Planetary Boundary Layer Height Modulates Aerosol—Water Vapor Interactions During Winter in the Megacity of Delhi

The Indo‐Gangetic Plain (IGP) is one of the dominant sources of air pollution worldwide. During winter, the variations in planetary boundary layer (PBL) height, driven by a strong radiative thermal inversion, affect the regional air pollution dispersion. To date, measurements of aerosol‐water vapor interactions, especially cloud condensation nuclei (CCN) activity, are limited in the Indian subcontinent, causing large uncertainties in radiative forcing estimates of aerosol‐cloud interactions. We present the results of a one‐month field campaign (February‐March 2018) in the megacity, Delhi, a significant polluter in the IGP. We measured the composition of fine particulate matter (PM1) and size‐resolved CCN properties over a wide range of water vapor supersaturations. The analysis includes PBL modeling, backward trajectories, receptor models and fire spots to elucidate the influence of PBL and air mass origins on aerosols. The aerosol properties depended strongly on PBL height and a simple power‐law fit could parameterize the observed correlations of PM1 mass, aerosol particle number and CCN number with PBL height, indicating PBL induced changes in aerosol accumulation. The low inorganic mass fractions, low aerosol hygroscopicity and high externally mixed weakly CCN‐active particles under low PBL height ( < $< $ 100 m) indicated the influence of PBL on aerosol aging processes. In contrast, aerosol properties did not depend strongly on air mass origins or wind direction, implying that the observed aerosol and CCN are from local emissions. An error function could parameterize the relationship between CCN number and supersaturation throughout the campaign.

The topographical features and meteorological processes in Delhi are different from those of other megacities around the world. For example, the shallow PBL during winter in Delhi is typically caused by surface radiative cooling (Dumka et al., 2019;Ojha et al., 2020;Raatikainen et al., 2011;Tiwari et al., 2013) unlike in other megacities. Therefore, to capture the nuanced interaction of aerosols with atmospheric water vapor, which leads to cloud, fog and haze formation, visibility deterioration and health impacts, field measurements are essential. At present, we have a fair understanding of CCN activity from size-resolved CCN measurements under diverse environmental conditions (Schmale et al., 2018). This includes the clean Amazonian rainforest Pöhlker et al., 2016Pöhlker et al., , 2018, the coastal background site, Mace Head (Paramonov et al., 2015) and highly polluted megacities like Beijing  and Guangzhou (Rose et al., 2010 in China. Other locations where measurements are reported include the Thuringian forest in Germany (Henning et al., 2014), the coastal region in California (Asa-Awuku et al., 2015), the continental city of Kanpur in India (Bhattu et al., 2016), a peninsular region in Japan (Iwamoto et al., 2016), the South China Sea (Atwood et al., 2017) and the Yangtze River delta in China (Che et al., 2017;Ma et al., 2017). Here we present a comprehensive study of aerosols using size-resolved CCN measurements, supplemented by concomitant non-refractory PM 1 (NR-PM 1 ) composition and black carbon (BC) observations conducted within the metropolitan area of Delhi. From the 27-day-long campaign in February and March 2018, we investigated and identified the meteorological factors playing a major role in transforming aerosol chemistry and retrieved CCN activity parameterizations validated against field measurements. During this short period, we observed two contrasting meteorological conditions characterized by the extent of radiative thermal inversion, which are in the following termed strong and weak inversion periods for the ease of reading. The resulting contrast in PBL height revealed interesting insights into the interaction of aerosols with water vapor, leading to cloud activation within the confined atmosphere of Delhi under varied anthropogenic emissions.
Previous studies on atmospheric aerosols in general, but not related to CCN, in Delhi have reported that primary sources of NR-PM 1 such as biomass burning (BB) and fossil fuel combustion  are dominant in the overall PM mass burden, along with a major contribution from secondary aerosol formation all year round (Gani et al., 2019;Jaiprakash et al., 2017). Episodic peaks in primary organic aerosols along with ammonium chloride were reported previously by Gani et al. (2019) and Bhandari et al. (2020), as well as the companion study by Gunthe et al. (2021). Studies on the influence of meteorology on aerosol properties show that under uniform and unchanged sources of emissions, the extreme lowering of the PBL below 100 m during winters builds up extreme particulate matter concentrations, driving the Air Quality Index to hazardous levels Dumka et al., 2019;Gani et al., 2019;Mandal et al., 2014;Murthy et al., 2020;Ojha et al., 2020). The influence of air mass history has also been reported in Delhi using back trajectory analysis Jaiprakash et al., 2017). A comprehensive analysis provided in this study combining remote sensing and modeling approaches with measurements of CCN activity at 11 different supersaturations is unprecedented.
Delhi is enclosed by the Thar desert (Rajasthan) to the west and the Deccan plateau to the south and is located in front of the Himalayan mountain range, which extends from the north of Delhi to the far north-east. The climate of Delhi is semi-arid with very hot summers, moderate monsoon and cold winters. The winter is from early November to mid-February, characterized by cool days and cold and humid nights resulting in a distinct radiative thermal inversion (Arun et al., 2018;Dumka et al., 2019;Kumar et al., 2017;Ojha et al., 2020;Raatikainen et al., 2011;Thomas et al., 2019). The average diurnal temperature (T) cycled between 10 and 25°C and the average diurnal RH from 26 to 90%. In the present study we used the wind data collected at the Central Pollution Control Board (CPCB, https://app.cpcbccr.com) operated station in Jawaharlal Nehru Stadium, which is at a distance of less than 1.5 km from the campaign location. The meteorological data at the CPCB site showed a strong correlation (R 2 0.97; see Figures S1 and S2 in Supporting Information Information S1) with the quasi-continuous data measured at the observation site. Therefore, as the wind data from the CPCB site is continuously available, we used those in the present study and for all analysis requiring wind data.
The aerosol measurement instruments were placed in an air-conditioned container, fitted with a stainless-steel inlet ∼5 m above the ground. The inlet tubing was smoothly bent so that the cover of the meshed opening was upside down to bar the entry of rainwater and other debris. The aerosol samples were dried so that the moisture content was below ∼25% RH using a diffusion drier containing silica gel (Merck, Germany; ∼1-3 mm size). Thereafter, the polydisperse aerosol flow was divided using a `Y'-shaped splitter and used to supply (i) an aerosol number size distribution and size-resolved CCN setup; and (ii) a chemical composition measurement setup. The temperature inside the container was maintained at ∼28°C throughout the campaign to ensure a stable working environment for the instruments.

