Variation in chemical composition and sources of PM$_{2.5}$ during the COVID-19 lockdown in Delhi

The Government of India (GOI) announced a nationwide lockdown starting 25th March 2020 to contain the spread of COVID-19, leading to an unprecedented decline in anthropogenic activities and in turn improvements in ambient air quality. This is the first study to focus on highly time-resolved chemical speciation and source apportionment of PM$_{2.5}$ to assess the impact of the lockdown and subsequent relaxations on the sources of ambient PM$_{2.5}$ in Delhi, India. The elemental, organic, and black carbon fractions of PM$_{2.5}$ were measured at the IIT Delhi campus from February 2020 to May 2020. We report source apportionment results using positive matrix factorization (PMF) of organic and elemental fractions of PM$_{2.5}$ during the different phases of the lockdown. The resolved sources such as vehicular emissions, domestic coal combustion, and semi-volatile oxygenated organic aerosol (SVOOA) were found to decrease by 96%, 95%, and 86%, respectively, during lockdown phase-1 as compared to pre-lockdown. An unforeseen rise in O$_3$ concentrations with declining NO$_x$ levels was observed, similar to other parts of the globe, leading to the low-volatility oxygenated organic aerosols (LVOOA) increasing to almost double the pre-lockdown concentrations during the last phase of the lockdown. The effect of the lockdown was found to be less pronounced on other resolved sources like secondary chloride, power plants, dust-related, hydrocarbon-like organic aerosols (HOA), and biomass burning related emissions, which were also swayed by the changing meteorological conditions during the four lockdown phases. The results presented in this study provide a basis for future emission control strategies, quantifying the extent to which constraining certain anthropogenic activities can ameliorate the ambient air.

However, in terms of the percentage species across factors, vehicular emissions accounted for 60% of the total Mn content, followed by 33% of Ba, 30% of V, 22% of Zn, 15% of Ca, 13% of K, and 11% of S.
Sulfur and Vanadium is known to occur naturally in crude oil, while pollution control measures have remained focused on reducing sulfur content in fuels, studies have pointed out the use of sulfur in engine oil anti-wear additives (Fitch, 2019). Multiple studies in the past have attributed Mn, Fe, Zn, and Ba to vehicular emissions, brake wear, and engine wear, in particular, recognizing them as abundant trace elements in brake pads and brake lining (Gianini et al., 2012;Grigoratos and Martini, 2015;Rai et al., 2020b;Thorpe and Harrison, 2008). Ti and V have also been attributed to brake and tire wear in some studies in the past (Gerlofs-Nijland et al., 2019). Potassium is noted to be used as an anti-freeze inhibitor and as an additive in engine oils. Also, K is known to be present in all unleaded fuels (Spencer et al., 2006).
Calcium and Chlorine are known to be added to engine lubricants, Ca-compounds serves as a base to neutralize acids, while Cl-based additives act as dispersants, to retain dirt in suspension, to protect the engine (Dyke et al., 2007;Lyyränen et al., 1999;Rudnick, 2017).
In terms of the time variation (Figure S3(a)), this factor is significantly affected by the lockdown with a 96% reduction in average concentration from PLD to eLD-1 ( Figure S6(b)), the time series and the composition pie-chart (Figure 2(b)) show a steady rise in the concentration of this factor, while the factor concentration in phase-4 remains 70% lower than its pre-lockdown value.
As an additional proxy, mobility trends (Google LLC, 2020) (Figure 2(c)) quantifying the percentage change in transit station mobility w.r.t PLD, was compared with the time variation of this factor, and a significant correlation (Pearson R = 0.81) between the two was noted.
In addition to the characteristic species noted above, this factor displayed a sharp diurnal peak coinciding with the morning rush hour and evening rush hour (Figure S4) during PLD and LD-4, when there was comparatively normal traffic load During eLD-1 to LD-3, the vehicular movement has remained extremely restricted; thus, no diurnality in traffic-related emissions was found. The secondary chloride factor seems to have no dependence on the lockdown event as the average concentration of this factor increases marginally from PLD to eLD-1 (3.94 g/m 3 to 4 g/m 3 ) (Figure 2(b)).
From Figure S3(a), secondary chloride time series, we can note that almost all peaks for this factor correspond to a local wind direction in the sector of 302-333 degrees or NW direction. Also, the factor concertation is found to decrease noticeably in eLD-2 (1.25 g/m 3 ) and remain low in LD-3 (1.26 g/m 3 ) (Figure 2(b)). This again seems to have a dependence on the wind direction, starting from the beginning of eLD-2 up till the middle of LD-3, a shift in wind direction towards the southeast, can be noted. However, the wind direction again shifts to the northwest towards the end of phase-3, which marks the rise in the secondary chloride concentrations again.
The possible sources of Cl or Br emissions may include a variety of sources like waste burning or industries, however, to account for the lack of dependence of this factor on the lockdown and following relaxations, the source of HCl/HBr must also remain unaffected/minimally affected by the lockdown.

