2.1 Site description

Here, we report the measurements and satellite observations from a suburban site, Mohali, located in the north-west Indo-Gangetic Plain. Figure 1 shows the location of Mohali, major cities, and a terrain map around North India. We also show the spatial distribution of mean NO₂ tropospheric vertical column densities probed by the TROPOMI satellite (van Geffen et al., 2017; Veefkind et al., 2012) for the period December 2017–October 2018 in a 25 km × 25 km box around Mohali. The Himalayan mountain range starts at ∼35 km in the north. The in situ and ground-based remote-sensing measurements were performed at the IISER (Sinha et al., 2014) Mohali atmospheric chemistry facility (30.667^∘ N, 76.739^∘ E; 310 m a.m.s.l.). A detailed description of the site including the seasonal characteristic of meteorological parameters, characteristic wind sectors, and major emissions can be found elsewhere (Kumar et al., 2016; Pawar et al., 2015; Sinha et al., 2014). The urban city Chandigarh is located in the wind sector spanning from north to east, while the west and south are comprised mostly of agricultural lands and small cities or towns. Two major power plants are located within 50 km of the measurement site. At ∼45 km to the north-west (∼342^∘) is the 1260 MW Guru Gobind Singh super thermal power plant (PP1), Rupnagar, which was operational with 90 % of its capacity until 2014. Since February 2014, a 1400 MW power plant (PP2) has been functional in Rajpura, which is ∼18 km south-west (∼230^∘) of the measurement site.

Figure 1Left: terrain map showing the location of Mohali (red triangle) in North India, along with major cities (crosses) and two major thermal power plants near Mohali (blue circles: PP1 – Guru Gobind Singh super thermal power plant; PP2 – Larsen & Toubro super thermal power plant (NPL), Rajpura) and province boundaries. Right: mean tropospheric NO₂ VCD measured by TROPOMI overlaid on the terrain map of a 0.5^∘ × 0.5^∘ box around Mohali (shown in the left panel) for the period December 2017–October 2018. The black arrow indicates the viewing direction of the MAX-DOAS instrument. Terrain maps are adapted from Stamen (http://maps.stamen.com/terrain, last access: 17 November 2020).

For the climate of India, the year can be divided into the following four seasons: winter (November to February), summer (March–June), monsoon (July–August), and post-monsoon (September–October). Large-scale crop residue burning events occur in the late summer and late post-monsoon, which strongly perturb the atmospheric chemistry and composition (Kumar et al., 2018; Sarkar et al., 2013). In order to account for these perturbations, Kumar et al. (2016) have recommended a further classification of summer and post-monsoon months into clean and polluted periods. The primary fetch region of the air masses is north-west throughout the year except in the monsoon season, when the wind direction is primarily south-east. Figure F1 shows the wind rose plots indicating the wind speed and wind direction frequencies around Mohali in the four major seasons over the measurement period. Over the years 2012–2017, rapid urbanization has happened in Mohali and nearby regions (e.g. the commissioning of new international airport terminal and highways, extended construction activities for residential (e.g. Aero city, ECO city) and institutional (Knowledge City, Medicity) purposes). According to the census of 2011, the district of Mohali holds the top rank in urban population growth among all districts in the state of Punjab at a rate of 90.2 % for the period between the years 2001 and 2011 (Tripathi and Mahey, 2017).

2.2 MAX-DOAS measurement setup and spectral analysis

A MAX-DOAS instrument (Hoffmann Messtechnik GmbH) was installed at ca. 20 m above ground level with an azimuth viewing direction of 10^∘ anticlockwise from the north. The instrument primarily consists of a Czerny Turner spectrometer (Ocean Optics USB 2000+), an optical assembly consisting of a quartz lens that collects the scattered sunlight, quartz optical fibre that transmits the light to the spectrometer, and electronics in a sealed metal box. The box is mounted on a stepper motor which can be programmed to set the elevation viewing angle of the instrument. The spectral resolution of the spectrometer was ∼0.7 nm in the spectral range of 318–465 nm with a field of view (FOV) of 0.7^∘. In order to avoid light outside the telescope's FOV being scattered onto the fibre, a black tube (ca. 6 cm long) was mounted in front of the lens. The scattered sunlight spectra were recorded for elevation viewing angles 1, 2, 4, 6, 8, 10, 15, 30, and 90^∘ at a total integration time (number of scans × acquisition time for one scan) of 60 s each. Since the complete MAX-DOAS instrument was mounted outside, it was important to adjust the detector temperature so that the following two conditions are met:

1. The detector temperature is lower than the ambient temperature.

2. The difference between the ambient temperature and detector temperature is not more than 20^∘ C. This ensures that the workload on the Peltier cooler is manageable and the detector temperature is stable.

Hence, depending on the seasons, the detector temperature was adjusted. Figure F2 shows the nominal detector temperatures (T_set) and actual detector temperature (T_cold) for the various periods during the measurement period. The dark current and offset spectra were recorded every night, and while performing the spectral analysis, these were subtracted from the measured spectra recorded at a similar detector temperature. Additional offset was corrected from the measured spectrum accounting for the mean intensity recorded by the dark pixels (pixel nos. 1–6 among the 2048 pixels) of the spectrometer. Wavelength-to-pixel calibrations were performed in the QDOAS software (http://uv-vis.aeronomie.be/software/QDOAS/, last access: 20 November 2020) (Danckaert et al., 2012) every time the detector temperature was changed, by matching the structures in a measured spectrum in the zenith direction at around noontime with those in a highly resolved solar spectrum. Horizon scans were performed every day at around 12:00 local time (LT). Figure F3 shows a typical variation in measured intensity and its derivatives at three different wavelengths over an elevation angle range from −3 to 3^∘. A steep increase was observed in the measured intensity, which was centred between 0 and 0.3^∘ for various wavelengths chosen for analysing the horizon scan. During the period of measurement, the horizon in the viewing direction was determined by a residential building with a height of about 40 m at a distance of 3 km. The viewing angle of this visible horizon would be about 0.38^∘ in good agreement with the results from the elevation scan. Thus we did not need to perform any correction for the true horizon in further analyses (Donner et al., 2020). We also see from Fig. F3 that the FOV of the instrument is rather large (>0.7^∘), and typically the root mean square (rms) of the spectral analysis for the measurements at 1^∘ elevation is substantially larger than those for the higher-elevation angles. Hence, we excluded the measurements at 1^∘ elevation angle from further analyses.

The measured spectra of the scattered sunlight were analysed for NO₂, HCHO, and the oxygen dimer (O₄) using the QDOAS software. Table 1 lists the wavelength intervals, included cross sections and other relevant details pertaining to the different retrievals, and Fig. 2 shows example DOAS fits and residuals for these retrievals. The typical values (peak of the frequency distribution) of the rms of the DOAS fit residuals are around 5 × 10⁻⁴, 7 × 10⁻⁴, 6 × 10⁻⁴, and 6 × 10⁻⁴ for O₄, NO₂ (UV), NO₂ (VIS), and HCHO, respectively. In order to retain analysis results corresponding to good-quality fits, we have excluded the O₄, NO₂, and HCHO differential slant column densities (dSCDs) corresponding to an rms greater than 2 × 10⁻³ and solar zenith angles higher than 85^∘ (Wang et al., 2019). The rms threshold removes 1.1 %, 1.4 %, 0.7 %, and 1.3 % of the O₄, NO₂ (UV), NO₂ (VIS) and HCHO dSCDs, respectively, of all the measured dSCDs at solar zenith angles less than 85^∘.

Table 1Spectral analysis settings and considered cross sections in QDOAS for the retrieval of dSCDs of NO₂ (in UV and Vis), HCHO, and O₄.

¹ Vandaele et al. (1998). ² Meller and Moortgat (2000). ³ Serdyuchenko et al. (2014). ⁴ Fleischmann et al. (2004). ⁵ Thalman and Volkamer (2013). ⁶ Rothman et al. (2010).

Download Print Version | Download XLSX

Figure 2Example DOAS fits and residuals for NO₂ in the visible (dSCD $=\mathrm{2.91}\times {\mathrm{10}}^{\mathrm{16}}\pm \mathrm{2.89}\times {\mathrm{10}}^{\mathrm{14}}$ molecules cm⁻²) and UV (dSCD $=\mathrm{2.09}\times {\mathrm{10}}^{\mathrm{16}}\pm \mathrm{8.17}\times {\mathrm{10}}^{\mathrm{14}}$ molecules cm⁻²), HCHO (dSCD $=\mathrm{7.23}\times {\mathrm{10}}^{\mathrm{16}}\pm \mathrm{4.32}\times {\mathrm{10}}^{\mathrm{15}}$ molecules cm⁻²), and O₄ (dSCD $=\mathrm{2.06}\times {\mathrm{10}}^{\mathrm{43}}\pm \mathrm{4.25}\times {\mathrm{10}}^{\mathrm{41}}$ molecules² cm⁻⁵) for a typical spectrum measured on 26 July 2015 at a solar zenith angle of 20^∘ and 2^∘ elevation angle.

