Significant Climate Impact of Highly Hygroscopic Atmospheric Aerosols in Delhi, India

Hygroscopicity of aerosol (κchem) is a key factor affecting its direct and indirect climate effects, however, long‐term observation in Delhi is absent. Here we demonstrate an approach to derive κchem from publicly available data sets and validate it (bias of 5%–30%) with long‐term observations in Beijing. Using this approach, we report the first estimation of κchem in Delhi and discuss its climate implications. The bulk‐averaged κchem of aerosols in Delhi is estimated to be 0.42 ± 0.07 during 2016–2018, implying a higher activation ability as cloud condensation nuclei in Delhi compared with Beijing and continental averages worldwide. To activate a 0.1‐μm particle, it averagely requires just a supersaturation of ~0.18% ± 0.015% in Delhi but ~0.3% (Beijing), 0.28%–0.31% (Asia, Africa, and South America) and ~0.22% (Europe and North America). Our results imply that representing κchem of Delhi using Asian/Beijing average may result in a significant underestimation of aerosol climate effects.

Traditionally, the hygroscopic property of aerosol can be described as the enhancement of light extinction/scattering (Wright, 1939) and the growth of geometrical size (Köhler, 1936) due to water uptake. The enhancement factor of aerosol light extinction/scattering coefficient (σ), defined as f (RH) = σ (RH)/ σ (RH ref ), is a common way to describe aerosol hygroscopicity (Brock et al., 2016;Titos et al., 2016). In this definition, σ (RH) and σ (RH ref ) represent the σ at a certain RH and at the reference RH in low/dry humid condition (RH ref ), respectively. Humidified nephelometer system is commonly used to directly measure f (RH) (Covert et al., 1972;Pilat & Charlson, 1966). In term of geometrical growth, Petters and Kreidenweis (2007) introduced the κ-Köhler;hler theory to describe hygroscopic growth of particle diameter using a single parameter (κ), on the basis of the original Köhler theory (Köhler, 1936). This single parameter represents the dependence of hygroscopicity on chemical composition of particles, referred to as κ chem in the following. The κ chem of a multicomponent particle can be calculated as volume-weighted average of each component, that is, the Zdanovskii-Stokes-Robinson rule (Stokes & Robinson, 1966;Zdanovskii, 1948). The parameter κ chem is widely used in laboratorial, field observational, and modelling studies, because it harmonizes the comparisons of hygroscopicity derived from different techniques and environments. The parameter κ chem can be derived from diameter growth factor measured by Hygroscopic Tandem Differential Mobility Analyser (HTDMA) or CCN activity following the κ-Köhler theory Liu et al., 2018;Petters & Kreidenweis, 2007;Wang et al., 2018;Wex et al., 2010) and can also be calculated with measurements of chemical components (Petters & Kreidenweis, 2007). A drawback of HTDMA method is missing the information of coarse particles (Titos et al., 2016), which could be highly hygroscopic (e.g., sea salt) and greatly contribute to hygroscopic growth (Chen, Wild, et al., 2018). The previous closure studies usually show reasonable agreements between HTDMA-derived, CCN-derived, and chemical-derived κ chem values (Hansen et al., 2015;Wu et al., 2016;Yeung et al., 2014). The strong relationship between f (RH), hygroscopicity (κ chem ), particle composition, and CCN activation has been investigated in lots of previous studies since the works of Charlson et al. (1967), Covert et al. (1972), Ervens et al. (2007), and Pilat and Charlson (1966).
Hygroscopicity (κ chem ) measurements have been carried out worldwide during the past two decades, the observational results are compiled in previous works (Bhattu et al., 2016;Kreidenweis & Asa-Awuku, 2014;Swietlicki et al., 2008). Hygroscopicity of aerosols was mostly measured during short-intensive field campaigns due to high financial cost and complicated maintenance of instruments. A few previous longterm observational studies mainly focused on clean environments (Fors et al., 2011;Holmgren et al., 2014;Kammermann et al., 2010) and one long-term study focused on Beijing . To the best of our knowledge, no long-term observation of aerosol κ chem in Delhi and National Capital Region of India was reported. Given the intensive solar radiation and the strong influence of the South Asia monsoon over Indian subcontinent, aerosol hygroscopicity assessment, especially based on long-term observations, is urgent and critical for the studies of radiative forcing and hydrologic cycle.
