9 km to 12 km (named 9-12km). Note that only the morning overpasses of IASI are considered for this study in order to remain in thermal conditions with a better sensitivity to the lower troposphere. To cover a larger period, we also consider only IASI on Metop-A.

Surface measurements
Observational data are issued from a freely available dataset accessible at https://quotsoft.net/air/ (last access: 3 May 2021).
The dataset provides hourly data of criteria pollutants SO2, O3, NO2, CO, PM2.5 and PM10 consolidated every day in near real time from May 2014. Only national level data are available in this dataset for about 1300 stations through mainland China. In 10 this study we consider only the stations with more than 50% of measurements available to ensure a good temporal coverage for the entire period (2014)(2015)(2016)(2017). In the domain shown in Fig. 1, this corresponds to 685 stations. We classified the stations in different types of environment: mountain, rural, suburban, urban and traffic based on the approach developed by Flemming et al. (2005) for Europe. This method has the advantage of not requiring any additional information other than the pollutant concentration. The relative amplitude of the diurnal cycle of O3 observations is used to evaluate the representative environment 15 of the station: the larger the amplitude of the diurnal ozone cycle is, the more the station is in an environment close to anthropogenic sources. In our case, each station has been evaluated on the studied period (i.e. 2014-2017).

IAGOS data
IAGOS (In-Service Aircraft for Global Observing System, http://www.iagos.org) is a European Research Infrastructure dedicated to measure air composition (Petzold et al., 2015). The program counts more than 62,000 flights between 1994 and 20 2021 with ozone measurements. For the purpose of this study, we used all profiles of ozone at any time of day available above Northeast China / Korea between 2011 and 2017. On board the IAGOS commercial aircraft, ozone is measured using dualbeam ultraviolet absorption monitor (time resolution of 4 s) with an accuracy and a precision estimated at about 2 nmol mol-1 and 2% respectively. Further information on the instrument is available in (Thouret et al., 1998;Nédélec et al., 2015). Longterm quality and consistency have been assessed by Blot et al. (2020). 25

OMI/MLS
OMI/MLS tropospheric column ozone is described by Ziemke et al. (2019). The OMI/MLS ozone product represents monthly means for October 2004-present at 1 o × 1.25 o resolution and latitude range 60 o S -60 o N. Tropospheric column ozone is determined by subtracting co-located Microwave Limb Sounder (MLS) stratospheric column ozone from OMI total column ozone each day at each grid point. Tropopause pressure used to determine MLS stratospheric column ozone invoked the WMO 2 K.km -1 lapse-rate definition from NCEP re-analyses. OMI total ozone data are available from https://ozonewatch.gsfc.nasa.gov/data/omi/ (last access: 22 April 2021). MLS ozone data can be obtained from https://mls.jpl.nasa.gov/products/o3_product.php/ (last access: 22 April 2021). Estimated 1σ precision for the OMI/MLS monthly-mean gridded TCO product is 1.3 DU.

LMDZ-OR-INCA model 5
The LMDZ-OR-INCA global chemistry-aerosol-climate model (hereafter referred to as INCA) couples on-line the LMDZ (Laboratoire de Météorologie Dynamique, version 6) General Circulation Model (Hourdin et al., 2006) and the INCA (INteraction with Chemistry and Aerosols, version 5) model (Hauglustaine et al., 2004). The interaction between the atmosphere and the land surface is ensured through the coupling of LMDZ with the ORCHIDEE (ORganizing Carbon and Hydrology In Dynamic Ecosystems, version 9) dynamical vegetation model (Krinner et al., 2005). In the present configuration, 10 the model includes 39 hybrid vertical levels extending up to 70 km. The horizontal resolution is 1.25° in latitude and 2.5° in longitude. The primitive equations in the GCM are solved with a 3 min time-step, large-scale transport of tracers is carried out every 15 min, and physical and chemical processes are calculated at a 30 min time interval. For a more detailed description and an extended evaluation of the GCM we refer to Hourdin et al. (2006). INCA initially included a state-of-the-art CH4-NOx-CO-NMHC-O3 tropospheric photochemistry (Hauglustaine et al., 2004;Folberth et al., 2006). The tropospheric photochemistry 15 and aerosols scheme used in this model version is described through a total of 123 tracers including 22 tracers to represent aerosols. The model includes 234 homogeneous chemical reactions, 43 photolytic reactions and 30 heterogeneous reactions.
Please refer to Hauglustaine et al. (2004) and Folberth et al. (2006) for the list of reactions included in the tropospheric chemistry scheme. The gas-phase version of the model has been extensively compared to observations in the lower-troposphere and in the upper-troposphere. For aerosols, the INCA model simulates the distribution of aerosols with anthropogenic sources 20 such as sulfates, nitrates, black carbon, particulate organic matter, as well as natural aerosols such as sea-salt and dust.
Ammonia and nitrates aerosols are considered as described by Hauglustaine et al. (2014). The model has been extended to include an interactive chemistry in the stratosphere and mesosphere (Terrenoire et al., 2021). Chemical species and reactions specific to the middle atmosphere were added to the model. A total of 31 species were added to the standard chemical scheme, mostly belonging to the chlorine and bromine chemistry, and 66 gas phase reactions and 26 photolytic reactions. 25 In this study, meteorological data from the European Center for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis have been used to constrain the GCM meteorology and allow a comparison with measurements. The relaxation of the GCM winds towards ECMWF meteorology is performed by applying at each time step a correction term to the GCM u and v wind components with a relaxation time of 2.5 h (Hauglustaine et al., 2004). The ECMWF fields are provided every 6 hours and interpolated onto the LMDZ grid. 30 The historical global anthropogenic emissions are taken from the Community Emissions Data System (CEDS) inventories (Hoesly et al., 2018)  acetone and methanol as well as NO soil emissions as described by Messina et al. (2016).
3 Evaluation of the IASI O3 satellite product and model simulations 5