Download

The spectral analysis is performed with respect to a Fraunhofer reference spectrum (FRS) measured in the zenith direction of each complete elevation angle measurement sequence in order to account for the Fraunhofer lines and the stratospheric contribution of the absorbers (Hönninger et al., 2004). For analysing the off-axis spectra measured at time “t”, we calculate the FRS at the time of the measurement by interpolating the zenith spectra measured before and after the complete measurement sequence. Thus, the primary retrieved quantity from MAX-DOAS spectral analysis is the so-called dSCD. dSCD of a trace gas (absorber) at an elevation angle α can be regarded as the difference between the absorber concentration integrated along the photon path at elevation angle α (SCD_α) and zenith direction (SCD₉₀).

dSCD is related to the tropospheric VCD through differential air mass factors (dAMF or AMF_α−AMF₉₀):

Unless specifically mentioned, we will use VCD to refer to tropospheric VCDs in the paper hereafter. The air mass factors are related to the light path of the photons reaching the telescope of the instrument and depend on several parameters, e.g. elevation angle, solar zenith angle, relative solar azimuth angle with respect to the instrument, surface reflectance, aerosol and trace gas vertical profile, and aerosol optical properties. In Sect. 2.4, we provide the details of the calculation of air mass factors and subsequent profile inversion to retrieve the VCDs and vertical profiles.

DOAS retrievals of NO₂ can be performed both in the UV and visible wavelength windows but have their respective advantages and limitations. NO₂ has a stronger absorption in the visible as compared to UV. Hence, the DOAS fit in the visible results in smaller fit uncertainties as compared to that in the UV (Fig. A1). This is an important aspect, especially for instruments with rather low quantum efficiencies like the detector of the MAX-DOAS instrument used in this study. However, the profile inversion methods (please refer to Sect. 2.4) to retrieve NO₂ vertical profiles and VCDs also require information about aerosol extinction profiles at the same wavelength as that used for NO₂ retrieval. The aerosol extinction profile retrieval for DOAS relies on the O₄ measurements, which have a rather weaker absorption at the visible wavelengths (in the spectral range of our instrument). Hence an alternative approach is to use an Ångström exponent to scale the aerosol extinction profiles derived at UV wavelengths to visible wavelengths. The aerosol extinction profiles calculated for the visible wavelength are subsequently used as an input parameter for the NO₂ profile inversion in the visible window. In Appendix A, we compare the performance and internal consistency of the NO₂ profiles and VCDs retrieved in the UV and visible by the profile inversion algorithm and by the geometric approximation under various sky conditions. Briefly, we found very good agreement between the NO₂ VCDs retrieved in the UV and visible under clear-sky conditions with a low aerosol load (slope=0.95 and r=0.9). Even in the clear-sky case with high aerosol load and cloudy-sky conditions, a reasonable agreement for NO₂ VCDs between the retrieval in UV and visible was observed (slope=0.75 and 0.78 for high aerosol and cloudy cases, respectively; r=0.82 for both). The NO₂ dSCDs from the retrieval in the UV wavelength window were found to be in good agreement with those retrieved in the visible, but they were systematically lower (r=0.9, slope=0.95 and a negative offset of 1.1 × 10¹⁴ molecules cm⁻²).

2.3 Cloud classification

Clouds have a strong impact on MAX-DOAS measurements and subsequent profile inversion as they alter the light path and intensity (Wagner et al., 2014, 2011). Clouds are generally not included within the radiative transfer models for profile inversion. The cloud classification scheme is based on the measured radiances at 360 nm, the colour index (ratio of measured radiances at 330 and 390 nm), and the measured O₄ air mass factors (O₄ slant column density (SCD) ∕ O₄ VCD) (for details see Wagner et al., 2016). Besides the absolute values of these quantities, their temporal variation and their elevation dependencies are also considered. The threshold for these quantities (the spread of O₄, the normalized CI, the spread of the CI, and the temporal variation in the CI) are parameterized as polynomials of the SZA as provided in Wagner et al. (2016). We can classify the sky conditions into the following seven categories: (1) clear sky with low aerosol load, (2) clear sky with high aerosol load (AOD >0.85 at 330 nm), (3) broken clouds, (4) cloud holes, (5) continuous clouds, (6) fog, and (7) optically thick clouds (Wagner et al., 2014, 2016). In the first step, using the colour index (CI), its variation across the zenith spectrum for an adjacent elevation sequence, and its variation within an elevation sequence, the primary cloud classification is performed to retrieve information about primary conditions (types 1–5). The identification of fog and thick clouds is performed in the second step using the O₄ air mass factor (AMF) and measured radiance at 360 nm. Fog is identified if there is very little variation in O₄ dSCD for different elevation angles within the measurement sequence. The details pertaining to the calculation of thresholds for radiance for the identification of thick clouds are provided in Appendix B. While the classification of aerosols might be slightly affected by the specific properties of the local aerosol, the cloud classification is robust to the variability of aerosol properties. However, this is not critical here because the main aim – the cloud classification – is hardly affected by these specific aerosol properties.

Figure 3 shows the percentage of the mean monthly sky condition for the complete measurement period. The most prominent sky condition in all the seasons is “clear sky with high aerosol load”, comprising ∼48 % of the total. March and April are marked by the maximum occurrence of clear-sky conditions with low aerosol load at ∼18 %. Continuous clouds and optically thick clouds are most abundant in July–August, which is marked by the monsoon season. Please note that due to the widespread crop residue burning and suppressed meteorological conditions, the months of October and November witness severe smog events in the north-west Indo-Gangetic Plain leading to very poor air quality and low visibility. The poor air quality conditions extend until December due to emission from domestic burning for heating and similar meteorological conditions. The prevalent high aerosol load conditions are also marked by the cloud classification algorithm as seen by occurrences greater than 55 % in October, November, and December. Fog is observed in December and January, when winter is at its peak.

Figure 3Relative frequencies of occurrence of various sky conditions in different months of the year over Mohali as derived from 4.5 years of MAX-DOAS observations. Note that the secondary cloud classifications of fog and optically thick clouds are not mutually independent and exclusive of the primary classification (clear sky, clear sky with high aerosol load, broken clouds, cloud holes, and continuous clouds). Hence, these are shown separately above the 100 % mark.

Download

2.4 Profile inversion to retrieve vertical profiles and vertical column densities from slant column densities

In order to account for the complex dependence of air mass factors on viewing geometry and measurement conditions, radiative transfer models (e.g. McARTIM; Deutschmann et al., 2011) are employed. dAMFs are calculated for various combinations (nodes) of viewing geometry and profiles (of trace gases and aerosol), which are stored offline as multi-dimensional lookup tables (LUTs) for various wavelengths (e.g. separate LUTs for 343, 360 and 430 nm). Profile inversion techniques use these LUTs to determine the scenarios which best match the measured dSCDs. From these scenarios, the aerosol and trace gas profiles, VCDs, and AOD are derived.

We have used MAPA (Mainz Profile algorithm) (Beirle et al., 2019) version 0.98 for this purpose. The vertical profiles of trace gas concentrations (or aerosol extinction) can be parameterized using three profile parameters namely column parameters (c) (VCD for trace gases and AOD for aerosol), height parameter (h), and shape parameter (s) (Beirle et al., 2019; Wagner et al., 2011). In the first step, the aerosol profiles are retrieved using the measured O₄ dSCDs. A Monte Carlo approach is utilized to identify the best ensemble of the forward model parameters (h, s and c) which fit the measured O₄ dSCDs for the sequence of elevation angles. Generally, a scaling factor (0.8 in most of the cases) is applied to the measured O₄ dSCDs before they are used in the profile retrieval (see Wagner et al., 2019, and references therein). The reason for this scaling factor is still not understood, and in Sect. 3.3, we also investigate the effect of the different scaling factors on the intercomparison of the retrieved AOD with satellite observation from MODIS. In the second step, the aerosol profiles retrieved from the O₄ inversion are used as input to retrieve similar model parameters (h, s, and c) for the trace gases (e.g. NO₂ and HCHO). To assess the quality of the retrievals, MAPA also provides “valid”, “warning”, or “error” flags for each measurement sequence, which are calculated based on pre-defined thresholds for various fit parameters (Beirle et al., 2019). We have used the lookup tables calculated at 360 nm for the inversion of O₄ and NO₂ in the UV and 343 nm for HCHO and 430 nm for NO₂ in the visible window. For the HCHO profile inversion, we observed unrealistic h and s at high solar zenith angles (SZA >60^∘), which are probably related to spectral interferences with the ozone absorption within the DOAS analysis. Therefore, we only consider HCHO profile results for measurements with an SZA of less than 60^∘. For the retrieval of NO₂ in the visible wavelength window and HCHO, the aerosol extinction profiles retrieved at 360 nm were scaled to those at 430 and 343 nm using an Ångström exponent of 1.54. This value was derived as the mean of the Ångström exponent (AE) between 470 and 550 nm measured by MODIS for the measurement period, where we do not observe a strong intra-annual variation (Fig. F4). We have calculated the AE using the measured AOD at 470 and 550 nm according to

We also investigated the effect of the choice of Ångström exponent on the profile inversion for a smaller subset of our data spanning 15 d. We found that AE values of 1.25 and 1.75 (minimum 5th percentile and maximum 95th percentile in Fig. F4) resulted in the same retrievals and the difference in the mean NO₂ VCD was less than 0.1 %. The surface NO₂ concentrations were slightly higher (4 %) for an AE value of 1.25 and were 3 % lower for an AE value of 1.75 as compared to those for an AE value of 1.54.

