Samples of PM_2.5 were collected at Jawaharlal Nehru University (JNU), New Delhi (28.539°N, 77.167°E), India on the rooftop (16 m height) of the School of Environmental Sciences (SES) in the year 2019. The map of the sampling site is shown in Figure 1 (source: Google maps and ArcGIS Pro version 1.2.0) and the locations of all the sites (sampling site and CPCB sites) have been marked. JNU is located in the south-west region of Delhi, in an environmentally sensitive area and extends over a large space of natural vegetation in an area of ~800 acres. The sampling site is distant from any principal industrial activities; however, it is surrounded by major roads with high traffic density. Inside the campus the traffic density is relatively very low. The major landform features around Delhi are the Himalayas lying in north-northeast of Delhi approximately at a distance of 160 km, the Thar Desert of Rajasthan lying in the west and the alluvial plains in the south and east of Delhi. Delhi, with a population of more than 16 million people and over 10 million registered vehicles (Rai et al., 2020) remains choked with heavy air pollution, especially in the winter season due to temperature inversion and shallow boundary layer height which entraps a massive amount of pollutants near the ground level (Hazarika et al., 2015).

PM_2.5 sampling during the monsoon (21 June−13 July, 2019; n = 20) and the post-monsoon (11 October−20 November, 2019; n = 40) seasons was done using a Fine Particulate Air Sampler (APM 550, Envirotech, India). The mean flow rate of the sampler was maintained at 1 m³ h⁻¹ (accuracy ±2%). The sampler was operated for 24 h (8:00 a.m.−8:00 a.m.), and the volume of the air filtered during each sample collection was used to deduce the mass concentrations of PM in the ambient atmosphere. Standard protocol for the sample collection and storage until chemical analysis has been followed as prescribed by the Central Pollution Control Board (CPCB), India.

The samples were collected on the hydrophobic PTFE filter (Fluoropore, FHUP04700) of 47 mm diameter. The moisture of the filters (before and after sampling) was removed by desiccating them in a silica based desiccator. The initial and final weight of the filter substrate was determined on a microbalance and the PM_2.5 concentration was calculated using the standard gravimetric method. The meteorological data of temperature, humidity and wind speed was taken for both the seasons from Safdarjung airport (IMD: Indian Meteorological Department) site.

For the quantification of metals in PM_2.5, an Energy Dispersive X-ray fluorescence spectrometry (ED-XRF) was used for its non-destructive mechanism and quick analysis (Khodeir et al., 2012; Moriyama et al., 2014; Shaltout et al., 2017). In the present study, a total of 19 heavy and trace elements (Si, S, Na, Mg, Al, Cl, Ca, K, Ti, V, Cr, Mn, Fe, Ni, Cu, Br, Rb, Zr, and Pb) were analyzed using ED-XRF (PANalytical Epsilon 5 analyzer). The quality control and analysis (QC and QA) have been carried out during the elemental analysis on ED-XRF. The apparatus was set up with standard reference material (NIST SRM 2783). The instrument was pre-calibrated using the XRF standard BRPC3 and it performs a semi quantitative analysis. Semi-quantitative analysis allows the users to compare spectral data from samples in order to ascertain relative elemental concentration between samples. Accordingly, the uncertainty in the concentration of elements on ED-XRF was found to be within 10%. The lower limit of detection (LLD) of each element for ED-XRF is provided in Supplementary Table 1. The description of elemental analysis using ED-XRF and detailed methodology for calculating the concentration of individual elements are provided in the previous literatures (Maciejczyk et al., 2005; Hazarika et al., 2015). In the present study, the concentration of each element in ng/m³ was calculated using the following equation (Hazarika et al., 2015):

where EDXRF value denotes the concentration of element x in ppm, difference in filter weight (pre- and post-sampling) is in milligrams and the volume of air sampled is in m³.

