Ozone seasonal evolution and photochemical production 1 regime in polluted troposphere in eastern China derived 2 from high resolution FTS observations

22 The seasonal evolution of O3 and its photochemical production regime in a 23 polluted region of eastern China between 2014 and 2017 has been investigated using 24 observations. We used tropospheric ozone (O3), carbon monoxide (CO) and 25 formaldehyde (HCHO, a marker of VOCs (volatile organic compounds)) partial 26 columns derived from high resolution Fourier transform spectrometry (FTS), 27 tropospheric nitrogen dioxide (NO2, a marker of NOx (nitrogen oxides)) partial 28 column deduced from Ozone Monitoring Instrument (OMI), surface meteorological 29 data, and a back trajectory cluster analysis technique. A broad O3 maximum during 30 Correspondence to: Cheng Liu (chliu81@ustc.edu.cn)


Introduction
Ozone in the troposphere is a key ingredient in a number of atmospheric physical and chemical processes.These include radiative forcing as ozone is an infrared absorber (greenhouse gas).It is also a precursor to the formation of the hydroxyl radical which affects the oxidizing (cleansing) capacity of the atmosphere.In addition, human health, terrestrial ecosystems, and materials degradation are impacted by poor air quality resulting from high photochemical ozone levels (Oltmans et al., 2006).In Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.polluted areas, tropospheric ozone generates from a series of complex reactions involving CO, NOx (NO+NO2), and VOCs in the presence of sunlight (Haagen-Smit, 1952).The ozone production efficiency is nonlinearly related to concentrations of its precursors, and thus tropospheric ozone levels depend on both absolute and relative levels of NOx and VOCs in the troposphere (Duncan et al., 2010).
Ozone production can be suppressed by reducing either emissions of NOx or VOCs depending on it is NOx-limited or VOC-limited photochemical regimes (Duncan et al., 1998).In order to determine the regime, the total reactivity with OH of the myriad of VOCs in the polluted area has to be estimated, as the reaction with OH is often the rate-limiting step of many oxidation pathways (Sillman, 1995;Duncan et al., 1998).In the absence of such information, the HCHO concentration can be used as a proxy for VOC reactivity because it is a short-lived oxidation product of many VOCs and is positively correlated with peroxy radicals (Duncan et al., 1998).
The air suffers increasing pollution along with the development of modernization in China (https://en.wikipedia.org/wiki/China).China issued the new air quality (AQ) standard in 2013 to address current air pollution problems.Tropospheric O3 was Europe and Northern America, whereas the number of sites in Asia, Africa, and South America is very sparse, and there is still no NDACC station that covers the vast Chinese area (http://www.ndacc.org/).
In this study, we first investigate ozone seasonal evolution and photochemical production regime in the polluted troposphere in eastern China with tropospheric ozone, CO and HCHO derived from an FTS site in Hefei, China, tropospheric NO2 deduced from overpass Ozone Monitoring Instrument (OMI), surface meteorological data, and the back trajectory cluster analysis technique.It is presented as follows: Section 2 presents site description and instrumentation.Section 3 investigates the possibility of the ground based FTS measurements in Hefei, China for providing reliable time series of tropospheric ozone and ozone related gases.Section 4 presents comparisons with OMI, global GEOS-Chem model simulation, and regional WRF-Chem model simulation.Section 5 investigates ozone seasonal evolution and photochemical production regime.Section 6 summarizes the conclusions.

Site description and instrumentation
The FTS observation site (117°10′E, 31°54′N, 30 m a. and azimuth angle over time on a typical observation day.In the morning (sunrise), the light path travels over the downtown with a large SZA, and then decreases over time.In the afternoon (sunset), the light path gets out of the city, travels over the cultivated land or wetland, and then increases over time.Consequently, the measurements in the morning and in the afternoon would be dominated by city emissions and emissions from cultivated land or wetland, respectively.So the special observation scenario makes this site also crucial for providing data to constrain regional sources and sinks in the vicinity of Hefei city (Sun et al., 2017;Tian et al., 2017).
The observation system consists of a high resolution FTS spectrometer

FTS retrievals for ozone and ozone related gases
In order to determine tropospheric ozone seasonal evolution and photochemical production regime, tropospheric O3, CO, NO2, and HCHO should be known with adequate accuracy.O3, CO and NO2 are routinely observed in the NDACC and HCHO is currently investigated.It confirms that tropospheric O3, CO and HCHO are robust in Hefei, but NO2 has a low information content in the troposphere which is in good agreement with the NO2 retrievals in other NDACC stations (Kerzenmacher et al., 2008;Robles-Gonzalez et al., 2012;Hendrick et al., 2012).So in this study, the tropospheric O3, CO, and HCHO columns were derived from FTS measurements, while the tropospheric NO2 columns were deduced from overpass OMI measurements.