Validation of the IASI O3 product with ozonesondes and IAGOS
We present here a short validation of the version v3.0 of the IASI O3 product developed by LISA. A detailed validation will be provided in a dedicated paper. The coincidence criteria used for the validation are 1° around the station, a time difference smaller than +/-6 hours and a minimum of 10 clear-sky pixels matching these criteria. No correction factor has been applied on ozonesonde measurements as our main concern is the (lower) troposphere. The results of the comparison between IASI 10 ozone retrievals and ozonesonde measurements are summarized in Table 1 for different partial columns in the troposphere.
Normalized mean biases (NMB) for the different partial columns remain very small (<2%). The estimated errors given by the normalized root mean square error (NRMSE) range between 10 and 20% depending on the partial columns and the Pearson correlation coefficient (R) is larger or equal to 0.79. Note that these results are based on the comparison with ozonesonde profiles smoothed by the averaging kernels of the IASI retrieval. If we compare with the raw sonde profiles without any 15 smoothing, the results are slightly degraded but remain good with the normalized biases within +/-5%, the errors smaller than 30% and the correlations larger than 0.6. Version v3.0 of the O3 IASI product reduces biases and increases the correlation with the ozone sondes measurements. The bias reduction is the most effective in the upper troposphere.   In addition, we compared IASI ozone partial columns with IAGOS ozone partial columns, calculated from profiles measured above Chinese and Korean airports for the period 2011-2017. In this region, the IAGOS coverage was too sparse before this 25 period. We use coincidence criteria similar to those of the ozonesondes for the IASI pixels taking the latitude and longitude of the center of the IAGOS profile as reference. One difficulty for comparing IASI and IAGOS arises from the top of the IAGOS profiles which are much lower in altitude than for the sondes for example. We select IAGOS profiles with top measurements not lower than 500 hPa: 213 profiles are then selected for 2011-2017 in coincidence with IASI over about 1000 IAGOS profiles available before filtering. We extend the IAGOS profiles with the a priori profiles used in the IASI retrieval from the top of IAGOS measurements to 60 km altitude in order to apply the averaging kernels. Then, the comparison between IASI and 5 IAGOS is the most meaningful below 500hPa and for the partial columns representative of the lower free troposphere. We focus more on the lower free tropospheric column (3-6km) where the IASI retrievals are the most sensitive (Dufour et al., 2012). Results are summarized in Table 1. The normalized mean bias and the normalized root mean square error estimate between IASI and IAGOS are -1.8% and 17% respectively when AK are applied to IAGOS profiles (-5.6% and 18% without AK applied) for this column in agreement with global ozonesonde validation. The correlation is smaller (0.62) than the one 10 with the sondes (0.79). Statistics for the other columns are also reported in Table 1. The agreement is still good in terms of bias in the LMT (0-3km) column but degrades in terms of correlation. The temporal coincident criterion (+/-6 hours) combined with a non-negligible influence of the ozone diurnal cycle (Petetin et al., 2016) on the LMT (0-3km) column might explain the worser correlation for this column (0.45). For the 6-9km column, only 50 profiles over 213 reach 300 hPa, then the IAGOS profiles are largely mixed with the a priori profile used in the IASI retrieval. The evaluation of IASI using IAGOS is then 15 difficult for this column.