The fine PM has relatively a high residence time, and it can undergo long-range transport in the ambient atmosphere (Maenhaut et al., 2016). Therefore, to understand the origin of PM_2.5 and its long-range transport pattern to the sampling site, the air-mass backward trajectory analysis was done using the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model. Daily GDAS (1 degree, global) meteorological files were accessed to compute the 72-h back trajectories at 500 m and 1,000 m above the ground level (AGL) for each day and night of the sampling campaign. These heights were chosen to represent winds in the boundary layer and to eliminate local land cover and topographic effects. AIRS (Atmospheric Infrared Sounder) dust score data from NASA website (URL: https://earthdata.nasa.gov/labs/worldview/) and MODIS Moderate Resolution Imaging Spectroradiometer fire count data (URL: https://earthdata.nasa.gov/active-fire-data) was overlaid on a true color image to locate the geographical regions responsible for enhancing the aerosol concentrations at the sampling site. AIRS dust score layer shows the level of dust aerosols in the Earth's atmosphere and the areas affected by it. Whereas, the MODIS thermal anomalies data is a fire product in which active fires and other thermal anomalies are identified using thermal anomalies algorithm (Giglio et al., 2003).

The source apportionment analysis of PM_2.5 was carried out only for the post-monsoon season using the Principal Component Analysis (PCA). PCA is a statistical tool which explains the variance of a large amount of data having inter-correlated variables and transforms it into a smaller dataset of independent variables called principal components (PC) (Thurston and Spengler, 1985; Sharma et al., 2016b). In PCA, the factor loadings or PC identify the sources associated with pollution based on the correlation of individual pollutant species with each component. The pollutant species highly correlated with individual PCs indicates the association of that PC with the source emission composition (Johnson et al., 2015). In the present analysis, PCA has been performed with a statistical software SPSS (Version 25) following the Varimax Rotation method. This method maximizes the variance of the squared elements in the column of a factor matrix. In PCA, the first PC explains the most significant fraction of the original variables, while the second PC estimates a reduced fraction of the original variable in comparison with the first PC and so on (Sousa et al., 2007). PCA begins by normalizing the set of variables as Z_ij using equation (1), so that the variance of this set of variables is unity.

Where C_ij is the concentration of jth species in the ith sample; Cj¯ and δ_j are the average concentration and standard deviation of that species j. The fundamental operation of PCA can be expressed by equation (2) indicating that it splits the data matrix into two matrices G_ik (factor loading) and H_kj (factor score), as shown below:

Here, i, j, and k represent the index for sample, species and factors, respectively. E_ij represents the residual matrix. The two vectors G and E are unknown in the factor analysis (FA) and are obtained by assuming various covariance relationships between the vectors H and E and finally Varimax rotation of matrix is applied to minimize the datasets of elements having high loading factor (Kumar et al., 2001). As the factor load of the variable increases, the identification of the possible source of components also increases (Henry, 2003). The detailed methodology for air pollutant's source identification and apportionment using PCA is provided in Chavent et al. (2009). In the present study PCA was carried out utilizing elemental concentration data of 40 samples for the post-monsoon season. Numerous other studies have employed PCA for source apportionment using limited data points viz; Hazarika et al. (2015) utilized data from 12 samples for each season (summer, winter and monsoon) to perform PCA analysis, Liu et al. (2018) utilized data from 15 day samples, similarly Zhang et al. (2019) and Kanellopoulos et al. (2020) also utilized limited data set to carry out source apportionment analysis using PCA.

In total, 60 samples were collected to ascertain the seasonal variation in PM_2.5 concentrations during monsoon (n = 20 samples) and post-monsoon (n = 40 samples) seasons. Average mass concentrations were found to be 42.2 ± 15.5 μg m⁻³ (range: 21.8–72.5 μg m⁻³) and 121.4 ± 53.6 μg m⁻³ (range: 45.9–298.1 μg m⁻³) in monsoon and post-monsoon seasons, respectively. The PM_2.5 data monitored by the Central Pollution Control Board (CPCB) at four nearby sites (Aya Nagar, IGI Airport, PUSA and Sri Aurobindo Marg) was also taken into account to assess the spatial heterogeneity in PM_2.5 mass concentrations. Figure 2 shows the temporal variability of PM_2.5 concentrations at the sampling site along with the mean concentration for the aforementioned nearby CPCB sites. The average PM_2.5 concentration at the sampling site i.e., JNU is lower than the other adjoining areas particularly in the post-monsoon season. The temporal variation of PM_2.5 mass concentrations for all sites during both the seasons are provided in Supplementary Figure 1. The average PM_2.5 mass concentrations at JNU and that integrated for the four CPCB sites were 42.2 μg m⁻³ and 44.3 μg m⁻³ in the monsoon season and 121.4 μg m⁻³ and 183.3 μg m⁻³ in the post-monsoon season. This can be attributed to the vast tract of natural vegetation and the low traffic density in the University campus site (JNU).