Retrieval strategy
The latest version of SFIT4 (version 0.9.4.4) algorithm is used in the profile retrieval (http://www.ndacc.org/).Theoretical basis for SFIT4 algorithm and error analysis is described in Appendix A. The retrieval settings for O3, CO, and HCHO are listed in Table 1.All spectroscopic line parameters are adopted from HITRAN 2008 (Rothman et al., 2009).A priori profiles of pressure, temperature and H2O for the measurement days are interpolated from the National Centers for Environmental Protection and National Center for Atmospheric Research (NCEP/NCAR) reanalysis (Kalnay et al., 1996).A priori profiles of the target gases and the interfering gases except H2O from a dedicated WACCM (Whole Atmosphere Community Climate Model) run.For O3 and CO, we follow the NDACC standard convention with respect to micro windows (MWs) selection and the interfering gases consideration (https://www2.acom.ucar.edu/irwg/links;Vigouroux et al., 2008).HCHO is not yet an official NDACC species but has been retrieved at a few stations (e.g.Vigouroux et al., 2009;Jones et al., 2009;Viatte et al., 2014;Franco et al., 2015) For all retrievals, the instrument line shape (ILS) is described with the LINEFIT14.5output using routine HBr cell measurements (Hase et al., 1999;Hase et al., 2011).

Profile information in the FTS retrievals
The averaging kernel matrix A = dx_r/dx, with x_r the retrieved profile and x the true profile, can be used to characterize the profile information contained in the FTS retrievals.The rows of A represent the sensitivity of the retrieved profile to the true profile, and are referred to as averaging kernels.Their FWHM (full width at half maximum) indicates the vertical resolution of the retrieval at a specific altitude.The area of averaging kernels represents sensitivity of the retrieval to the measurement (Rodgers, 2000).Fig. 2 shows the averaging kernels and their areas of O3, CO, and HCHO.The altitude ranges with sensitivity larger than 0.5 and the corresponding DOFs are summarized in Table 2. Within these sensitive ranges, more than 50% of the retrieved profile information comes from measurement rather than the a priori information.Each gas has a different sensitive range.The sensitive range for CO and HCHO is mainly tropospheric, and for O3 is both tropospheric and stratospheric.The typical DOFS over the total atmosphere (TC DOFS) obtained in Hefei for each gas are also included in In order to separate independent partial column amounts in retrieved profiles, we have chosen the altitude limit for each independent layer such that the DOFS in each associated partial column is not less than 1.0.Fig. 3 shows the partial column averaging kernels of O3, CO, and HCHO within each chosen layer.Their averaging kernels are resolved at their FWHM and peak at the middle of the chosen layers.The retrieved profiles of O3, CO, and HCHO can be divided into four, three, and one independent layers, respectively.The troposphere is well resolved by O3, CO, and HCHO, where CO exhibits the largest resolution which contains more than two independent layers in the troposphere.
The chosen altitude ranges of the tropospheric layers for each gas and the corresponding DOFS are listed in Table 2.We have chosen the same upper limits for all gases and they are not equivalent to the real tropopause heights, but are about 3 km lower than the mean value which, derived from the NCEP database, is 15.1 km with a standard deviation (1σ) of 1.1 km for Hefei.This manner not only ensured the accuracies of tropospheric O3, CO, and HCHO retrievals, but also minimized the influence of transport from stratosphere, i.e., the so called STE process (stratosphere-troposphere exchange).were listed in Table 3.The total errors were calculated as the sum in quadrature of each random error and systematic error.In the troposphere associated error calculation, the elements of gain matrices were set to zero for the altitudes larger than the upper limit of tropospheric partial column (see Eqs. ( A6) -(A8) in Appendix A).The total errors in the tropospheric partial columns for O3, CO, and HCHO, have been evaluated to be 8.7%, 6.8%, and 10.2%, respectively.

