2.1 MAX-DOAS measurements

2.1.2 Spectral retrieval

All the measurement spectra were first corrected for offset and dark current and then analyzed using the software QDOAS version 3.2 (http://uv-vis.aeronomie.be/software/QDOAS/, last access: 7 August 2019). The MAX-DOAS spectral fit was performed at two different wavelength ranges of 338–370 nm for NO₂ retrieval and 324.5–359 nm for HCHO retrieval. In this study, the zenith spectrum ($\mathit{\alpha }=\mathrm{90}{}^{\circ }$) taken in the same measurement cycle is used as reference spectrum in the spectral analysis to retrieve the differential slant column densities (DSCDs). The differential slant column density (DSCD) is defined as the difference between the slant column density (SCD) of the measured spectrum and the corresponding zenith reference spectrum. The broadband spectral structures caused by Rayleigh and Mie scattering are removed by including a low-order polynomial in the DOAS fit. Absorption cross section of several trace gases were included in the DOAS fit at different fitting ranges, details of the DOAS retrieval settings for each wavelength range are listed in Table 1. The DOAS fit settings follow the one used in the Quality Assurance for Essential Climate Variables (QA4ECV) project (http://www.qa4ecv.eu/, last access: 7 August 2019). These settings have also been adopted for the Second Cabauw Intercomparison campaign for Nitrogen Dioxide measuring Instruments (CINDI-2) campaign (Kreher et al., 2019). A Small shift or squeeze of the wavelengths is allowed in the wavelength mapping process in order to compensate small uncertainties caused by the instability of the spectrograph. Figure 1 shows an example of the spectral fit retrieval of NO₂ and HCHO DSCDs from a MAX-DOAS spectrum taken on 12 May 2013 at 08:48 LT (local time) with a viewing elevation angle of 1^∘.

Figure 1An example of the DOAS retrieval of NO₂ and HCHO DSCDs from a MAX-DOAS spectrum taken on 12 May 2013 at 08:48 LT (local time) with viewing elevation angle of 1^∘. Panels (a, b, c) show the DOAS fit in the wavelength range of 338–370 nm, while panels (d, e, f) show DOAS fit in the wavelength range of 324.5–359 nm.

Download

Fleischmann et al. (2003)Meller and Moortgat (2000)Vandaele et al. (1998)Vandaele et al. (1998)Serdyuchenko et al. (2014)Serdyuchenko et al. (2014)Thalman and Volkamer (2013)

Table 1The DOAS retrieval settings for different wavelength bands.

^a I₀ correction is applied with SCD of 10¹⁷ molec cm⁻² (Aliwell et al., 2002). ^b I₀ correction is applied with SCD of 10²⁰ molec cm⁻² (Aliwell et al., 2002).

Download Print Version | Download XLSX

As clouds are not included in the radiative transfer calculation of the aerosol and trace gas profile retrieval, the retrieval result can be significantly influenced by the presence of cloud in the atmosphere. Therefore, the retrieved DSCDs were first filtered by removing cloudy scenes before proceeding to the aerosol and trace gas profile retrieval. The vertical profile of the oxygen collision complex O₄ only varies in a small range with atmospheric pressure and temperature, the retrieved O₄ DSCDs and (relative) intensities are assumed to vary smoothly with time with the solar and viewing geometry. Rapid change in O₄ DSCDs and intensities indicates a sudden change in the radiative transport condition, which is likely due to the presence of clouds. Therefore, we applied a locally weighted regression smoothing filter (LOWESS) (Cleveland, 1981) with a regression window of 3 h to the O₄ DSCDs and intensity time series at each elevation angle to filter data influenced by inhomogeneous and/or rapid changes in radiation transport conditions. Data with rapidly varying O₄ DSCDs and intensities were filtered out. Only data with slowly varying O₄ DSCDs and intensities were used in the aerosol and trace gas retrieval. The limitation of this cloud screening algorithm is that the algorithm is not able to distinguish between continuous and homogeneous cloud conditions. However, it is rare that the cloud does not change for a long time (within an hour) and the cloud layer is homogeneous for all viewing directions.

2.1.3 Aerosol and trace gas profile retrieval

Previous studies show that there is a systematic discrepancy between observation and model simulation of O₄ DSCDs (Wagner et al., 2009, 2011; Clémer et al., 2010; Chan et al., 2015, 2018; Wang et al., 2016; Zhang et al., 2018). A scaling correction factor has to be applied to the measured O₄ DSCDs in order to bring measurement and model results into agreement. However, the physical meaning of this observation is still not well understood (Ortega et al., 2016; Wagner et al., 2019). Theoretically, the optical path should be the longest under aerosol-free conditions for off-zenith measurement. Thus, the MAX-DOAS measurement of O₄ DSCDs should be smaller than the one simulated with pure Rayleigh atmosphere. In this study, we compared the forward simulation of O₄ DSCDs under aerosol-free conditions to the O₄ DSCDs retrieved from the MAX-DOAS observations to examine the necessity of O₄ correction. Our result shows that the measured O₄ DSCDs occasionally exceeded the forward simulation results which implies that correction of O₄ DSCDs is necessary. The ratio between simulated and measured O₄ DSCD varies from 0.75 to 1.0 for cases where the measured O₄ DSCDs exceeded the forward simulation results. Our finding agrees with previously reported scaling correction factors of O₄ DSCDs, which are ranging from 0.7 up to 1.0. In order to avoid over-correction due to the outliers, we take the 10th percentile instead of the minimum value of the simulated and measured O₄ DSCD ratio as the correction factor which is ∼0.8. All MAX-DOAS observations of O₄ DSCD are corrected by multiplying the correction factor of 0.8. From hereafter, all O₄ DSCDs are referring to the corrected O₄ DSCDs.