For the profile inversion using MAPA, we compare the number of valid and those retrievals flagged as warning and error in Table 2. We note that the sky conditions associated with thick clouds and fog are mostly flagged as errors in the profile inversion. For the analysis results shown in the paper hereafter, we only retained the DOAS measurements corresponding to sky conditions without thick clouds and fog.

Table 2Frequency of MAPA retrievals flagged as “valid”, “warning”, or “error” for various species and different sky conditions.

^* The number of valid retrievals for NO₂ (UV) is calculated using relaxed rms flag criteria (Appendix A).

Download Print Version | Download XLSX

2.5 Satellite data

2.6 Ancillary measurements

In situ measurements of NO_x (NO and NO₂) were performed using a model 42i trace level analyser (Thermo Fischer Scientific) based on the chemiluminescence technique. Ambient NO₂ is first converted into NO using a heated molybdenum converter, which further reacts with excess ozone generated inside the instrument to produce a chemiluminescence signal proportional to the available NO. Checks for zero drifts were performed every week, and five point calibrations were performed every month. Meteorological parameters (e.g. ambient temperature, rainfall, wind speed, wind direction, and relative humidity) were measured using collocated met sensors (Met One Instruments Inc.). The details pertaining to the measurements' principle and calibration protocol can be found elsewhere (Sinha et al., 2014; Kumar et al., 2016).

In situ measurements of formaldehyde were performed at Mohali using a high-sensitivity proton transfer reaction mass spectrometer (PTR-MS) (Lindinger et al., 1998). Inside the PTR-MS, HCHO (proton affinity = 170.4 kcal mol⁻¹) is chemically ionized by hydronium ions (H₃O⁺) because of its higher proton affinity than that of water vapour (proton affinity = 165.2 kcal mol⁻¹), prior to its detection using a quadrupole mass analyser. Using a PTR-MS, HCHO is detected at a protonated mass-to-charge ratio (m∕z) of 31. The measured signal (counts s⁻¹) is converted to VMRs using the m∕z dependent sensitivity factors, which are usually determined using calibration experiments. Due to the degradation of the detector, the sensitivity of PTR-MS might also change with time, which is evaluated using routine calibration performed using a gas standard of known VMR (Chandra and Sinha, 2016; Sinha et al., 2014). However, such a calibration for HCHO could not be performed due to the unavailability of a calibration standard. Hence, we have calculated theoretical sensitivity factors for HCHO, similar to the method incorporated by Kumar et al. (2018). The major sources of uncertainty in the HCHO measurements, when a theoretical sensitivity factor is used, are from (1) uncertainty in the proton transfer reaction rate constant of the reaction between HCHO and H₃O⁺ (∼15 %) and (2) the ratio of transmission efficiencies of HCHOH⁺ and H₃O⁺ (∼25 %) (Zhao and Zhang, 2004; de Gouw and Warneke, 2007). Systematic uncertainties due to the degradation of the detector would increase with time. Ambient humidity is also known to interfere with the formaldehyde measurements with a PTR-MS, and we have performed an absolute humidity-based correction, according to Cui et al. (2016). HCHO VMRs increase by ∼30 % on average after the application of the absolute humidity-based correction.

3.1 Seasonal and annual trends of AOD, NO₂, and HCHO

Figure 4 shows the time series of monthly mean AOD, NO₂ VCD, and HCHO VCD for the complete measurement period from January 2013 until June 2017. The vertical error bars show the monthly variability as the interquartile range. The gaps in the time series are due to instrument malfunction (primarily due to the stepper motor and connection to the measurement computer). The mean AOD for the complete measurement period was 0.63 ± 0.51, with the monthly means varying between 0.30 in March 2013 and 1.25 in August 2014. We quantify the seasonal variability of the measured AOD to be 187 %, which is calculated as the difference between the maximum and minimum of the 30 d running mean divided by the mean over the measurement period. For the same months, very small inter-annual variability in the monthly means (<0.2) was observed for all the months except April–July. We observe the maximum AOD (0.8–1.2) during the monsoon months (June–August) at Mohali, which is most probably caused by the hygroscopic growth of the aerosol particles (Altaratz et al., 2013). Previous studies comparing various MODIS AOD data products and AERONET measurements over the Indo-Gangetic Plain have also found the maximum AOD in the monsoon months but also significant differences among the different data products (Mhawish et al., 2019; Tripathi et al., 2005). The retrieval of the aerosol size distribution from AERONET measurements has shown that during the monsoon, the coarse-mode fraction increases by more than 50 % of the annual average over the Indo-Gangetic Plain (Tripathi et al., 2005). In order to further confirm that the high AOD during the monsoon is not an artefact caused by the persistent cloud cover over the IGP, we have investigated the seasonality of the AOD under cloud-free conditions measured at different AERONET sites in the IGP nearest to Mohali (New Delhi ∼250 km south and Lahore ∼250 km west). Both at Lahore and New Delhi, high AOD values are observed in the monsoon months (June–August) (Fig. F5). Relatively high monthly mean AOD (0.6–0.9) values are also observed in May and October, which are characterized by crop residue fire emissions. We have also compared the MODIS AOD (converted to 440 nm using the AE derived from Eq. (3) to the AERONET AOD at these two stations and found very good agreement in the daily measured values with Pearson correlation coefficients (r) >0.84 (Fig. F5) and an overall bias <10 % for both sites.

Figure 4Monthly and annual variation in (a) AOD, (b) NO₂, and (c) HCHO vertical column densities derived from MAX-DOAS measurements. The lower and upper vertical error bars represent the 25th and 75th percentiles, respectively.

Download

The mean NO₂ VCD for the complete measurement period was (6.7 ± 4.1) × 10¹⁵ molecules cm⁻² with a variability in the monthly means between 4.7 × 10¹⁵ molecules cm⁻² in July 2015 and 8.9 × 10¹⁵ molecules cm⁻² in November 2013. The observed NO₂ VCDs are comparable to those observed in long-term measurements in rural and suburban environments (Kramer et al., 2008; Drosoglou et al., 2017) and satellite observations in the Indian metropolitan city Mumbai (Hilboll et al., 2013). These are much smaller than those observed in the urban areas worldwide (Mendolia et al., 2013; Drosoglou et al., 2017) or rural, suburban, and urban locations of China (Ma et al., 2013; Wang et al., 2017a; Chan et al., 2019; Jin et al., 2016; Vlemmix et al., 2015), where the mean monthly levels are generally higher than 1 × 10¹⁶ molecules cm⁻². The observed seasonal variability of 145 % in the 30 d running means can be explained by the seasonality of emissions and the changing lifetimes.

For NO_2, we see that the monsoon months (July and August) are cleanest followed by early pre-monsoon months (March–April). The primary emission sources active throughout the year in and around the region are vehicular emissions, garbage burning, and biomass burning for cooking and construction activities. In the suburban and rural regions around the measurement site, biomass (e.g. wood, domestic, and agricultural residue) and biofuel (coal) burning serve as the primary source of heating. Increased emissions from biomass burning for domestic heating in winter when the atmospheric lifetime of NO₂ is at a maximum, marks the highest observed NO₂ VCDs in the year. Also, during the crop residue burning active periods of summer (May–June) and post-monsoon (October–November), we observe enhanced NO₂ VCDs. Notwithstanding the strong urbanization trends, a significant annual trend is not observed in either of the three parameters (AOD, NO₂, and HCHO) measured by MAX-DOAS, indicating the dominance of non-organized emission sources, e.g. domestic heating, garbage burning, and crop residue burning for the ∼4.5-year measurement period. This is further ascertained by the fact that we do not see any noticeable weekday–weekend dependence in either of NO₂, HCHO, and AOD (Fig. F6).