Comparisons of FTS tropospheric ozone with satellite and model data
The FTS tropospheric O3 data has been compared with the correlative satellite data, global chemical model data, and regional chemical model data.Our data filtering scheme, designed to minimize the impacts of significant weather events and instrument problems, is described in Appendix B. In order to accurately compare the FTS data with the correlative data, the method of Rodgers and Connor (2003) are used to consider the effects of differences in altitude grid, the a priori profile and vertical resolution (i.e., averaging kernel).First, the correlative ozone profiles were interpolated to the FTS altitude grid to ensure a common altitude grid.The interpolated profiles were then smoothed by the corresponding FTS averaging kernels and a priori profiles by Eq. ( 1).Finally, we compared the smoothed correlative ozone profiles with the FTS profiles.
where xs represent the smoothed correlative profiles, xa and A represent the FTS a priori profile and averaging kernel matrix, respectively.xc is the interpolated correlative profiles.The tropospheric partial column was then calculated by integrating the smoothed profile by Eq. ( 2).In comparison, we have taken the FTS retrievals as the reference, and the fractional difference (D%) is defined here as Eq.
(3), where Am is the air-mass profile derived from FTS retrievals.X is a vector which can include multiple elements such as gas profile or only one element such as tropospheric partial column.Xref is the same as X but for the FTS retrievals.
In this study, the Pearson correlation coefficient (PPC) is used to measure the linear similarity or correlation between two coincident variables.We regard the two variables as completely correlated if the PCC is between 0.8 and 1.0; as well correlated if the PCC is between 0.6 and 0.8; as moderately correlated if the PCC is between 0.35 and 0.6; as weakly correlated if the PCC is between 0.2 and 0.35; and as no correlation if the PCC is between 0.0 and 0.2.

Comparison with OMI data
The FTS tropospheric ozone data were compared with the OMI PROFOZ product selected within the  0.7° latitude/longitude rectangular area around the Hefei site and with uncertainty for tropospheric ozone column of less than 10%.PROFOZ contains the retrieved ozone profile, total, stratospheric, and tropospheric ozone columns, other retrieved auxiliary parameters, random noise and total retrieval errors for all of the retrieved quantities, the retrieval averaging kernels for the ozone profile, and the random-noise measurement error covariance matrix for ozone (https://avdc.gsfc.nasa.gov/pub/data/satellite/Aura/OMI/V03/L2/OMPROFOZ/).The algorithm for OMI ozone profile retrieval has been described in detail in Liu et al. (2010).The total, stratospheric, and tropospheric ozone columns, corresponding errors are integrated from the profile and error covariance matrices.FTS data is -0.19×10 17 molecules*cm -2 .In section 5, this difference is subtracted to minimize the difference between OMI and FTS data.

Comparison with GEOS-Chem model data
The GEOS-Chem model (v10-01, http:// wiki.seas.harvard.edu/ geos-chem/ index.php/ GEOS-Chem_v10-01) is a global 3-D chemistry transport model (CTM) driven by the GEOS-FP assimilated meteorological data from the Global Modeling Assimilation Office (GMAO) at NASA Goddard Space Flight Center.GEOS-FP is the latest meteorological data product from GMAO.The GEOS-FP 3-D products are updated every 3 h at a native horizontal resolution of 0.3125° longitude × 0.25° latitude and a 72-layer vertical grid extending to 0.01hpa (Bey et al., 2001).In this study, we simulated ozone at 2.0° × 2.5° horizontal resolution and 47 vertical pressure levels.The tagged O3 simulation uses archived ozone production and loss rates to perform a simulation for geographically tagged ozone tracers.There needs a restart file before the tagged O3 simulation.The initial tracer concentrations used for January 2013 were derived from a ten year spin up from the year 2004 to 2013.The outputs of GEOS-Chem simulations were ozone mixing ratio profiles cover the period from January 2013 to September 2017 with 1 h time frequency, in which the nearest grid point to FTS site coordinates were selected and interpolated to the FTS vertical grid, and then convolved with FTS averaging kernels and a priori profiles through Eq. ( 1) (Rodgers and Connor, 2003).data is -0.55×10 18 molecules*cm -2 .This difference is mainly attributed to the difference between GEOS-Chem input files (including ozone production and loss rates and emission inventory) and the actual ones.In section 5, we used this difference to correct the bias between GEOS-Chem and FTS data.

