CALIOP is a dual-wavelength polarization lidar measuring attenuated backscatter radiation at 532 and 1064 nm since June 2006. The vertical resolution of aerosol extinction profiles is 30 m below 8.2 km and 60 m up to 20.2 km (Winker et al., 2013), with a total of 399 sampled altitudes. The horizontal resolution of CALIOP scenes is 335 m along the orbital track and is given over a 5 km horizontal resolution in level 2 data.

As detailed in Appendix B, we use the daily all-sky version 3 CALIOP level 2 aerosol profile product (https://search.earthdata.nasa.gov/search?q=CALIOP aerosol&ok=CALIOP, last access: April 2017) aerosol at 532 nm from 2007 to 2015 to construct a monthly level 3 climatological dataset of aerosol extinction profiles over China and nearby regions. This dataset is constructed on the GEOS-Chem model grid (0.667∘ long × 0.5∘ lat) and vertical resolution (47 layers, with 36 layers or so in the troposphere). The ratio of climatological monthly CALIOP to monthly GEOS-Chem profiles represents the scaling profile to adjust the daily GEOS-Chem profiles in the same month (see Sect. 2.2)

The NO2 retrieval consists of three steps. First, the total NO2 SCD is retrieved using the differential optical absorption spectroscopy (DOAS) technique (for the 405–465 nm spectral window in the case of OMI). The uncertainty of the SCD is determined by the appropriateness of the fitting technique, the instrument noise, the choice of fitting window, and the orthogonality of the absorbers' cross sections (Bucsela et al., 2006; Lerot et al., 2010; Richter et al., 2011; van Geffen et al., 2015; Zara et al., 2018). The NO2 SCD in DOMINO v2 has a bias at about 0.5–1.3 ×1015 molec. cm-2 (Dirksen et al., 2011; Belmonte Rivas et al., 2014; Marchenko et al., 2015; van Geffen et al., 2015; Zara et al., 2018), which can be reduced by improving wavelength calibration and including O2–O2 and liquid water absorption in the fitting model (van Geffen et al., 2015; Zara et al., 2018). The tropospheric SCD is then obtained by subtracting the stratospheric SCD from the total SCD. The bias in the total SCD is mostly absorbed by this stratospheric separation step, which may not propagate into the tropospheric SCD (van Geffen et al., 2015). The last step converts the tropospheric SCD to VCD by using the tropospheric AMF (VCD = SCD/AMF). The tropospheric AMF is calculated at 438 nm by using look-up tables (in most retrieval algorithms) or online radiative transfer modeling (in POMINO) driven by ancillary parameters, which act as the dominant source of errors in retrieved NO2 VCD data over polluted areas (Boersma et al., 2007; Lin et al., 2014b, 2015; Lorente et al., 2017).

Our POMINO algorithm focuses on the tropospheric AMF calculation over China and nearby regions, taking the tropospheric SCD (Dirksen et al., 2011) from DOMINO v2 (Boersma et al., 2011). POMINO improves upon the DOMINO v2 algorithm in the treatment of aerosols, surface reflectance, online radiative transfer calculations, spatial resolution of NO2, temperature and pressure vertical profiles, and consistency between cloud and NO2 retrievals (Lin et al., 2014b, 2015). In brief, we use the parallelized LIDORT-driven AMFv6 package to derive both cloud parameters and tropospheric NO2 AMFs for individual OMI pixels online (rather than using a look-up table). NO2 vertical profiles, aerosol optical properties, and aerosol vertical profiles are taken from the nested GEOS-Chem model over Asia (0.667∘ long × 0.5∘ lat before May 2013 and 0.3125∘ long × 0.25∘ lat afterwards), and pressure and temperature profiles are taken from the GEOS-5- and GEOS-FP-assimilated meteorological fields that drive GEOS-Chem simulations. Model aerosols are further adjusted by satellite data (see below). We adjust the pressure profiles based on the difference in elevation between the pixel center and the matching model grid cell (Zhou et al., 2010). We also account for the effects of surface bidirectional reflectance distribution function (BRDF) (Zhou et al., 2010; Lin et al., 2014b) by taking three kernel parameters (isotropic, volumetric, and geometric) from the MODIS MCD43C2 dataset (https://search.earthdata.nasa.gov/search?q=MODIS MCD43C2&ok=MODIS 20MCD43C2, last access: December 2015) at 440 nm (Lucht et al., 2000).