Planetary Boundary Layer Modeling
The Weather Research and Forecasting model (WRF version 4.0) was used to simulate the diurnal variations in planetary boundary layer height (H BL ). The center of the model domain was at 76E, 29N and there were 240 grid points in the east-west direction and 147 grid points in the north-south direction on a Mercator projection, along with 51 vertical levels. Model simulations were conducted at a resolution of 12 × 12 km. Initial and lateral boundary conditions for the meteorological fields were prepared using the Era Interim data (https://www.ecmwf. int/en/forecasts/datasets/). The model simulation was conducted for the period of January 28 to March 3, 2018 at a time step of 72 s and the model output was stored every hour for analysis. The first 4 days of model output have been discarded to account for model spin-up. Physics schemes used in the model to parameterize different processes were: (i) Lin et al. scheme for cloud microphysics; (ii) Grell 3D ensemble scheme for cumulus parameterization; (iii) Unified Noah land surface model for land surface option; (iv) Rapid radiative transfer model for long wave radiation and (v) Goddard shortwave scheme for short wave radiation. Two PBL schemes were used in two different simulations: (i) Mellor-Yamada-Janjic scheme (representing the turbulence kinetic energy) and (ii) Yoinsei University scheme (based on Bulk Richardson number).
We also analyzed the latest ECMWF satellite reanalysis data set, ERA5 and the mixing depth from the HYSPLIT model for the same period. On comparison of the above data sets with the quasi-continuous ceilometer measurements carried out at the same location as part of this campaign (Murthy et al., 2020), the WRF modeled results and ERA5 data showed good correlation (Figures S3 and S4; Table S1 in Supporting Information S1). The WRF modeled data with the Bulk Richardson number method was used in this study because of the finer spatial resolution compared to ERA5 (0.5° × 0.5°). This data are provided in the Data Set S1.

Back Trajectory and Fire Spot Analysis
The spatiotemporal pattern of air masses over Delhi was studied using 3-day back trajectory (BT) data, retrieved using the HYSPLIT (The Hybrid Single-Particle Lagrangian Integrated Trajectory) model (Stein et al., 2015). A fixed height of 1000 m above the ground was used as top of the model for an analysis to identify the different air masses over a regional scale. Three years of data (January 1, 2016 to December 31, 2018), which constituted 26 ,304 BTs, with each BT containing 73 latitude-longitude data pairs were clustered spatially using the Quickbundles algorithm (Garyfallidis et al., 2012) in Python (Van Rossum & Drake, 2009). It is a fast clustering method, for simplifying complex and large sets of tractography data. Quickbundles has a built-in metric called the minimum average direct-flip distance (MDF), which selects the minimum among the Euclidean distance between trajectories determined using two methods, one that considers endpoints that lie in the same sequential position and another that considers endpoints lying in opposite sequence in both trajectories. MDF efficiently separates the BTs based on length as well as direction. Based on this algorithm and subsequent analysis results, we identified nine major BT cluster directions in a year. Out of this, only three could be mapped for the time of the year we conducted the measurements. They were north-west (NW), south-east (SE) and mixed regional pollution (MRP).
The most frequently sampled air mass during the campaign belonged to NW, which gave us 540 samples in the size-resolved CCN experiment (or CCN scans), followed by MRP which gave 149 scans and SE which gave 44 scans, out of the total 733 good-quality CCN scans. The density of BTs in a grid of 0.1°× 0.1° was also calculated and plotted ( Figure 1) using a program written in Python. The HYSPLIT model was also run using H BL modeled by WRF (see Section 2.2) as top of the model to study the effect of local air masses on measured aerosol and CCN parameters.
To complement the BT analysis, fire intensity (W m −2 ) maps were retrieved from Copernicus Atmosphere Monitoring Service (CAMS) -Global Fire Assimilation System (GFAS). The data were in NetCDF3 format and the graphs were visualized in QGIS overlaid on a geo-referenced map layer of India. During the measurement period, the fire spot data analysis revealed a series of small fire events much farther than the consistent fire spots present over Punjab and Haryana, which generally affect air quality in Delhi during the stubble burning period of October-November. Our fire spot analysis further confirms the presence of fire spots in October and November-2017 preceding our campaign.

Potential Source Contribution Function (PSCF) and Concentration Weighted Trajectory (CWT) Analysis
The spatial distribution of potential aerosol emission sources was identified by employing the potential source contribution function (PSCF) and concentration weighted trajectory (CWT) models. The models combined measured aerosol number and mass concentrations at the receptor site with the HYSPLIT back trajectories (Fan et al., 1995). For the analysis, 3-day back trajectories were calculated using the WRF modeled H BL as top of the model. An Igor based tool, Zefir (Petit et al., 2017), was used for the PSCF and CWT analyses. To identify potential local sources of pollution, the analysis was also carried with 1-day BT data using Zefir.
For both analyses, the area around the receptor site was sub-gridded into 0.1° × 0.1°grids. Assuming that there is no loss through diffusion, chemical or physical processes, the probability of a grid cell through which the air mass passes, causing a concentration above a threshold is given by the PSCF (M. D. Cheng et al., 1993;Fan et al., 1995;Polissar et al., 2001). CWT takes into account the residence time of air masses and the concentration caused at the receptor site by passage over the respective grid cells to estimate the potential source concentration in each grid cell (Bao et al., 2017;Rai et al., 2020). Both methods can, however, show unreasonably high probability in grid cells that contain very few trajectory end points. Therefore, a sigmoidal weighting function (W) (Petit et al., 2017) was used to down-weight those cells with low trajectory density (x) (Equation 1). (1)

Size-Resolved CCN Measurements
The size-resolved CCN measurements were carried out based on Frank et al. (2006) and Rose et al. (2008), by the coordinated controlling of three instruments, viz., a Cloud Condensation Nuclei Counter (CCNC, model CCN-100, DMT; Roberts & Nenes, 2005) to determine CCN number concentration, an electrostatic classifier (EC, model 3080, TSI) with a differential mobility analyzer (long DMA, model 3081, TSI) to select monodisperse aerosol particles and a condensation particle counter (UCPC, model 3776, TSI) to count total aerosol particles. The entire setup was controlled externally by a computer which runs a dedicated, in-house developed and well tested LabVIEW (National Instruments, Munich, Germany) program (Pöhlker et al., 2016) to continuously sample ambient aerosols of the size range relevant to cloud processing (26 size bins or mobility diameters (D) in the range, 10-370 nm) and measure cloud droplet activation at different supersaturation (S) levels.
The dried polydisperse aerosol samples were drawn into the EC through an inertial impactor that prevents particles larger than ∼370 nm from entering the system. Inside the EC, the polydisperse aerosol flow was passed through a radioactive neutralizer (Kr-85, model 3077A, TSI) to attain a known charge distribution. The charged particles were then passed through an electric field inside the long-DMA, where they were segregated based on their electrical mobility. The DMA was controlled by the LabVIEW program to set the appropriate voltage in order to select the desired D and produce a near-monodisperse aerosol flow. This flow was further split, using a `Y'-shaped connector tube, between CCNC and UCPC, to measure CCN and total aerosol number concentration respectively for the chosen diameter. The DMA maintained the near-monodisperse flow of selected D for 30 s, then switched to the next D after flushing out the DMA column for 40 s to make it particle-free. Meanwhile, the CCNC was controlled to maintain a specific S level for a duration of 40 min, to let all the successive D steps pass through the instruments. The CCNC measured the particles of D activated as cloud droplets (N CCN (S,D)) employing the empirical approach of counting the droplets greater than 1 μm (Roberts & Nenes, 2005;Rose et al., 2008) at the S level attained inside the CCNC column. This method was adopted since larger particles are not sampled and therefore the probability of counting unactivated particles as CCN was low (Y. Wang et al., , 2021. Simultaneously, the UCPC measured the total particle number concentration for a given D (N CN (D)). A relevant set of 11 selected S levels were cycled in the CCNC column by providing an equilibration time of 5 min between each S, taking ∼8 h to finish a complete cycle of S levels.
The sample flow through the DMA was 0.8 L min −1 , which includes the 0.3 L min −1 to the UCPC and 0.5 L min −1 to the CCNC. The sheath to sample flow ratio inside the DMA was 7.5 and the total flow to aerosol flow ratio inside the CCNC was 10. The liquid supply pump was working in the low flow mode with a supply of 4 mL h −1 . The size-resolved CCN experiment setup measured aerosol samples at temperature 299 2.3 K, pressure 959 8.2 hPa and 25% 5% RH (arithmetic mean standard deviation).
The CCNC was systematically calibrated based on Rose et al. (2008) for different S levels before and after the 27day campaign (February 02 and March 03, 2018). Calibrations were performed using standard ammonium sulfate aerosols, whose cloud droplet activation is well explained using classic Köhler theory. Standard ammonium sulfate aerosols were generated by nebulizing an aqueous salt solution (0.3 g L −1 ) of ammonium sulfate ((NH 4 ) 2 SO 4 , purity 99.5%) in a TSI Aerosol Generator. Both calibration experiments gave a similar relationship between the S level estimated experimentally (or effective supersaturation, S eff ) and the corresponding measured temperature gradient (dT, K) inside the CCNC column. This ensured the stable functioning of the CCNC during the campaign period. All S levels for which parameters are reported in this paper refer to experimentally derived S eff .