Comparison with WRF-Chem model data
In contrast to GEOS-Chem model, WRF-Chem is a regional and tropospheric chemical model.In this study, the WRF-Chem version 3.7 was used to simulate air pollutants and meteorological parameters.The input setting was described in detail in Liu et al.(2016).Briefly, the model domain centers at 35.0° N, 110.0°E with a grid resolution of 20×20 km and covers East China and its surrounding area.The Multi-resolution Emission Inventory for China (MEIC, http://www.meicmodel.org/)(Li et al., 2014;Liu et al., 2015) were used to provides anthropogenic emissions.The data is -0.27×10 18 molecules*cm -2 .This difference is mainly attributed to uncertainties of WRF-Chem input files including ozone production and loss rates and MEIC inventory.In section 5, we used this difference to correct the bias between WRF-Chem and FTS data.
5 Tropospheric ozone seasonal evolution and production regime 5.1 Tropospheric ozone seasonal variability Fig.12 (a) shows the tropospheric ozone column time series recorded by FTS from 2014 to 2017.We implemented the bootstrap resampling method as described in Gardiner et al. (2008) to determine the annual trend and intra-annual variations (Gardiner et al., 2008).In the present work, we followed Gardiner's method and used where photochemical ozone production associated with anthropogenic sources (CO, NOx, and VOCs) occurs.In this study, we observed a broad maximum within both spring and summer mainly because photochemical ozone production is active in both season.The selection of troposphere limit in this work minimized but cannot avoid the influence of transport from stratosphere, so the STE process may also contribute to high level of tropospheric ozone column in spring.
In order to determine where the air masses came from and thus contributed to the observed tropospheric ozone levels, we have used the HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) model to calculate the three-dimensional kinematic back trajectories (Draxler et al., 2009).In the calculation, the GDAS (University of Alaska Fairbanks GDAS Archive) meteorological fields were used with a spatial resolution of 0.25°× 0.25°, a time resolution of 6 h and 22 vertical levels from the surface to 250 mbar.All daily back trajectories at 12:00 UTC, with a 24 h pathway at 3000 m a.s.l., have been grouped into clusters (Stunder, 1996).
The results show that, in spring and summer, air masses arriving at Meanwhile, the direction shifts from east China to southeast China, and the percentage decreases by 10%.Thus, the air pollutions from the key Yangtze river delta area would decrease dramatically.The same as spring and summer, the air masses from north China and south China are also very small (0.28%) in autumn and winter, and thus the contribution due to air pollutions in Beijing-Tianjin-Hebei and Pearl River Delta areas are still small.Considering a fact that the air pollution in east China is typically heavier than that in west China because of densely population and industrialization, air pollutions in spring and summer in the observed area would be larger than those in autumn and winter, and thus have more contributions to the observed tropospheric ozone levels.A relatively larger difference in spring and summer for both GEOS-Chem and WRF-Chem is probably due to the two reasons: First, compared with autumn and winter, the larger air pollution in spring and summer would increase the uncertainties of emission inventories that are used as input files of model simulations.Second, ozone production is more active in spring and summer than that in autumn and winter (see section 5.3), which would increase the uncertainties of ozone production regime modeled in GEOS-Chem and WRF-Chem.

Tropospheric ozone production regime
Tropospheric ozone is not an emission pollutant, but produced by photochemical oxidation of CO, NOx, and VOCs under certain meteorological condition.This process is complicated and thus shows regional representativeness.This section presents an investigation of the tropospheric ozone production regime in the observed area in eastern China.are more favorable to ozone production (higher sun intensity, higher temperature and lower pressure) than in autumn and winter, which consolidates a fact that tropospheric ozone in spring and summer are larger than those in autumn and winter.
In order to determine the relationship between tropospheric ozone production and its precursors, the chemical sensitivity of ozone production (PO3) relative to CO, HCHO, and NO2 changes was investigated.Fig. 17 shows daily mean time series of tropospheric CO, HCHO, and NO2 columns that have tropospheric O3 counterparts.factors, such as air pollution levels and meteorological conditions (e.g., Kleinman et al., 2005), the transition threshold estimated by Sillman (1995) or Martin et al. (2004a) or Duncan et al. (2010) would vary both geographically and temporally.
Following Duncan et al. (2010)'s technique, chemical sensitivity of PO3 in Hefei was thus investigated.Since the obtainment of HCHO product was not the same as that of Duncan et al. (2010), and the air pollution and meteorological condition in Hefei were also different from those in Los Angeles.Thus, the transition thresholds (HCHO/NO2 ratio at 1 and 2) are not straightly applied here.In order to determine reasonable transition thresholds, we iteratively altered the HCHO/NO2 ratio threshold and judged the correlations of tropospheric O3 to HCHO or NO2.We found that the correlation between tropospheric O3 and HCHO is larger than 0.6 if HCHO/NO2 is less than 1.3, and the correlation between tropospheric O3 and NO2 is larger than 0.6 if HCHO/NO2 is larger than 2.8.So we regard PO3 as VOC-limited if HCHO/NO2 < 1.3, NOx-limited if HCHO/NO2 > 2.8, and mix VOC-NOx-limited if HCHO/NO2