Figure 3 shows the daily variation of meteorological parameters (temperature, humidity and wind speed) during both the sampling seasons. It is apparent from the figure that the monsoon season was warmer than the post-monsoon season. The wind speed in the monsoon season was also higher than the post-monsoon season. However, the relative humidity appeared to be higher in the post-monsoon season as compared to the June–July period of the monsoon season. It is evident from Figure 2 that the mean concentration of PM_2.5 was substantially higher in the post-monsoon season than the monsoon season. The previous studies conducted in Delhi assessing seasonal variations in PM_2.5 also showed an increased concentration in post-monsoon and winter season with respect to the monsoon season (Mandal et al., 2014; Gopalaswami, 2016; Panda et al., 2016; Sharma and Mandal, 2017; Jain et al., 2020). The most plausible explanation for lower PM_2.5 mass concentrations during the monsoon season relates to the high ventilation and dispersion of pollutants as well as occurrence of frequent precipitation leading to washout of the air pollutants during the monsoon (Jain et al., 2020). In the post-monsoon season, activities such as stubble (agriculture residues) burning in the fields of Haryana, Punjab, and western part of Uttar Pradesh, firecrackers burst during Diwali festival along with prevailing meteorological conditions such as minimal wind speed, shallow boundary layer height, and geographical settings (Himalayan range in the north and Deccan Plateau in the south) results into entrapment of a large amount of PM_2.5 near the ground level over the study region (Perrino et al., 2011; Rai et al., 2020). Kulshrestha and Kumar (2014) in their review report highlighted the need and significance of trajectory analysis for identifying the sources of PM pollution in South Asia.

The results from HYSPLIT back trajectory analysis indicating for the long-range transport of air masses at the sampling site are illustrated in Figures 4A–E. Air-mass back trajectories for a particular season do not look very different at two different altitudes, i.e., 500 and 1,000 m. During the monsoon season, the 72-h back trajectory indicated that the air masses traveled longer distances due to higher wind speed (Figures 4A,B). Most of the air masses in this period advanced to the receptor site from Indo-Gangetic Plain (IGP), Rajasthan, Gujarat, Pakistan, and the northern Arabian Sea, along with traces of air masses from Middle-East and the northern Bay of Bengal. The AIRS dust score overlaid on a true color image from MODIS shown for 24 June 2019 (Figure 4C) indicated for a strong dust layers in the atmosphere over Pakistan, Middle East and parts of the Arabian Sea which indicates that maybe the dust particles from these places would have transported over the receptor site. However, clean air mass from the Arabian Sea and Bay of Bengal would have diluted the overall dust load impact from these places at the receptor site. Sharma and Mandal (2017) made similar observations with the air-mass trajectory analysis during the monsoon season of 2013 over Delhi. Sharma et al. (2010) in their study of SO₂ variation over Delhi also highlighted the long-range transport of air-masses from sources pointing toward western and southwestern regions in the monsoon season.

In the post-monsoon season (Figures 4D,E), the back-trajectory showed that the air masses have covered shorter distance, attributable to the lower wind speed, before arriving at the receptor site. The air masses in this period were approaching the receptor site mainly from states of Haryana, Punjab, IGP region, Rajasthan, and Pakistan. The MODIS fire count data plot (Figure 4F) for 04 November 2019 exhibits an intensive stubble burning in Punjab, Haryana and western part Uttar Pradesh indicating that high concentration of PM (and gaseous pollutants) from biomass burning emission could severely impact the air quality at the receptor site. Moreover, emissions from numerous industries operating in Delhi would also contribute to PM loading at the receptor site. Therefore, higher concentration of PM_2.5 in the post-monsoon season can be attributed to the combined effect of lower wind speed, shallower boundary layer height, and the high incidence of stubble burning which increases the aerosol load over Delhi (Kanawade et al., 2020). For post-monsoon season similar results mentioning stable meteorological conditions leading to accumulation of local and transboundary pollutants have been mentioned in the previous studies conducted in Delhi (Tiwari et al., 2013; Panda et al., 2016; Cusworth et al., 2018; Kulkarni et al., 2020; Nair et al., 2020). In sharp contrast, the precipitation and high wind speed lead to lowering the concentration of atmospheric particles during the monsoon season (Sharma et al., 2016a). It is worth mentioning that the aerosols sampling was performed for 60 days, spread over both the seasons, and we found that the PM concentrations exceeded the NAAQS (National Ambient Air Quality Standard) standard value for more than 70% of the days. This is coherent with the other research work carried out over the Delhi (Pachauri et al., 2013; Tiwari et al., 2013; Sahu and Kota, 2016; Das et al., 2020; Jain et al., 2020).