In this study, we demonstrate an approach for assessing long-term bulk-averaged aerosol hygroscopicity, based on data sets publicly available in a large spatial and temporal coverage. The bulk-averaged κ chem of aerosols in Delhi is reported based on 3-year (2016-2018) ground observations. The corresponding climate implications are also discussed. The approach demonstrated here is also valuable for studies in the other regions where high-quality long-term observations of aerosol hygroscopicity are not available.

Observations
PM 2.5 mass loading is measured by a beta attenuation monitor (BAM-1020, MetOne) at the U.S. Embassy in Delhi during 2016-2018. BAM is a U.S. EPA (Environmental Protection Agency) equivalent reference method for continuous PM 2.5 monitoring and is used for over 80% of the state and local level observations in U.S. (EPA, 2015;Mukherjee & Toohey, 2016). PM 2.5 measured with BAM is not strongly influenced by aerosol associated water (Mukherjee & Toohey, 2016). The instruments are well maintained and calibrated; details of instrument technique, operation, and calibration are given in EPA (2009EPA ( , 2015. Hourly PM 2.5 concentrations in Delhi are available from the AirNow platform (https://www.airnow.gov/) maintained by the U.S. EPA.
The hourly visibility and meteorological conditions are recorded at the Indira Gandhi International Airport (DEL) in Delhi. The hourly visibility is observed by a transmissometer (Drishti, Council of Scientific and Industrial Research-National Aerospace Laboratories; Khare et al., 2018), which is well calibrated and performs well at the airport as reported by India Meteorological Department (http://metnet.imd.gov.in/mausamdocs/16644_F.pdf). RH is calculated as the ratio between water vapor pressure and saturation vapor pressure, which are respectively derived from dew point temperature and temperature using the Magnus formula (World Meteorological Organization (WMO), 2008). As one of the Integrated Surface Database stations, the measurements at DEL are well calibrated and quality controlled according to the regulation of National Oceanic and Atmospheric Administration, National Climatic Data Center (NOAA-NCDC; Neal Lott, 2004). These data sets are available from the NOAA-NCDC website (https://www.ncdc.noaa.gov/).
A limited spatial inhomogeneity is expected in PM 2.5 concentrations and visibility between the U.S. Embassy and DEL. As shown in Figure S1 in the supporting information, the distance between them is only~7 km, which is in the visibility measuring range. Furthermore, there is very slight variation in topography and anthropogenic PM 2.5 emission flux over the region between DEL and the U.S. Embassy in Delhi (Marrapu et al., 2014;Sahu et al., 2011).

Assessment of Aerosol Hygroscopicity
The f (RH) and κ chem are parameters describing aerosol hygroscopicity. Here we briefly describe the approach in this study for deriving f (RH) and κ chem using publicly available long-term data sets. The approach consists of two steps. First, estimate bulk-averaged f (RH) as a function of RH from the data sets of PM 2.5 loading and meteorology (Mukherjee & Toohey, 2016). Second, derive κ chem from the function between f (RH) and RH (Brock et al., 2016;Kuang et al., 2017). We firstly validate the approach by measurements in Beijing, where extensive data sets of field campaigns have been published in recent years. And then the approach is applied to conduct the first estimation of aerosol hygroscopicity in Delhi.