Evaluation of the INCA simulations
The model is evaluated using the Chinese surface network described in section 2.3. As the model resolution is coarse and not representative of urban situations, we compare the model only with the rural type stations. The daily O3 concentrations simulated by the model are compared to the daily averages calculated from the hourly surface measurements provided by the 20 Chinese surface network. The normalized mean bias between INCA and the surface stations is 12% over the Chinese domain considered, INCA being larger. The correlation and the normalized root mean square error (NRMSE) are 0.42 and 50% respectively. On average, the model shows relatively good performances, especially in terms of bias. However, the performances of the model to reproduce the ozone concentration depend on the region. A good agreement is observed in the CEC region with bias of 3%, correlation of 0.49, and NRMSE of 48% respectively (Table 2). In the BTH region, north of the 25 CEC, the modeled O3 concentrations are smaller than the observed ones (-13%) but the correlation is higher (0.68). In the south of the CEC, the comparison in the YRD region remains satisfying in terms of bias (18%) but is degraded for the correlation (0.34) and the NRMSE (53%). In the coastal region of PRD, the available stations are within one model grid cell including land and sea. The coarse resolution of the model likely limits its capability to reproduce correctly the O3 concentrations of the coastal stations: the model overestimates the surface measurements (41%) with large NRMSE (60%) and 30 poor correlation (0.44). In the SCB, too few stations are available to provide statistics for the comparison. Figure A1 (Appendix A) shows the comparison station by station. Similar results are shown with a very good agreement in the northern part of the domain: biases within 10%, correlation larger than 0.6 and NRMSE smaller than 40%. In the southern part of the domain, the model has some difficulties to reproduce the observations with biases ranging from 30% to 60% for most of the stations, larger for some stations. The correlation is limited and the NRMSE is larger than 50%.   We use also IAGOS ozone profiles above Chinese and Korean airports for the period 2011-2017 to evaluate the model above 950 hPa (Fig. 2). The selected IAGOS data correspond to the lowermost troposphere (950-700 hPa), the lower (700-470 hPa) and upper (470-300 hPa) free troposphere and the UTLMS (<300 hPa) above the Chinese coast (east of 110°E and between 15 30 and 50°N) and South Korea. In order to assess more precisely the model abilities to reproduce the observed ozone behaviour, the IAGOS data are projected onto the model daily grid using the Interpol-IAGOS software (Cohen et al., 2020) and averaged every month. The subsequent product is called IAGOS-DM (Distributed onto the Model grid) hereafter. We derive monthly means from the INCA daily output by selecting the sampled grid cells. These monthly fields are called INCA-M (the M suffix referring to the IAGOS Mask). The two products IAGOS-DM and INCA-M are thus consistent in space and time, and can be 20 compared together. It is important to note that the regional averages calculated here do not account for the tropopause altitude, in contrast to Cohen et al. (2020). Last, as in Cohen et al. (2018), the statistical representativeness of the observations is enhanced by filtering out the regional monthly means either with less than 300 data, or less than 7 days separating the first and the last measurements. A very good agreement is observed between the INCA model and the IAGOS observations with small biases ranging from 1.6% in the lower free troposphere to 12% in the lowermost troposphere. The INCA and IAGOS timeseries 25 are well correlated with correlation coefficients equal or larger than 0.76. Looking in details the time series shows that the model tends to underestimate O3 in the lowermost troposphere and to underestimate the largest O3values in the lower and the upper free troposphere.

