Chemical source profiles of fine particles for five different sources in Delhi.

Increasing emissions from sources such as construction and burning of biomass from crop residues, roadside and municipal solid waste have led to a rapid increase in the atmospheric concentrations of fine particulate matter (≤2.5 μm; PM2.5) over many Indian cities. Analyses of their chemical profiles are important for receptor models to accurately estimate the contributions from different sources. We have developed chemical source profiles for five important pollutant sources - construction (CON), paved road dust (PRD), roadside biomass burning (RBB), solid waste burning (SWB), and crop residue burning (CPB) - during three intensive campaigns (winter, summer and post-monsoon) in and around Delhi. We obtained chemical characterisations of source profiles incorporating carbonaceous material such as organic carbon (OC) and elemental carbon (EC), water-soluble ions (F-, Cl-, NO2-, NO3-, SO42-, PO43-, Na+ and NH4+), and elements (Mg, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Br, Rb, Sr, Ba, and Pb). CON was dominated by the most abundant elements, K, Si, Fe, Al, and Ca. PRD was also dominated by crustal elements, accounting for 91% of the total analysed elements. RBB, SWB and CPB profiles were dominated by organic matter, which accounted for 94%, 86.2% and 86% of the total PM2.5, respectively. The database of PM emission profiles developed from the sources investigated can be used to assist source apportionment studies for accurate quantification of the causes of air pollution and hence assist governmental bodies in formulating relevant countermeasures.


a b s t r a c t
Increasing emissions from sources such as construction and burning of biomass from crop residues, roadside and municipal solid waste have led to a rapid increase in the atmospheric concentrations of fine particulate matter ( 2.5 mm; PM 2.5 ) over many Indian cities. Analyses of their chemical profiles are important for receptor models to accurately estimate the contributions from different sources. We have developed chemical source profiles for five important pollutant sources -construction (CON), paved road dust (PRD), roadside biomass burning (RBB), solid waste burning (SWB), and crop residue burning (CPB)during three intensive campaigns (winter, summer and post-monsoon) in and around Delhi. We obtained chemical characterisations of source profiles incorporating carbonaceous material such as organic carbon (OC) and elemental carbon (EC), water-soluble ions (F À , Cl À , NO 2 À , NO 3 À , SO 4 2À , PO 4 3À , Na þ and NH 4 þ ), and elements (Mg, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Br, Rb, Sr, Ba, and Pb). CON was dominated by the most abundant elements, K, Si, Fe, Al, and Ca. PRD was also dominated by crustal elements, accounting for 91% of the total analysed elements. RBB, SWB and CPB profiles were dominated

Introduction
Exposure to ambient fine particulate matter ( 2.5 mm; PM 2.5 ) is a major threat to human health Landrigan et al., 2018), linked to nearly 1.1 million premature mortality cases per year across India (Cohen et al., 2017;Conibear et al., 2018). In particular, the northern region of India suffers the most severe particulate matter air pollution (Guo et al., 2017;Jethva et al., 2018;Kumar et al., 2014;Schnell et al., 2018). The underpinning factors that increase pollution levels include increasing levels of population, industrialisation, urbanisation and energy consumption Shukla et al., 2020). Delhi currently has a population of 30.29 million (WPR, 2020) with annual average PM 2.5 concentration levels reaching up to 15-times the WHO guideline of 10 mg m À3 (WHO, 2018). PM 2.5 formed from both natural and anthropogenic sources is considered an important air pollutant, which is responsible for air quality deterioration in cities, such as Delhi (e.g. Kumar et al., 2020). Anthropogenic sources, including transport, industrial processes, power generation, commercial and residential energy use (heating and cooking), diesel generators, construction and road dust emissions contribute to the total ambient concentrations of PM 2.5 directly , and indirectly through gas-to-particle reactions . PM 2.5 can consist of a complex mixture of organic and inorganic chemical constituents (Heal et al., 2012). Chemical composition influences the toxicity of PM 2.5 (Atkinson et al., 2015), and is associated with emission source and atmospheric physical and chemical processing (Kim et al., 2018). Carbonaceous aerosols and water-soluble ionic species are typically the major components of PM 2.5 , together with trace levels of elements which can be useful for source apportionment. Carbonaceous aerosols consist of organic (OC) and elemental carbon (EC). EC is emitted directly from the incomplete combustion of fossil fuels and biomass. OC can either be emitted directly from sources, such as fossil fuel combustion, traffic emissions and biomass burning (primary organic carbon) or produced via a chemical reaction of its precursor gases and/or their condensation into particles as secondary organic carbon. Inorganic ions in the PM 2.5 fraction in urban areas occur mainly as ammonium sulphate ((NH 4 ) 2 SO 4 ), and ammonium nitrate (NH 4 NO 3 ), originating from the neutralisation of sulfuric acid (H 2 SO 4 ) and nitric acid (HNO 3 ), respectively, with ammonia in the atmosphere (Squizzato et al., 2013). Metal compounds are also an important part of PM since they can impact human health and the environment (Izhar et al., 2016), and affect heterogeneous chemical processing (SO 2 oxidation). Emission sources of metal elements in PM 2.5 include a variety of natural and anthropogenic sources, such as crustal and soil dust, construction activities, traffic emissions (exhaust and non-exhaust), industry, municipal waste incineration, and biomass burning (Cheng et al., 2013;Das et al., 2015;Pant and Harrison, 2012). These metals have also been reported to affect numerous cellular oxidation-reduction reactions by influencing key enzymes; an excess of toxic metals in the body can lead to cellular and tissue damage and a variety of adverse impacts on human health (Tchounwou et al., 2012).
An important input to receptor models such as the Chemical Mass Balance (CMB) approach for PM source apportionment is the chemical fingerprint, also referred to as 'source profile', of sources. Source profiles are defined as "fractional mass abundances and uncertainty values of chemical species relative to primary PM emission sources "; these are typically produced for different sources, and can be used as input for the CMB apportionment process, and the calculation and validation of emission inventories . Source profiles used in CMB are typically obtained from direct emission measurements, which do not consider atmospheric processing (Pernigotti et al., 2016). Further, previous studies have suggested that the lack of location/region-specific source profiles in India (Table 1), severely limits the ability of receptor models to apportion sources correctly (Pant and Harrison, 2012). Thus, locally derived aerosol source profiles are vital for source apportionment models. Limited studies have reported PM source profiles in India (Table 1). To date, no studies have reported the source profiles of PM 2.5 in Delhi in detail. Therefore, this work develops an in-depth understanding of profiles for important PM diffused sources in Delhi, based upon direct real-world sampling, and establishes a PM emission profile database that can assist source apportionment studies and allow governmental bodies to formulate relevant countermeasures. The objective of this work is thus to characterise the chemical composition of PM for five different major sources in Delhi, evaluate the similarities and differences among the profiles, and reconstruct mass closures using measured species, calculated enrichment factors and source signatures of PM.