The conversion of the MAX-DOAS observations to aerosol extinction and trace gas profile requires inversion of the underlying radiative transfer equations. These equations cannot be linearized; therefore, it is suggested to fit the measurement quantities to the forward calculation of radiation transfer (Wagner et al., 2004; Hönninger et al., 2004; Sinreich et al., 2005; Frieß et al., 2006; Hartl and Wenig, 2013; Chan et al., 2018). As the vertical distribution profile of O₄ is very stable and it has several absorption bands in the ultraviolet and visible spectral range, it is commonly used as a fitting quantity for the aerosol retrieval.

In this study, aerosol vertical profiles are retrieved at the 360 nm O₄ absorption band using the Munich Multiple wavelength MAX-DOAS retrieval algorithm (M³; Chan et al., 2018). The algorithm is developed based on the optimal estimation method (Rodgers, 2000) and utilizes the Library for Radiative Transfer (LibRadTran) model (Emde et al., 2016) as the forward model. In this study, the U.S. Standard Atmosphere midlatitude profiles for winter (January) and summer (July) are temporally interpolated to each month of the year for the radiative transfer simulations. A brief description of the aerosol and trace gas vertical profile retrieval algorithm and the parameterization used in this study are presented in the following. A more detailed description of the M³ retrieval algorithm can be found in Chan et al. (2018).

In this study, all valid MAX-DOAS observations within a single measurement cycle are grouped together for the aerosol vertical profile retrieval. Assuming the set of measurements can be reproduced by the forward model and the forward model results are dependent on the aerosol extinction profile, we can then retrieve the aerosol extinction profile by fitting the forward model results to the MAX-DOAS O₄ observations using the iterative Newton–Gauss method. The forward simulations of O₄ DSCDs are performed at 360 nm due the strong O₄ absorption at this wavelength. As the information contained in the MAX-DOAS observations are most likely not enough to retrieve a unique aerosol extinction profile, we have to supply the necessary information to the aerosol inversion in a form of a priori aerosol profile. As the aerosol load in Nanjing varies over a wide range, using a fixed a priori could result in over-regularizing the retrieval under higher aerosol load conditions. Therefore, we have implemented an iterative approach to avoid over-regularizing the retrieval. We first use a fixed initial a priori to retrieve the aerosol extinction profile. The fixed a priori profile is then scaled to have the same aerosol optical depth retrieved from the previous run. Then the new a priori is used in the next run to retrieve the aerosol extinction profile. This procedure is repeated until the difference between the retrieved and a priori aerosol optical depth is less than 10 % or the number of iterations reaches the limit.

Figure 2An example of vertical profile retrieval of aerosol extinction, NO₂ and HCHO from MAX-DOAS measurements taken on 4 July 2013 at around 08:47 LT. The left panels show the (a) aerosol extinction, (d) NO₂ and (g) HCHO profiles. Panels in the middle column indicate the average kernel of (b) aerosol extinction, (e) NO₂ and (h) HCHO profile retrieval. The measured (blue curves) and modeled (green curves) DSCDs of (c) O₄, (f) NO₂ and (i) HCHO are shown in the right column.

Download

As aerosols are typically emitted and formed close to the surface in urban areas, we assume the a priori aerosol extinction profile follows an exponentially decreasing function with a scale height of 1.0 km. The aerosol optical depth (AOD) of the a priori aerosol profile is set to 0.5 for the retrieval at 360 nm. The uncertainties in the a priori aerosol profile are set to 100 % and the correlation length of the aerosol inversion is assumed to be 0.5 km. As MAX-DOAS measurements are more sensitive to the aerosol and trace gases close to the instrument, we therefore divided the lowest 3.0 km of the troposphere into 15 layers with a thickness for each layer of 200 m. A fixed set of single scattering albedo of 0.92, asymmetry parameter of 0.68 and ground albedo of 0.04 is assumed in the radiative transfer calculations. An example of aerosol extinction profile retrieval is indicated in Fig. 2a–c.

The aerosol information obtained from the procedure described above was used for the differential box air mass factor (ΔDAMF) calculation for the trace gas profile inversion. The ΔDAMFs were calculated at a single wavelength for the retrieval of trace gas profile using the radiative transfer model LibRadTran with the Monte Carlo simulation module MYSTIC (Emde et al., 2016). The ΔDAMF was assumed to be constant within the DOAS spectral fitting windows. As NO₂ DSCDs were retrieved in the same fitting window as O₄, we therefore adapted the same wavelength of 360 nm for NO₂ ΔAMF simulations. An example of NO₂ profile retrieval is indicated in Fig. 2d–f. HCHO vertical distribution profile retrievals were retrieved at 340 nm which is close to the central wavelength of the DOAS fitting window of 342 nm. In addition, this wavelength is commonly used in HCHO retrieval, e.g., De Smedt et al. (2018). Aerosol extinction profiles obtained at the 360 nm O₄ band are converted to 340 nm assuming a fixed Ångström coefficient (Ångström, 1929) of 1 for the calculation of ΔAMF. The Ångström coefficient can vary significantly with time. Sun photometer measurements in Nanjing (see Sect. 2.2) show that the Ångström coefficient mostly varies between 0.7 (10th percentile) and 1.4 (90th percentile). Based on the sun photometer observations, we estimated that the error caused by fixed Ångström coefficient on aerosol extinction is less than 2 %. The ΔDAMFs are subsequently calculated at 340 nm for the HCHO profile retrieval. An example of HCHO profile retrieval is indicated in Fig. 2g–i.