As a prerequisite to the POMINO NO2 retrieval, clouds are retrieved through the O2–O2 algorithm (Acarreta et al., 2004; Stammes et al., 2008) with O2–O2 SCDs from OMCLDO2, and with pressure, temperature, surface reflectance, aerosols, and other ancillary information consistent with the NO2 retrieval. Note that the treatment of cloud scattering (as an “effective” Lambertian reflector, as in other NO2 algorithms) is different from the treatment of aerosol scattering and absorption (vertically resolved based on the Mie scheme).

POMINO uses the temporally and spatially varying aerosol information, including AOD, SSA, phase function, and vertical profiles from GEOS-Chem simulations. POMINO v1.1 (this work) further uses CALIOP data to constrain the shape of the aerosol vertical extinction profile. We run the model at a resolution of 0.3125∘ long × 0.25∘ lat before May 2013 and 0.667∘ long × 0.5∘ lat afterwards, as determined by the resolution of the driving meteorological fields. We then regrid the finer-resolution model results to 0.667∘ long × 0.5∘ lat, to be consistent with the CALIOP data grid. We then sample the model data at times and locations with valid CALIOP data at 532 nm to establish the model monthly climatology.

For any month in a grid cell, we divide the CALIOP monthly climatology of aerosol extinction profile shape by model climatological profile shape to obtain a unitless scaling profile (Eq. 1) and apply this scaling profile to all days of that month in all years (Eq. 2). Such a climatological adjustment is based on the assumption that systematic model limitations are month dependent and persist over the years and days (e.g., a too strong vertical gradient; see Sect. 3.3). Although this monthly adjustment means discontinuity on the day-to-day basis (e.g., from the last day of a month to the first day of the next month), such discontinuity does not significantly affect the NO2 retrieval, based on our sensitivity test.

In Eqs. (1) and (2), EC represents the CALIOP climatological aerosol extinction coefficient, EG the GEOS-Chem extinction, EGr the post-scaling model extinction, and R the scaling profile. The subscript i denotes a grid cell, k a vertical layer, d a day, m a month, and y a year. Note that in Eq. (1), the extinction coefficient at each layer is normalized relative to the maximum value of that profile. This procedure ensures that the scaling is based on the relative shape of the extinction profile and is thus independent of the accuracies of CALIOP and GEOS-Chem AOD. We keep the absolute AOD value of GEOS-Chem unchanged in this step. Ri,k,m=Ei,k,mC/max(Ei,k,mC)Ei,k,mG/max(Ei,k,mG)Ei,k,d,m,yGr=Ei,k,d,m,yG×Ri,k,m In POMINO, the GEOS-Chem AOD values are further constrained by a MODIS/Aqua Collection 5.1 monthly AOD dataset (https://search.earthdata.nasa.gov/search?q=MODIS AOD&ok=MODIS AOD, last access: December 2016) compiled on the model grid (Lin et al., 2014b, 2015). POMINO v1.1 uses the Collection 5.1 AOD data before May 2013 and Collection 6 data afterwards. For adjustment, model AODs are projected to a 0.667∘ long × 0.5∘ lat grid and then sampled at times and locations with valid MODIS data (Lin et al., 2015). As shown in Eq. (3), τM denotes MODIS AOD, τG GEOS-Chem AOD, and τMr post-adjustment model AOD. The subscript i denotes a grid cell, d a day, m a month, and y a year. This AOD adjustment ensures that in any month, monthly mean GEOS-Chem AOD is the same as MODIS AOD while the modeled day-to-day variability is kept. τi,d,m,yGr=τi,m,yMτi,m,yG×τi,d,m,yG Equations (4–5) show the complex effects of aerosols in calculating the AMF for any pixel. The AMF is the linear sum of tropospheric layer contributions to the slant column weighted by the vertical sub-columns (Eq. 4). The box AMF, amfk, describes the sensitivity of NO2 SCD to layer k, and xa,k represent the sub-column of layer k from the a priori NO2 profile. The variable l represents the first integrated layer, which is the layer above the ground for clear sky, or the layer above cloud top for cloudy sky. The variable t represents the tropopause layer. POMINO assumes the independent pixel approximation (IPA) (Boersma et al., 2002; Martin, 2002). This means that the calculated AMF for any pixel consists of a fully cloudy-sky portion (AMFclr) and a fully clear-sky portion (AMFcld), with weights based on the cloud radiance fraction (CRF =CF⋅Icld1-CF⋅Iclr+CF⋅Icld, where Iclr and Icld are radiance from the clear-sky part and fully cloudy part of the pixel, respectively) (Eq. 5). AMFcld is affected by above-cloud aerosols, and AMFclr is affected by aerosols in the entire column. Also, aerosols affect the retrieval of CRF. Thus, the improvement of aerosol vertical profile in POMINO v1.1 affects all three quantities in Eq. (5) and thus leads to complex impacts on retrieved NO2 VCD. AMF=∑ltamfkxa,k∑ltxa,kAMF=AMFcld⋅CRF+AMFclr⋅1-CRF