Sampling sites
In order to develop profiles of major pollutant sources in and outside Delhi (Supplementary Information, SI, Section S1), we selected specific sampling locations ( Fig. 1) for each of the five sources considered (Table 2), as discussed below: Construction (CON): Samples were collected at the Indian Institute of Technology Delhi (IITD,28.32 N,77.11E) campus. Sampling was carried out for 8 h (1000e1800h; local time) when construction activity was ongoing. During the sampling period, the construction site was at the initial phase of substructure construction when groundwork activities such as the excavation, filling and soil cutting activities were in progress. Paved road dust (PRD): Samples were collected on the main road (inside IITD, 28.32 N, 77.11E); Ring road (outside IITD,28.33 N,77.11E); Mathura Road at the Central Road Research Institute (CRRI,28.33 N,77.16E);and Okhla area (28.33 N,77.17E). Dust samples were collected from paved roads at the above locations; further details are given in Section 2.2. Sampling was carried out for 30 min periods (ensuring adequate PM mass was collected on the filter in the resuspension chamber (Fig. S1). Roadside biomass burning (RBB): Samples were collected within the IITD campus (28.32 N, 77.11E), where people used biomass for cooking in the evening. Again, sampling was carried out for 30 min periods at the site.   (Kaskaoutis et al., 2014).

Sampling
PM samples were collected during winter (JanuaryeFebruary 2018), summer (April 2018), and post-monsoon seasons (October 2018) ( Table 2). Two Mini-Vol portable air samplers (AirMetrics, M.T., 2006), equipped with a PM 2.5 impactor sampling with a flow rate of 5 L min À1 , were simultaneously located close to the sources at each site (in-line with wind direction and directly into the source dominated volume). It is worth noting that the wind speed during the sampling duration was consistently low (<1 m s À1 ; Fig. S2) and thus the influence of the meteorological conditions on the measured concentrations can be expected to be negligible. Following the US EPA protocols for PM collection on filters (Chow and Watson, 1998), a total of 72 filter samples (52 of PM 2.5 , and 20 of PM 10 , Table 2) were sequentially collected, which included 36 samples on 47 mm Quartz filters and the same number on 47 mm Teflon filters. We calculated PM mass concentrations from the mass collected on each filter by the samplers gravimetrically. Paved road dust sampling followed procedures reported in previous studies (Kong et al., 2012;Patil et al., 2013). Paved road dust samples were collected by sweeping 0.5e1 kg of the dust using a clean plastic dustpan and brush from the surface and storing in zipped polyethylene bags. Samples were dried, sieved (Tyler 400mesh screen) by using the method described by Matawle et al. (2015), and mechanically re-suspended in a laboratory chamber (Fig. S1) and sampled through PM 10 inlets at an average flow rate of 5 L min À1 . The sampler was mounted at the bottom of the resuspension chamber to collect PM 10 samples on the quartz and Teflon filters separately.

Chemical analysis
All the filter samples were analysed for three types of chemical species: (i) carbonaceous material including OC and EC (including 8 sub-fractions); (ii) 8 water-soluble ions (F À , Cl À , NO 2 À , NO  (Cavalli et al., 2010). The carbonaceous material collected was exposed to different thermal conditions, so that the samples were thermally desorbed from the filter medium under an oxygenfree inert helium atmosphere, followed by an oxidising atmosphere containing an oxygen/helium mixture. Firstly, OC was vaporised in an inert atmosphere, and then EC was converted into vapour upon exposure to a higher temperature and an oxidation atmosphere of 98%He and 2%O 2 . Quartz filters were analysed for eight individual carbon fractions, OC1, OC2, OC3 and OC4 at 200, 300, 450 and 650 C, respectively. EC1, EC2, EC3 and EC4 were obtained at 500, 550, 700, and 850 C, respectively (Table S1). After extraction of a fraction of the samples in ultra-pure water, chromatography (Thermo Dionex; Sunnyvale, CA) was used for analysing quartz filters for ionic species. The Teflon filters were analysed for elements from Na to Bi using energy dispersive X-ray fluorescence (ED-XRF) (PANalytical Epsilon 5), which was calibrated using MicroMatter thin-film standards (Watson et al., 1999).