The atmosphere layer settings of the trace gas profile retrieval are the same as the one used in the aerosol profile retrieval. NO₂ and HCHO concentrations above the retrieval height (3 km) are assumed to follow the U.S. Standard Atmosphere (Anderson et al., 1986). As the trace gas profile cannot be fully reconstructed by the small number of measurements, we therefore use the optimal estimation method (Rodgers, 2000) with iterative approach to regularize the inversion (Chan et al., 2015, 2018). The first step uses a fixed initial a priori to retrieve the trace gas distribution. The fixed a priori profile is then scaled to have the vertical column retrieved from the first run. The scaled a priori is used in the next run to retrieve the trace gas profile. The process repeats until the difference between the retrieved and a priori trace gas vertical column is less than 10 % or the number of iteration reaches the limit. In this study, the a priori is assumed to follow an exponential decrease function with a scale height of 1.0 km. The uncertainty in the a priori profile is set to 100 % of the a priori and correlation length is set to 0.5 km in the trace gas profile inversion. The NO₂ vertical column density (VCD) of the a priori is set to 2 × 10¹⁶ molec cm⁻², while the a priori HCHO VCD is set to 1 × 10¹⁶ molec cm⁻².

2.2 Sun photometer measurements

A sun photometer was installed on the roof of a building of Nanjing University of Information Science and Technology (32.20^∘ N, 118.70^∘ E), which is located at the north bank of the Yangtze River. The sun photometer site is ∼25 km northwest of the MAX-DOAS measurement site. The two instruments are separated by the Yangtze River. An industrial park is about 3 km east of the sun photometer site. Heavy industry factories, i.e., steel and cement plants, are located in the industrial park. In this study, AODs measured by the sun photometer were used for the intercomparison study to validate the MAX-DOAS aerosol retrieval. Cloud screened data were used. The data consist of AOD measurements at seven different wavelength channels from 340 to 1020 nm. AOD measurements at 340 nm and 380 nm are interpolated to 360 nm for comparison.

2.3 Meteorological data

Meteorological parameters such as temperature, wind speed and wind direction are taken from a weather station in Nanjing, which is located about 20 km south of the MAX-DOAS measurement site. The weather station is operated by the National Meteorological Center of the China Meteorological Administration. These data are available on the web page of the China Meteorological Administration (https://data.cma.cn/site/index.html, last access: 7 August 2019). The meteorological data are mainly used for the analysis of meteorological effects on air quality in Nanjing during the Youth Olympic Games in 2014.

2.4 OMI satellite observations

The ozone monitoring instrument (OMI) is a passive nadir-viewing satellite-borne imaging spectrometer (Levelt et al., 2006) on board the Earth Observing System's (EOS) Aura satellite. The EOS Aura satellite orbit at an altitude of ∼710 km with a local Equator overpass time of 13:45 LT (local time). The instrument consists of two charge-coupled devices (CCDs) covering a wavelength range from 264 to 504 nm. A scan provides measurements at 60 positions across the orbital track covering a swath of approximately 2600 km. The spatial resolution of OMI varies from ∼320 km² (at nadir) to ∼6400 km² (at both edges of the swath). The instrument scans along 14.5 sun-synchronous polar orbits per day providing daily global coverage.

The NASA OMI NO₂ and HCHO standard product version 3 (Krotkov et al., 2017; González Abad et al., 2015) is used in this study. In the NO₂ product, the slant column densities (SCDs) of NO₂ are derived from Earth's reflected spectra in the visible range (402–465 nm) using an iterative sequential algorithm (Marchenko et al., 2015). Previous studies show the updated SCDs are on average 10 %–40 % lower compared to the previous version of retrieval (Marchenko et al., 2015). The OMI NO₂ SCDs are converted to VCDs by using the concept of air mass factor (AMF) (Solomon et al., 1987). The AMFs are calculated using NO₂ profiles simulated by the Global Modeling Initiative (GMI) chemistry transport model. The horizontal resolution of GMI is 1^∘ (latitude) × 1.25^∘ (longitude) (Rotman et al., 2001). Separation of stratospheric and tropospheric columns is achieved by the local analysis of the stratospheric field over unpolluted areas (Bucsela et al., 2013).

The OMI HCHO retrieval algorithm uses the direct fit of radiances in the spectral range from 328.5 to 356.5 nm for the SCD retrieval. In the current version of the HCHO product, OMI radiance measurement over the remote Pacific is used as reference in the fitting process. This approach is reported to reduce the interferences from unresolved spectral structures in the retrieval of weak absorbers like HCHO (De Smedt et al., 2018). The retrieved SCDs are then converted to VCDs using the AMF approach. The AMFs are calculated based on climatological HCHO profiles.