Measurement Specifications
It should be noted that the measurements made in the study are not strictly ambient, due to the conditioning of the aerosol samples taken from the atmosphere to a moisture content below 25% RH. It is strongly possible that there are losses of volatile species during the passage of the air through the chambers of the various instruments, which are operated at a fixed temperature. Nevertheless, the measurements reported should be considered as the intrinsic properties of aerosols at the following atmospheric conditions, T = 299 2.3 K, P = 959 8.2 hPa and RH = 25% 5% (arithmetic mean standard deviation measured during campaign). To convert to standard conditions (STP: 273 K, 1000 hPa, 0% RH), the reported values can be multiplied by a factor of 1.08 (derived using the ideal gas law). However, this correction does not account for the change in particle size, mass and hygroscopicity due to loss of volatile species, interaction with gaseous phase components, and gas -particle partitioning.

Data Analysis of Size-Resolved CCN Measurements
Size-resolved CCN measurements provide the cloud activated fraction (N CCN (S,D)/N CN ) for a selected D at the S level attained in the CCNC column. The values of N CCN (S,D)/N CN plotted against the corresponding D for the entire particle size range sampled at a single measured S level gives a CCN activation curve, which constitutes a measured scan. The average CCN activation curves measured at each S level are shown in Figure 2. The most important parameter obtained from these curves is the midpoint activation diameter (D a (S)), which is the minimum diameter required for activation of 50% of condensation particles in the sample at given S level. Over the course of the entire campaign, 744 scans of entire particle size range at a single S level were measured and analyzed to calculate CCN properties based on the theories put forth in Petters and Kreidenweis (2007) and Rose et al. (2008). Based on the quality of data, 733 scans  were selected for further analyses and representation in this study.

Errors and Corrections in Size-Resolved CCN Data
Practical limitations of the instruments generally introduce errors in the measured mobility diameter of particles and activated fractions. The outlet slit of the DMA has a finite width and hence it permits a size bin, rather than a fixed size in the outflow (Knutson & Whitby, 1975). Moreover, some of these particles carry multiple charges and attain enhanced electrical mobility inside the DMA. These multiple-charged particles flow along with smaller single-charged particles of similar electrical mobility resulting in enhanced activated ratios and induce errors.

of 25
The measured activated fractions were corrected for multiple-charged particles based on Frank et al. (2006), by taking into account the presence of up to quadruple-charged particles. Following Rose et al. (2008) these corrected fractions were used to back-calculate the measured diameter by applying the transfer function, which is the probability of a certain D passing through the DMA's slit (Knutson & Whitby, 1975). The corrected activation curves were fitted with a three-parameter cumulative distribution function (CDF) using the non-linear least squares fitting routine (Gauss-Newton method, Matlab, MathWorks, Inc.) defined by the parameters: D a (S); N CCN (S,D)/N CN at 50% activation or a(S,D); and the width of the activation curve or σ(S) (Rose et al., 2008).

Size Data Inversion for Particle Number Size Distribution
Due to instrumental limitations the routinely measured aerosol particle number size distribution could not be used for further analysis. Therefore, alternatively we used the aerosol number size distribution obtained by the inversion of N CN (D) data retrieved during size-resolved CCN measurements. A dedicated inversion routine was developed (Matlab, MathWorks, Inc.) following the method and approximations described by Wiedensohler (1988).
The measured N CN (D) is the concentration of particles in a narrow size bin, D, governed by the electrical mobility of the particles inside the DMA. Since the DMA used had a negatively charged electrode, N CN (D) was the concentration of only the positively charged particles. Moreover, the UCPC measures only half the original concentration of positively charged particles, owing to the triangular nature of the transfer function (Knutson & Whitby, 1975). Hence two times the value of N CN (D) is used to account for this loss. The inertial impactor at the sampling inlet of DMA ensures that only those particles less than a given cut diameter enter the instruments and the largest size bin (D max ) sampled was approximately equal to this cut diameter. Therefore, the concentration at D max , N CN (D max ) corresponds to only +1 charged particles, so it required no multiple charge correction. In order to calculate near-ambient particle concentrations, the measured 2*N CN values are divided by the probability of the radioactive neutralizer generating particles of charge e = +1 in each D (Wiedensohler, 1988). This quantity, when calculated for all D except D max , includes particles carrying multiple charges in the range, e = [−4, +4]. So the multiple charge correction is applied for up to 2 charges using Wiedensohler's coefficients (Wiedensohler, 1988) and for 3 to 4 charges using Gunn's equation (Gunn, 1954).
The inverted size distribution was compared with quasi-continuous parallel measurements of aerosol number size distribution using an SMPS consisting of an electrostatic classifier (EC, model 3082, TSI) with Nano DMA (model 3085, TSI) and a condensation particle counter (CPC, model 3772, TSI). Although the measured size range of the Nano-SMPS was 8-105 nm, it served as a quality check for the inverted particle concentration in that size range. The qualitative match of the inverted size distribution was in good agreement, with about a 23.1% reduction in number, which is expected due to the size-resolved mode of measurement, where the mobility diameter was selected step-wise unlike the continuous scanning mode done in the SMPS. Hence in order to take this loss into account, the inverted particle number size distribution (dN/dlogD) and particle concentrations in the size range 10-370 nm (N CN,10 ) reported in this paper have been up-scaled by a factor of 1.3 .