Quality control and assurance
Prior to sampling, quartz filters were baked at 550 C for 12 h to remove any artefacts present. Teflon filters were equilibrated and desiccated at a temperature of 20e25 C and RH of 40e50% in a controlled room before gravimetric analysis; they were weighed before and after sampling using a microbalance (Sartorius CPA2PeF) with a sensitivity of ±1 mg. After sampling, the filters were sealed in aluminium foil bags and stored in a freezer (À20 C) prior to analysis. All analytical procedures were performed in line with strict quality control (QC) and quality assurance (QA) measures. The method detection limits (MDL) for ions were in the range 0.004 mg mL À1 to 0.533 mg mL À1 . The MDL for elements analysed by XRF ranged from 1.3 ng cm À2 to 180.3 ng cm À2 , and the MDL for OC and EC was 2.7 and 0.1 mg cm À2 , respectively. The measurement for each sample was subtracted from the ambient values to obtain the final result which represented the real profile for each source. The concentrations of all species near to source and in ambient condition are summarised in Table S2. These values indicate that the source profiles are not highly affected by ambient concentration (The ambient correction was <2% for all profiles except CON). In order to develop the understanding of the local ambient concentrations around the monitoring sites, source profile measurements were carried out simultaneously both at the upwind and downwind of each site. All source profile samples were corrected by subtracting these upwind measurements of local ambient conditions during the measurements. CON profile was an exception since the site area was large and upwind measurements were not possible. An approach to identify and subtract ambient background concentrations for the CON sample is summarised in SI Section S2.

Chemical mass closure
PM mass reconstruction was performed by taking the sum of mineral dust (MD), organic matter (OM), elemental carbon (EC), inorganic soluble ions, and other metals (except mineral dust metals). The mineral dust (i.e. soil, dust, or mineral) was estimated using Eq. (1): where the numbers correspond to mass adjustment for the principal oxides (SiO2, Al2O3, TiO2, FeO/Fe2O3, CaO, MgO) of the elements quantified (Chan et al., 1997;Marzaccan et al., 2001). OM was calculated as 1.6✕OC (Chow et al., 2015) to express the hydrogen, oxygen, and other elements present in the ambient organic aerosol. OM is estimated from an organic carbon (OC) multiplier that can range from 1.2 to 1.8, based on of OM oxidation and secondary organic aerosol formation (Chow et al., 2015). Since we focus on more than one source here, we used 1.6 (for CON, RBB, SWB, and CPB) and 1.2 (for PRD) in our case to obtain better mass closure. The EC contribution was reported as analysed by the Thermal Optical Method (Section 2.3).
The inorganic water-soluble ions contribution was calculated as the sum of concentrations of all anions and cations (Eq. (2)), which have a significant contribution to each source profiles as follows: Trace elements represent a small percentage of total PM mass concentrations. However, they have a significant impact on human health (Moreno et al., 2004) and the environment, owing to their toxicity and anthropogenic origin (Rees et al., 2004). We estimated the contribution of trace elements by summation of all other metals, except those used in Eq. (1).

Enrichment factors
Enrichment factors (EFs) are concentrations of elements normalised as ratios to another constituent present. In order to further understand the sources of the elements (crustal versus anthropogenic), EFs were calculated for crustal elements in CON and PRD using Eq. (1) based on continental crust concentrations using Al is the reference element Crilley et al., 2017;Hama et al., 2018). Al is used as a reference element due to its high relative abundances in the earth-crust and soil (Cesari et al., 2012) and its conventional use for this purpose. The crustal composition given by Wedepohl (1995), the EFs were calculated using Eq. (3): where C X and C Al represent the mass concentrations of element and aluminium (Al) in the sample and upper continental crust, respectively. EFs were calculated based on the upper continental crust composition (Hama et al., 2018). EF values > 1 typically indicate enrichment of the sample by anthropogenic sources.