We exclude OMI pixels affected by row anomaly (Schenkeveld et al., 2017) or with high albedo caused by icy/snowy ground. To screen out cloudy scenes, we choose pixels with a CRF below 50 % (effective cloud fraction is typically below 20 %) in POMINO.

The selection of CRF threshold influences the validity of pixels. The effective CRF in DOMINO implicitly includes the influence of aerosols. In POMINO, the aerosol contribution is separated from that of the clouds, resulting in a lower CRF than for DOMINO. The CRF differs insignificantly between POMINO and POMINO v1.1 because the same AOD and other non-aerosol ancillary parameters are used in the retrieval process. Using the CRF from POMINO instead of DOMINO or QA4ECV for cloud screening means that the number of valid pixels in DOMINO increases by about 25 %, particularly because many more pixels with high pollutant (aerosol and NO2) loadings are now included. This potentially reduces the sampling bias (Lin et al., 2014b, 2015), and the ensemble of pixels now includes scenes with high “aerosol radiative fractions”. Further research is needed to fully understand how much these high-aerosol scenes may be subject to the same screening issues as the cloudy scenes. Nevertheless, the limited evidence here and in Lin et al. (2014b, 2015) suggests that including these high-aerosol scenes does not affect the accuracy of NO2 retrieval.

We use MAX-DOAS measurements at three suburban or urban sites in East China, including one urban site at the Institute of Atmospheric Physics (IAP) in Beijing (116.38∘ E, 39.38∘ N), one suburban site in Xianghe County (116.96∘ E, 39.75∘ N) to the south of Beijing, and one urban site in Wuxi City (120.31∘ E, 31.57∘ N) in the Yangtze River Delta (YRD). Figure 1 shows the locations of these sites overlaid with POMINO v1.1 NO2 VCDs in August 2012. Table 1 summarizes the information of MAX-DOAS measurements.

The instruments in IAP and in Xianghe were designed at BIRA-IASB (Clémer et al., 2010). Such an instrument is a dual-channel system composed of two thermally regulated grating spectrometers, covering the ultraviolet (300–390 nm) and visible (400–720 nm) wavelengths. It measures scattered sunlight every 15 min at nine elevation angles: 2, 4, 6, 8, 10, 12, 15, 30, and 90∘. The telescope of the instrument is pointed to the north. The data are analyzed following Hendrick et al. (2014). The Xianghe suburban site is influenced by pollution from the surrounding major cities like Beijing and Tianjin. At Xianghe, MAX-DOAS data have been continuously available since early 2011, and data in 2012 are used here for comparison with OMI products. At IAP, MAX-DOAS data are available in 2008 and 2009 (Table 1); thus for comparison purposes we process OMI products to match the MAX-DOAS times.

Located on the roof of an 11-story building, the instrument at Wuxi was developed by the Anhui Institute of Optics and Fine Mechanics (AIOFM) (Wang et al., 2015, 2017a). Its telescope is pointed to the north and records at five elevation angles (5, 10, 20, 30, and 90∘). Wuxi is a typical urban site affected by heavy NOx and aerosol pollution. The measurements used here are analyzed in Wang et al. (2017a). Data are available in 2012 for comparison with OMI products.