CCN Properties
The parameters of the CDF fit to the activation curves, along with S eff and particle number size distribution were used to calculate key CCN properties (as enumerated below) based on literature.

Maximum activated fraction of the aerosol population at S: MAF(S):
The fit parameter of CDF, a(S,D) gives the ratio of CCN to CN particles at S when 50% activation has occurred. From this, MAF is calculated as 2*a(S,D). If less than 100% of the aerosol population is activated, MAF will be less than 1 and indicates the presence of externally mixed CCN inactive particles Pöhlker et al., 2016;Rose et al., 2008). 3. Total number concentration of CCN at S: N CCN (S): The measured CN distribution was multiplied by the corresponding activation curve or CDF at given S to obtain the CCN distribution. Then, total CCN is calculated by integrating the particles under the CCN number size distribution Pöhlker et al., 2016;Rose et al., 2008). 4. CCN efficiency of the aerosol population at S: N CCN (S)/N CN : The ratio of total CCN number concentration at a given S to the corresponding total CN number concentration gives the CCN efficiency of the sampled aerosol population Pöhlker et al., 2016;Rose et al., 2008).
Average CCN properties measured during the field campaign are tabulated as a function of experimentally derived S levels in Table 1. Since the time taken for measuring a complete set of S levels was long (∼8 h), we have examined the frequency of measurements of all S levels in a full diurnal cycle for the entire data, to assess the impacts on CCN measurements. The distribution was similar, indicating that the CCN properties reported are not biased toward a period in the diurnal cycle for the S levels and therefore, the CCN measurements are not much affected by the long time taken by one complete S cycle.

Aerosol Chemical Composition and Black Carbon Measurements
Simultaneous measurements of non-refractory PM 1 (NR-PM 1 ) aerosols were carried out using an Aerosol Chemical Speciation Monitor (ACSM), which uses a quadrupole mass spectrometer (Ng et al., 2011). The calibration and measurement technique of this instrument are explained in detail elsewhere .