2.5 Backward trajectory modeling

The National Oceanic and Atmospheric Administration Air Resources Laboratory (NOAA ARL) developed the HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015) (https://www.arl.noaa.gov/hysplit/, last access: 7 August 2019) used to investigate the transportation of pollutants over the Yangtze River Delta. Meteorological data from the Global Data Assimilation System (GDAS) with a spatial resolution of $\mathrm{0.5}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ and 24 vertical levels were used in the model for trajectory simulations. Backward trajectories are computed for air masses arriving at the midpoint of each MAX-DOAS retrieval layer.

3.1 Comparison of MAX-DOAS and sun photometer AODs

Aerosol optical depths (AODs) retrieved from the MAX-DOAS observations are compared to the sun photometer measurements. As the sampling resolution of the MAX-DOAS and the sun photometer are different, individual measurements are averaged to hourly, daily and monthly values for comparison. Figure 3 shows the scatter plot of AOD measured by the sun photometer and MAX-DOAS. Both datasets are cloud filtered. AODs measured by both MAX-DOAS and sun photometer are in general in good agreement. The Pearson correlation coefficient (R) of daily averaged data is 0.73. The slope of the total least squares regression between the two datasets is 0.56 with an offset of 0.13. The discrepancies between the two datasets can be explain by the difference in measurement technique and measurement location. The sun photometer derives AOD from direct sun measurements, while the MAX-DOAS retrieves AOD using the O₄ absorption information from scattered sunlight. The sun photometer is sensitive to the entire column, while the MAX-DOAS is mostly sensitive to aerosol at the lowest few kilometers of the troposphere. Therefore, the MAX-DOAS is likely to underestimate the AOD when there is an elevated aerosol layer in the upper troposphere. This explanation matches with the observations that the MAX-DOAS agrees better with the sun photometer under low aerosol conditions (AOD < 1) and underestimates the AOD during high AOD conditions. Another source of error is the assumption of aerosol optical properties in the MAX-DOAS retrieval. In order to quantify the error caused by the fixed asymmetry parameter and single scatter albedo used in the aerosol retrieval, we have performed a sensitivity analysis by retrieving the aerosol profile with different asymmetry parameter and single scattering albedo. As the sun photometer in Nanjing does not provide full inversion of aerosol optical properties, aerosol optical properties are taken from the Taihu AERONET (AErosol RObotic NETwork) station which is about 150 km southeast of Nanjing. The 10th and 90th percentiles of asymmetry parameter and single scatter albedo are used for the sensitivity study. The result shows that the error caused by the fixed asymmetry parameter on AOD retrieval can be up to 4 %, while the single scattering albedo shows a smaller impact of up to 1.5 %. In addition, the MAX-DOAS and sun photometer are separated by a distance of ∼25 km and the sun photometer site is very close to an industrial park, where the heavily polluting industries are located, i.e., steel and cement plants. Therefore, the sun photometer measurements are expected to be higher than the MAX-DOAS observations. Accounting for these effects, the MAX-DOAS and sun photometer observations agree well with each other and the MAX-DOAS-derived aerosol profiles are reliable for the retrieval of NO₂ and HCHO profiles.

Figure 3Comparison of AOD measured by the MAX-DOAS and sun photometer. The total least squares regression line is calculated based on the daily data.

Download

3.2 Comparison of MAX-DOAS and OMI observations

Vertical column densities of NO₂ and HCHO retrieved from the ground-based MAX-DOAS measurements are compared to OMI satellite observations. MAX-DOAS data are temporally averaged around the OMI overpass time from 12:00 to 14:00 LT, while OMI VCDs are spatially averaged for pixels within 20 km of the MAX-DOAS measurement site. Time series of MAX-DOAS and OMI observations of NO₂ and HCHO VCDs are shown in Figs. 4 and 5, respectively. Both OMI NO₂ and HCHO datasets are filtered for cloud radiance fraction smaller than 0.4. Daily and monthly averaged data are shown. Missing data for some months are due to cloud filtering or instrumental issues of the MAX-DOAS.

Figure 4(a) Time series of the MAX-DOAS (red markers) and NASA OMI (blue markers) tropospheric NO₂ VCDs over Nanjing from 2013 to 2017. MAX-DOAS data are temporally averaged around the OMI overpass time. OMI measurements are spatially averaged for pixels within 20 km of the MAX-DOAS measurement site. OMI NO₂ VCDs retrieved using MAX-DOAS profile as a priori information are indicated as green markers. Tropospheric NO₂ VCDs taken from a KNMI product are also shown as gray marker for reference. (b) Comparison of tropospheric NO₂ VCDs between the MAX-DOAS and OMI satellite observations. The total least squares regression line is calculated based on the monthly average data.

Download

Figure 5(a) Time series of the MAX-DOAS (red markers) and OMI (blue markers) HCHO VCDs over Nanjing from 2013 to 2017. MAX-DOAS data are temporally averaged around the OMI overpass time. OMI measurements are spatially averaged for pixels within 20 km of the MAX-DOAS measurement site. OMI NO₂ VCDs retrieved using MAX-DOAS profile as a priori information are indicated as green markers. (b) Comparison of HCHO VCDs between the MAX-DOAS and OMI satellite observations. The total least squares regression line is calculated based on the monthly average data.