First step, a recent study (Mukherjee & Toohey, 2016) demonstrated a method to derive the bulk-averaged f (RH) based on publicly available data sets: (i) PM 2.5 loading (units: μg/m 3 ) from U.S. Embassy and (ii) RH (unit: %) and visibility (unit: km) from NOAA-NCDC. The total light extinction coefficient can be derived using Koschmieder's equation from visibility (Koschmieder, 1924). As shown in equation (1), the PM 2.5 associated extinction coefficient (σ PM , with units of km −1 ) can be estimated as total σ deducted by air extinction (σ air ) and other factors (σ other ). As recommended by Mukherjee and Toohey (2016), (i) a constant empirical factor σ other = 0.064 km −1 is adopted to represent the influences of gaseous pollutants and coarse particles, and (ii) σ air = 0.056 km −1 is adopted in our study, corresponding to a maximum visibility of 70 km under clear-sky condition (Mukherjee & Toohey, 2016). Therefore, the data set consisting pairs of RH, PM 2.5 , and σ PM can be prepared for analysis. Although the value of σ other is adopted from an estimation for Beijing (Mukherjee & Toohey, 2016), this only introduces uncertainty to κ chem estimation by less than 5% in general (details in Text S1). In the study of Mukherjee and Toohey (2016), Beijing data set during 2009-2014 was prepared and projected to 10 RH bins with 280-320 pairs per bin. The slope between σ PM and PM 2.5 (σ PM /PM 2.5 with units of m 2 /g) can be obtained for each RH bin using least squares fit linear regression, referred to as slope (RH) in the following. The slope at RH ref (median RH at the lowest RH bin) is used to assess dry mass extinction efficiency of PM 2.5 . The ratios between slope (RH ref ) and the slopes of higher RH bins represent the enhancements of light extinction by aerosol liquid water. Finally, the unitless light extinction enhancement factors are derived by normalizing the slopes with slope at RH ref , that is, f (RH) = slope (RH)/slope (RH ref ). In our study and Mukherjee and Toohey (2016), we use median RH in the bin between 30% and 40% as RH ref , since WMO/GAW (2016) recommends a reference RH of 30%-40% for nephelometer and 40% as a maximum RH for the sampling flow. Mukherjee and Toohey (2016) validated this approach with other independent observation-based estimations. The slope at RH ref (3.7 ± 0.4 m 2 /g) is in a good agreement with an independent estimation  using IMPROVE algorithms I (3.2 m 2 /g) and II (4.1 m 2 /g) (Pitchford et al., 2007). The derived f (RH) values are also in a good agreement with the estimations in other studies, details shown in the Figure 6d of Mukherjee and Toohey (2016).
Second step, we further derive κ chem from f (RH), following the works of Brock et al. (2016) and Kuang et al. (2017). Recently, Brock et al. (2016) proposed a single parameter (κ opt, refer to κ value directly derived from optical method/data sets) to describe f (RH), and Kuang et al. (2017) further developed this parameterization with RH ref included, as shown in equation (2). They demonstrated that κ opt can better describe f (RH) than the widely used "gamma" power-law approximation (Kasten, 1969). Following the works of Brock et al. (2016) and Chen et al. (2014), which are based on κ-Köhler;hler and Mie theories, Kuang et al. (2017) proposed a physically based approach to derive the equivalent κ chem from κ opt with R 2 = 0.97. The derived κ chem values (κ f(RH) in Kuang et al. (2017)) agree well (R 2 = 0.77) with measurements in Beijing using HH-TDMA, which is similar to HTDMA with capability of operating under higher RH. The ratio between κ opt and κ chem (R κ ) is influenced by particle number size distribution and chemical composition to some extent. R κ is in a range of 0.58-0.77 (0.69 on average) based on Beijing observations (Kuang et al., 2017). Furthermore, they simplified the influences of particle number size distribution and chemical composition on R κ as a function of Ångström exponent and κ opt and provided a 2-D look-up table for R κ ( Figure S2).