Results and Discussion
The temporal evolution of the measured aerosol and CCN properties, meteorological parameters and model results are shown in Figure 3. The top strip of the compendium plot shows the 3-day BT analysis of sampled air masses, using the HYSPLIT model (see Section 2.3). It marks the duration of occurrence of three distinct directions derived using the clustering algorithm (see Section 2.3), indicating the origin and path of the air mass: NW, SE and MRP consisting of both, north-west and south-east branches. Panel (a) shows the wind speed color-scaled by the wind direction (see Section 2.1). The wind speed was generally low ( 5 m s −1 ), indicating the potential influence of surface roughness (Jacobson et al., 2019) on winds in Delhi during late winter at the local scale. The wind speed also exhibited a systematic diurnal cycle with lower wind speeds ( 3 m s −1 ) during nighttime and relatively faster wind during daytime. As evident from Figure 3a, the wind direction was north-westerly during most of the campaign period consistent with the air mass BT analysis. The changes in wind direction were abrupt and brief ( Figure 3a) and hence, lacked pronounced diurnal variation. The temperature and relative humidity (T and RH, Figure 3b) cycled diurnally with high T and low RH during daytime, particularly at late afternoon hours. The PBL height (H BL ) simulated using the WRF model (see Section 2.2) is shown in panel (c). The alternating occurrence of low wind speed, low T and high RH during nighttime contributed to a shallow nocturnal PBL, resulting in pronounced diurnal variations in H BL .
A strong influence of the shallow nocturnal boundary layer on the measured aerosol mass and number concentrations was observed, with an inverse relationship between concentrations and H BL (see H BL in Figure 3c, M BC,e in Figure 3c and N CN,10 in Figure 3d). To further investigate the dependencies between the modeled H BL and measured meteorological parameters and characteristic aerosol properties, the diurnal variations of H BL , RH and T, along with the aerosol properties that show strong correlation to H BL are shown in Figure 4. We found that the relations between H BL and M BCe (see Figures 4a and 4b); H BL and N CN,10 (see Figures 4a and 4b) were the strongest at the diurnal scale. The average diurnal variation of the modeled H BL (Figure 4a) is in good agreement (R 2 = 0.95; see Figure S3a in Supporting Information S1) with the observed ceilometer data measured at the same location and period as reported by Murthy et al. (2020). An increase in H BL is observed after 08:00 LT (Local Time, i.e., UTC +05:30) in the morning, corresponding to the rise in temperature and drop in RH. This indicates that the convective mixing after 08:00 LT leads to the breaking of the nocturnal stable layer, resulting in an increase in H BL , which sustained until 18:00 LT. As night falls, the ground surface cools faster than the All other values are expressed as average ± standard deviation. Parameters tabulated are: midpoint activation diameter (D a (S)), hygroscopicity from size-resolved CCN measurements (κ(S, D a )), width of the CCN activation curve (σ(S)), maximum activated fraction (MAF(S)), total CCN concentration (N CCN (S)), total particle concentration in the size range ∼10-370 nm (N CN,10 ), CCN efficiency (N CCN (S)/ N CN,10 ) and number of samples for each S level (n).  (Gopalakrishnan et al., 1998;Singh, 2016;Stull, 2012). Hence, the nocturnal atmosphere is static, stable, cold and humid and persisted for an extended duration during nighttime (∼18:00 to ∼08:00 LT).
The mass concentration of BC e , the total particle number concentration (Figure 4b) and the mass concentrations of the NR-PM 1 (Figure 4c) followed a similar trend, with a strong diurnal variation, indicating a dominant contribution of locally emitted aerosol from sources like fossil fuel combustion, solid biomass burning for domestic cooking and heating to the total aerosol mass burden. This is consistent with previous studies by Bhandari et al. (2020) and Reyes-Villegas et al. (2021). All parameters consistently exhibited a characteristic morning traffic rush hour peak between 06:00 and 09:00 LT, subsequently showing a drop thereafter. This drop strongly coincides with the PBL rise and therefore, is further pronounced due to dilution by convective mixing. The nighttime peak starts at 19:00 LT coinciding with the evening rush hour as well as the drop in H BL , which results in the reduction of PBL mixing volume and facilitates enhanced nighttime concentration. While N CN,10 drastically decreased after the rush hour at 22:00 LT, indicating either particle removal or coagulation processes (to be discussed in follow-up studies), the BC e and NR-PM 1 mass concentrations continued to rise until 00:00 LT. The prominent late night BC e peak has been attributed to diesel engine exhaust emissions from trucks and other heavy vehicles, which are permitted passage through Delhi only after 21:00 LT (Guttikunda & Gurjar, 2012). The mass concentrations of organic and major inorganic components also followed the typical diurnal trend in concentration variations (Figure 4c). These observations are consistent with the diurnal trend of PM 1 concentrations previously reported by Gani et al. (2019) in Delhi. The average diurnal cycling of aerosol number size distribution (Figure 4d) further shows that the dominant size range of aerosol particles emitted during the morning and evening rush hours consistently remained in the range of 80-100 nm. It is interesting to note that a growth in particle modal diameter from ∼80 to ∼140 nm was observed starting in the evening hours (18:00 LT) and lasting until almost early morning (06:00 LT), consistent with the observations reported by Gani et al. (2020). The detailed processes and mechanisms with possible implications of such a growth will be discussed in a follow-up study.
The Positive Matrix Factorization (PMF) analysis of organic aerosols and source apportionment of BC e performed on the chemical composition dataset used in this study reported by Reyes-Villegas et al. (2021) have shown that traffic is the main primary aerosol source in Delhi. The spatial characteristics of the emission sources were studied using PSCF and CWT analyses of 1-day ( Figures S6 and S7 in Supporting Information S1) and 3-day ( Figure S8 in Supporting Information S1) BTs and measured aerosol concentrations at the receptor site. The analyses showed the presence of strong potential emission sources on the north-west direction from the campaign site, consistent with the companion study by Gunthe et al. (2021). The source locations of N CN,10 (Figures S6a and S6b in Supporting Information S1) and BC e from fossil fuel combustion (M BC-FF,e ) (Figures S6c and S6d in Supporting Information S1) using PSCF and CWT analyses were similar, indicating the influence of traffic emissions on N CN,10 . PSCF and CWT analyses using BC e from BB (M BC-BB,e ) ( Figure S7 in Supporting Information S1) also showed similar qualitative characteristics to M BC-FF,e , however, with lower quantity consistent with Reyes-Villegas et al. (2021). Regional sources identified using PSCF and CWT analyses of 3-day back trajectories also showed potential emission sources in the north-westerly direction ( Figure S8 in Supporting Information S1). Further PSCF and CWT analyses were done to understand the influence of H BL diurnal cycling on aerosol number and mass concentrations and hygroscopicity at the receptor site, which will be discussed later.
Although the average diurnal cycling of H BL was pronounced, a closer look at panel (c) in Figure 3 reveals multiple periods with weak diurnal cycling of H BL . These periods of weak diurnal cycling coincided with a low solar radiation and a reduced difference between the maximum and minimum temperature during the following diurnal cycle ( Figure S5 in Supporting Information S1; see Text S1 in Supporting Information S1). Low solar radiation can cause relatively faster radiative cooling of the atmosphere when night falls, which leads to increase in nighttime H BL , resulting in a more ventilated nocturnal atmosphere (Stull, 2012). These periods were classified as weak inversion periods (blue background in Figure 3; see Table S2 in Supporting Information S1 for time of occurrence). The rest of the campaign, with stronger solar radiation, were classified as strong inversion periods (white background in Figure 3; see Table S2 in Supporting Information S1 for time of occurrence). The strong inversion periods were marked with higher near-surface RH levels during the nighttime as shown in Figure 5b. Since moist air is characterized by higher heat transfer (Still et al., 1998), this could cause warming of the atmosphere above ground, which would result in reduced inversion layer thickness (Pasricha et al., 2003) and hence, could induce a positive feedback to lowering of H BL . The above observations were substantiated by the PBL simulations that showed lower nighttime H BL during strong inversion periods in contrast to the higher nighttime H BL during weak inversion periods (Figure 5c; see Text S1 in Supporting Information S1). Overall, the statistical properties of RH and H BL were different during nighttime between strong and weak inversion periods, whereas they were comparable during daytime for both periods.
To better understand the role of meteorology on aerosol extensive and intensive properties and their interdependence, we plotted real-time N CN,10 values against H BL for strong and weak inversion periods. We further scaled it by wind speed and direction and RH ( Figure 6). Based on the minimum, maximum and quartile values, the H BL was divided into four classes (28-50, 50-158, 158-685 and 685-1374 m). The categorization of observations and naming of each class (see top strip of Figure 6) was done based on the probability of occurrence of nighttime observations during strong inversion periods (further referred to as 'strong inversion-nighttime'), nighttime observations under weak inversion periods (further referred to as 'weak inversion-nighttime') and daytime observations in each inversion periods (Table S3 in Supporting Information S1). The transition class (H BL = 50-158 m) exhibited a similar probability of occurrence of both strong and weak inversion-nighttime. Since H BL values during daytime of both strong and weak inversion periods did not exhibit any difference, they were merged into a single class, daytime. Only the first class (H BL = 28-50 m), which has the smallest PBL mixing volume and the lowest wind speeds, showed high N CN,10 . This class accommodated 60% of the observations during strong inversion-nighttime, while only 5% of weak inversion-nighttime and 9% of daytime observations fall in this class. Hence, strong inversion-nighttime is characterized by a shallow, stable and static boundary layer with enhanced aerosol concentration close to the ground. Faster winds contributing to dilution and hence low N CN,10 are observed mostly during the weak inversion-nighttime periods and the daytime periods irrespective of the different inversion strength. During strong inversion-nighttime however, the faster winds caused lower N CN,10 as shown in Figure 6a.