Figure 5 shows the percentage elemental composition of fine particulates for both seasons [(a) monsoon and (b) post-monsoon]. The daily variations in the percentage elemental composition of PM_2.5 for both the seasons are shown in Supplementary Figure 2. The temporal variations in concentration of the analyzed elements associated with PM_2.5 for both the seasons are depicted in Supplementary Figures 3A–C. The result of the current study showed that the average concentrations of elements were found in the decreasing order of Si>Al>Na>Mg>S>Ca>Cl>K>Fe for the monsoon season and S>Al>Na>Si>Cl>K>Mg>Ca>Fe for the post-monsoon season. The presence of Si, Al, Na, Mg, S, Ca, Fe and K as major elements associated with PM_2.5 is in agreement with other studies conducted in Delhi by Jain et al. (2017) and Sharma and Mandal (2017). Hazarika et al. (2015) also observed Na, Ca, Si and K as abundant elements in PM_2.5 followed by Ni, Cu and Pb. However, in the present study, the mean elemental concentration of individual elements was comparatively less than the previous studies conducted in Delhi. The relative contribution of elements associated with crustal or natural origin such as Si, Al, Na and Mg (Pipal et al., 2014; Ali et al., 2019; Rai et al., 2020) accounted for 87% of the elemental composition of PM_2.5 in the monsoon season, whereas in the post-monsoon season these elements accounted only for 57% (Figure 5). The element which contributed the most to the fine particulate was Silicon (Si) 39% in the monsoon season and Sulfur (S) 30% in the post-monsoon season. The high concentration of Si could be attributable to uplifted mineral dust contribution, whereas elevated S content may have contribution from coal combustion and agricultural-residue (biomass) burning (Sternbeck et al., 2002; Perrino et al., 2011). The sources of soil/road dust in the atmosphere include transboundary transport from deserts or entrainment from paved or unpaved roads, construction activities, and agricultural practices (Kulshrestha et al., 2009; Tiwari et al., 2013).

Moreover, the contribution of Potassium (K), Chlorine (Cl), and Lead (Pb) have been found to increase in the post-monsoon season due to different reasons. Higher concentration of K can be attributed to local biomass burning (in addition to some fraction from upper continental crust) for space heating and agricultural residue burning (Liu et al., 2018; Das et al., 2020); Cl can have contributions from lubricants, diesel fuels, coal combustion, biomass burning, and plastic and paper burning (Singhai et al., 2017; Chang et al., 2018); Pb to ore and metal processing, lead-acid battery production/recycling as well as waste incineration (Bukowiecki et al., 2009; Kothai et al., 2011). Summing up, the post-monsoon season witnessed a substantial increase in PM_2.5 concentrations predominantly due to the anthropogenic emissions (as suggested by elevated levels of S, K, Cl, Pb) with a relatively low contribution of mineral dust as compared to the scenario in monsoon season. Similar remarks have been noticed before in the previous research works (Khodeir et al., 2012; Das et al., 2020; Jain et al., 2020; Rai et al., 2020). Within the trace elements, three carcinogenic heavy metals were identified, i.e., Ni, Cr and Pb, that may pose substantial risk to humans (Liu et al., 2015). The estimated concentration of these elements is depicted in Supplementary Figure 3B. However, their concentration in this study was found well within the limits prescribed by the World Health Organization (WHO). The comprehensive details on source apportionment of PM_2.5 based on associated metals profile are discussed in the following section.