The mean HCHO VCD for the complete measurement period was (12.1 ± 7.5) × 10¹⁵ molecules cm⁻², with a strong seasonality with monthly means ranging from 5.3 × 10¹⁵ molecules cm⁻² in March 2017 to 17.7 × 10¹⁵ molecules cm⁻² in October 2015. The seasonal variability was found to be strongest at 284 % for HCHO among the three parameters measured by MAX-DOAS. The mean and monthly variability is comparable to those observed in the urban areas of China (Vlemmix et al., 2015; Wang et al., 2017a). In contrast to these long-term measurements of formaldehyde in China, the minimum VCDs are not observed in the winter but rather observed in March. Globally photochemical production from biogenic and anthropogenic hydrocarbons dominates the formaldehyde sources while having a minor fraction being directly emitted from biomass burning and vegetation (Fortems-Cheiney et al., 2012). At complex suburban environments, e.g. Mohali, different sources can dominate the formaldehyde production in different periods of the year. At an urban site Kolkata in the IGP, the contribution of primary sources to ambient formaldehyde was observed to be 71 % and 32 % during summer and winter, respectively (Dutta et al., 2010).

At Mohali two distinct formaldehyde enhancement periods are observed; first in May–June and second in October, both of which are the periods when crop residue burning is practised around the region. Several identified and unidentified chemical compounds are formed in the atmosphere due to crop residue fires, which readily react with OH radicals and form second- and higher-generation oxidation products and potentially form formaldehyde as a by-product (Kumar et al., 2018; Sarkar et al., 2013). The pre-monsoon period of May and June provides favourable conditions (e.g. long daytime hours, uninterrupted solar radiation, and high temperature) for the photochemical production of secondary pollutants from the precursors emitted from wheat residue fires (Kumar et al., 2016; Sinha et al., 2014). This is reflected in very high formaldehyde VCDs observed in May and June in the range of 16.9 × 10¹⁵–17.4 × 10¹⁵ molecules cm⁻². Similar levels of HCHO VCDs are also observed in October, when paddy residue fires are active in the region. Also, the photochemical production of ozone is attributed to the emission from crop residue fires by Kumar et al. (2016). The observed maximum NO₂ VCDs in the winter months indicates very high primary emissions, which could eventually lead to formaldehyde production from its co-emitted precursors. However, low ambient temperatures and lack of ample sunlight hours result in HCHO VCDs that are not so high compared to May, June, and October.

The agricultural lands in the north-west fetch region of Mohali practice agroforestry, where poplar (most common) and eucalyptus trees are planted in the periphery of fields (Sinha et al., 2014; Pathak et al., 2014; Mishra et al., 2020). Over the Indo-Gangetic Plain, biogenic sources contribute to ∼40 % of the total VOC emission flux annually (Chaliyakunnel et al., 2019). In the early post-monsoon season, soil moisture availability and daytime temperatures between 30 and 35 ^∘C provide a favourable condition for isoprene emissions (Guenther et al., 1991). Generally, a strong variability is observed in the number of rainfall events and the total rainfall over the IGP between different years (Fukushima et al., 2019), which also affects the soil moisture availability and in turn the biogenic emissions from plants. For the period discussed in this study with available MAX-DOAS HCHO measurements, the years 2014 and 2015 were quite different with respect to the monsoon rainfall. During the monsoon months, 2014 witnessed 18 rainy days (total rainfall = 378 mm), while 2015 witnessed 32 rainy days (total rainfall = 435 mm) in Mohali. Following the number of rainfall events, the early post-monsoon months (August–September) of 2015 witnessed higher HCHO VCDs, which can be attributed to the photo-oxidation of stronger biogenic emissions of isoprene (Mishra and Sinha, 2020). Poplar trees in the Indian subcontinent show little emissions in the months from December to March (Singh et al., 2007) due to loss of leaves. Minimal biogenic and anthropogenic emissions of formaldehyde and its precursors in March resulted in the minimum VCDs during the year. From these observations, we conclude that anthropogenic emissions (primarily due to biomass burning) and their oxidation dominate the formaldehyde seasonality in most of the year except early post-monsoon where biogenic emissions have a major contribution to the measured formaldehyde.

3.2 Diurnal variation and vertical profiles of AOD, NO₂, and HCHO

Figure 5 shows the diurnal variation in the mean vertical profiles of aerosol extinction, NO₂ concentration and HCHO concentration for different months of the year retrieved using MAX-DOAS measurements at Mohali. The corresponding diurnal variations in the VMR of NO₂ and HCHO and aerosol extinction close to the surface are shown as Figs. F7–F9.

Figure 5Hourly mean vertical profiles of aerosol extinction (top three rows, shades of olive), NO₂ concentrations (middle three panels, shades of blue) and HCHO concentrations (bottom three panels, shades of red) in different months retrieved from 4.5 years of MAX-DOAS measurements over Mohali.

Download

The vertical profile of aerosol extinction is expected to be primarily driven by the boundary layer height (BLH) and to some extent, the photochemistry, which eventually drives secondary aerosol formation (Wang et al., 2019). At Mohali, the diurnal evolution of the aerosol extinction profile heights reaches its maximum during afternoon hours. In Fig. 6a, we show the typical diurnal evolution of BLH from the ERA5 reanalysis data for the four major seasons. We observe a growth of the BLH from morning until noon with a maximum at 14:00 LT and a subsequent decline. The maximum BLH up to 3 km is observed in summer. Shallow daytime BLH up to 1.2 km are observed in the monsoon period due to overcast sky conditions, stronger wind, and high surface moisture and in winter due to low surface temperature and low surface heat flux (Sathyanadh et al., 2017). We observe that the aerosol is trapped in the bottom layers (within 400 m) in winter, whereas during the afternoon hours in summer, monsoon, and early post-monsoon months, aerosol extinction up to 0.2 km⁻¹ is observed even at around 1.5 km altitude. Though the ERA5 BLH is shallow in monsoon, yet we observe similar aerosol profiles during that period as during summer, which indicates that at Mohali, the vertical distribution of aerosol does not follow ERA5 BLH transition from summer to monsoon. Over India, the monsoon months are characterized by strong convective activity, which can bring the surface air aloft to several kilometres despite a shallow ERA5 BLH (Lawrence and Lelieveld, 2010). The convection is rather strong in the Himalayan foothill region (which also includes Mohali) and even pumps the surface pollutants into the UTLS (upper troposphere–lower stratosphere) (Fadnavis et al., 2015). The evidence of pollutant transport associated with deep convection is crucial for PAN (peroxyacetyl nitrate) formation in the UTLS, which is observed by the modelling studies over the IGP and Himalayan region. Long-lived non-methane VOCs (e.g. ethane) can be transported to the UTLS where both convective NO_x transported from the surface and exchanged from the stratosphere serve as fuel for PAN formation. High aerosol extinction (>1.5 km⁻¹) is observed in the surface layer in the winter months, with the maximum in December. In the winter months, biomass burning contributes to primary aerosol, the formation of secondary aerosol and particle growth as a result of coating on existing aerosol particles. Moreover, the high ambient relative humidity in winter (Kumar et al., 2016) further contributes to the growth of the existing aerosol particles, which further increases the aerosol extinction and in the extreme cases can lead to intense fog.

Figure 6(a) Diurnal evolution of the hourly means ERA5 boundary layer height (BLH) at Mohali for the four major seasons of the year. (b) Mean afternoon time (between 12:00 and 15:00 LT) profile height (with 75 % of the total amount below) for aerosols, NO₂, and HCHO and the ERA5 BLH for different months. The upper and lower vertical error bars represent the monthly variability as 75th and 25th percentiles, respectively.

Download

In order to quantitatively describe the mixing altitude, we define a characteristic profile height H₇₅, as the height below which 75 % of the trace gas column (or AOD) is located (Vlemmix et al., 2015). The profile parameter h used in the MAPA inversion algorithm represents the height below which the concentration (or extinction) of the trace gas (or aerosol) remains constant. For elevated profiles (for details, please refer to Wagner et al., 2011), h refers to the height above the trace gas (or aerosol) layer where the concentration (or extinction) becomes zero. Using h and shape parameter s, we calculate a profile height (H₇₅) by employing Eqs. (2–5) of Beirle et al. (2019).