Comparison between IASI O3 product and INCA simulation in North East Asia
We compare IASI and INCA O3 partial columns over the East Asia domain (100-145°E, 20-48°N) averaged over the 2008-2017 period, the model being smoothed (Fig. A2) or not (Fig. 3) by IASI averaging kernels. Spatial distribution and spatial gradients of ozone are in good agreement for the 4 partial columns. On average, the differences are smaller than 5% for the 0-3km and 3-6km columns (1.5% and 2.7% respectively without AK smoothing). The difference is larger for the upper columns 5 -6-9km and 9-12km columns -with a mean negative difference of about 15 %, INCA being larger than IASI (-14.9% and -16.4% respectively). The agreement is improved when the model is smoothed by the IASI averaging kernels, the mean difference is reduced to -7% for both 6-9km and 9-12km columns. For the 0-3km partial column, it is worth noting that the IASI retrieval is not highly sensitive to these altitudes and that the a priori contribution is larger (Dufour et al., 2012). However, the agreement between IASI and INCA remains largely reasonable accounting for the observation and model uncertainties. 10 IASI is systematically smaller than INCA over China ranging from -5% to -25% (Fig. 3a) and the agreement improves within +/-5% when applying the AK to the model (Fig. A2a). A difference smaller than 10% is observed over Korea, Japan, and the surrounded seas. Larger differences are seen for tropical maritime regions largely reduced when AK are applied. This reflects the reduced sensitivity and larger a priori impact of IASI retrievals in the lowest layers. For the 3-6km partial columns, where the IASI retrievals are the most sensitive, a very good agreement between IASI and INCA, within +/-10%, is observed for a 15 large part of the domain (Figs 3b and A2b). It is the partial columns for which the agreement is the best. For the upper columns (6-9km and 9-12km), IASI is almost systematically smaller than INCA over the domain (Figs 3c-d and A2c-d). IASI is always smaller than INCA over the most part of China whatever the partial columns considered. IASI is mainly larger than INCA in the lower troposphere and smaller in the upper troposphere elsewhere. In the desertic northwestern part of the domain, even if the emissivity is included in the IASI retrievals, the quality of the retrievals can be affected and confidence in the data reduced. 20 This region should then not be considered here. The retrieval in the tropical-type airmasses have been shown to reinforce the natural S-shape of the ozone profiles, leading to some overestimations of ozone in the lower troposphere and an underestimation in upper troposphere (Dufour et al., 2012). This likely explains the positive and negative differences with the model in the southeastern part of the domain (Fig. 3). This translates even stronger to the model when AKs are applied: the model is then smaller than IASI in the upper troposphere (Fig. A2). Globally, the differences between IASI and INCA are the 25 smallest over the Central East China (CEC). Figure 4 shows the IASI and INCA monthly timeseries of the different O3 partial columns between 2008 and 2017 for this region. The correlation between the IASI and INCA timeseries is good: larger than 0.8 except for the 6-9km column (0.75). The high correlation is partly driven by the seasonal cycle, but the correlation remains quite high for deseasonalized (anomalies) series -0.65, 0.63, and 0.68 for 0-3km, 3-6km and 9-12km columns respectivelyexcept for the 6-9km column (0.44). Biases ranging from 8% to 14%, INCA being larger, are observed between IASI and 30 INCA for the 0-3km, 6-9km and 9-12km columns respectively. The highest values are larger with INCA for the 0-3km and 9-12km columns and the lowest values larger for the 6-9km columns (Fig. 4). A smaller bias (-3.4% on average) better balanced between small and large values is observed for the 3-6km column (Fig. 4b). The seasonal cycle observed with IASI is reasonably reproduced by the model for the different partial columns with a better agreement in the 3-6km and 9-12km columns. However, the summer drops observed with IASI in the lower troposphere (0-3km and 3-6km) is not systematically reproduced by the model and the summer maximum is shifted for the 6-9km column.
In the following, after presenting the trend analysis globally over the Asian domain for the different partial columns, the discussion will focus more on the 3-6km partial column where IASI and INCA agree well. The CEC region will also be 5 privileged in the discussion as the model and observation operate better and they are in rather good agreement in this region.
Some other highly populated and polluted regions such as the Sichuan Basin (SCB) and the Pearl River Delta (PRD) will be also discussed keeping in mind the largest differences between model and observations. As we show here, comparison between IASI and INCA are satisfying with and without applying the AK to the model. For the trend analysis, we will consider the model without AK applied to avoid introducing retrieval a priori information in the model and have a model fully independent 10 of the observations. This will allow us to exploit the sensitivity tests conducted with the model to determine the processes that drive the trends.