Download

NO₂ VCDs measured by the MAX-DOAS and the OMI satellite show a similar seasonal pattern with higher NO₂ columns during winter and lower NO₂ values over summer. Higher NO₂ levels are due to a longer atmospheric lifetime of NO₂ during winter and higher emissions in winter, e.g., higher power consumption and emissions from individual domestic heating. MAX-DOAS and OMI observations show good temporal consistency with each other with Pearson correlation coefficient (R) of 0.91. Despite the strong correlation between the two datasets, the OMI observations are systematically underestimating the NO₂ columns. The slope and offset of the total least squares regression between the two datasets are 0.50 and $-\mathrm{1.89}\times {\mathrm{10}}^{\mathrm{15}}$ molec cm⁻², respectively. Averaged OMI NO₂ column over Nanjing is 5.67 × 10¹⁵ molec cm⁻², which is about 60 % lower than the MAX-DOAS averaged VCD of 14.96 × 10¹⁵ molec cm⁻² (see Fig. 4). The underestimation of NO₂ VCDs from OMI measurements agree with previous studies in the Yangtze River Delta (Chan et al., 2015). The discrepancies can be related to the differences in spatial coverage between the ground-based and satellite observations and the uncertainties related to the NO₂ vertical distribution profile shape used in the air mass factor calculation. We quantified the influence of NO₂ vertical distribution profile on the air mass factor calculation by recomputed the OMI tropospheric NO₂ air mass factor using the MAX-DOAS NO₂ profile as a priori information. Seasonal average of OMI a priori NO₂ profiles are shown in Fig. 6a. MAX-DOAS observations of NO₂ profiles are shown in Sect. 3.3. The MAX-DOAS observations show that most of the NO₂ resided below 500 m, while OMI a priori NO₂ profiles shoed a larger fraction of NO₂ above 500 m. OMI NO₂ VCDs retrieved with MAX-DOAS NO₂ profiles are also indicated in Fig. 4. Using the MAX-DOAS profile as a priori information in the OMI NO₂ VCDs retrieval on average increased the OMI NO₂ VCDs by ∼30 % with correlation nearly unchanged. The averaged OMI NO₂ column over Nanjing increased to 7.29 × 10¹⁵ molec cm⁻². In addition, the slope of the regression line increases from 0.50 to 0.57, while the offset reduces from −1.89 to $-\mathrm{1.35}\times {\mathrm{10}}^{\mathrm{15}}$ molec cm⁻². The result indicates that a better estimation of NO₂ vertical distribution in the satellite VCD retrieval reduces the discrepancies. Our findings agree with previous studies that using better estimated NO₂ vertical profile in the satellite air mass factor calculation enhanced the OMI NO₂ columns and reduced discrepancies between OMI and ground-based observations (Chan et al., 2012, 2015; Lin et al., 2014; Wang et al., 2017). However, the improved OMI NO₂ VCDs are still about 50 % lower than the MAX-DOAS observations indicating that there are still remaining issues. A previous study shows that the OMI NO₂ tropospheric vertical columns processed by the Royal Netherlands Meteorological Institute (KNMI; DOMINO version 2.0; Boersma et al., 2011) only underestimated the tropospheric NO₂ VCDs by ∼20 % compared to ground-based MAX-DOAS observations in Wuxi (another city in the Yangtze River Delta; Wang et al., 2017). In order to investigate the differences between different OMI NO₂ products, we have also plotted the KNMI OMI NO₂ product in Fig. 4. The average NO₂ column reported from the KNMI) product is 10.10 × 10¹⁵ molec cm⁻², which is about a factor of 2 higher than the NASA product. The result shows the KNMI product only underestimate the NO₂ columns in Nanjing by ∼30 %, which is consistent with a previous study in the Yangtze River Delta (Wang et al., 2017). The difference between the two OMI products can be related to different spectral analysis techniques, stratospheric tropospheric separation algorithms and a priori profiles used in the retrievals. Investigation of the differences between the two algorithms is, however, beyond the scope of this study. Another source of discrepancy is related to the difference in spatial coverage of OMI and MAX-DOAS observations. The spatial coverage of the OMI measurement footprint is ranging from 330 up to 4600 km² with an average of 920 km². Measurements with large pixel size are probably difficult to capture the spatial gradient of NO₂ and result in an underestimation over pollution hot spots due to the averaging over a large OMI footprint. This effect is especially significant over Nanjing, as it is a local pollution hot spot surrounded by rather clean areas. The underestimation of NO₂ columns is coherent with previous measurements over urban pollution hot spots, i.e., Washington DC and Shanghai (Wenig et al., 2008; Chan et al., 2015). In addition, previous measurements over a suburban area in Shanghai show better agreement with OMI observations compared to the measurements in the city (Chan et al., 2015), indicating the effect of spatial inhomogeneity of NO₂ on the satellite data comparison. However, the impact of this effect is difficult to quantify due to a lack of high spatial resolution data. Cloud contamination could be an important source of error in the satellite retrieval. A previous study shows that the cloud effect on OMI NO₂ VCD retrieval is only significant for cloud radiance fraction > 0.4 (Wang et al., 2017). The OMI NO₂ data used in this study are filtered for cloud radiance fraction smaller than 0.4. Therefore, cloud contamination should only show a negligible effect in the comparison.