Calculation of source marker
Source markers of PM are mostly described by a specific size distribution, suite of elements and ratios of chemical compounds, elements or isotopes (Mitra et al., 2002). To further understand the sources of the different chemical species, relative source signatures were calculated for all developed source profiles by the following equation (Yang et al., 2002;Kong et al., 2012): where Xi is the ith individual chemical species concentration for each profile; (Xi/SX)j is the quotient of the ith individual species divided by the sum of 39 chemical species concentrations for source j; (Xi/SX)min is the minimum quotient of the ith individual species divided by the sum of 39 chemical species concentrations (Chen et al., 2003;Yang et al., 2002). In addition, normalisation was applied according to Mitra et al. (2002) and Kong et al. (2012) to minimise the effect of physical parameters and decrease variation between all the chemical species. Normalised individual chemical species concentrations were calculated by dividing the ith individual chemical species concentration by the sum of ith individual species concentration for all the source profiles (Kong et al., 2012). Table 3 shows the composite profiles of measured chemical species; their standard deviations (SD) were calculated to indicate the representative distributions of chemical abundances and their variabilities . Organic aerosols accounted for a significant fraction of PM mass in all five sources, accounting for~9 4%, 86%, 86.2%, 26% and~18% of PM 2.5 mass for RBB, CPB, SWB, CON and PRD, respectively. The highest percentage of carbonaceous materials were found in combustion source profiles (SWB, RBB, and CPB). In CON, after the carbonaceous species, K (12.1 ± 9.3%), Si (11.5 ± 8.9%) and Na þ (7.3 ± 2.9%) showed the highest percentage among other species. In PRD, Si (13.1 ± 2.1%), Ca (11.2 ± 1.4%) and Al (6.1 ± 0.9%) showed the highest fractions when compared with other species. This is consistent with a previous study for different Indian regions in which they found the highest percentages of Si, Ca, and Al (~40%) in a similar source profile study (Patil et al., 2013;Matawle et al., 2015). Interestingly, we observed the highest percentages for Cl (SWB: 7.1 ± 2.6%; CPB: 4.8 ± 1.8%; RBB: 0.5 ± 0.4%) in source profiles of combustion sources (RBB, SWB, and CPB). The above findings and these abundances are comparable to those from previous studies Ni et al., 2017;Sun et al., 2019b), which also found a similar proportion for Cl À ,0.4 ± 0.2% for RBB and 6.4 ± 4.4% for CPB profiles. The profiles for individual sources are discussed in detail in the following sub-sections.  Table S4 summarises the mean, SD and median values of OC, EC and their fractions for the CON profile. OC and EC concentrations were 7.7 ± 5.9 and 6.5 ± 0.15 mg m À3 , respectively. In addition, OC and EC fractions, OC4 (4.6 ± 0.5 mg m À3 ) and EC4 (3.8 ± 0.4 mg m À3 ), were relatively higher than OC3 (2.7 ± 0.8 mg m À3 ) and EC3 (3.0 ± 0.8 mg m À3 ) ( Table S4). The OC (13%) and EC (18%) levels were higher in our CON profile than those reported by Patil et al. (2013) who found lower OC (~9%) and EC (~1%). This may be as a result of the sampling method used and the analysis protocol applied to the samples. For example, we followed the EUSAAR protocol as opposed to the IMPROVE (Interagency Monitoring of Protected Visual Environment) protocol used by Patil et al. (2013) for OC and EC determination. In addition, Na þ (4.1 ± 0.5 mg m À3 ), NO 3 À (3.9 ± 1.6 mg m À3 ) and SO 4 2À (1.5 ± 1.1 mg m À3 ) were the dominant ions ( Fig. 2; Table S4) found. CON dust is mainly a mixture of construction materials such as cement, grit, and sand, alongside soil and road dust. CON profile was found to be dominated by Na þ , NO 3 À and SO 4 2À which are characteristic of cement dust emissions (Yatkin and Bayram, 2008). In previous studies, SO 4 2À and Ca were found to be dominant for the CON profile (Patil et al., 2013;Sun et al., 2019a). The most abundant elements were K, Si, Ca, Al, Fe and Mg in CON which accounted for 84% of the total analysed elements. The crustal elements Si, Fe, Al, and Ca may be attributed to activities such as aggregate processing and geological dust. This is consistent with earlier studies reporting construction profile studies Patil et al., 2013) with high proportions of crustal elements (~85% of the total analysed elements) in India, and (~88% of the total analysed elements) in China. The trace and heavy elements Cr, Mn, Cu, Zn, As, and Pb, found relatively lower fractions and accounted for 6% of the total analysed elements. These elements play a key role in atmospheric pollution impact on human health (Meng et al., 2013). The above observations allow us to conclude that the CON profile was dominated by OC, EC, Na þ , NO 3 À and SO 4

2À
, and the most abundant elements were K, Si, Fe, Al, and Ca.