O3 trends: satellite and model comparison
To derive the trends, we first calculate the monthly timeseries either at the INCA resolution -gridding IASI at this resolution -or averaging the model or observation partial columns over the regions reported in Fig. 1. The monthly mean ozone values are used to calculate a mean 2008-2017 seasonal cycle. This cycle is then used to deseasonalize the monthly mean timeseries by calculating the anomalies. The linear trend is then calculated based on the monthly anomalies. It is provided either in DU/yr 10 or in %/yr. The trends uncertainties correspond to the 95% confidence interval, the p-values are also calculated and reported when possible. An example is given in Fig. 4 for the CEC with monthly timeseries on the left and anomalies on the right.  Trends derived from IASI are negative with p-value < 0.05 for most of the domain and the different partial columns, except in the upper troposphere (9-12km column). They range between -0.2 %/yr and -0.6 %/yr for the 0-3km column, and between -0.4 %/yr and -1 %/yr for the 3-6km and 6-9km columns. The trends derived from the model are rather uniform over the domain for the 0-3km and 3-6km columns, being smaller than -1%/yr, except in the CEC region where the trends tend to zero in the 5 lowermost troposphere (0-3km). It is worth noting that the model shows positive trends at the surface level in this region (not shown) in agreement with surface measurement studies (e.g. Li et al., 2020). A residual positive trend is observed up to 1 km altitude in the model and becomes negative higher (not shown). The trends in the mid-upper troposphere (6-9km and 9-12 km columns) are mainly negative (p<0.05) north to 30°N latitude and can be more variable in the subtropics (Fig. 5). To evaluate the impact of the IASI sampling (representative of clear-sky conditions), we calculate the model monthly mean including the 10 model grid cells on the days when IASI observations are available in these cells. The trends derived from the model resampled to match IASI observations are reported on Fig. 5. The resampling changes only slightly the trends derived from the model. In the following, we consider the model without matching the IASI sampling. Table 3 summarizes O3 trends derived from IASI and INCA for different partial columns and for the different regions reported in Fig. 1. We choose the most populated Chinese areas where significant pollutant reductions have occurred since 2013 (Zheng 15 et al., 2018), such as CEC -including BTH and YRD -and PRD and SCB. We also consider the KJ region as a region influenced by the pollution export from China. We bold the trend values in the table when both IASI and INCA have trends with p<0.05 and when the trends agree within 40% between the model and the observations. The trend values corresponding to p<0.05 and a poorer agreement are in italic. The CEC region shows the best agreement between the trends derived from IASI and INCA for all the columns except the upper tropospheric columns (9-12km). The anomalies and calculated linear 20 trends are shown in detail on Fig. 4 (right panels). For this region, where both the observations and the model are the most reliable, trends are in very good agreement (<15%) for the 0-3km, 3-6km and 6-9km columns. Trends derived from IASI for the UTLMS columns (9-12km) are very small with large p-values for all the regions (Table 3). It is then difficult to compare and conclude for the upper tropospheric columns -the trends calculated from the model are mainly negative with p<0.05. For the PRD and the SCB, the model and the observations are less reliable for different reasons explained in section 3. This leads 25 to a poor agreement of the derived trends and a lack of reliability of the trends for these two regions (large uncertainties on the trends values, Table 3). For the BTH and YRD, included in the CEC, and for the KJ, the trends calculated from the observations and the model are in good agreement for the 3-6km and 6-9km columns with p<0.01.  Fig. 1 and the 0-3km,   3-6km, 6-9km, and 9-12km Table 4 for details). White crosses are displayed when p-values are smaller than 0.05.