To validate our approach for deriving κ chem from data sets of PM 2.5 loading and meteorology, we estimate a bulk-averaged κ chem of 0.18-0.24 (0.2 on average, considering the variation of R κ ) using the estimated f (RH) values in Beijing 2014, which is adopted from Mukherjee and Toohey (2016). Our results agree well with a long-term observation of κ chem in Beijing 2014 . They conducted a 9-month HTDMA field measurement and reported that the averaged κ chem in Beijing is in a range of 0.14-0.23 for dry particles with diameters of 50-350 nm, details in the Table 2 of Wang et al. (2018). An increase of κ chem as particle size increases was found in their study. This may explain the slight overprediction of κ chem (bias of 0.01-0.04, about 5%-30%) in our approach. Since HTDMA can only measure the κ chem of particles at a certain size (usually smaller than 350 nm), however, our approach estimates a bulk κ chem of the whole PM 2.5 population. These results strongly suggest that the approach we demonstrated here can estimate κ chem value in a reasonable range.
The PM 2.5 and meteorological data sets during 2016-2018 in Delhi are used in this study for the assessment of κ chem . We conduct the analysis using the visibility records in the range of 0-9 km, as recommended by Mukherjee and Toohey (2016). This makes the analysis of f (RH) more reliable, since all visibility with values greater than 10 km is recorded as 10 km. The data pairs with wind speed larger than 6.5 m/s (Kurosaki & Mikami, 2007;Tegen & Fung, 1994 alongside PM 2.5 concentration higher than 500 μg/m 3 are excluded from analysis to minimize the uncertainties induced by dust. Additionally, we exclude the period with RH higher than 90%. This can minimize the uncertainties from noise signals caused by fog, cloud, precipitation, and low accuracy of RH sensor under high RH conditions. We project the data pairs of RH, PM 2.5 , and σ PM to eight RH bins (with borders of 30%, 40%, 50%, 60%, 70%, 75%, 80%, 85%, and 90%) and estimate the bulk-averaged f (RH) of each RH bin where more than 300 pairs of data are available. Then κ opt can be derived from the function between f (RH) and RH (equation (2)), and κ chem can be estimated as R κ is given. We identify the R κ value for Delhi using the 2-D look-up table ( Figure S2; Kuang et al., 2017) and perform Monte Carlo calculation (1 million random samples) to estimate the uncertainties of R κ ( Figure S3a) and κ chem ( Figure S3b). Uncertainty estimation is detailed in Text S2 (long-term Ångström exponent for Delhi refer to Lodhi et al., 2013). Finally, the potential of CCN activation in Delhi is estimated using κ chem and κ-Köhler theory (Petters & Kreidenweis, 2008).

Results and Discussion
As shown in Figure 1, increased PM 2.5 loading and RH can lead to higher light extinction. The σ PM shows a clear increase trend with increase of PM 2.5 and also progressively increases as increase of RH for a given PM 2.5 . This is because hygroscopic growth of particle significantly enhances the light extinction. In order to estimate this enhancement effect, we derive the f (RH) as a function of RH (see Method) as shown in  These results indicate that the urban pollutants may moderate the hygroscopicity of marine aerosols, however, may enhance the hygroscopicity over inland regions, such as Delhi and Beijing. The higher hygroscopicity of aerosols in Delhi may also imply a more severe anthropogenic pollution than Beijing. This is in line with the database of WHO (http://www.who.int/airpollution/data/cities/en/), which shows a twice higher PM 2.5 loading in Delhi compared with Beijing. Furthermore, lots of previous studies (e.g., Titos et al., 2016;Wang et al., 2007;Zhang, Sun, et al., 2015) reported that f (RH = 80%-85%) is inversely proportional to the mass fraction of organic matter (F OM ). Higher f (RH = 80%-85%) in Delhi may indicate a lower F OM compared to Beijing. This is consistent with a recent long-term observational study in Delhi (Sharma et al., 2018), which reported an annual averaged F OM in PM 2.5 is in a range of 15%-20% during 2012-2016 (mass of organic matter is usually calculated as 1.4 times of organic carbon). However, the F OM in Beijing is usually in a range of 20%-40% Huang et al., 