To distinguish the possible sources of fine fraction particulate matter, the principal component analysis (PCA) was carried out. Based on the eigenvalues >1, PCA segregates the data into several clusters known as principal components (PC), which also indicate the contribution of each dependent variable (which in this study is the element concentration) in terms of factor loadings. The PCA was performed using the data set consisting of 19 elemental species in 40 PM_2.5 samples (during the post-monsoon season) collected at JNU site in Delhi. Table 1 shows the output of PCA analysis using SPSS (IBM, SPSS, version 25) software for the post-monsoon season. The correlation matrices for all 19 elements during both the seasons are shown in Supplementary Tables 2, 3. Figure 6 shows the possible sources of PM_2.5 pollutions in the post-monsoon season as identified using PCA. The total variance explained for the sources of PM_2.5 during post-monsoon season (81.7%) using PCA are provided in Supplementary Table 4.

During the post-monsoon season, the PCA analysis revealed three factors, accounting for 81.7% of the total variance. The first PC explained 33.7% of the overall variance with an eigenvalue of 6.39 with high loadings of Na, Mg, Al, K, Ca, Ti, V and Si. Perrino et al. (2011) and Sharma et al. (2016a) reported that crustal dust is the major source of Si, Na, Mg, Ca, Ti, and Al in particulate matter. Therefore, this factor can be well-associated with crustal suspension. The % contribution of these elements (Figure 5) to fine particulate matter was relatively low in the post-monsoon season with respect to the monsoon season possibly due to low wind speed and boundary layer height leading to a lower resuspension of mineral dust in the ambient atmosphere. The Potassium (K) content, however, increased substantially owing to stubble/agriculture residue burning in the surrounding regions [as K is a tracer of biomass burning (Khare and Baruah, 2010)]. Similar results have been perceived in several other studies over Delhi (i.e., Jain et al., 2020; Rai et al., 2020). Therefore, this factor should be identified as crustal suspension and biomass burning. The second component with eigenvalue 5.21 represented 27.4% of the total variation with high loadings of S, Mn, Ni, Fe, and Pb. While Ni is related to vehicular emission especially with heavy diesel based vehicles (Khanna, 2015; Das et al., 2020), the elements such as Ni, Pb, Fe, and Mn are well associated with vehicular emissions mixed with road dust resuspension (Sternbeck et al., 2002; Khanna, 2015; Liu et al., 2015; Pant et al., 2015). The high concentration of S could be associated with coal combustion and secondary aerosols formation (Gummeneni et al., 2011). Therefore, this factor can be attributed to the vehicular emission, road dust resuspension, and secondary aerosols formation.

The third factor explains 20.6% of the overall variance with an eigenvalue of 3.9 and is strongly correlated with the elements Br, Cl, Cu, Rb, and Zr. These elements can have origin from industrial activities and solid-waste burning (Khodeir et al., 2012). Br in previous studies has been identified as the element associated with industrial emissions probably from various drug and chemical manufacturing industries (Kothai et al., 2011). Halogens can also be produced from solid-waste burning. In Delhi, most of the incineration of dumped waste occurs in three locations viz, Okhla (also a major industrial area in South Delhi), Bhalswa (North Delhi) and Ghazipur (East Delhi) (Ghosh et al., 2019). Moreover, in post-monsoon and winter season, brick kilns have been reported to function in areas encompassing Delhi, and they can possibly elevate the Cl concentrations in the atmosphere. These brick kilns are operating in large numbers around Delhi-NCR (National Capital Region) due to heavy demand in the infrastructure sector (Rai et al., 2020). These traditional kilns use coal and biomass as fuel to bake bricks which could emit a large amount of Cl and Br (Rai et al., 2020). Cu, Rb and Zr are associated with industrial emissions from different electroplating and other alloy manufacturing industries (Hazarika et al., 2015). Chowdhury et al. (2017) in their study in Delhi have indicated major point sources of emissions from coal based thermal power plants as well as major industrial areas located in central, northern, and eastern part of Delhi. The total number of coal based thermal power plants in Delhi-NCR is 6. These power plants and numerous industrial areas emit large amount of PM and other pollutants (Mittal et al., 2012). Therefore, this factor can be represented as coal combustion, industrial emissions, and waste incineration.