Sensitivity tests to evaluate the processes contributing to the trends
In this work focused on China and 2008-2017, both IASI observations and INCA simulations show negative O3 trends of similar magnitude in the lower (3-6km) and upper (6-9km) free troposphere in large parts of China and its downwind region. 10 Dufour et al. (2018) suggest that negative O3 trends derived from IASI in the lower troposphere over the North China Plain for a slightly shorter time period can be explained for almost half by NOx emissions reduction in China since 2013. They argue the negative impact on the trend of these reductions compared to the positive one at the surface is due to changes in the chemical regime with the altitude. To go further on this and quantify the processes contributing to the O3 trends, we performed, as described in Section 2, several sensitivity simulations with the INCA model. The objective is to remove one-by-one the 15 interannual variability and trend induced by the different processes (emissions, meteorology). The different sensivivity simulations are summarized in Table 4 (Fig. 5), especially when p<0.05. Then, we calculate the contribution of each process as described in Table 4. The contribution of the different processes is shown in Fig. 6 for the 3-6km and 6-9km columns. We focus on these two columns as the trends are in good agreement between IASI and INCA. The main contributions to the trends are the local 15 Chinese emissions and the meteorology, with contributions larger than 20%. The other tested variables (global emissions, biomass burning emissions and methane) contribute to the trends within 20%, with a negative contribution of methane for most of the domain. This means that the increase in methane concentrations and then the associated ozone production counteracts the ozone reduction due to the other processes (emissions and meteorology). The different contributions for the regions where the model and the observations are the most reliable are detailed in Table 5. Chinese emissions contribute to 20 60% in the main source region, the CEC, with variations inside the regions for the 3-6km column: the Chinese emissions contribute to 40% in BTH and more than 70% in YRD. The Chinese contribution to the trends in the export region (KJ) remains high with 47% contribution. The meteorological contribution ranges from 34 to 38%. Methane and biomass burning emissions contributions are rather stable over the different regions around -15% and +14% respectively, biomass burning contribution being slightly higher in the export region (19% for KJ). Surprisingly, the global emissions contribute the most to the trends in the highly polluted region of the BTH (22%). For the 6-9km column, the meteorological contribution to the trends increases (about 50% or larger) as the Chinese emissions contribution decreases. The biomass burning contribution is globally larger, especially in the export region where it reaches 30%. The global anthropogenic emissions contribution remains small in absolute value, except in the YRD region where it becomes negative and reaches -20%. The prevailing contribution of Chinese 5 emissions changes in the negative O3 trends in the lower and upper free troposphere seems to confirm the previous outcomes of Dufour et al. (2018). the global biomass burning emissions and the CH4 to the trends calculated for the 3-6km and 6-9km partial columns (see Table 4 10 for details on the sensitivity tests).
https://doi.org/10.5194/acp-2021-476 Preprint. Discussion started: 5 July 2021 c Author(s) 2021. CC BY 4.0 License. Table 5: Contributions (in %) of the meteorological variability, the Chinese anthropogenic emissions, the global anthropogenic emissions, the global biomass burning emissions and the CH4 to the trends calculated for the 3-6km and 6-9km partial columns for each individual region in Fig. 1

Limitation of the study 5
The results presented in our study are not fully in line with the recent Tropospheric Ozone Assessment Report (Gaudel et al., 2018) and related works (Cooper et al., 2020;Gaudel et al., 2020), which states a general increase in tropospheric ozone during the last decades. If negative trends are observed at the surface in developed countries for example in summer, positive trends for the free troposphere are reported using mainly IAGOS as a reference (Cooper et al., 2020). Even if our study seems to show consistent trends derived by IASI and the model in the free troposphere, we stress, in this subsection, vigilant points for 10 the interpretation of the results.

Length of the period
In this study, we derived trends over a limited 10-year period. Calculating short-term trends leads to an increased sensitivity to the inter and intra-annual variations, the length of the period and to the starting and ending point of the time series. Due to 15 the availability of the satellite measurements and the simulated period with the model, it was not possible to extend the time period further. We tested the impact of the starting and ending point of the time series by removing one and two years at the beginning and at the end of the period. The trends derived from IASI and INCA for the 3-6km column in the CEC for the different periods are summarized in Table B1 (Appendix B). They remain consistent with the trend derived for 2008-2017, respectively for IASI and INCA, and within its confidence interval. These results seem to comfort the consistency between the 20 modelled and observed trends and their robustness. However, it is worth noting that the calculated trends seem more sensitive to the end of the period, corresponding to a strong El Nino period, than to the beginning of the period. Removing the last two years of the period leads to a decrease of the IASI trend and to an increase of the INCA trend. This apparent inconsistency, which should be evaluated when longer simulations with consistent emissions and longer observation time series will be available, stress the difficulty of working with short-term trends and the caution to take to not overinterpret the results.