We show the diurnal evolution of characteristic profile heights (H₇₅) in Fig. F10 for the four major seasons. Figure 6b shows the mean afternoon time characteristic profile heights (H₇₅) for aerosol, NO₂, and HCHO for different months, together with the mean ERA5 BLH. Due to their short atmospheric lifetime (<6 h) during the daytime, H₇₅ for NO₂ and HCHO is lower than that for aerosol. H₇₅ for the measured species is observed to be smaller than the typical boundary layer heights. In the monsoon season, we observe H₇₅ comparable to that in summer, even though the boundary layer height is shallow and comparable to that in winter. Trace gases and aerosol from the surface are lofted up due to deep convection in the monsoon leading to high H₇₅. This indicates that the vertical mixing of aerosol during the monsoon is not driven by the parameters used to calculate the ERA5 BLH but rather follows the trend of ambient daytime temperature, which does not show such a large difference between summer and monsoon (e.g. Fig. S2 of Kumar et al., 2016). The profile heights for aerosols and HCHO in summer months are similar to those observed in Beijing (Vlemmix et al., 2015), but we observe a much stronger seasonal dependence. H₇₅ up to 1.1 km for aerosols are observed for the summer (except March) and monsoon months, while in winter H₇₅ is usually less than 500 m. In all months, we observe the minimum H₇₅ among the three measured species for NO₂ (∼200 to 450 m), due to its short lifetime and production close to emission sources near the surface. H₇₅ for NO₂ is generally much smaller than for Beijing (urban) and Xingtai (suburban) in China (Wang et al., 2019; Vlemmix et al., 2015).

In all the months and hours of the day, the major fraction of the NO₂ column is located in the bottom-most layer extending from the surface until 200 m. Until 11:00 LT, more than 60 % of the NO₂ column is located in the bottom-most layer in all the months. In winter months (November–February), the same is true for all hours of the day, but in the morning time (until 11:00 LT) the fraction in the bottom layer is even >80 %. During the late afternoon of the summertime, monsoon, and early post-monsoon, when the BLH grows deeper due to the heating of the surface, we observe a considerable fraction (∼20 %–30 %) of NO₂ in the layer extending from 200 to 400 m. There is a very small fraction (<5 %) of NO₂ column in the layers above 600 m.

We observe the maximum NO₂ columns in the morning hours between 08:00 and 11:00 LT, subsequently decreasing during the day. The major factors driving the NO₂ columns are emissions and lifetime with respect to OH radicals. For the surface concentration, boundary layer dynamics also play an important role. Emissions from traffic and biomass burning for heating and cooking peak in the morning and evening hours. The lifetime of NO₂ is at a minimum in the afternoon when OH radial concentration peaks, which explains the decrease during the day. The amplitude of the diurnal profiles of the NO₂ surface VMR (Fig. F7) is at a maximum during winter months when there is little diurnal variability in the BLH (dilution effect) and NO₂ lifetime (sink effect) between morning and late afternoon hours and the shape is driven primarily by the emissions (source). The amplitude is at a minimum during the monsoon when biomass burning ceases (Mishra and Sinha, 2020). In Sect. 3.7, we will discuss the diurnal variation in the NO₂ surface VMR retrieved from MAX-DOAS and in situ measurements for different months.

For formaldehyde, we observe a comparable distribution among the 0–200 and 200–400 m layers for all seasons except winter. In winter months, the bottom-most layer contains up to 50 % of the total column, which is smaller as compared to aerosols and NO₂. During the late afternoon hours of summer, monsoon, and the post-monsoon period, we also observe a larger fraction of the HCHO column in the 200–400 m layer as compared to the 0–200 m layer. In the presence of high NO_x close to the surface, OH concentration is depleted, which might result in slowing down of the formaldehyde production from precursor VOCs. The higher characteristic profile heights of HCHO as compared to those of NO₂ can be attributed to the secondary photochemical formation from primary precursor emissions. A considerable fraction of primary emissions is transported to intermediate layers (similar to NO₂) during summer, monsoon, and early post-monsoon, where secondary products (e.g. formaldehyde) are formed due to active photochemistry. The months in which primary anthropogenic emissions of formaldehyde and its precursors are stronger (e.g. months except for March, April, July, August, and September), the gradients of vertical profiles of HCHO are stronger in the layers from the surface to 1 km altitude. For the surface VMR, we observe maxima in the morning hours in all seasons except winter and late post-monsoon (Fig. F8). Even though formaldehyde is formed photochemically, which should increase during the day, the VMR close to the surface is reduced due to the vertical mixing in the afternoon hours. We observe the highest daytime HCHO surface VMR in winter months since the major fraction of HCHO is trapped in the bottom-most layer.

3.3 Intercomparison and temporal trends of aerosol optical depth

Figure 7a shows the time series of the AOD at 360 nm retrieved from MAX-DOAS measurements of O₄ (scaled by 0.8) and the AOD at 360 nm calculated from MODIS MAIAC data product over Mohali. The corresponding scatter plot is shown in Fig. 7b. Similar plots for the OMAERUV AOD product at 354 nm are shown in Figs. 7c and d. The solid line represents the monthly mean values, while the individual dots in the background represent the daily measurements. The vertical error bars represent the standard error of the mean ($\mathit{\sigma }/\sqrt{N}$), where σ represents the standard deviation and N represents the number of daily measurements for the month.

Figure 7Intercomparison of daily (dots) and monthly mean (lines and markers) AOD at 360 nm retrieved from ground-based MAX-DOAS measurements and from the MODIS MAIAC data product (a, b) and OMI AERUV data product (c, d). The monthly mean of the MAX-DOAS and satellite data products were calculated by considering only the days of the month when both the measurements were available, causing different MAX-DOAS monthly means in (a) and (c).

Download

The mean MAX-DOAS AOD at 360 nm for the measurement period if averaged around the MODIS overpass time (between 9:30 and 11:30 and between 12:30 and 14:30 LT) was 0.59 ± 0.39 as compared to the MODIS AOD of 0.81 ± 0.53. The correlation coefficients (r) for the linear regressions of daily and monthly mean data between the MAX-DOAS and MODIS AOD are found to be 0.78 and 0.85, respectively. For the monthly mean, we also performed the orthogonal distance regression (ODR) weighted by σ⁻² (where σ is the monthly standard deviation) between MAX-DOAS and MODIS AOD, for which the slope and offset were 1.13 and 0.12, respectively.

Initially, we performed a comparison of AOD retrieved from MAX-DOAS measurements without applying any scaling factor to the measured O₄ dSCD. While there was a general agreement between the trends in the AOD retrieved from MAX-DOAS and MODIS, the MAX-DOAS AOD showed a strong underestimation (Fig. F11). Several MAX-DOAS measurement studies and comparison with independent datasets (e.g. sun photometer) found that a scaling factor (less than 1) was necessary to bring MAX-DOAS results and independent measurements into agreement (Wagner et al., 2009, 2019; Beirle et al., 2019). However, a similar number of studies did not find the need to apply such a scaling factor (Wang et al., 2017a; Franco et al., 2015). Currently, the reason for the scaling factors is not understood (Wagner et al., 2019, and references therein). In most cases, when scaling factors are used, values between about 0.8 and 0.9 are found best. In our case, the application of a scaling factor was found to be necessary to bring MAX-DOAS and satellite measurements into an agreement. In order to further confirm the choice of the scaling factor, the profile inversion was performed with a variable scaling. Figure F12 shows the distribution of the scaling factor which concludes that an O₄ scaling factor of 0.8 fits best for our measurements.

The NASA OMAERUV data product provides the AOD at 354 nm, but the spatial resolution and temporal coverage are not as good as for the MODIS MAIAC product. From Fig. 7c and d, we observe that OMAERUV generally underestimates the AOD over Mohali, and the level of agreement is also worse both for daily and monthly mean values. For OMAERUV, the mean AOD was 0.53 ± 0.26. Over central and east Asia, independent comparison of OMI with AERONET measurements also found a ∼50 % underestimation by OMAERUV and a poor agreement for a 10-year period (Zhang et al., 2016). We think that the much coarser spatial resolution of OMAERUV as compared to the MAIAC data product is among the probable reasons for the worse agreement. In order to evaluate this hypothesis, we created a time series of the MODIS MAIAC data product, spatially averaged over 5 and 25 km around Mohali.