Conclusion
We have investigated the possibility of the ground based Fourier transform spectrometer (FTS) measurements in Hefei, China for providing reliable time series of tropospheric ozone and ozone related gases, CO and HCHO.The investigation showed that the FTS can retrieve robust tropospheric O3, CO, and HCHO columns.
We used the FTS tropospheric ozone to validate the Ozone Monitoring Instrument The agreement with OMI is better than that with GEOS-Chem model and with WRF-Chem model.Besides, the agreement with regional WRF-Chem model is better than that with global GEOS-Chem model.
We used tropospheric ozone, CO and HCHO derived from FTS measurements, tropospheric NO2 deduced from overpass OMI, the surface meteorological data, and the back trajectory cluster analysis technique to investigate ozone seasonal evolution shows that the tropospheric ozone column increases over time at the first half year and reaches the maximum in June, and then it decreases over time at the second half year.
Tropospheric ozone columns in June are, on average, 0.5×10 18 molecules*cm -2 higher than those in December which has a mean value of 1.05×10 18 molecules*cm -2 .The OMI time series shows similar behaviour.The measured features can basically be reproduced by GEOS-Chem and WRF-Chem data but with slight shifts in the timing of the seasonal maximum.
Back trajectories computed with the HYSPLIT show that: air pollutions in megacities in central-southern China, northwest China, and the key pollution area, i.e., Yangtze River Delta area in eastern China, dominates the contributions to the observed tropospheric ozone levels, while the contributions from the other two key pollution areas, i.e., Beijing-Tianjin-Hebei in north China and Pearl River Delta in south China, are very small; Air masses generated from polluted areas have more transportations to the observed area in spring and summer than in autumn and winter, and hence have more contributions to the observed tropospheric ozone levels.
Correlations between tropospheric ozone and meteorological data disclosed that spring and summer is more sensitive to photochemical ozone production than in autumn and winter.Finally, we used the HCHO/NO2 ratio as a proxy to investigate the chemical sensitivity of ozone production (PO3  The basic principle of SFIT4 is using an Optimal Estimation Method (OEM) to fit the calculated-to-observed spectra with an iterative Newton scheme (Rodgers, 2000;Hannigan and Coffey, 2009).The retrieved profile x is expressed as, where x is retrieved profile, xa is a priori profile, y is measured spectra, and F(x, b) is forward model calculated spectrum.The m×n matrix Ki=F(xi, b)/xi is weighting function matrix or Jacobian matrix for the i-th iteration.Gy is the contribution function matrix, where Sε and Sa are measurement noise covariance matrices and a priori profile covariance matrices, respectively.The averaging kernel matrix A can be calculated as (Rodgers, 2000), The degrees of freedom for signal (DOFs) is calculated as the trace of A, An error analysis and characterization is of great importance for a retrieval algorithm.Errors are traditionally classified as systematic or random according to whether they are constant between consecutive measurements, or vary randomly.The total error covariance matrix (E) can be expressed as the sum of the contributions from (a) the measurement error due to measurement noise (Em), (b) the smoothing error due to the limited altitude resolution of the FTS system (Es), (c) the model parameter error due to uncertainties of forward model parameters (Emodel), and (d) forward model error due to the error of the model in physical process simulation (Ef): The forward model error is hard to evaluate because it requires a model which includes the correct physics.In this study, we neglect the forward model error.The smoothing error Es is calculated via equation (A6), the measurement error Em is calculated via equation (A7), and the model parameter error Emodel is calculated via equation (A8) (Rodgers, 2000).