Figure 6Seasonal average of (a) NO₂ and (b) HCHO a priori profiles used in OMI retrieval. A priori profile of spring (March, April, May), summer (June, July and August), autumn (September, October, November) and winter (January, February and December) are shown.

Download

Figure 7Vertical distribution of (a) NO₂ and (b) HCHO for different seasons. Shaded areas represent 1σ standard deviation of variation.

Download

Figure 8The upper panels show spatial distribution of NO₂ in (a) summer and (b) winter reconstructed from the MAX-DOAS measurements using backward trajectories calculated by the HYSPLIT model. Backward trajectories are calculated over 6 h for summer measurements and 12 h for winter measurements. The lower panels indicates the average OMI NO₂ vertical column densities over the Yangtze River Delta during (c) summer and (d) winter.

We have also compared the MAX-DOAS and OMI observations of HCHO and both datasets show a similar seasonal variation pattern. The seasonal variation pattern of HCHO is opposite to NO₂ with higher HCHO columns during summer and lower in winter. Higher HCHO levels in summer are related to the increases in biogenic emission of precursor VOCs from plants and higher oxidation rate of VOCs. The MAX-DOAS measurements of HCHO VCDs agree well with the OMI observations with Pearson correlation coefficient (R) of 0.75. In addition, the absolute value of the columns show a perfect agreement. The average HCHO VCD measured by the MAX-DOAS is 8.04 × 10¹⁵ molec cm⁻², while the averaged OMI HCHO VCD is 7.89 × 10¹⁵ molec cm⁻². The slope of the total least squares regression between the two datasets is 0.99 with an offset of 0.31 × 10¹⁵ molec cm⁻² (see Fig. 5b). In order to quantify the influence of HCHO vertical distribution profile on the air mass factor calculation, we have recomputed the OMI HCHO VCDs using the MAX-DOAS HCHO profiles as a priori information. OMI HCHO VCDs retrieved with MAX-DOAS HCHO profiles are also indicated in Fig. 5. Using the MAX-DOAS profile as a priori information in the OMI HCHO VCDs retrieval on average increased the OMI HCHO VCDs by ∼25 % with correlation nearly unchanged. The averaged OMI HCHO column over Nanjing increased to 9.95 × 10¹⁵ molec cm⁻². The slope of the regression line also increases from 0.99 to 1.39, while the offset reduces from 0.31 to $-\mathrm{0.72}\times {\mathrm{10}}^{\mathrm{15}}$ molec cm⁻². Higher HCHO VCDs retrieved from OMI using MAX-DOAS profiles as a priori is mainly due to ignoring HCHO in the upper altitudes. Seasonal average of OMI a priori HCHO profiles are shown in Fig. 6b. The a priori show a considerable amount of HCHO above 3 km especially during summer, while the MAX-DOAS only reports HCHO mixing ratios up to 3 km above ground level. Sources of HCHO in the troposphere include the oxidation of various VOCs, including methane. Some of these VOCs have relatively long atmospheric lifetimes, e.g., methane; therefore, they are rather well mixed in the atmospheric and result in a larger portion of HCHO in the upper troposphere. The MAX-DOAS could not capture HCHO at higher altitudes. Therefore, using the MAX-DOAS profiles for OMI HCHO retrieval likely underestimates the air mass factors and overestimates the HCHO total columns.

Figure 9The upper panels show spatial distribution of HCHO in (a) summer and (b) winter reconstructed from the MAX-DOAS measurements using backward trajectories calculated by the HYSPLIT model. Backward trajectories are calculated over 6 h for summer measurements and 12 h for winter measurements. The lower panels indicates the average OMI HCHO vertical column densities over the Yangtze River Delta during (c) summer and (d) winter.

3.3 Seasonal variation in aerosol, NO₂ and HCHO vertical profiles

The seasonal variations in pollutants are closely related to meteorological conditions as well as the characteristics of different emission sources. Analyzing the seasonal variation patterns of different atmospheric pollutants can provide further information on the emission source characteristic as well as the atmospheric processes. Figure 7 shows the vertical profiles of NO₂ and HCHO for all four seasons. Significant seasonal patterns are observed from the NO₂ and HCHO data. The NO₂ vertical profiles show higher NO₂ maximum ratios in autumn and winter and lower in spring and summer. NO₂ profiles from all seasons show that the NO₂ mixing ratios decrease with increasing altitude. The result indicates most of the measured NO₂ is produced close to the surface which agrees with the fact that traffic emission is one of the largest sources of atmospheric NO₂ in Nanjing. HCHO measurements show a different seasonal variation pattern with lower values during winter and higher mixing ratios in summer. The seasonal pattern of HCHO indicates the significant contribution from biogenic emissions from vegetation. Although HCHO profiles for all seasons also show decreasing mixing ratio with height, the decreasing rate is in general much lower than that of NO₂. A larger fraction of HCHO is located at higher altitudes. This is probably related to the source characteristic of HCHO. A majority of the tropospheric HCHO is secondary produced from the oxidation of VOCs and results in a larger portion of HCHO in the upper altitudes.