Paved road dust
Paved road dust is one of the major sources of particulate matter (particularly PM 10 ) in Delhi (Patil et al., 2013). Road dust consists primarily of coarse-sized particles derived from different sources such as vehicular emissions, brake and tyre wear and mineralogical dust (Pant and Harrison, 2013). It is interesting to observe that the carbonaceous species and their fractions in PRD were dominated by OC (2937.9 ± 1583.3 mg m À3 ), OC4 (1351.0 ± 651.3 mg m À3 ), OC3 (726.3 ± 431.6 mg m À3 ) and EC1 (303.8 ± 228.8 mg m À3 ) ( Table S5).
The abundance of EC and OC in the PRD profile may be as a result of traffic emissions including brake and tyre wear and oil drips as reported by Ho et al. (2003). Previous source profile studies in India reported similar results (Patil et al., 2013;Samiksha et al., 2017). A study in China  found lower levels of OC and EC in the road dust profiles, which may be due to the different chemical composition of road materials in Chinese cities. The PRD profiles developed in the present study were observed to have high levels of water-soluble ions such as PO 4 3À (198.2 ± 175.4 mg m À3 ), SO 4 2À (126.4 ± 96.4 mg m À3 ) and Na þ (76.5 ± 57.0 mgm À3 ). Interestingly, SO 4 2À was found to be abundant, accounting for 23.3% of the total water-soluble ion content in PRD. This may be associated with the chemical reactions of secondary aerosols with crustal elements (for example, calcium); this is consistent with previous studies (Patil et al., 2013;Pervez et al., 2018;Sun et al., 2019a), which also found high levels of SO 4 2À (10e20% of total ions) in road dust profiles. In addition, the elemental composition of PRD is dominated by crustal elements such as Si, Ca, Al, K, Fe, Ti and Mg (Table S5, Fig. 3). The crustal elements accounted for 91% of the total analysed elements. Si and Ca showed the highest fraction in PRD which accounted for 30%, and 27% of the total elements, respectively (Table S5). A spill of building materials such as the cement, sand or mixture of such materials on the construction site (Yu et al., 2020) are expected on the construction sites as an important source of PM. Such processes contribute more to coarse fraction of PM compared with their smaller counterparts. Ca has been widely used as a marker for crustal dust and construction activity (Pant and Harrison, 2013). In Delhi, a higher abundance of Ca-rich dust is attributed to the dust from the Thar Desert . Si and Ca were comparable and dominant, reflecting combined effects of building materials and soil dust . The proportions of Si and Ca is in line with those reported in Matawle et al. (2015) and Samiksha et al. (2017), who found high crustal fractions 43.5e83.8% in Raipur city, India, and Si (~17%) and Ca (~6.5%) of PM 10 mass in Bhopal, India, respectively. It is interesting to note that the crustal road dust fractions show differences between this study and previous work, with the proportions of crustal elements found here being consistently much higher than the above cities. Road dust composition may vary with location, and so obtaining such different source profiles is plausible, and highlights the importance of making regionally localised measurements. In order to explore the origins of these crustal elements in both CON and PRD profiles, we calculated their enrichment factors (Section 4.2). The other elements accounted for only 4.3% of the total elements. Of these, the PM 10 profiles for PRD were found to have notable levels of Pb (3.2 ± 1.6 mg m À3 ). Most Indian cities, including Delhi, phased out leaded petrol after the year 2000 (Singh and Singh, 2006). Since Pb is still persistent in road dust samples implies that this is likely to be due to anthropogenic sources such as coal burning and possibly brake dust Zhang et al., 2014) and ongoing emission of Pb from industrial activities, etc . PRD was dominated by OC, with major OC fractions being OC4 and OC3. We also found high levels of water-soluble ions such as PO 4 3À , SO 4 2À , and Na þ . Overall, PRD was dominated by crustal elements which accounted for 91% of the total analysed elements. Si and Ca showed the highest fractions in PRD, which accounted for 30%, and 27% of total elements in road dust, respectively. Table 3 Composite PM source profiles in percent weight by the mass of emissions from studied sources. We averaged concentrations of profiles collected over the two different seasons Session-specific profiles are presented in SI Table S3. Note that 'BDL' denotes below the detection limit.

Species
Source type (percent weight by mass)

Roadside biomass burning
Despite the ban on RBB, it can still commonly be seen across Delhi during the winter for heating and cooking purposes. This is mostly due to the inefficient collection of roadside waste such as the dried leaves, papers, plastics, or clothes that results in accumulating over a period and is generally burned by the local people as an easy way of clearing it up. The RBB samples collected were dominated by OC (9265.3 ± 5207.7 mg m À3 ). Previous studies  reported that combustion and biomass burning are major sources of organic carbon (Denier van der Gon et al., 2015;Gao et al., 2018;Vicente and Alves, 2018). Diagnostic ratios of chemical species can be used as source indicators (Ni et al., 2017;Sun et al., 2019b). The OC/EC ratio is often applied to distinguish between combustion sources (Han et al., 2016), with (lower temperature) biomass burning associated with higher OC/EC ratios than coal combustion and vehicle exhaust . The OC/EC ratios found in RBB (16.9 ± 6.7), were higher than those reported by Matawle et al. (2014) in Raipur, India and by Zhang et al. (2012) in Guizhou, China, who found lower OC/EC ratios of 6.9 and 10.7 for domestic wood burning. Our results suggest that OC was dominated by primary organic aerosol in the RBB profile. It can be inferred from the wood emissions that OC/EC ratios vary considerably, probably because of variation in the nature of the fuel, the combustion process (temperature, oxygen supply, fuel moisture level) sampling treatment or the analysis protocol e hence the approach adopted here of realworld sampling in situ. ). In the present study, the K þ /K ratio was 0.66, which is consistent with previous studies (Patil et al., 2013;, reporting comparable K þ /K ratio values, of 0.68 and 0.8, respectively. As expected, the most abundant elements from biomass burning were Cl, S, and K, which accounted for 88% of the total elements (Table S6). The abundance of tracers such as, K, Cl and S confirms this source to be that of biomass burning. In fact, K is globally considered as a biomass marker (Pant and Harrison, 2012;Jain et al., 2020). Crustal elements (Si, Mg, and Na) were not found in the RBB profile. Relatively lower proportions of the other elements were found, for 11% of the total elements present. Overall, RBB was dominated by OC which accounted for 58.7% of PM 2.5 mass. The main ions found in the RBB profile were SO 4 2À , NO 2 À , PO 4 3À and Cl À . The most abundant metals found were Cl, S and K which accounted for 88% of the total elements.