MAX-DOAS measurements are spatially representative of a few kilometres in the field of view, depending on the ambient aerosol load and elevation angle, whereas the ground footprints of individual OMI pixels are 13 × 24 km² in the best case. We have calculated the horizontal sensitivity distance (HSD) of MAX-DOAS for low-elevation angles as the e-folding distance of O₄ dAMF from the instrument location (Wagner and Beirle, 2016). Figure F13 shows that the mean afternoon time (between 12:00 and 15:00 LT) HSD ranges between 5 and 7 km for clear-sky condition with low aerosol load and between 3 and 6 km for high aerosol conditions. Here it is important to note that this estimate is mainly representative of the near-surface layers. While for the trace gas inversions, the VCD is constrained by all elevation angles, the determination of the AOD is mostly constrained by the high-elevation angles. For high-elevation angles, the sensitivity range is much closer to the instrument (at distances up to 1 and 2 km for layer height of 0.5 and 1 km, respectively) (see Fig. 6a). Comparing the spatially degraded time series with MAX-DOAS AOD resulted in a worse agreement (r=0.75 and 0.79 for the daily and monthly means, respectively) for 25 km but did not change significantly for 5 km (Fig. F14) as compared to the original comparison when only a 2 km area around Mohali was considered for spatial averaging. In addition to the effect of horizontal gradients, the poor agreement of the OMAERUV product might also be caused by residual cloud contamination. Further, non-representative assumptions about the aerosol types between smoke, dust, and non-absorbing aerosols used for the inversion of the measured reflectances might play a role, as highlighted by Zhang et al., 2016.

3.4 Intercomparison and temporal trends of NO₂ vertical columns

Figure 8 shows the time series of the NO₂ vertical column densities measured by MAX-DOAS and by OMI for the three different data products. Please note that for calculating the monthly means, we only considered the days of the month when both cloud-screened and quality-controlled satellite and MAX-DOAS data were available. Within the complete study period, the DOMINO, QA4ECV, and OMNO2 data products had 60 %, 47 %, and 46 % days of cloud- and quality-screened data, respectively. We observe a general agreement in the trends of the NO₂ VCDs between the MAX-DOAS and OMI datasets. However, all OMI data products generally underestimate the NO₂ VCDs. The mean MAX-DOAS NO₂ VCD for the measurement period if averaged between 12:30 and 14:30 LT (around the OMI overpass) was (5.4 ± 3.0) × 10¹⁵ molecules cm⁻². The mean VCD for OMI was (3.7 ± 2.4) × 10¹⁵ molecules cm⁻², (3.7 ± 1.3) × 10¹⁵ molecules cm⁻², and (3.7 ± 1.7) × 10¹⁵ molecules cm⁻², respectively, for the DOMINO, QA4ECV, and OMNO2 data products. The reasons for the systematic difference between the MAX-DOAS and OMI measurements can be attributed to several factors including (1) difference in spatial representation and (2) differences in vertical sensitivity of MAX-DOAS and OMI. Previous validation studies over China also found systematic underestimation of ∼60 % over Nanjing by the OMNO2 and ∼30 % over Beijing and Nanjing by the DOMINO product (Chan et al., 2019; Ma et al., 2013). The observed discrepancies were attributed to differences in spatial representativeness, which introduces smoothing of the measured NO₂ VCD over a large satellite ground pixel and to the shielding effect of aerosols. At Mohali, the maximum disagreement of the OMI products is observed during the late post-monsoon and winter months, where all satellite data products significantly underestimate the NO₂ VCDs. Note that a large amount of aerosol and trace gases are emitted from the crop residue and domestic fires in these months, a major fraction of which is trapped close to the surface due to suppressed ventilation.

Figure 8Intercomparison of time series of daily (dots) and monthly mean (lines and markers) NO₂ VCDs retrieved from ground-based MAX-DOAS measurements and the OMI DOMINO data product (a), the OMI QA4ECV data product (b), and the OMI OMNO2 data product (c). The vertical error bars represent the monthly variability as the standard error of the mean. The monthly mean of the MAX-DOAS and satellite data products were calculated by considering only the days of the month when both the measurements were available, causing different MAX-DOAS means in (a), (c), and (e). Scatter plots (panels b, d, and f for DOMINO, QA4ECV, and OMNO2, respectively) using the daily and monthly mean values are shown adjacent to the time series.

Download

Figure 8 also shows the linear regression fits of the three OMI data products versus the MAX-DOAS NO₂ VCDs. We observe smaller scatter in the QA4ECV and OMNO2 products for both daily and monthly values, as compared to DOMINO product. An ODR fit was performed between the monthly means of MAX-DOAS and OMI NO₂ VCDs weighted by σ⁻², where σ is the monthly standard deviation. The slopes of the ODR fit between the MAX-DOAS and OMI monthly mean NO₂ VCDs were 0.94, 0.59, and 0.78, respectively, for DOMINO, QA4ECV, and OMNO2. The offsets of the ODR fits were −8.1 × 10¹⁴, 8.4 × 10¹⁴, and −1.7 × 10¹⁴, respectively, for DOMINO, QA4ECV, and OMNO2, respectively. Over Mohali, we observe excellent consistency between the QA4ECV and OMNO2 products with a slope and correlation coefficient (r) of 0.94 and 0.72, respectively, between both datasets.

Since we have retained the OMI pixels whose centre points lie within 0.25^∘ × 0.25^∘ of the MAX-DOAS measurement site, differences can arise due to the smoothing effect across the OMI ground pixels. With the pristine regions of the Himalayan mountain range only ∼35 km from Mohali, smaller NO₂ VCD from OMI measurements are expected due to systematic gradients towards the mountain range. The effect of smoothing over a large area can be minimized if only collocated pixels are retained for the intercomparison. However, this significantly reduced the number of available days for intercomparison. If only collocated pixels were considered, we were left with only 35 %, 25 %, and 22 % of the measurement days for DOMINO, QA4ECV, and OMNO2, respectively. Due to the poor statistics, we did not observe improvements in the correlation coefficient (r) of the linear regression of the daily data and these changed from 0.38, 0.50 and 0.43 to 0.38, 0.56, 0.43, respectively, for DOMINO, QA4ECV, and OMNO2. In the absence of higher-resolution NO₂ data for the study period, we could not quantify the effect of the different spatial representativeness of the MAX-DOAS and OMI measurements.

One of the major reasons for the disagreement between satellite and MAX-DOAS measurement is the difference in vertical sensitivity of the two measurements. Satellite instruments have limited sensitivity close to the ground. In contrast, MAX-DOAS measurements have the highest sensitivity close to the ground, while it becomes virtually zero above 3–4 km. The limited sensitivity of the satellite instruments is addressed using the box air mass factors (bAMFs) and the a priori profiles of the trace gases to be retrieved (Eskes and Boersma, 2003). In Appendix D we compare the bAMFs used in the DOMINO and QA4ECV retrievals with those calculated by employing the radiative transfer model (RTM) McARTIM over Mohali using the mean aerosol extinction profiles retrieved from the MAX-DOAS measurements. A discrepancy is found between the calculated bAMFs and those used for OMI retrievals (DOMINO and QA4ECV) (Fig. D1), such that the calculated bAMFs show systematically higher values close to the surface. In such a case, attribution of a smaller fraction of NO₂ in layers close to the surface in the a priori profiles cause a systematic underestimation of the VCDs (please see Appendix D for details). We found that the a priori NO₂ profiles used in the DOMINO v2 retrievals strongly differ from those retrieved using the MAX-DOAS measurements for winter and polluted post-monsoon when a large fraction of NO₂ is present in layers close to the surface (Figs. 5 and D2).

In order to eliminate the difference caused by the non-representative a priori NO₂ profiles, we calculated the “modified MAX DOAS VCDs” (called VCD_mod hereafter), which represent the MAX-DOAS NO₂ VCDs as observed by OMI. The application of the OMI tropospheric averaging kernels and a priori profiles also to the MAX-DOAS profiles makes the comparison independent of the a priori profiles used for the OMI retrieval. In order to do so, we apply the tropospheric averaging kernels of OMI DOMINO (AK_trop) data product to the NO₂ vertical profiles (x_doas) retrieved from MAX-DOAS in layers (i) from ground (i=0) to h=4 km (i=20), according to the following equation (Rodgers and Connor, 2003):

Here, x_ap,i represents the DOMINO a priori NO₂ profiles. Please note that the total averaging kernels (AK_tot) provided in the DOMINO level 2 data product are converted to tropospheric averaging kernels using the ratio of total AMF (AMF_tot) and tropospheric AMF (AMF_trop):

Figure 9 shows the time series of original MAX-DOAS VCDs (red), OMI DOMINO VCDs (blue), and modified MAX-DOAS VCDs (black). We observe that the bias between the OMI and MAX-DOAS measurements is smaller if the averaging kernels and a priori profiles are applied to the MAX-DOAS NO₂ profiles. However, in contrast to the MAX DOAS VCDs, the VCD_MOD are systematically lower than the DOMINO NO₂ VCDs.

Figure 9Time series of daily (dots) and monthly means (lines and markers) of MAX-DOAS NO₂ VCDs, OMI DOMINO NO₂ VCDs, and modified MAX-DOAS VCDs modified by using the DOMINO averaging kernels and a priori profiles.