Figure 10Spatial distribution of NO₂ in summer (left panels) and winter (right panels) reconstructed from the MAX-DOAS measurements using different assumed lifetime. The average OMI NO₂ vertical column densities during summer and winter are shown in (k) and (l), respectively. The corresponding lifetimes used in the calculation and the spatial correlation between the reconstructed maps and OMI observations are indicated each plot.

3.4 Regional pollution transport

Air pollutants released to the atmosphere do not only cause local impacts, but also can influence regions far from the source through regional transport. In order to investigate the influences from regional transportation of air pollutants, we use the backward trajectories calculated by the HYSPLIT model to back-correlate the possible source regions of the MAX-DOAS measured NO₂. Similar approaches have also been applied for other source appointment studies (Stohl et al., 2009, 2011; Brunner et al., 2012, 2017; Wang et al., 2019). Backward trajectories are calculated at the center point of each MAX-DOAS retrieval layer below 2 km. The MAX-DOAS measurement of NO₂ partial columns are then assigned to the grid points along the backward trajectories. We assume the partial columns represent a 200 m thick layer of the assigned grid point and the vertical distribution of NO₂ in the assigned grid point is assumed to be homogeneous from the surface to 2 km above ground. In this study, the backward-propagated NO₂ data are binned in a resolution of $\mathrm{0.2}{}^{\circ }\times \mathrm{0.2}$^∘ grid. As NO₂ is a relatively short-life pollutant in the atmosphere, it is less likely to stay in the atmosphere for a long time or be transported far from the sources. Therefore, NO₂ measured by the MAX-DOAS is less correlated to pollutant concentrations far from the site, which take longer time to reach the measurement site. In addition, model error accumulates over time. Results with shorter simulation time (or backward time) are considered more reliable. In order to account for this effect, an age weighting factor (w_τ) is used in the backward propagation. The age weighting factor is defined by Eq. (1) where τ is the assumed lifetime of NO₂ and t represents the time for the air mass to travel to the measurement site. This weighting scheme is useful when multiple trajectories are overlapping each other within a single grid point.

Figure 11Time series of (a) aerosol optical depths, (b) HCHO and NO₂ VCDs and (c) HCHO and NO₂ surface mixing ratios measured by the MAX-DOAS around the Youth Olympic Games. Hourly and daily averaged values are shown. The gray shadowed area indicates the Youth Olympic Games period (16 to 28 August 2014).

Download

As the lifetime of NO₂ has a strong seasonal variability, we separate the measurements into summer (June, July and August) and winter (December, January and February) in our analysis. Figure 10 shows the spatial distribution of NO₂ reconstructed from the MAX-DOAS measurements using different assumed lifetime. The corresponding lifetime used in the calculation and the spatial correlation between the reconstructed maps and OMI observations are indicated on the plots. Backward trajectories are calculated over 2 times the assumed lifetime. As the assumed lifetime is only used for the calculation of the age factor, the accuracy of the lifetime does not show a strong impact on the reconstructed spatial pattern. NO₂ lifetimes of 3 and 6 h are chosen to reconstruct the spatial distribution of NO₂ during summer and winter, respectively. Backward trajectories are calculated over 6 h for summer measurements and 12 h for winter measurements. These values are chosen based on the consideration of attaining a balance between having better spatial coverage and the reliability of the reconstructed pollution maps. Note that the simulation time (or backward time) is changing along the trajectory, and the weighting of each point along the trajectory is depending on its simulation time according to Eq. (1). Points along the trajectory with shorter simulation time are weighted higher in the calculation. NO₂ fields reconstructed from the MAX-DOAS measurements in summer and winter using backward trajectories calculated by the HYSPLIT model are shown in Fig. 8a and b. Figure 8c and d shows the averaged OMI satellite observations of NO₂ during summer (June, July and August) and winter (December, January and February) from June 2013 to February 2017. The OMI measurements show that Nanjing has a serious NO₂ pollution problem in winter.

Figure 12(a) Time series of wind speed (blue curve) and wind direction (red curve). (b) Time series of daily minimum (blue curve), daily average (green curve) and daily maximum temperatures. The gray shadowed area indicates the Youth Olympic Games period (16 to 28 August 2014).

Download

Figure 13Backward trajectories calculated for (a) pre-Olympic, (b) Olympic and (c) post-Olympic periods. Backward trajectories are calculated over 12 h.

The reconstructed NO₂ fields show a good spatial agreement with OMI satellite observations retrieved by NASA. However, the reconstructed NO₂ fields are on average 3 times higher than the OMI observations. Large discrepancies of the absolute values between the reconstructed data and OMI observations are expected, as the differences are also observed in the direct comparison of VCDs measured by the MAX-DOAS and OMI as presented in Sect. 3.2. Elevated NO₂ levels are observed during winter when air mass coming from the east, e.g., Shanghai, Suzhou and Nantong, where OMI observations also show high NO₂ levels. The summertime NO₂ levels are in general lower. However, elevated values can still be observed along the Yangtze River and over some cities, e.g., Nantong and Wuhu.