When comparing the four OMI products against MAX-DOAS observations, temporal and spatial inconsistency in sampling is inevitable. The spatial inconsistency, together with the substantial horizontal inhomogeneity in NO2, might be more important than the influence of temporal inconsistency (Wang et al., 2017b). The influence of the horizontal inhomogeneity was suggested to be about 10 %–30 % for MAX-DOAS measurements in Beijing (Ma et al., 2013; Lin et al., 2014b) and 10 %–15 % for less polluted locations like Tai'an, Mangshan, and Rudong (Irie et al., 2012). Following previous studies, we average MAX-DOAS data within 1 h of the OMI overpass time, and we select OMI pixels within 25 km of a MAX-DOAS site whose viewing zenith angle is below 30∘. To exclude local pollution events near the MAX-DOAS site (such as the abrupt increase in NO2 caused by the pass of consequent vehicles during a very short period), the standard deviation of MAX-DOAS data within 1 h should not exceed 20 % of their mean value (Lin et al., 2014b). We elect not to spatially average the OMI pixels because they can reflect the spatial variability in NO2 and aerosols.

We further exclude MAX-DOAS data in cloudy conditions, as clouds can cause large uncertainties in MAX-DOAS and OMI data. To find the actual cloudy days, we use MODIS/Aqua cloud fraction data, MODIS/Aqua level 3 corrected reflectance (true color) data at 1∘ ×  1∘ resolution, and current weather data observed from the nearest ground meteorological station (indicated by the black triangles in Fig. 1b). Since there is only one meteorological station available near the Beijing area, it is used for both IAP and Xianghe MAX-DOAS sites. We first use MODIS/Aqua corrected reflectance (true color) to distinguish clouds from haze. For cloudy days determined by the reflectance checking, we examine both the MODIS/Aqua cloud fraction data and the meteorological station cloud records, considering that MODIS/Aqua cloud fraction data may be missing or have a too coarse of a horizontal resolution to accurately interpret the cloud conditions at the MAX-DOAS site. We exclude MAX-DOAS NO2 data if the MODIS/Aqua cloud fraction is larger than 60 % and the meteorological station reports a “broken” (cloud fraction ranges from five-eighths to seven-eighths) or “overcast” (full cloud cover) sky. For the three MAX-DOAS sites together, this leads to 49 days with valid data out of 64 days with pre-screening data.

We note here that using cloud fraction data from MODIS/Aqua or MAX-DOAS (for Xianghe only, see Gielen et al., 2014) alone to screen cloudy scenes may not be appropriate on heavy-haze days. For example, on 8 January 2012, MODIS/Aqua cloud fraction is about 70 %–80 % over the North China Plain and MAX-DOAS at Xianghe suggests the presence of thick clouds. However, both the meteorological station and MODIS/Aqua corrected reflectance (true color) products suggest that the North China Plain was covered by a thick layer of haze. Consequently, this day was excluded from the analysis.

The aerosol layer height (ALH) is a good indicator to what extent aerosols are mixed vertically (Castellanos et al., 2015). As defined in Eq. A1 in Appendix B, the ALH is the average height of aerosols weighted by vertically resolved aerosol extinction. Figure 2a shows the spatial distribution of our CALIOP ALH climatology in each season. At most places, the ALH reaches a maximum in spring or summer and a minimum in fall or winter. The lowest ALH in fall and winter can be attributed to heavy near-surface pollution and weak vertical transport. The high values in summer are related to strong convective activities. Over the north, the high values in spring are partly associated with Asian dust events, due to high surface winds and dry soil in this season (Huang et al., 2010; Wang et al., 2010; Proestakis et al., 2018), which also affects the oceanic regions via atmospheric transport. The springtime high ALH over the south may be related to the transport of carbonaceous aerosols from Southeast Asian biomass burning (Jethva et al., 2016). Averaged over the domain, the seasonal mean ALHs are 1.48, 1.43, 1.27, and 1.18 km in spring, summer, fall, and winter.