Discrepancies between different satellite sounders and products
The TOAR points toward a major discrepancy between the different satellite ozone products available for the report: the OMI UV sounder showing mainly positive trends over 2008-2016 and the IASI IR sounder showing mainly negative trends over the same period (Fig B2 -Appendix B). Over North East Asia, the discrepancy in the sign of the trend calculated from the different satellite products is more contrasted. The OMI/MLS and OMI-RAL products still show positive trends as all over the 5 globe. The IASI-SOFRID product shows positive trends all over China, and the IASI-FORLI product positive trends only in the southeastern part of China. The IASI O3 product used in this study was not included in the comparison as it is not a global product. It is worth noting that for the TOAR, the tropospheric columns derived from the different satellite products were not based on the same tropopause height definition and each product was considered with its native sampling. This might contribute partly to the differences between the trends, in addition to the fundamental differences in the measurement techniques (UV 10 and IR) and the retrieval algorithms used. Possible drifts over the time have not been systematically studied in the TOAR.
Some individual studies exist but once again they do not allow one to conclude. Indeed, Boynard et al. (2018) noticed a significant negative drift in the Northern Hemisphere in the IASI-FORLI product, which is not detected in the most recent IASI-SOFRID product (Barret et al., 2020). The OMI/MLS product shows a small positive drift when compared to ozonesondes but not significant when based upon a difference t-test (Ziemke et al., 2019). For this study, we compare our IASI 15 O3 product with the OMI/MLS one. To conduct a proper comparison, we used the same definition of the tropopause height to calculate the tropospheric columns. As the OMI/MLS product provides directly tropospheric columns without ozone profiles, we selected the tropopause height used for OMI/MLS, derived from the NCEP re-analyses, as the reference tropopause height.
We calculated the tropospheric columns from the IASI O3 profiles retrieved up to the defined tropopause height. We calculated the monthly time series at the resolution of 1°x1°. Only the days for which IASI and OMI/MLS are both available in the 20 considered grid cell are used to calculate the monthly means and anomalies for the given grid cell. The derived trends are shown in Fig. 7. OMI/MLS shows large positive trends all over the domain except in the southern part of the BTH region. On its side, IASI shows trends close to zero with positive trends over central China, the East China Sea and over the Pacific in the southeastern part of the domain, and negative trends over North China and Korea. The TOC trends derived from IASI show completely different spatial patterns from the trends derived in the free troposphere (Fig. 5) and seem to reflect more the trends 25 of UTLS column (9-12 km). Work is still needed to understand the differences in the trends derived from different satellite instruments. Especially, one important question is to identify from which part of the troposphere the TOC is the most representative in the different products and how the vertical sensitivity of the different instruments and retrieval algorithms influence the calculated columns and trends. Answering this question is one of the objectives of the satellite working group of the TOAR phase II, which started in 2021. 30

Impact of the sampling: comparison with IAGOS
The IAGOS observations are considered as a reference for free tropospheric ozone trends in the TOAR framework. Then, we try to evaluate the IASI and INCA trends using IAGOS in addition to its use for validation reported in section 3. As already mentioned for the validation of IASI using IAGOS, the comparison is somehow difficult due to the limited top altitude of IAGOS profiles to properly apply the AK to the profiles. We limit our comparison to the 3-6km columns and explore the 10 impact of the sampling on trend calculations. We use the same coincidence criteria than the one used in section 3 to select pairs of IAGOS and IASI profiles. Based on the selected profiles, we consider the daily simulated profiles of INCA for the same day and the grids cells of the model corresponding to the latitudes and longitudes of the profiles. Partial columns are then calculated for the subset of observed and modeled profiles and the trends are derived from the monthly anomalies. As mentioned in section 3, the number of IAGOS profiles in the Chinese region is not very high (about one thousand) and reduces 15 to 315 profiles in coincidence with IASI with 213 profiles covering the pressure range from 1000 hPa to 500 hPa. This number even falls to 26 profiles for profiles within 1000 hPa and 250 hPa. The trends that can be calculated from this set of profiles are then not robust enough to conclude. Then, we consider the European region for which more profiles are available. 9185 IAGOS profiles are initially available for 2008-2017 with almost no measurements in 2010. Looking for the subset of profiles in coincidence with IASI strongly reduces the number of available profiles: 3276 are selected. We consider IAGOS profiles 20 covering the 1000-250hPa range for a proper comparison with IASI. This allows one to reduce the proportion of a priori information potentially introduce in the IAGOS profiles when they are completed up to 60 km and smoothed by the AKs (see section 3 for details). This leads to reduce the subset of profiles to 1103 profiles. Table 6 provides the trends derived from this subset of profiles over Europe for IASI, INCA and IAGOS. IAGOS trends are calculated from both raw and smoothed profiles.
We also provide the IAGOS trends calculated from the initial set of IAGOS profiles to evaluate the sampling impact. We 25 calculate the trends for 2008-2017 and 2011-2017 as 2010 was not sampled by IAGOS and that 2008-2009 were associated to a negative anomaly (Cooper et al., 2020) which might perturb the short-term trend calculation. It is interesting to note that the difference in the sampling between the initial set of IAGOS profiles and the set in coincidence with IASI changes the calculated trends from a positive trend (0.05 DU/yr, p<0.01) to a trend close to zero for both the raw and smoothed IAGOS columns INCA trends remain negative for this time period, -0.14 DU/yr (p<0.01) and -0.06 DU/yr (p=0.04) respectively. The IASI negative trend is more than twice larger in absolute value compared the ones derived from IAGOS and INCA. These results 10 do not allow one to clearly conclude wether the negative trends derived from the model and IASI are realistic or not. They mainly show the strong sampling issue for trend calculation and stress the need to compare different datasets using, as far as possible, similar sampling to evaluate the derived trends. These results highlight again the difficulty to draw firm conclusions on short-term trends derived from different datasets in addition to sampling differences.   observed and modeled trends in the CEC region are -0.07 ± 0.02 DU/yr and -0.08 ± 0.02 DU/yr respectively for the lower free troposphere, and -0.05 ± 0.02 DU/yr and -0.06 ± 0.02 DU/yr respectively for the upper free troposphere. A good agreement is also observed in the region including Korea and Japan and corresponding to the region of pollution export from China. Based 5 on sensitivity studies conducted with the INCA model, we quantify the contribution of the Chinese anthropogenic emissions, the global anthropogenic emissions, the global biomass burning emissions, methane, and meteorology to the ozone trends. In the CEC region, 60% of the negative trend derived from the model in the lower free troposphere can be attributed to the Chinese anthropogenic emissions and 52% in the upper free troposphere. The second contribution to explain the negative trend is the meteorological variability (34% and 50% respectively). The background ozone produced from methane globally 10 counteracts the decrease in ozone with a contribution of about 15 % to the trends in the lower and upper free troposphere. The global anthropogenic emissions changes account for less than 10% in the ozone trends and biomass burning emissions changes between 10 and 20 %. These results suggest that the reduction of NOx anthropogenic emissions that occurs since 2013 in China leads to a decrease in ozone in the Chinese free troposphere, contrary to the increase in ozone at the surface. However, too few independent measurements such as IAGOS or ozonesondes are available in the region during 2008-2017 to fully validate the 15 decreasing trends calculated by both the IASI observations and the model. A comparison done in Europe where more independent IAGOS measurements are available show that trend calculation can be strongly affected by the sampling of the considered datasets and the time period considered when analyzing short-term trends. Particular caution should be taken to not overinterpret short-term trends and when comparing trends derived from different datasets with different sampling. In addition, comparisons between the trends calculated from the OMI/MLS O3 tropospheric columns and the IASI ones, calculated using 20 the same tropopause height and sampling, show large discrepancies as already stated by the TOAR (Gaudel et al., 2018), and point toward a need to better understand how the differences in vertical sensitivity of the satellite observations impact the observed tropospheric columns and the derived trends.