Download

While the application of the DOMINO averaging kernels and a priori NO₂ profiles to the MAX-DOAS profiles according to Eq. (4) accounts for the reduced OMI sensitivity close to the surface, it does not account for the limited sensitivity of MAX-DOAS at higher altitudes (above 3–4 km). VCD_MOD hence represents the MAX-DOAS NO₂ VCD from the ground to up to ∼4 km altitude, whereas the DOMINO NO₂ VCDs represents the NO₂ VCDs from the ground until the tropopause. For a qualitative estimate of the NO₂ column at high altitudes, we have calculated the fraction of the NO₂ column between 4 km altitude and the tropopause by only considering the NO₂ partial VCDs of the TM4 a priori profiles in various layers. The NO₂ partial columns in the 4 km to tropopause altitude range account for 7 %–18 % (interquartile range) of the total NO₂ a priori VCDs. Hence, due to the limited sensitivity of MAX-DOAS at higher altitudes, the VCD_MOD is systematically smaller than DOMINO NO₂ VCDs.

Please note that a similar comparison with OMI QA4ECV and OMNO2 products is not possible as a priori profiles and averaging kernels, respectively, are not available for these data products. For qualitative evaluation, we have used the TM4 NO₂ a priori profiles with QA4ECV averaging kernels and calculated VCD_MOD. This also results in an improvement of the bias between the MAX-DOAS and QA4ECV NO₂ VCDs (Fig. F15). A different approach for improved agreement between MAX-DOAS and satellite VCDs is by using the MAX-DOAS NO₂ profiles as a priori profiles for the calculation of air mass factors for the satellite retrieval (Chan et al., 2019; De Smedt et al., 2015). We discuss this approach and its limitations in Appendix D.

3.5 Intercomparison and temporal trends of HCHO vertical columns

Figure 10 shows the time series of HCHO VCDs measured by MAX-DOAS and by OMI for the QA4ECV (panel a) and OMHCHO products (panel c) with the respective scatter plots (panels b and d). Within the chosen quality and cloud filters, the QA4ECV and OMHCHO datasets have 42 % and 67 % days, respectively, out of the complete study period. The mean MAX-DOAS HCHO VCD for the measurement period if averaged between 12:30 and 14:30 LT (around OMI overpass) was (11.3 ± 6.9) × 10¹⁵ molecules cm⁻². The mean VCDs from OMI observations were (14.9 ± 11.3) ×  10¹⁵ molecules cm⁻² and (11.3 ± 11.7) × 10¹⁵ molecules cm⁻² for the QA4ECV and OMHCHO data products, respectively. A small negative bias was expected in the MAX-DOAS HCHO VCDs, as its sensitivity is limited at higher altitudes where a background HCHO may be present due to the oxidation of long-lived hydrocarbon (mainly methane). Though the non-methane VOCs dominate the formaldehyde production over land, methane oxidation is a ubiquitous source of formaldehyde across the globe. At high altitudes (between 3.6 and 8 km) in a pristine location (Jungfraujoch), the background HCHO VCDs have been observed between 0.75 × 10¹⁵ and 1.43 × 10¹⁵ molecules cm⁻² (Franco et al., 2015) which is equivalent to ∼10 % of the total column as measured over Mohali.

Figure 10Intercomparison of time series of daily (dots) and monthly mean (lines and markers) HCHO VCDs retrieved from ground-based MAX-DOAS measurements and the OMI QA4ECV data product (a) and the OMI OMHCHO data product (c). The respective vertical error bars represent the 1σ monthly variability as the standard error of the mean. The monthly means of the MAX-DOAS and satellite data products were calculated by considering only the days of the month when both the measurements were available, causing different MAX-DOAS means in (a) and (c). Scatter plots (b, d) using the daily and monthly mean values are shown adjacent to the time series.

Download

In general, we see slightly higher VCDs by the OMI QA4ECV product compared to MAX-DOAS, except for the post-monsoon months, when we observe a better agreement between the two datasets. For the OMHCHO product, the monthly mean HCHO VCDs agree well with MAX-DOAS except for the post-monsoon of 2015, where MAX-DOAS VCDs were higher. The generally good agreement of the OMHCHO VCDs with MAX-DOAS is in line with previous works (Chan et al., 2019; Wolfe et al., 2019). The OMHCHO product was also shown to have a good agreement with airborne measurements in the remote troposphere (Wolfe et al., 2019). However, the accountability in the range of the monthly mean HCHO VCDs was much better for the QA4ECV product (slope=0.77) as compared to OMHCHO (slope=0.41). In comparison to the QA4ECV NO₂ dataset, we observe a higher noise in both spatial and temporal patterns of the HCHO VCDs, which arises due to the relatively small atmospheric absorption. The larger uncertainty in the QA4ECV HCHO dataset is also evident from the scatter of daily measurements where sometimes VCDs close to zero are observed. For some months (e.g. January 2015, January 2016, March 2017, April 2017, and June 2017) when less than 6 d of QA4ECV data is retained for calculating the monthly means for intercomparison with MAX-DOAS measurements, the values should be considered carefully. If only collocated pixels are considered for the QA4ECV HCHO product, the statistics get poorer (only 17 % of valid observations) and we observe a worse coefficient of correlation (r=0.19 and 0.17 for daily and monthly means) with MAX-DOAS measurements.

The finding that in contrast to NO₂, no general underestimation is observed for the comparison of the satellite HCHO VCDs to the MAX-DOAS HCHO VCDs can be attributed to the different vertical profiles and the different vertical sensitivities of MAX-DOAS and OMI observations. Since in general, the HCHO profiles reach higher altitudes than NO₂, the satellite observations capture a larger fraction of the total HCHO column. However, we cannot perform a comparison of the modified MAX-DOAS VCDs for HCHO similar to that calculated for DOMINO NO₂, as the total AMFs and averaging kernels (needed for Eq. 5) are not available for the QA4ECV and OMHCHO data products, respectively.

3.6 Discerning the sensitivity of ozone production on NO_x and VOCs

Martin et al. (2004) recommended the use of the ratio of the formaldehyde and NO₂ columns from satellite observations as an indicator for the ozone production regime. HCHO∕NO₂ ratios of less than 1 represent a VOC sensitive regime, whereas values greater than 2 indicate a NO_x sensitive regime. Intermediate values of the HCHO∕NO₂ ratio indicate a strong sensitivity towards both NO_x and VOCs. The threshold for this indicator was initially calculated for afternoon time (between 13:00 and 17:00 LT) but was later extended to also include morning period by Schroeder et al. (2017). However, Schroeder et al. (2017) also indicated that the upper limit of the intermediate regime might vary spatio-temporally. Nonetheless, higher HCHO∕NO₂ indicates that a reduction in NO_x emissions would be more effective for ozone reduction. While the in situ measurements of the total OH reactivity are a more robust method to evaluate the ozone production regime, due to the experimental constraints, these measurements are reported only rarely and for short time periods (e.g. a few weeks or months) (Kumar et al., 2018; Kumar and Sinha, 2014). Mahajan et al. (2015) evaluated the ozone production regime over India using the ratio of HCHO and NO₂ VCDs observed from SCIAMACHY for the mean of the years 2002–2012. Over the north-west IGP, the HCHO∕NO₂ was observed to be less than 1 in the winter months and between 1 and 2 in all other months. From our intercomparisons in the previous sections, we note that while the OMI NO₂ VCDs are generally underestimated, the HCHO VCDs are generally well accounted for. Hence the true HCHO∕NO₂ will be smaller than those indicated by satellite observations, which indicates that the estimated sensitivity of the ozone production regime towards NO_x should be smaller and shifted towards VOCs. Using the WRF-CMAQ model simulations at 36 × 36 km² resolution over India for 2010, Sharma et al. (2016) have evaluated the ozone production to be strongly sensitive to NO_x emissions throughout the year and recommended reduction in transport emissions which account for 42 % of the total NO_x emissions. However, with an increase in transport and power plant emissions (strong NO_x sources) over India, the regimes are susceptible to shift away from being NO_x limited and need to be re-evaluated. Figure 11 shows the monthly mean HCHO∕NO₂ ratio calculated using the MAX-DOAS measurements for the morning (09:30–11:30 LT), noontime around the OMI overpass (12:30-14:30 LT), and late afternoon (15:30–17:30 LT). We observe a stronger (smaller) sensitivity towards NO_x during the late afternoon (morning) as compared to noontime similar to other urban locations in the USA (Schroeder et al., 2017). VOCs contribute to ozone production via their oxidation by OH radicals and subsequent formation of peroxy radicals. During the build-up hours of ozone (between sunrise until noontime) at Mohali, the radicals' abundance is expected to be limited. Hence, the ozone production is more sensitive to VOC (or “radicals”) during morning, which shifts towards NO_x later during the day. In winter months, mean daytime HCHO∕NO₂ ratios between 1 and 2 are observed, which represent sensitivity towards both NO_x and VOCs. The sensitivity of the ozone production regime changes towards NO_x with the onset of summer and stays like that until the end of the post-monsoon season. Over the Indo-Gangetic Plain, the strongest ozone pollution episodes are observed in the summer and post-monsoon months during the afternoon hours between 12:00 and 16:00 LT (Kumar et al., 2016; Sinha et al., 2015). Surface ozone measurements from Mohali have shown enhancement in its ambient concentrations during the late post-monsoon as compared to the early post-monsoon even though the daytime temperature drops by 6 ^∘C. During summer, enhanced precursor emission from fires lead to an increase in ∼19 ppb ozone under similar meteorological conditions. Considering the stronger sensitivity of daytime ozone production towards NO_x, the ozone mitigation strategies should focus on NO_x emission reductions.