As expected, the wind direction does not show any trend during the different inversion conditions (Figure 6b), indicating that the contributions to total aerosol burden were dominated by local city emissions over any long range transported aerosols. Further, the high number concentration resulting from stagnant and local winds during strong inversion-nighttime with high RH (Figure 6c) can facilitate the multiphase processes and heterogeneous reactions leading to formation of secondary aerosols affecting the aerosol hygroscopicity . Hence, such an overall scenario can not only affect the extensive properties of the aerosols (higher aerosol mass burden associated with stagnant and local winds) but can also under favorably high RH conditions modulate the intensive aerosol properties. The above observations suggest that aerosol concentrations drastically increase only under low H BL during nighttime characterized by low wind speeds and high RH caused by the preceding lower solar radiation and therefore, highlights the influence of meteorology at the measurement site on aerosol properties. The dependency observed in this study between H BL and aerosol number and mass concentration is in good agreement with previous studies reporting aerosol properties over Delhi Gani et al., 2019Gani et al., , 2020.  In contrast to previous studies, however, we found no correlation between the air mass origin and aerosol properties for the period we performed the measurements. The clustering of HYSPLIT BTs using a constant height of 1000 m as top of the model showed that majority of the BTs belonged to the NW air mass consisting of longer trajectories and brief periods of MRP and SE air mass with shorter trajectories (Figures S9a-S9c in Supporting Information S1). All three categories of air masses had overlapping paths in the proximity of the megacity. Since SE was observed for a shorter period and does not cover a complete cycle of S levels, only NW and MRP were used for further analysis. Aerosol populations measured under a given air mass (both NW and MRP) did not show consistent diurnal trends of M BC,e , N CN,10 and N CCN during its period of occurrence (see Figures 3 top  strip, 3c and 3d); moreover, a reduction in concentration and changes in κ(S,D a ) were observed for a given air mass when weak inversion conditions overlapped with its occurrence (see Figure 3 top strip, 3 blue graph background, 3c, 3d and 3f). The average aerosol number size distribution was qualitatively similar during NW, MRP and SE with NW alone exhibiting slightly higher number concentrations and the average aerosol hygroscopicity distribution showed similar characteristics during the different air masses ( Figure S9d in Supporting Information S1). Further investigation of the trend in aerosol properties during the daytime and nighttime of the respective air masses showed that the average aerosol properties were different between daytime and nighttime only for NW air masses ( Figure S10 in Supporting Information S1). This, however, cannot be attributed to the influence of the specific air mass, as the differences in aerosol properties between daytime and nighttime was discernible for NW air masses when they coincided with strong inversion periods ( Figure S11a in Supporting Information S1). This was consistent for MRP air masses as well ( Figure S11c in Supporting Information S1). It should be noted that the frequency of strong inversion periods during NW was higher, whereas weak inversion periods were more frequent during MRP (see top strip in Figure 3). As a result, the combined average of strong and weak inversion periods for NW showed differences in characteristics between daytime and nighttime ( Figure S10a in Supporting Information S1) and that for MRP showed similar characteristics during daytime and nighttime ( Figure S10b in Supporting Information S1). From the above analyses, we conclude that the regional air masses, in particular their origin and direction, have no influence on aerosol properties in Delhi during late winter and that they are instead dependent on the meteorological conditions such as wind speed, RH and H BL at the measurement site (further referred to as local meteorology).
Since the megacity is a dominant source of pollution, the different air masses would cause enhanced aerosol concentration only if they originate from another distant and major pollution source. The stubble burning spots in the neighboring states of Punjab and Haryana during the crop burning season, predominantly during Oct and Nov (Kulkarni et al., 2020) are an example for such a pollution source. Fire spot analysis showed that such a major pollution event was absent during the campaign (see Section 2.3) and hence, any air mass originating from outside the city, irrespective of their different locations around the city would have caused dilution effects. This would have contributed to the lower aerosol concentration observed under the more ventilated weak inversion periods (Figures S11b and S11d in Supporting Information S1), in contrast to the strong inversion periods (Figures S11a and S11c in Supporting Information S1) when mixing would have been limited under the shallow PBL.
Air mass back trajectory analysis done using H BL values from WRF as the top of HYSPLIT model ( Figure S12 in Supporting Information S1, see Section 2.3) showed that the air mass path was shorter and that they overlapped during most of the time (see Text S2 in Supporting Information S1), except when there was a south-east branch during daytime and nighttime of weak inversion periods (Figures S12f and S12h in Supporting Information S1). The south-east branch, however, showed no effect on the aerosol properties, because the strong inversion-daytime air masses ( Figure S12g in Supporting Information S1), which lacked this branch of air mass, showed comparable aerosol number and hygroscopicity distribution to that during the weak inversion periods (Figures S12a and S12b in Supporting Information S1). This shows that, irrespective of the origin and direction of air masses, higher H BL and faster winds facilitate mixing, resulting in similar aerosol properties under such meteorological conditions. Hence, the air masses did not have any strong influence on the locally emitted aerosols within Delhi during the measurement period. The above observations lead to the conclusion that the diurnal evolution of H BL is the most important meteorological parameter that locally influences aerosol properties in Delhi, in the absence of any secondary and major source of pollution around the city.
With the exception of the CCN efficiency (Figure 3e), all aerosol and CCN properties exhibited a strong dependence on local meteorological parameters, predominantly the H BL . A strong correlation was observed between H BL and the binned averages of N CN,10 , PM 1 mass concentration (Figure 7; Table 2) and N CCN for three different S levels ( Figure 8; Table 3 top strip of Figures 7 and 8) were selected based on the three most frequently occurring H BL in each class. The width of each bin was determined based on the similarity in trend of the nearest observations. A simple power-law fit provides an effective parameterization for the correlations of binned average concentrations of all the absolute aerosol and CCN properties with H BL over the entire range of height variations. Moreover, the binned first and third quartiles of N CN,10 and PM 1 mass showed similar variation as their respective median (Figure 7). Analogously, N CCN at different S levels showed a simple power-law relationship with H BL (Figure 8). These correlations suggest that at different levels of aerosol loading the relationship of H BL with N CN,10 , PM 1 mass and N CCN remains unaffected.
The resulting fit parameters are summarized in Tables 2 and 3. The parameterization relates the changes in aerosol concentration to the fluctuations in PBL mixing volume. While the mass concentration decreases to the power of 0.4, the number concentration decreases to the power of 0.3 of H BL . It may also be noted that the variations in H BL are tightly coupled with other meteorological parameters (for example T and RH), which in turn can affect underlying processes and mechanisms of secondary aerosol formation.
It is important to note that the parameterization (Equation 2) presented in this study was used to illustrate the effect of H BL on aerosol and CCN loading during late winter under the given local aerosol emissions in Delhi. For effective parameterizations to be used in regional climate models, however, parameterization of large-scale multidimensional data including other meteorological factors using artificial intelligence (Czernecki et al., 2021;Jia et al., 2019) is recommended. Similar comprehensive measurements over larger spatial extent, both vertical and horizontal, representing diverse environmental conditions   Figure 7 and seasons as long-term measurements are important to further validate and prove the relevance of such parameterizations in prognostic modeling.
Summarizing, firstly, the analysis of average aerosol properties during the daytime and nighttime under different inversion conditions for a given air mass ( Figure S11 in Supporting Information S1) indicates that aerosol properties show considerable difference between daytime and nighttime under the strong inversion periods, irrespective of the air mass origin and path. Secondly, the correlation of absolute aerosol and CCN properties with H BL (Figures 7 and 8) indicates that the difference in aerosol and CCN loading between the daytime and nighttime is more pronounced during the strong inversion periods compared to the weak inversion periods. To better understand the influence of low H BL , low wind speed and high RH, characteristic of the strong inversion-nighttime on aerosol properties, CWT analyses were done using the respective air masses during the daytime and nighttime of the different inversion conditions and were compared with the respective average size and hygroscopicity distribution. The results are summarized in Figure 9.
A significant difference in aerosol size and hygroscopicity distribution was observed between the daytime and nighttime during strong inversion periods (Figure 9a). During weak inversion periods, this difference between the daytime and nighttime aerosol properties was either weak or minimal (Figure 9b). It should be noted that except strong inversion-nighttime, all other periods showed qualitatively and quantitatively similar characteristics in the average aerosol number size and hygroscopicity distribution. Comparison with the CWT analysis further shows that, although the daytime and nighttime under strong and weak inversion periods showed similar spatial characteristics of potential emission sources (Figures 9c-9e and 9f), enhanced contributions from source regions were observed during the strong inversion-nighttime. This difference exhibited between strong inversion-nighttime and strong inversion-daytime, in spite of having similar trajectory density characteristics (Figures S13a and S13c in Supporting Information S1), indicates that the distinct meteorological conditions prevalent during this period could be a potential reason for changes in aerosol size and hygroscopicity distribution. Figure 9a shows that relatively low κ values were recorded during strong inversion-nighttime (see also the κ column in the bottom panel of Tables S4 and S5 in Supporting Information S1) and in addition, the MAF across all S levels was lower by 13% during strong inversion-nighttime than that during weak inversion nighttime (see the MAF column in the bottom panel of Tables S4 and S5 in Supporting Information S1). This further indicates that the aerosols contain enhanced proportions of externally mixed weakly-CCN active particles under low H BL . During daytime, when H BL is high irrespective of N CCN (S) Average 20 ,093 0.3 0.79 47 ,913 0.3 0.91 56 ,559 0.3 0.86 Table 3 The Figure 8 the inversion periods, the MAF across all S levels between strong inversion-daytime and weak inversion-daytime showed only a difference of 3%. Complementing the above observations, the average MAF across all S levels during strong inversion periods was up to 18% lower during nighttime than daytime, whereas during weak inversion periods, nighttime showed 9% lower values than daytime.