Solid waste burning
Open burning of municipal solid waste is an important contributor to PM 2.5 in Delhi and other Indian cities (Patil et al., 2013). OC was found to be the most abundant species (9940.8 ± 5812.8 mg m À3 ). The OC/EC ratio of 43.7 and a K þ /K ratio of 0.86, were higher than those reported in the previous study (OC/ EC ¼ 19.4, and K þ /K ¼ 0.69) for solid waste burning in India . The reason for higher ratios may be due to different sampling methods and site locations. In addition, another reason may be due to the type of waste generated and burnt. While it was not feasible to characterise the waste burned in detail, various types of solid waste material such as plastics, glass, paper, rubber, clothes, cardboard and packaging material can be expected to be burned at the site (Malav et al., 2020). OC/EC ratios also depend on the analysis protocol applied to the collected samples (Chow et al., 2004;Han et al., 2016). Our results suggest that primary organic aerosol also contributed to the OC in the SWB profile. SWB was dominated by Cl À , NH 4 þ , and PO 4 3À ions which accounted for 48%, 20% and 16% of total analysed ions, respectively (Table S7, Fig. 5).
Previous studies have reported similar profiles for SWB Matawle et al., 2014;Patil et al., 2013). In addition, the ratios given in Table S7 are known to be tracers of open solid waste and biomass combustion emissions. SWB's K/EC ratio (0.07) was higher than RBB 0 K/EC ratio (0.03), but lower than CPB's K/EC ratio (0.5). It is interesting to note that the Cl/EC ratio (5.7) for the SWB profile was the highest (Table S7), which was 3e4 times higher than the reported values for crop residue burning (1.1e1.3). The ratio K/OC was found to be similar to that in the RBB profile, but lower (15e20 times) than the CPB profile (Table S7). This might be linked to the emission of K in CPB as it showed the highest levels and could be a better indicator for differentiating these two profiles. The Cl/OC ratio (0.1) was relatively higher than the two biomass burning profiles, RBB and CPB. This indicates that the Cl level showed the highest value in the SWB profile and can be used as a good source marker for SWB profiles. The SWB profile was dominated by three elements such as Cl, S and K which accounted for 87.5%, 4.6% and 1.3%, of total measured elements, respectively (Table S8). Interestingly, Zn (2.9 ± 0.7 mg m À3 ) and Pb (2.5 ± 1.8 mg m À3 ) showed the highest mass concentrations among other trace elements (Table S8). This is consistent with previous source profile studies Matawle et al., 2014), reporting similar levels of Zn and Pb in SWB profiles in India. Zn and Pb are also linked to waste incineration (Duan and Tan, 2013). Rai et al. (2020) found a strong correlation between Zn and Pb in waste incineration in Delhi, thus supporting the high concentrations for Zn and Pb found in our case for the SWB. It can be concluded that the OC was the most abundant species and accounted for 53.8% of measured PM 2.5 mass. OC1 showed the highest, followed by OC2, OC3, and OC4. EC1 showed the highest, followed by EC2, EC3, and EC4, indicating that biomass burning had the highest contribution in the SWB. The Cl/ EC ratio for the SWB profile was the highest among all other diagnostic ratios, suggesting these two ratios (Cl/OC and Cl/EC) may be used as a source indicator for the SWB source profile. The elemental composition of the SWB was dominated by Cl, S, and K.