Appendix B: Data filtering
For the tropospheric O3, CO, and HCHO columns derived from FTS measurements, we established a specific filter criterion to remove the outliers by setting certain thresholds for measurement intensity, fitting error, DOFS, and fitting residuals.Measurements satisfying the criteria as follows were classified as valid and were subsequently used in the analysis.
1) Spectra recorded with too low incident signals are discarded to ensure adequate SNRs.Meanwhile, spectra recorded with too high incident signals are discarded because of non-linearity in the detector.Specifically, for O3 spectra recorded with MCT detector, the signal intensity should lie in between 5,000 and 11,000 ADCs, and for CO and HCHO spectra recorded with InSb detector, the signal intensity should lie in between 10,000 and 20,000 ADCs.
2) The auxiliary data such as solar intensity and meteorological data (at least surface pressure and temperature) should be recorded synchronously with the measurement.
Otherwise, the measurements are screened out.
3) The observed scene must be nearly cloud-free and not seriously affected by smog or haze.The spectra recorded with a solar intensity variation (SIV) of larger than 10% were not used in this study.The SIV within the duration of a spectrum is the ratio of 4) The root mean square error (RMS) of the residual difference (relative difference between measured and calculated spectra after the fit) in all fitting windows has to be less than 2.5%.
5) The retrievals should be converged and the concentrations of the target and interfering gases at each sub layer should be positive.
6) The tropospheric DOFs should be larger than 0.8, the SZA should be less than 85°.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.already included in the new AQ standard as a routine monitoring component (http://kjs.mep.gov.cn/).According to AQ data released by Chinese Ministry of Environmental Protection, tropospheric O3 has displaced PM 2.5 as the primary pollutant in many cities during summer (http://kjs.mep.gov.cn/).High precision and accuracy of measurements for tropospheric O3 and its precursors by FTS in China can not only improve the understanding of photochemical ozone production regime, but also greatly contribute to ozone pollution controls and the validation of satellite data and model simulations.Unfortunately, most global NDACC sites are located in s.l.(above sea level)) is located in the western suburbs of Hefei city (the capital of Anhui Province) in central-eastern China.This site is a NDACC site candidate, its location relative to the other sites and the enlarged view is shown inFig.1 (a).The observatory is currently the only site in China that has a continuously-operating solar FTS, making it crucial to calibrate and validate the satellite data or model simulations in this important region.Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.In this area prevail southeast winds in summer and northwest winds in winter.The regional landscape is mostly flat with a few hills.Downtown Hefei is located to the southeast of this site with only a 10 km distance and is densely populated with seven million people.The same as other megacities, serious air pollution is common in Hefei throughout the year (http://kjs.mep.gov.cn/).The site is surrounded by wetlands or cultivated lands in other directions.For a FTS solar absorption geometry used in NDACC, the position of the sun may affect the measurements especially for short lived compounds.Fig.1 (b) demonstrated the variations of solar zenith angle (SZA)

(
IFS125HR, Bruker GmbH, Germany), a solar tracker (Tracker-A Solar 547, Bruker GmbH, Germany), and a weather station (ZENO-3200, Coastal Environmental Systems, Inc., USA).The instrument has been operating almost continuously since its installation in April 2014, and the near infrared and MIR solar spectra were alternately acquired in routine observations.The MIR spectra used in this study are recorded over a wide spectral range (about 600 -4500 cm -1 ) with a spectral resolution of 0.005cm -1 .The instrument is equipped with a KBr beam splitter & a MCT detector for O3 and a KBr beam splitter & an InSb detector for other gases, and it has seven optical filters to avoid detector non-linearity.The weather station includes sensors for air pressure (± 0.1hpa), air temperature (± 0.3° C), relative humidity (± 3%), solar radiation (± 5% under daylight spectrum conditions), wind speed (± 0.5 m/s), wind direction (± 5°), and the presence of rain.Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.

Figs. 4
Figs.4 and 5show the error components contributing to the systematic error and random error covariance matrices of O3, CO, and HCHO, as well as the combined errors (theoretical basis is described in Appendix A).The error profile shape reflects the propagation of different errors in the retrieval process.In the troposphere, three main sources of systematic errors for each gas are evident: smoothing error, line intensity error and line pressure broadening error.The dominant systematic error for O3 and CO is smoothing error, and for HCHO is line intensity error.The main sources of random errors are gas dependent.The main random errors for O3 and HCHO are measurement error, temperature profile error and interference error, and for CO are zero baseline level error, measurement error, and temperature profile error.The dominant random error for O3 and HCHO is measurement error, and for CO is zero baseline level error.When taken all error items into account, the summarized errors in O3, CO, and HCHO for 0 -12 km tropospheric partial column and for the total column 3) Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.