Figure 3a, b further show the climatological monthly variations in ALH averaged over northern East China (the anthropogenic source region shown in orange in Fig. 1a) and northwest China (the dust source region shown in yellow in Fig. 1a). The two regions exhibit distinctive temporal variations. Over northern East China, the ALH reaches a maximum in April (∼1.53 km) and a minimum in December (∼1.14 km). Over northwest China, the ALH peaks in August (∼1.59 km) because of the strongest convection (Zhu et al., 2013), although the springtime ALH is also high.

Figure 4a shows the climatological seasonal regional average vertical profiles of aerosol extinction over northern East China. Here, the aerosol extinction increases from the ground level to a peak at about 300–600 m (season dependent), above which it decreases gradually. The height of peak extinction is lowest in winter, consistent with a stagnant atmosphere, thin mixing layer, and increased emissions (from residential and industrial sectors). The large error bars (horizontal lines in different layers, standing for 1 standard deviation) indicate strong spatiotemporal variability in aerosol extinction.

Over northwest China (Fig. 5a), the column total aerosol extinction is much smaller than that over northern East China (Fig. 4a), due to lower anthropogenic sources and dominant natural dust emissions. Vertically, the decline of extinction from the peak-extinction height to 2 km is also much more gradual than the decline over northern East China, indicating stronger lifting of surface emitted aerosols. In winter, the column total aerosol extinction is close to the high value in dusty spring, whereas the vertical gradient of extinction is strongest among the seasons. This reflects the high anthropogenic emissions in parts of northwest China, which have been rapidly increasing in the 2000s due to relatively weak emission control supplemented by growing activities of relocation of polluted industries from the eastern coastal regions (Zhao et al., 2015; Cui et al., 2016).

Overall, the spatial and seasonal variations in CALIOP aerosol vertical profiles are consistent with changes in meteorological conditions, anthropogenic sources, and natural emissions. The data will be used to evaluate and adjust GEOS-Chem simulation results in Sect. 3.2. A comparison of our CALIOP dataset with NASA's official level 3 data is presented in Appendix C.

Figure 2b shows the spatial distribution of seasonal ALHs simulated by GEOS-Chem. The model captures the spatial and seasonal variations in CALIOP ALH (Fig. 2a) to some degree, with an underestimate by about 0.3 km on average. The spatial correlation between CALIOP (Fig. 2a) and GEOS-Chem (Fig. 2b) ALH is 0.37 in spring, 0.57 in summer, 0.40 in fall, and 0.44 in winter. The spatiotemporal consistency and underestimate are also clear from the regional mean monthly ALH data in Fig. 3 – the temporal correlation between GEOS-Chem and CALIOP ALH is 0.90 in northern East China and 0.97 in northwest China.

Figures 4a and 5a show the GEOS-Chem-simulated 2007–2015 monthly climatological vertical profiles of aerosol extinction coefficient over northern East China and northwest China, respectively. Over northern East China (Fig. 4a), the model (red line) captures the vertical distribution of CALIOP extinction (black line) below the height of 1 km, despite a slight underestimate in the magnitude of extinction and an overestimate in the peak-extinction height. From 1 to 5 km above the ground, the model substantially overestimates the rate of decline in extinction coefficient with increasing altitude. Across the seasons, GEOS-Chem underestimates the magnitude of aerosol extinction by up to 37 % (depending on the height). Over northwest China (Fig. 5a), GEOS-Chem has an underestimate in all seasons, with the largest bias by about 80 % in winter likely due to underestimated water-soluble aerosols and dust emissions (J. Wang et al., 2008; Li et al., 2016).

Since the POMINO v1.1 algorithm uses MODIS AOD to adjust model AOD, it only uses the CALIOP aerosol extinction profile shape to adjust the modeled shape (Eqs. 1 and 2). Figures 4a and 5b show the vertical shapes of aerosol extinction, averaged across all profiles in each season over northern East China and northwest China, respectively. Over northern East China (Fig. 4b), GEOS-Chem underestimates the CALIOP values above 1 km by 52 %–71 %. This underestimate leads to a lower ALH, consistent with the finding by van Donkelaar et al. (2013) and Lin et al. (2014b). Over northwest China (Fig. 5b), the model also underestimates the CALIOP values above 1 km by 50 %–62 %. These results imply the importance of correcting the modeled aerosol vertical shape prior to cloud and NO2 retrievals.