Figure 11Monthly mean HCHO VCD∕NO₂ VCD ratios (triangles) calculated from MAX-DOAS measurements for the morning (09:30–11:30 LT, red), noon around the OMI overpass time (12:30–14:30 LT, black), and late afternoon (15:30–17:30 LT, blue) over Mohali. The lines at the centres of the boxes represent the median; the boxes show the interquartile ranges whereas the whiskers show the 5th and 95th percentile values.

Download

3.7 Surface volume mixing ratios of NO₂ and HCHO

The surface VMRs of NO₂ and HCHO can be derived from the MAX-DOAS measurements using the retrieved profiles. For MAPA, the profiles are saved using output grids of a uniform thickness of 200 m. The mean concentrations in the bottom-most layer have been used to calculate VMR by considering the measured ambient temperature.

Figure 12 shows the time series of the NO₂ VMRs measured with the in situ chemiluminescence analyser and that in the lowest 200 m grid obtained from the profile inversion from the MAX-DOAS measurements. The measurement frequency of the in situ measurements was 1 min, whereas, from MAX-DOAS, the mean concentrations are retrieved from individual complete sequences (∼10 min). Hence, we averaged the in situ measurements to coincide with the exact MAX-DOAS measurement time. We observe a reasonable agreement between the two measurements in terms of temporal variability, but the MAX-DOAS surface VMRs are systematically higher until the end of 2014. The differences between the two can probably be explained by the difference in the horizontal and vertical representativeness and resolution and have been discussed in detail in Appendix E. Briefly, the surface VMR from MAX-DOAS represents the mean in the bottom-most 200 m grid of MAPA output, while those from in situ measurements are more sensitive to air mass sampled closed to the inlet. Hence MAX-DOAS VMRs are representative of a larger area around the measurement site, and we infer from Fig. 1 that the measurement location is relatively cleaner compared to the surroundings in terms of NO₂ levels. The mean NO₂ surface VMR from MAX-DOAS measurements was 8.2 ± 6.7 ppb, whereas that from the concurrent in situ measurements was 6.0 ± 5.0 ppb. If we consider all in situ observations (which also include the night-time observation and periods when MAX-DOAS measurements were unavailable), the mean NO₂ VMR was 8.6±6.9 ppb. As compared to previous MAX-DOAS measurements from India (Biswas et al., 2019), at a rural site (0.8 ± 0.2 ppb), we observe much higher NO₂ VMR in Mohali. However, these are comparable to previous in situ NO₂ VMR measured for a period of more than 1 year at urban and suburban locations (distant from traffic) of India (e.g. Mohali: 8.9 ppb; Pune: ∼9.5/8.7 ppb; Kanpur: 5.7 ppb) (Gaur et al., 2014; Kumar et al., 2016; Beig et al., 2007; Debaje and Kakade, 2009) but smaller than near traffic urban measurement (e.g. New Delhi: 12.5/18.6 ppb; Agra: 15–35 ppb) (Saraswati et al., 2018; Tiwari et al., 2015; Singla et al., 2011). Please note that we have used an NO₂∕NO_x ratio of 0.9 to estimate NO₂ VMR for comparison with the previous measurements which reported NO_x VMR and hence have a larger uncertainty (Kunhikrishnan et al., 2006). The inset of Fig 12a shows the monthly variability of surface NO₂ VMR as a box-and-whiskers plot. Similar to the VCDs, we observe maximum NO₂ VMR in winter followed by the crop residue burning active periods of post-monsoon and summer.

Figure 12(a) Daily mean NO₂ mixing ratios measured using an in situ chemiluminescence analyser (black line) and average mixing ratio in the lowest layer (0–200 m) retrieved by the profile inversion of MAX-DOAS measurements (red line). The green and orange dots in the background show the mixing ratios corresponding to the individual MAX-DOAS elevation sequences flagged as valid and warning, respectively. The inset of (a) shows the monthly variability of the MAX-DOAS surface NO₂ VMR as a box-and-whiskers plot; (b) shows the frequency distribution of the BIAS (MAX-DOAS – in situ) for the individual surface VMR measurements; (c) shows the scatter plot between the daily mean surface NO₂ VMR from MAX-DOAS and an in situ analyser colour-coded according to the profile height (H₇₅). The individual measurements are shown as black dots in the background of (c).

Download

Figure 13 shows the time series of the HCHO VMR measured with the PTR-MS and those derived for the lowest 200 m layer obtained from profile inversion of the MAX-DOAS measurements. The mean HCHO surface VMR from MAX-DOAS measurements was 8.7 ± 7.5 ppb, whereas that from concurrent in situ PTR-MS measurements was 3.3 ± 1.7 ppb. For HCHO, higher VMR from MAX-DOAS measurements can be explained by its photochemical formation at altitudes higher than the inlet of PTR-MS. This has been further discussed in Appendix E. The measured HCHO VMRs are comparable to previous MAX-DOAS measurements from India (Pantnagar: 2–6 ppb) but much lower than those measured previously in India using offline techniques (e.g. north Kolkata: 16 ppb; south Kolkata: 11.5 ppb) (Dutta et al., 2010; Hoque et al., 2018). From the monthly variation in HCHO VMRs shown in the inset of Fig. 13a, we observe that the maximum is observed in late post-monsoon and winter followed by late summer (crop residue burning active period). This monthly variability is slightly different from that of the HCHO VCDs (discussed in Sect. 3.1). A shallower profile (Fig. 5) in the late post-monsoon and winter leads to high surface VMR even though the VCD is smaller than that in late summer.

Figure 13(a) Daily mean HCHO mixing ratios measured using an in situ chemiluminescence analyser (black line) and average mixing ratio in the lowest layer (0–200 m) retrieved by the profile inversion of MAX-DOAS measurements (red line). The green and orange dots in the background show the mixing ratios corresponding to the individual MAX-DOAS elevation sequences flagged as valid and warning, respectively. The inset of (a) shows the monthly variability of the MAX-DOAS surface HCHO VMR as a box-and-whiskers plot; (b) shows the frequency distribution of the BIAS (MAX-DOAS – in situ) for the individual surface VMR measurements; (c) shows the scatter plot between the daily mean surface HCHO VMR from MAX-DOAS and PTR-MS colour-coded according to the profile height (H₇₅). The individual measurements are shown as black dots in the background of (c).

Download

Figure F7 shows the mean diurnal profiles of the surface NO₂ VMRs for different months using MAX-DOAS and in situ measurements. A typical diurnal feature representative of a suburban location was observed from the in situ measurements, which is explained in detail elsewhere for Mohali for different seasons (Kumar et al., 2016). For the daytime hours, when MAX-DOAS observations are also available, we observe a general agreement in the absolute values and temporal evolution between the two measurements in all the seasons except winter. The occurrence of a morning peak for the in situ measurements between 07:00 and 09:00 LT in all months is primarily driven by emissions (traffic and biofuel combustion). For the MAX-DOAS observations, a similar maximum occurs between 08:00 and 10:00 LT and between 09:00 and 11:00 LT in winter months. The time shift compared to the maximum in the in situ measurements can be explained by the accumulation of the surface emission in the boundary layer and breaking of the nocturnal boundary layer. The latter causes the NO_x present at high altitudes (e.g. from power plants >200 m, Fig. 5) to mix in with the surface layers. A modelling study performed using the CMAQ model has shown that the peak in the diurnal profiles of NO₂ columns occurs 2–3 h later than that for surface concentrations (Fishman et al., 2008). Since MAX-DOAS surface VMR represents the mean over a few hundred metres, we expect similar behaviour in its comparison with in situ measurements. In winter months, though the diurnal trends are similar in both observations, a general overestimation by MAX-DOAS is found, which is caused by a shallower boundary layer and the presence of high NO₂ mixing ratios above the inlet of the in situ analyser.