2014;Tao et al., 2017;Yang et al., 2017), where more than half of the organic matter originates from secondary organic aerosol (SOA; Hu et al., 2015;Huang et al., 2014;Jimenez et al., 2009). Stronger solar radiation in Delhi may increase photochemical reactions and oxidation of volatile organic compounds, therefore, may enhance SOA formation (Guo et al., 2014;Hu et al., 2019;McFiggans et al., 2019;Zhang, Wang, et al., 2015;Zhu et al., 2011). However, hotter weather in Delhi compared with Beijing could suppress the condensation of semivolatile organic compounds and compensate the enhancement of SOA formation. The lower F OM in Delhi may be due to less SOA, resulting from the competition between the two effects above; however, more observational evidences are required. Moreover, in contrast to the rapid decrease of SO 2 emission in China over the past decade, the significant increase of SO 2 emission in India  could lead to a great formation of highly hygroscopic particulate sulfate. This could be another reason of higher hygroscopicity and larger light extinction enhancement of aerosol in Delhi than in Beijing. The intensive field measurements of physicochemical properties of particulate matter and gaseous pollutants are scarce in Delhi; we highlight the urgency of these observational studies for better understandings of physical and chemical properties of aerosols in Delhi.
To facilitate the assessment of climate impact and comparison with other studies, we derive the κ chem of aerosols in Delhi from f (RH) using equation (2). The annual bulk-averaged κ chem in Delhi is about 0.42 ± 0.07 during 2016-2018. In line with above discussion, this value indicates higher (by~100%) hygroscopicity in Delhi than in Beijing. The long-term HTDMA field observation in Beijing reports an averaged κ chem in the range of 0.14-0.23 for particles within a size range of 50-350 nm . Given the absence of direct hygroscopicity measurements in Delhi, we compare our observation-based estimation with a global model study (Pringle et al., 2010). They show reasonable model results, with deviations between the modelled and observed κ chem values less than 0.05 at 10 out of the 14 locations over the world. In line with our study, their model result of κ chem in Delhi is about 50%-100% higher than the result in Beijing. Our estimated κ chem in Delhi is much higher than averaged values of Asia (0.22), Australia (0.21), S. America (0.17), and Africa (0.15), however much lower than the averaged values of N. Atlantic (0.59) and Southern Ocean (0.92; Pringle et al., 2010). The κ chem in Delhi is much higher (by about 100%) than Asian averages and Beijing observations. As discussed above, this is possibly resulting from less SOA or abundant anthropogenic sulfate aerosol in Delhi, which is also implied by Pringle et al. (2010).

Implication of Finding
Cloud formation exerts a significant impact on the radiative balance of the earth system (indirect radiative forcing) and hydrologic cycle. Cloud droplet number plays a crucial role in determining albedo and lifetime of cloud (Ming et al., 2006) and is very sensitive to κ chem (Reutter et al., 2009). To further investigate the impact of κ chem on aerosol-cloud interaction, we estimate the CCN activation ability of aerosols in Delhi using κ chem following the works of Kreidenweis (2007, 2008) and compare it with the activation ability of other regions over the world and some typical constituents of atmospheric relevance (Figure 3). It is worth noting that κ chem can be size-dependent, bulk-averaged κ chem values are adopted and could introduce uncertainty in the following estimation. Long-term size-resolved particle hygroscopicity observations are required in future studies to quantify this uncertainty. The activation ability of aerosols in Delhi is much higher than some organic matters of atmospheric relevance, for example, oxidized dihexylethyle sebacate, fractionated fulvic acid, fulvic acid, mixture of levoglucosan with succinic and fulvic, and pure levoglucosan (Figure 3a; Svenningsson et al., 2006). However, the activation ability is lower than some typical inorganic matters of atmospheric relevance, for example, ammonium nitrate (Figure 3a). The activation ability of aerosols in