Fig. 6
Fig. 6  shows the comparison between the OMI and FTS retrieved ozone profiles in the troposphere from 2015 to 2017 and the corresponding fractional differences calculated using Eq.(3).FTS profiles within 3 h of OMI overpass time (13:30 local time (LT )) were averaged and used for comparison.There were 53 FTS profiles that had OMI counterparts.The smoothed profiles have much better agreement with the FTS profiles than the unsmoothed profiles.Large discrepancies between the FTS profiles and unsmoothed OMI profiles were observed especially in middle troposphere, where the fractional difference varies about 15%.However, after

Fig. 8
Fig. 8  shows the comparison between the GEOS-Chem model profiles and FTS retrieved ozone profiles in the troposphere from 2015 to 2017 and the corresponding fractional differences calculated using Eq.(3).FTS profiles within 30 min of the GEOS-Chem model time were selected for comparison.There were 481 FTS profiles that had GEOS-Chem model data counterparts.As expected, the smoothed profiles biogenic emissions were calculated online with the Model of Emissions of Gases and Aerosols from Nature (MEGAN) embedded in WRF-Chem model.WRF-Chem outputs simulations with 45 vertical layers from ground to the height of 10 hPa.The WRF-Chem ozone profiles were also available with 1 h time frequency and treated with the same manner as that of GEOS-Chem model data.Fig. 10 shows the comparison between the WRF-Chem model profiles and FTS retrieved ozone profiles in the troposphere from 2015 to 2017 and the corresponding fractional differences Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.calculated using Eq.(3).The discrepancies between the FTS profiles and unsmoothed WRF-Chem model profiles were observed in boundary layer and middle troposphere, and they are smaller than the discrepancies between the FTS profiles and unsmoothed GEOS-Chem model profiles.After smoothing, the difference between WRF-Chem model profile and FTS profile is typically less than 12%.The coincident time series and correlation plot of tropospheric ozone column obtained by FTS and WRF-Chem are shown in Fig. 11.The WRF-Chem model data are smoother than the coincident FTS data, and the agreement of the two data sets has a correlation coefficient (r) of 0.65.Average difference between WRF-Chem and FTS a second-order Fourier series plus a linear component to determine the annual trends and intra-annual variations of tropospheric ozone at Hefei site.Even though it failed to determine the annual trend of tropospheric ozone column probably because the time series is much shorter than those inGardiner et al. (2008), the observed seasonal cycle of tropospheric ozone variations was well captured by the bootstrap resampling method.Typically, high levels of tropospheric ozone occur in spring and summer, and low levels of tropospheric ozone occur in autumn and winter.Day-to-day variations in spring and summer are in most cases larger than those in autumn and winter.Fig. 12 (b), which shows the monthly means of tropospheric ozone columns derived from FTS, smoothed and bias corrected OMI, GEOS-Chem, and WRF-Chem data, Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.illustrates this feature even better.At the same time, it shows that the tropospheric ozone column roughly increases over time at the first half year and reaches the maximum in June, and then it decreases over time at the second half year.Tropospheric ozone columns in June were, on average, 0.5×10 18 molecules*cm -2 (47.6%) higher than those in December which has a mean value of 1.05×10 18 molecules*cm -2 .The OMI time series shows similar behaviour.The measured features can basically reproduced by GEOS-Chem and WRF-Chem data but with slight shifts in the timing of the seasonal maximum.Vigouroux et al. (2015) studied ozone trends and variability with eight global NDACC stations that have a long-term time series of FTS ozone measurements, namely, Ny-Ålesund (79° N), Thule (77° N), Kiruna (68° N), Harestua (60° N), Jungfraujoch (47° N), Izaña (28° N), Wollongong (34° S) and Lauder (45° S).All these stations were located in non-polluted or relatively clean areas.The results showed a maximum tropospheric column in spring at all stations except at Jungfraujoch which extended into summer.This is because the STE process is most effective during late winter and spring(Vigouroux et al. 2015).However,Logan (1985) observed a broad summer maximum in the mid-latitude northern hemisphere, Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.