Crop residue burning
Burning of crop residue in agricultural fields leads to emissions of various air pollutants , which can play a dominant role in atmospheric composition locally, and regionally (Ravindra et al., 2019).
OC is a major species in the CPB source profile (17053.2 ± 6145.1 mg m-3). OC1 and OC4 showed the highest (6427.9 ± 2945.1 mg m À3 ) and the lowest (1452.5 ± 205.4 mg m À3 ) contributions (Table S9). EC concentrations were 700.6 ± 50.3 mg m À3 , with the highest and lowest EC fraction for EC1 (1984.7 ± 443.3 mg m À3 ) and EC4 (47.3 ± 21.4 mg m À3 ), respectively. This result suggests that these fractions have different emission sources. For example, OC1 represents the biomass burning contribution. The OC/EC ratio of 24.3 is higher for crop residue burning than other sources (construction (2.5), road dust (18.8)). This suggests that primary organic aerosol in the CPB profile contributed to the OC. In addition, major ions were Cl À (772.7 ± 201.4 mg m À3 ), PO 4 3À (452.8 ± 13.2 mg m À3 ) and K þ (253.8 ± 74.8 mg m À3 ). Cl À and K þ , which have been reported in previous source apportionment studies (Wang et al., 2020a;Kumar and Raman, 2020), are often used as tracers for crop residue burning emissions. Harvesting followed by the burning of the crop residue in the field during the post-monsoon season in the region of Haryana is common practice since it is an easy and economical way of coping with large quantities of crop residues; this could be one reason why Cle PO 4 3À and K þ are found in the Delhi atmosphere. In addition, K is mostly water-soluble, as indicated by the K þ /K ratio, the value was 0.74 in this study for the CPB profile. This is in agreement with results reported in Ni et al. (2017), who also found a very similar K þ /K ratio of 0.77.  also reported ratios of K þ /K ratios ranging from 0.1 in geological material to 0.9 in vegetative burning profiles (such as crop residue). Table S9 shows the source diagnostic ratios for the CPB profile. These ratios have been adopted in the literature to assess biomass burning contributions (Sun et al., 2019b). It is interesting to note that the K/EC ratios (Table S7) showed comparable values with those reported in Sun et al. (2019b), while they were lower than those measured in laboratory chamber experiments (Ni et al., 2017) values of 0.4 and 2.26, respectively. Thus, the diagnostic source ratios and K found in our studies can be used as good source markers for crop residue burning profiles. Other ratios for the CPB source (Table S7) were largely comparable with RBB and SWB profiles. However, the Cl/EC (1.3, Table S7) ratio for the CPB profile was found to be higher than that for the RBB profile, but lower than that for the SWB profile owing to the Cl level showed the highest value in the SWB profile. Interestingly, based on the Cl/EC ratio, Cl can also be used as a good marker for crop residue burning. The above source diagnostic ratios of chemical species can also be used as source indications and will help to distinguish all three types of biomass burning (RBB, SWB, and CPB) given in the present study. The CPB profile was dominated by Cl (68.2% of total elements) and K (24.8% of total elements) ( Table S9, Fig. 6). The CPB source profile developed shows that Cl and K elements are good source indicators for the vegetative burning profiles. Crop residue burning is a major emitter of K, which accounted for 1.7 ± 0.8% of PM 2.5 mass (Table S9). As expected, the crustal elements (Si, Mg, and Na) were not found in the CPB profile. The trace elements were the lowest among all source profiles, accounting only for 1.3% of total elements. We conclude that the OC is a major species in this profile with a contribution of~53.7% to the PM 2.5 total mass. OC1 showed the highest proportions as opposed to the lowest contributions shown by the OC4 and EC4, indicating that the most abundant source was the burning of biomass, meaning that the CPB profile was very well represented in this study. The CPB profile was also dominated by Cl and K. Fig. 7 shows the results of the chemical mass closure for all source profiles. The PM mass was reconstructed on a source profile basis using the method described in Section 2.5. For CON, the order of fractions was as follows: MD (28.3%) > OM (25.8%) > ions (22.5%) > EC (13.8%) > other metals (9.5%) (Fig. 7). The CON was dominated by MD, OM, water-soluble ions and EC. This result indicated that good mass reconstruction and source markers were obtained for the CON. In addition, these fractions were higher than those reported in a previous study in India, in particular OM (9e12%) and EC (1e2%) (Patil et al., 2013). This may be associated with higher re-suspended soil dust and fuel combustion activities at the site in our study. Moreover, during the sampling, the construction site was at the initial phase of substructure construction including activities such as excavation and digging. The excavator, which includes heavy-duty diesel engines, were operational during the monitoring period, as described in Section 2.1. Other sources such as biomass burning by workers for heating/cooking at sites is another possible source, but such events occurred during the wind direction away from the sampling site (Fig. S3) and thus did not make any direct contributions to measured concentrations. PRD was dominated by the MD (76.3%), followed by OM (18.1%), ions (2.8%), other metals (1.9%) and EC (0.9%) (Fig. 7). The MD proportion is comparable with previous studies, e.g. 58e75% (Matawle et al., 2015), and 65e76% , and relatively higher than the range (47%e56%) reported by Jiang et al. (2018). Interestingly, we found relatively higher OM (~18%) in PRD. This might be related to the traffic-related pollutants, including PAHs and possibly contributed by non-exhaust road materials such as asphalt , coal tar, tyre dust and organic-rich soil. However, further studies are needed to substantiate the non-exhaust component and the possible reasons for relatively high OM levels. In addition, the RBB, SWB, and CPB profiles were dominated by organic matter which accounted for 94.0%, 86.2% and 86.0%, respectively (Fig. 7). The percentages for other components (EC, ions, MD and other metals) accounted for~6.0%, 13.8% and 13.9% of the mass for RBB, SWB and CPB profiles, respectively (Fig. 7). This indicated that the biomass, open solid waste and crop residue burning emissions were dominated by organic compounds. Previous studies reported similar results for OM emitted from roadside biomass, solid waste burning and crop residue burning Barraza et al., 2020;Guo et al., 2017;Pandey et al., 2014). The OM proportion obtained in this study was relatively higher than previous studies, for example, 69e88%  and 80e84% (Matawle et al., 2014) in India. Notably, the proportion of the above fractions of reconstructed PM mass were higher than that found in urban areas (Hama et al., 2018;Galindo et al., 2020;Kubelova et al., 2015). This indicated that good mass reconstruction and source markers were achieved for all developed source profiles. It can be concluded that OM, MD and ions were major components in CON. PRD was dominated by the ions and MD. OM was the most abundant species in all combustion profiles (RBB, SWB, and CPB). It is worth noting that some of the factors used in this section to calculate mass closure are derived from the literature and can be directly representative of the Indian situation. The source profiles developed in this work are representative of the current situation and source characteristics in Delhi and may be applied in quantifying the contribution of sources by receptor modelling, such as that using the CMB model. Fig. 8 shows the EFs for crustal elements in CON and PRD profiles, computed using the approach described in Section 2.6. EFs are used to assess anthropogenic contributions to the elements in CON and PRD profiles. It should be noted that calculation of EFs provides qualitative insight, because of the wide variation of elemental concentration of the upper crust at different locations (Samiksha et al., 2017). Notably, among all elements in both profiles, Ca was highly enriched with EFs of 4.1 and 5.0 for CON and PRD, respectively (Fig. 8). This indicated that both profiles represent construction activity and road dust emissions as both profiles were rich in Ca. The calculated EFs were generally lower and close to one for Si and Mg in both CON and PRD, indicating that crustal material was the main source. The calculated EFs were higher than one for other elements in both profiles, indicating that anthropogenic sources contributed. The EFs for all elements in PRD were higher than their corresponding values for CON. This observation suggests that the anthropogenic source contributions to PRD were higher than those in the CON profile. Table 4 shows the source markers for all source profiles. Values were calculated using the method described in Section 2.7. The six chemical species with the highest ratio values for total relative source profiles are considered representative source markers, and compared with previous studies Chen et al., 2017;Kong et al., 2012;Matawle et al., 2014;Ni et al., 2017;Pervez et al., 2018;Watson et al., 2008;Wang et al., 2018). OC, EC, Na þ , NO 3 À , K and Si were identified as a source marker for CON in the current study (Table 4). This is slightly different from that reported in previous studies (Kong et al., 2012;Matawle et al., 2015;Pervez et al., 2018). This may be related to the sampling location in previous studies. It should be noted that the source markers identified are very varied across previous studies due to the challenging nature of obtaining real construction signatures. OC, Si, Ca, Al, Fe and K were found to be representative chemical species for PRD. The source-specific tracers for PRD were similar to those reported in previous studies (Table 4). This indicates the chemical composition of road dust is dominated by crustal elements (Si, Ca, Al, Fe and K) in cities (Sun et al., 2019a , NH 4 þ and S were identified as source markers for SWB. This is comparable to previous studies Watson et al., 2008). Notably, we identified NH 4 þ , PO 4 3À and EC species in SWB which were not highlighted in previous studies Matawle et al., 2014). This can be associated with the nature and type of materials burnt during sampling as very diverse materials (such as clothes, and plastics) were observed to have accumulated in the landfill and were burning on site. This is a typical behaviour of some cities in India and open burning of solid waste is a major contributor to air pollution in India Nagpure et al., 2015). Chemical species such as OC, Cl, EC, PO 4 3À , K and S were found as source markers for CPB. This is comparable with reported source markers in previous studies Chow et al., 2004;Ni et al., 2017;Wang et al., 2020b). It can be noted that OC and Cl were the major chemical species found in the CPB profile. Crop residue burning in the open field, of which 18% (by weight) is contributed by India (Ravindra et al., 2019). Emissions from crop residue burning in the states of Punjab and Haryana (after harvesting by farmers in post-monsoon season) is frequently advected towards Delhi (Jethva et al., 2018;Ravindra et al., 2019Ravindra et al., , 2020Shivani et al., 2019). New sets of source signatures for five different source profiles of Indian origin have been observed compared with those reported elsewhere (Table 4). Differences in source markers of PM 2.5 and PM 10 emissions from selected source profiles compared with those reported earlier (Table 4), are attributed to local factors and behaviours, highlighting the  importance of development of region-specific source profiles to obtain precise results in receptor modelling. These profiles can be used in receptor model studies, to estimate the contribution of different sources to particulate mass concentrations observed within the Delhi and Haryana regions.