Parameters and Goodness of Fit of the Correlation Between N CCN (S) Versus H BL Shown in
The low hygroscopicity and greater external mixing observed during strong inversion-nighttime can be attributed to less ventilation within the shallow PBL by the following mechanism. The low H BL during strong inversion-nighttime causes reduced ventilation as indicated by the low wind speeds during this period (Figure 6a). This would prevent the mixing of freshly emitted aerosols with gaseous pollutants within the city and aged and background aerosols from surrounding regions, which facilitates aging of aerosols (Pöschl, 2005;Riemer et al., 2019). Such mixing in megacities has been shown to enhance the internal mixing of aerosols with inorganic components, leading to higher κ values as reported by Gunthe et al. (2011) and Rose et al. (2010). In the absence of such mixing processes under the low H BL during the strong inversion-nighttime, the aerosols thus possess lower hygroscopicity and greater external mixing, as observed in this study.
Consistent with the above observations, the PM 1 composition also showed significant differences between daytime and nighttime during strong inversion periods (top panel of Figures 10a and 10b). The ratios of BC e mass to PM 1 during nighttime under both strong and weak inversion periods (right panel pie chart over gray background in Figures 10a and 10b; also see Figure S14 in Supporting Information S1) are comparable. A major fraction of BC e mass is associated with fossil fuel combustion and this emission scenario remains unchanged during strong as well as weak inversion periods (Figures S15 and S16 in Supporting Information S1). However, despite emission sources remaining similar, the quantitative behavior of the diurnal variation of PM 1 mass is different between the strong and weak inversion periods as shown in Figures 10a and 10b. The PM 1 mass concentrations exhibited a pronounced diurnal variation during the strong inversion periods. In contrast, the well-ventilated weak inversion periods showed reduced PM 1 mass concentrations of all species leading to weaker diurnal variation. This shows that local meteorological conditions could play an important role in determining the accumulation of PM 1 mass within the boundary layer. Furthermore, a lower inorganic mass fraction was observed during strong inversion-nighttime (average f inorg = 17% in Figure 10a right panel; see also Figure S14a in Supporting Information S1) compared to weak inversion-nighttime (average f inorg = 28% in Figure 10b right panel; see also Figure S14b in Supporting Information S1). This indicates that the low H BL during the strong inversion-nighttime, in the absence of any strong winds (Figure 6a) could have restricted the mixing of polluted air with gaseous emissions and aged aerosol influx from the nearby and sub-urban region (e.g., the less polluted state of Haryana during late winter [Lalchandani et al., 2021]), which otherwise would have resulted in a higher inorganic mass fraction Rose et al., 2010). This is consistent with the low κ and MAF observed during the strong inversion-nighttime. On the other hand, during the weak inversion-nighttime H BL is higher and the winds are relatively faster (Figure 6a), which could facilitate better mixing of freshly emitted aerosols with gaseous pollutants within the city (Pöschl, 2005) and aged aerosol influx from nearby regions to produce more internally mixed aerosols (Riemer et al., 2019) leading to higher f inorg . This is consistent with the high κ and MAF during weak inversion-nighttime as already discussed.
Under the cold and humid conditions persisting during the strong inversion-nighttime, feedback to total aerosol load by pollutants emitted due to increased biofuel use for domestic heating during winter (Dec to Jan) was reported by Hakkim et al. (2019). The aerosol load is expected to be consistently high if feedback through cold-weather-induced BB emission were taking place particularly during low H BL conditions. In contrast, we observed pronounced variations of N CN,10 corresponding to the traffic peak hours during nighttime (N CN,10 in Figure 10a bottom-right panel with gray background) indicating that there was no prominent feedback from biofuel use to aerosol burden. Instead, variations in H BL alone appeared to be the dominating factor for increased aerosol mass burden. This variation, although quantitatively unsubstantial, was also observed under weak inversion periods (N CN,10 in Figure 10b, bottom-right panel with gray background).
From the consistent observations in the ACSM, AE33 and size-resolved CCN measurements, we conclude that the shallow nocturnal PBL and smaller mixing volume during strong inversion periods were limiting the internal mixing of fresh city emissions with aged and background aerosols and other gaseous pollutants due to the stagnant conditions caused by lower wind speeds. Hence, we hypothesize that the freshly emitted aerosols possibly undergo rapid chemical transformation and aging under the well mixed layer existing throughout the weak inversion periods, whereas the confined boundary layer during the nighttime of strong inversion periods may inhibit such a transformation and aging due to limited mixing with gaseous and particulate ingredients for multiphase processes. Thus, H BL appears to be the primary factor influencing aerosol accumulation and chemical processing, followed by emissions.
The influence of local meteorology on aerosol properties in Delhi is in contrast with observations from other megacities, for example, Beijing, mainly due to the confined geography of Delhi and the variations in H BL driven by radiative cooling. Studies have shown that aerosol formation and processing over megacities in China are also related to H BL , however, the aerosol properties are not independent of the air mass origin and contributions from regional pollution (Garland et al., 2009;Rose et al., 2011;Zheng et al., 2015). In these cities, a positive feedback to aerosol pollution through aerosol-radiation interaction has also been observed (J. Wang et al., 2014;J. Wu et al., 2019;Tie et al., 2017), which appeared to be negligible during the late winter in Delhi. Moreover, the mechanism of PBL lowering in Delhi during late winter was predominantly radiative thermal inversion, unlike the megacities in China where synoptic meteorological processes like frontal inversion or temperature advection play a major role in modulating the H BL (T. Su, Li et al., 2020;Yu et al., 2020;Zheng et al., 2015).
The CCN efficiencies remained consistent throughout the campaign and the average diurnal CCN efficiencies during both strong and weak inversion periods could be fitted well to a single error function (erf) curve as described by Pöhlker et al. (2016) (Equation 3) (Figure 10c), which has been applied to measurements at the ATTO site .
This implies that CCN efficiencies are not affected by the change in aerosol accumulation and processing under the low H BL . This is because the aerosol size distribution during the strong inversion-nighttime had a pronounced accumulation mode (geometric mean diameter ∼132 nm) and the abundance of larger particles compensated for the effect of reduced hygroscopicity, yielding similar CCN efficiency spectra. This suggests the importance of aerosol particle number and size for effective CCN activation, consistent with previous studies Gunthe et al., 2009Gunthe et al., , 2011. The high CCN efficiency achieved at the highest measured S = 0.69% during both strong and weak inversion periods (70% and 74%,respectively) also verifies the high activated fractions estimated by Arub et al. (2020).
The combined CCN efficiency spectrum fit (Figure 10c) is used with the real-time measured N CN,10 during strong and weak inversion periods to reproduce the corresponding averaged CCN spectra (black dots and circles in Figure 10d), which shows that the sensitivity of N CCN to H BL is captured well using the combined CCN efficiency parameterization. For the reproduced CCN spectra, a modified erf fit (Equation 4) Based on Pöhlker et al. (2016) (red lines in Figure 10d) caused the CCN number concentration to converge against the respective N CN,10 at high S levels during strong and weak inversion periods as shown in Figure 10d (strong inversion periods −N CN,10 = 20,000 cm 3 and weak inversion periods −N CN,10 = 12,300 cm 3 ). This good agreement indicates that at high S levels all internally mixed particles activate, unlike the unrealistic increase in CCN number concentration exceeding the total available aerosol number concentration at higher S levels that would result from the traditional Twomey fit (Twomey, 1959) (gray lines in Figure 10d). Therefore, the modified CCN spectra introduced here give a much better physical representation of CCN number concentrations as a function of S levels than the traditional Twomey power-law fit. This effectively shows that the combined CCN efficiency spectra are sufficient for modeling CCN activity using total aerosol load under varying meteorological conditions. The fit parameters of the modified CCN spectra under strong and weak inversion periods are summarized in the caption of Figure 10. This parametrization is less data intensive and therefore is readily applicable to estimate CCN number concentrations, particularly as a firsthand approximation. It is recommended for further use in numerical modeling studies and requires validation using future long-term measurements under diverse environmental conditions and distinct seasons.