Fig. 15
Fig. 15 shows daily mean time series of temperature, pressure, humidity, wind direction, wind speed, and solar radiation recorded by the surface weather station that have tropospheric ozone counterparts.The correlation plots between FTS tropospheric ozone column and each meteorological element are shown in Fig. 16.The seasonal cycles of all coincident meteorological elements are not evident except temperature

Fig. 18
Fig. 18  shows the correlation plot between FTS tropospheric ozone column and the coincident tropospheric CO,HCHO, and NO2 columns.Tropospheric CO and HCHO    were derived from FTS spectra as tropospheric O3, while tropospheric NO2 was deduced from OMI product selected within the  0.7° latitude/longitude rectangular area around Hefei site and with retrieval uncertainty for tropospheric column of less than 30% (https://disc.gsfc.nasa.gov/datasets/OMNO2_V003/).The same as tropospheric O3, tropospheric HCHO exhibits a clear seasonal cycle and has a minimum in winter and maximum in summer.Pronounced tropospheric CO and NO2 variations were observed but the seasonal cycle is not evident because both emissions are not constant over seasons.When fitting all tropospheric O3 column to coincident CO, HCHO, and NO2 columns, we obtained a good correlation between tropospheric O3 and CO (Fi.g 19(a)).However, the correlations to NO2 (Fig.18 (b left)) and HCHO (Fi.g 19 (c left)) were only moderate.This confirms that the observed tropospheric ozone is indeed highly correlated to air pollution, but cannot be simply attributed to either NOx pollution or VOCs pollution.Sillman (1995) taken HCHO and NOy (total reactive nitrogen) as the 'indicator species' and investigated the sensitivity of PO3 to changes in VOCs and NOx, which is regarded as NOx-limited when the ratio of HCHO to NOy is high and VOC-limited Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.between 1.3 and 2.8.Fig. 18 (b right) exemplifies a correlation plot between O3 and NO2 within NOx-limited region, and Fig. 18 (c right) exemplifies a correlation plot between O3 and HCHO within VOC-limited region.Tropospheric O3 shows a good correlation with both NO2 and HCHO in the respective limited regions.With this transition criteria, 106 days of observations that have coincident O3, HCHO, and NO2 counterparts in the reported period are calculated.Fig. 19 (a) shows classified time series of tropospheric ozone column and HCHO/NO2 ratios.The HCHO/NO2 ratios in summer are typically larger than those in winter.The PO3 is mainly NOx limited in summer, while is mainly VOC or mix VOC-NOx limited in winter.Fig. 19 (b) listed the statistics for the 106 days of observations, which shows that NOx limited, VOC-NOx limited, and VOC limited PO3 accounts for 60.1%, 28.7%, and 11%, respectively.Considering most of tropospheric ozone are NOx limited or mix VOC-NOx limited, reductions in NOx would reduce most tropospheric ozone in eastern China.

(
OMI) data, global GEOS-Chem model simulation, and regional WRF-Chem simulation.The OMI time series shows a similar seasonal cycle.GEOS-Chem model and WRF-Chem model can generally reproduce the seasonal cycle observed by FTS.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.and photochemical production regime in eastern China from 2014 -2017.A pronounced seasonal cycle for tropospheric ozone is observed by FTS, where high levels of tropospheric ozone occurs in spring and summer, and low levels of tropospheric ozone occurs in autumn and winter.Day-to-day variations in spring and summer are in most cases larger than those in autumn and winter.At the same time, it

where
. Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License.Svar is the error covariance matrix of the model parameter vector var, a nd Kvar is the corresponding weighting function matrix.Here var refers to one of the error items listed in Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1176Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 18 December 2017 c Author(s) 2017.CC BY 4.0 License. the standard deviation to the average of the sun intensities.

Fig. 7 .
Fig.7.(a): Comparison of FTS tropospheric ozone column and smoothed OMI data.(b): Correlation plot between coincident FTS tropospheric ozone column and smoothed OMI data.The blue line is the linear fitted curve of the scatter points.The red line denotes one-to-one line.

Fig. 11 .
Fig.11.(a): Comparison of FTS tropospheric ozone column and smoothed WRF-Chem model data.975 (b): Correlation plot between coincident FTS tropospheric ozone and smoothed WRF-Chem 976 model data.The blue line is the linear fitted curve of the scatter points.The red line denotes 977 one-to-one line.The black scatters are the same comparison but a bias of 0.27×10 18 978 molecules*cm -2 from the smoothed WRF-Chem model data is included.979 980

Table 2
Table 1 except smoothing error and measurement error.In Eq. (A6), to estimate Es correctly, Sa should represent natural variabili ty of the target gas in the atmosphere, and thus should be evaluated from clim atological data.In this study, the selection of Sa as described in Table 1 tends to underestimate the error, which is in good agreement with most similar stud ies.The model parameter error contains the error from retrieved parameters, na mely the interference error, and the error from non-retrieved forward model par ameters.