Data and code availability
The IAGOS data set is available at https://doi.org/10.25326/20, and more precisely, the time series data are available at 25 https://doi.org/10.25326/06. The distribution of the IAGOS data onto the model grid is based on an updated version of the Interpol-IAGOS software, which can be found at https://doi.org/10.25326/81. The LMDZ, INCA and ORCHIDEE models are released under the terms of the CeCILL license. The mode codes, input data, and outputs are archived in the CEA (Commissariat à l'énergie atomique et aux énergies alternatives) high-performance computing centre TGCC and are available upon request.
The IASI observations (level 1C) are available from the AERIS data infrastructure (www.aeris-data.fr). The full archive of IASI ozone product retrieved from the level 1C data is available, upon requested to Gaëlle Dufour (gaelle.dufour@lisa.ipsl.fr), for the Asian domain considered here between 2008 and 2017.
In addition, the monthly gridded partial columns derived from IASI and INCA, used to calculate the trends in the project, will be made available through a DOI (DOI attribution under request). 5

Author contribution
GD managed the study from its conception, the analysis of data, the preparation of the manuscript and the funding acquisition.
DH and YZ performed the model simulations. ME performed the IASI ozone retrieval and managed the resulting level-2 product. YC, AG, and VT provided IAGOS observations and helped with their use and analyses in the study. BB was in charge of ground surface data processing and cleaning. GS, ML, and AB managed the model evaluation with the surface 10 measurements. JZ provided the OMI/MLS satellite data. All the authors participated in reviewing and editing the manuscript.

Competing interests
The authors declare that they have no conflict of interest. producers and providers used in this study: the ozonesonde data used in this study were mainly provided by the World Ozone 20 and Ultraviolet Data Centre (WOUDC) and are publicly available (see http://www.woudc.org). We acknowledge the Institut für Meteorologie und Klimaforschung (IMK), Karlsruhe, Germany, for a licence to use the KOPRA radiative transfer model. A part of this work was performed using HPC resources from GENCI-TGCC (Grant A0050106877 and grant GENCI2201