Summary
We conducted a comprehensive investigation of aerosol and CCN properties in Delhi, combining measurements, remote sensing data and model simulations. The study showed that H BL coupled with other meteorological variables plays a key role in modulating extensive and intensive aerosol properties. While aerosol properties such as hygroscopicity and mixing state were associated with the changes in H BL , the aerosol number, CCN number and PM 1 mass concentrations were particularly well explained by a simple function of H BL . We observed two distinct meteorological conditions characterized by high and low solar radiation, resulting in strong and weak radiative thermal inversions and subsequent strong and weak diurnal variations of the modeled H BL . The pronounced diurnal variation of aerosol and CCN concentration, hygroscopicity and mixing state during the strong inversion periods, despite exhibiting similar emissions to weak inversion periods, supports the hypothesis that H BL is the major factor affecting aerosol build up in the local atmosphere of Delhi. We suggest that the distinct meteorological processes in the geographically confined and poorly ventilated Delhi air basin may further enhance the significance of local meteorology in modulating aerosol accumulation and subsequently other aerosol multiphase processes. We present an error function-based parameterization for CCN number concentrations during this campaign, which yielded more realistic N CCN estimates than the widely used Twomey fit.
The air mass back trajectory analysis revealed weak influence of the air mass origin and path on the aerosol properties, since the air masses originated from and traversed over regions that were relatively less polluted than the local surroundings. Only during the crop residue burning season (Oct and Nov), another major pollution source outside the megacity is present. Without this source during the late winter, the surrounding regions do not signif-icantly affect the aerosol properties in Delhi. The local wind direction also had no effect on aerosols, implying that the measured aerosol and CCN properties are representative of the metropolitan area.
The H BL correlated well with the aerosol and CCN number concentrations and the PM 1 mass concentrations. These characteristic aerosol properties can be described as a simple power law function of H BL . The aerosol hygroscopicity, although less affected by H BL was lower during the low H BL experienced during the nighttime of strong inversion periods. During the same period, the maximum activated fractions were also lower, indicating enhanced external mixing compared to the other periods. This change in hygroscopicity, mixing state, and the lower inorganic mass fraction in PM 1 during strong inversion-nighttime under unchanged emissions, can be explained by the hypothesis that mixing of aerosols with gaseous pollutants within the city and aged aerosol influx from outside the megacity was inhibited by low wind speeds under the shallow nocturnal PBL. Therefore, such a major influence of PBL height on aerosol accumulation and processing appears to be specific to Delhi, making it distinct from comparably polluted megacities in China.
Regardless of the change in aerosol number and composition caused by the strong variations in H BL , the CCN efficiencies were unaffected and remained consistent for the whole measurement period. Hence, a single parameterization could model the CCN efficiency spectrum, which captured the variability in CCN concentration under varying meteorological conditions. The CCN efficiency spectrum is well represented by the error function fit (Equation 3; Pöhlker et al., 2016) and the CCN spectra are well represented by the modified Pöhlker et al. (2016) error function fit (Equation 4).
Our study shows that the CCN number variability, which relates to the interaction of aerosols with water vapor in the polluted atmosphere of Delhi primarily depends on the local meteorologically coupled processes during late winter. However, the CCN efficiency remains unaffected due to the ubiquitous nature of accumulation mode particles in the megacity's PBL. These particles in the accumulation mode not only can adversely affect public health (Andersen et al., 2008;Fonceca et al., 2018), but can also deteriorate atmospheric visibility by efficient light scattering and extinction (Jacobson, 2005;Waggoner et al., 1981). The PBL in Delhi contains abundant efficient CCN, which can form fog and haze under the high humidity conditions prevalent during late winter. Moreover, the elevated CCN particles can enter clouds and contribute to cloud microphysical effects (Stocker et al., 2013), implying changes to the aerosol-cloud interaction induced radiative forcing in Delhi. Our findings are expected to improve the representation of aerosols and CCN activity particularly over the Indian region, where accurate measurements are sparse. This will help to formulate control measures for improving regional air quality and mitigating public health impacts.
It is important to note that the interactions between meteorology and aerosol properties presented here are specific to the location and season of the year. Thus, more measurements under varying environmental conditions for longer duration are urgently required for validation and further improvement of the proposed parameterizations. Such measurements during the different seasons of the Indian sub-continent with special attention to the IGP region are recommended. Nevertheless, our measurements highlight the important relation between meteorology and aerosol properties and the effects it may have on fog and haze formation over Delhi. Further, the detailed information and CCN parameterization presented here could enable efficient description of the role of meteorology induced aerosol processes and implications for fog and haze formation over Delhi during this season in aerosol property and process models.

MAF(S)
Maximum activated fraction at given S level N CCN (S) CCN number concentration at given S level, cm −3 N CN,10 Aerosol number concentration (10 to 370 nm) in the measured sample, cm −3 RH Relative humidity, % S Effective supersaturation measured inside the CCNC column, % T Temperature, C κ(S,D a ) Effective particle hygroscopicity corresponding to size D a measured at S

Data Availability Statement
The data associated with key results have been deposited in associated data files in an open research repository. The time series of corrected data from size-resolved CCN experiments (named "CCN.dat"); time series of inverted particle number size distribution (named "SD.dat"); and campaign average of CCN properties (named "AvgCCN.dat") and time series of modeled planetary boundary layer height (named "PBL.dat") are available in NASA Ames format under https://dx.doi.org/10.17617/3.5y .