Delhi is close to continental-polluted aerosol represented by a mixture of inorganic (70%) and organic matters (30%); detailed information of mixture is given in Petters and Kreidenweis (2007) and Svenningsson et al. (2006). This result may imply that the aerosol in Delhi is a mixture containing majority of inorganic and minority of organic species, and this is consistent with long-term measurements in Delhi (Khare et al., 2018;Sharma et al., 2018). In order to emphasize the importance of climate impacts of aerosols in Delhi (Figure 3b), we compare its activation ability with averaged values of Beijing  and continental averages worldwide (Pringle et al., 2010). A 0.1-μm particle can activate as a cloud droplet under a supersaturation of~0.22% for Europe and North America, about 0.28%-0.31% for Asia, Australia, South America and Africa, and~0.3% for Beijing. However, only a supersaturation of~0.18% ± 0.015% is required to activate 0.1-μm particles in Delhi on average. To activate a smaller particle possessing a diameter of 0.05 μm, it requires a supersaturation of~0.51% ± 0.04% (Delhi),~0.70% (Europe and North America), 0.80%-0.92% (Asia, Australia, South America, and Africa), and~0.85% (Beijing), respectively. Therefore, the CCN activation ability of aerosols in Delhi is much higher than the continental averages and another Asian megacity, Beijing. This indicates a larger impact of aerosols in Delhi on climate and hydrologic cycle, even if under same meteorological conditions and same particle number concentration. Additionally, the frequent influence of monsoon and great PM 2.5 loading in Delhi make its climate impacts more remarkable (~125 μg/m 3 on average during 2016-2018 and~110 μg/m 3 in 2015 as details given in Figure S1; van  Donkelaar et al., 2015). Our results imply that using Asian average or measurements in other Asian megacities (e.g., Beijing) to represent the κ chem in Delhi would lead to significant underestimation of its climate impacts.

Summary
Hygroscopicity of aerosol is an important parameter affecting its climate effects; however, the long-term observation of it in Delhi, one of the biggest cities in the world, is absent. In this study, we demonstrate an approach to derive the hygroscopicity (κ chem ) of aerosol in Delhi from publicly available data sets. This approach is well validated and shows a good agreement (bias of 0.01-0.04, 5%-30%) with long-term observations in Beijing.
We analyze the Delhi observations during 2016-2018 and estimate a long-term bulk-averaged κ chem of 0.42 ± 0.07. This value is much higher (by about 100%) than the κ chem of Beijing as reported from previous modelling and observational studies. This implies the difference in aerosol chemical composition between these two Asian megacities, Delhi, and Beijing. The possible reasons could be higher contribution from anthropogenic sulfate or lower contribution from SOA in Delhi; however, further evidences are still needed from direct measurements. To activate particles of 0.1 μm (0.05 μm) as cloud condensation nuclei, a supersaturation of~0.18% ± 0.015% (0.51% ± 0.04%) is required in Delhi, which is much lower than that in Beijing and the Asian average. Furthermore, the hygroscopicity-induced light extinction enhancement of aerosols in Delhi, that is, f (RH = 80%-85%), is estimated to be in the range of 1.7-2.3, which is much higher than Beijing (1.3-1.5). The higher light extinction enhancement and easier cloud activation imply larger direct and indirect radiative forcing of aerosols in Delhi. These climate effects can be significantly underestimated if a hygroscopicity of Beijing or Asian average is used to represent the condition of Delhi. We highlight the urgency of direct hygroscopicity measurements in Delhi for a deeper understanding of human's influences on cloud formation, climate change, and global hydrologic cycle. The approach we demonstrated in this study is also valuable for estimating aerosol hygroscopicity and its climate effects in other parts of the world where high-quality direct measurements are not available.

Author contributions
Y. C. conceived the study. Y. C. and Y. W. performed the analysis and interpreted the results. All authors discuss the results and cowrite the manuscript.