We have also applied the same approach to the MAX-DOAS HCHO measurements with HCHO lifetime of 1 and 2 h for summer and winter measurements, respectively. Due to the considerable contribution of secondary formation of HCHO from its precursors and short atmospheric lifetime of HCHO, relatively short lifetimes are chosen to reconstruct the spatial distribution of HCHO. The reconstructed maps would be more related to the spatial distribution of precursors of HCHO instead of HCHO itself if a longer lifetime is applied in the analysis. HCHO fields reconstructed from the MAX-DOAS measurements are shown in Fig. 9a and b. OMI observations of HCHO are also shown in Fig. 9c and d for reference. Summertime HCHO data are in general higher than the observations during winter. However, the spatial distribution of the back-propagated HCHO data does not show a strong correlation with the OMI satellite observations as the NO₂ data. This is probably due to that fact that HCHO is mostly secondarily produced in the atmosphere. In addition, the atmospheric lifetime of HCHO is much shorter compared to NO₂ and therefore more dependent on the local productions. The result suggests that the MAX-DOAS measurements are sensitive to the regional transport of air pollutants. Despite the strong local contributions, the air quality of Nanjing can also be significantly influenced by the air pollution transportation, especially during winter.

Figure 14Boxplots of (a) aerosol optical depths, (b) NO₂ and (c) HCHO VCDs before, during and after the Youth Olympic Games.

Download

3.5 Assessments of emission reduction during the Youth Olympic Games

The summer Youth Olympic Games event was held from 16 to 28 August 2014 in Nanjing, China. During the event, the government implemented a series of air pollution control measures. Heavy emission vehicles were strictly prohibited. Local heavy industries, e.g., petrochemical and steel industries, were required to limit their production during the Youth Olympic Games. These air pollution control measures were often implemented when such international events were held in China (Wu et al., 2013; Wang et al., 2014; Chan et al., 2015; Liu et al., 2016; Sun et al., 2016). In order to study the influences of the emission control measures implemented during the Youth Olympic Games on the local air quality, we compare the MAX-DOAS observations of aerosol, NO₂ and HCHO taken before, during and after the Youth Olympic Games. We define the pre-Olympic period from a month before the Youth Olympic Games (16 July 2014) to 1 d before the Youth Olympic Games (15 August 2014). During the pre-Olympic period, the government gradually started to implement some of the emission control measures. During the Youth Olympic Games all those air pollution control measures were strictly implemented. The post-Olympic period is defined from 29 August to 28 September 2014 where the emission control was gradually relaxed back to normal.

Time series of aerosol optical depths, HCHO and NO₂ VCDs and surface mixing ratios measured by the MAX-DOAS around the Youth Olympic Games are shown in Fig. . The gray shadowed area indicates the Youth Olympic Games period (16 to 28 August 2014). Meteorological parameters such as temperature, wind speed and wind direction, measured by a weather station in Nanjing during the same period, are shown in Fig. 11. Backward trajectories over the three periods are shown in Fig. 12. The backward trajectories show that there are slightly less air masses coming from the southwest direction during the pre-Olympic period, while the Olympic and post-Olympic periods show very similar air mass transportation conditions. The AODs, HCHO and NO₂ measurements show a strong temporal variability. In order to investigate the effect of the Youth Olympic Games, we analyzed the statistics of AODs, HCHO and NO₂ VCDs for the pre-Olympic, Olympic and post-Olympic periods. Boxplots of the AODs, HCHO and NO₂ VCDs for the three periods are shown in Fig. 13. The results show that AOD, NO₂ and HCHO VCDs are the lowest during the Olympic period among those three periods. The MAX-DOAS measurements of AOD reduced from 0.9 during the pre-Olympic period to 0.6 for the Olympic period and bounced back to about 0.8 after the Youth Olympic Games. A reduction of NO₂ columns can also be observed during the Youth Olympic Games. Averaged NO₂ VCDs are 14, 11 and 19 × 10¹⁵ molec cm⁻² for periods before, during and after the Youth Olympic Games, respectively. A similar reduction pattern is also observed from the NO₂ surface mixing ratios. Averaged NO₂ surface mixing ratios are 7.5, 4.7 and 9.0 ppbv for periods before, during and after the Youth Olympic Games, respectively. The HCHO columns measured during the Youth Olympic Games also decreased by more than ∼50 % from 17 × 10¹⁵ molec cm⁻² before the Olympic down to 8 × 10¹⁵ molec cm⁻² during the Youth Olympic Games. Surface HCHO mixing ratios also show similar reduction. The surface HCHO mixing ratios measured also reduced from 7.0 ppbv before the Youth Olympic to 3.3 ppbv during the Youth Olympic. The emission and secondary formation of HCHO are closely related to the ambient temperature. Higher ambient temperature is usually associated with higher formation rate and natural emission of HCHO and its precursors. The meteorological data indicate that the ambient temperature is decreasing with time when the season changes from summer to autumn. The HCHO concentrations measured during the Youth Olympic Games are expected to be higher than those measured in the post-Olympic period if anthropogenic emissions are unchanged. However, the measurement shows that the HCHO concentration is slightly lower during the Youth Olympic compared to the post-Olympic period. The result implies that the emissions of HCHO (and its precursors) are reduced during the Youth Olympic Games. As the other meteorological conditions are very similar during the three periods, the decrease in pollutant concentrations are mainly attributed to the reduction of emissions during the Youth Olympic Games.