Conclusions and future outlook
We developed chemical source profiles of five principal PM sources in and around Delhi, based on three intensive field campaigns carried out in 2018. The key conclusions drawn from this work are as follows: The construction profile (CON) was dominated by OC, EC, Na þ , NO 3 À and SO 4 2À , and the most abundant elements were K, Si, Fe, Al, Na and Ca. The paved road dust profile (PRD) was dominated by OC, of which the major OC fractions were OC4 and OC3. We also found high levels of water-soluble ions such as PO 4 3À , SO 4 2À and Na þ .
Overall, PRD was dominated by crustal elements and accounted for 91% of the total analysed elements in PRD. Si and Ca showed the highest fraction in PRD which accounted for 30% and 27% of total elements in road dust, respectively. The roadside biomass burning profile (RBB) was dominated by OC which accounted for 58.7% of PM 2.5 mass. The major ions were SO 4 2À , NO 2 À , PO 4 3À and Cl À in the RBB profile. The most abundant metals were Cl, S and K which accounted for 88% of total elements. For the solid waste burning profile (SWB), OC was the most abundant species and accounted for 53.8% of PM 2.5 mass. OC1 showed the highest, followed by OC2, OC3 and OC4. EC1 showed the highest, followed by EC2, EC3, and EC4. The SWB profile also showed high Cl, S and K. For the crop residue burning profile (CPB), OC is a major species with a contribution of 53.7% to the PM 2.5 total mass. OC1 and EC1 showed the highest proportion within the OC and EC component, while OC4 and EC4 showed the lowest. The CPB profile also showed high Cl and K. For the CON source profile, the dominant fractions were MD (28.3%), followed by OM (25.8%), water-soluble ions (22.5%). PRD was dominated by MD (76.3%). In addition, RBB, SWB and CPB profiles were dominated by organic matter which accounted for 94%, 86.2% and 86.0% respectively. In the CON and PRD profiles, enrichment factors (EFs) of Si and Mg were lower or close to one, indicating that these elements originated from crustal material. In contrast, EFs for other elements were significantly higher than one, indicating that crustal material and anthropogenic sources were the main contributors. Six chemical species (OC, EC, Na þ , NO 3 À , K, and Si); (OC, Si, Ca, Al, Fe, and K); (OC, EC, SO 4 2À , NO 3 À , PO 4 3À , and Cl)); (OC, Cl, EC, PO 4 3À , NH 4 þ and S); and (OC, Cl, EC, PO 4 3À , K and S) were identified as source markers for CON, PRD, RBB, SWB and CPB profiles respectively.
The profiles developed and the associated new database are representative of current sources in Delhi. They may be instrumental in accurately estimating the contribution of different sources to PM through receptor modelling studies. Similar studies in different Indian cities and elsewhere are recommended to provide comparable databases that could also establish the extent to which source profiles can vary from city to city. Our source profile for the construction source is representative of substructure (i.e. belowground construction), such as excavation of soil, which may vary for superstructure development (i.e. above-ground construction) activities. Thus, further studies are recommended to develop the source profile covering different phases of the construction process for a holistic assessment.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.