Zn-K-Br Rich
In terms of % factor mass, this factor is mainly composed of Zn (42%), K (40%), and Fe (11%). However, in terms of % species across factors, these factors contribute to 68% of total Zn, 38% of total Br, and 16% of K (Figure 2(a)).
Multiple studies in the past have attributed a Zn-dominated factor to waste incineration (Gupta et al., 2012;Julander et al., 2014;Parekh et al., 1967;Sweet et al., 1993). These studies have mainly been associated with electronic or municipal waste burning, where a halide catalyzes the volatilization of metals to form metal halogenides, usual metals related to waste incineration in addition to Zn, include K, As, Fe and Pb. While Cl is a more abundant halide, but Vehlow et al. (2003) have discussed how Br may be high in plastics containing flame retardants and, in turn, drive the volatilization of heavy metals like Zn Fe and As. However, Zn and As have also been attributed to iron/steel industries and waste incineration (Duan and Tan, 2013). Also, past studies have attributed Zn-Pb-Cl to industrial emissions (Bullock and Gregory, 1991).
This ambiguity in the published literature regarding tracer for waste incineration/industrial activities has to lead us to define this factor as a Zn-K-Br rich factor only. In terms of the time variation ( Figure S3(a)) the factor in line with total PM2.5 decreases by 42% in eLD-1 w.r.t pre-lockdown, followed by 85%, 54%, and 50% in each of eLD-2 to LD-4 w.r.t PLD concentrations (Figure S6(b)).

Dust Related
The predicted dust-related source profile (Figure 2(a)) is dominated by Si, accounting for 48% of the factor mass, followed by 22% and 20% of the factor mass for Ca and Fe, respectively. In terms of % species across factors, dustrelated source accounts for 84% of total Sr, 80% of total Si, 76% of Ca, 68% each of Ti and V, 55% of Fe and 37% of Mn. Each of the above-noted species has been extensively used as tracers for road dust/crustal elements in multiple studies across the globe (Gupta et al., 2007;Kothai et al., 2011;Rai et al., 2020a;Sharma et al., 2016;Sun et al., 2019).
In terms of the time variation ( Figure S3(a)) of this factor, there seems to be no observable effect of the lockdown on dust-related particulate matter. However, we observe a significant correlation of the factor concentration with ambient temperature (Pearson R= 0.64) (Figure S2(a)) and an inverse correlation with RH (Pearson R= -0.67) ( Figure S2(a)), which is in agreement with several past studies (Csavina et al., 2014;Jayamurugan et al., 2013). Also, during LD-4, we observe an the increase in the average concentration of this factor, from 1.15 g/m 3 in LD-3 to 2.77 g/m 3 in LD-4 ( Figure 2(b)), which may be influenced by multiple meteorological parameters like WD, WS or gust events.

Power Plants
Considering the % factor mass, sulfur solely dominates this factor profile accounting for 93% of the total factor mass.
In terms of the % species across the factors, the power plants factor contributes to 73% of the total sulfur, 35 % of As, and 32% of total Ba (Figure 2(a)).
In this study, the power plants factor (Figure S3(a)) displays a significant correlation with the SOR (Figure 1 Observing the time series of this factor, we do note a 65% decrease in eLD-1 w.r.t the PLD concentrations ( Figure   S6(b)). A recent report from the Power System Operation Corporation (POSOCO, 2020) does indicate a significant reduction (44% reduction in April compared to last year) in the power demand due to the closure or scaled-down operations in almost all industries due to the lockdown, which in turn could lead to temporarily scaled down operations at some power plants. Thus, some order of reduction in source emission can also be partially responsible for the significant drop observed in the factor concentration at the receptor.
However, it would be implausible to attribute the entire reduction to the lockdown alone, as there has been significant variation in the factor concentration within the PLD (-54% to +89% w.r.t PLD average) ( Figure S6(b)), indicating some role of metrological or other transport variables rather than the source emission alone for the variation in the concentration. Again, during eLD-2, the average concentration is found to increase relative to PLD levels; however, the concentration again starts to fall during LD-3 and LD-4, thus advising of some external metrological/transport phenomena affecting the concentration values.

Coal Combustion
The coal combustion factor (Figure 2(a)) is dominated by Lead, Zinc, and Sulfur accounting for 66%, 21%, and 8.5% of the total factor mass, respectively. Considering the % Species across factors, coal combustion is responsible for 87% of Pb, 18% of Se, 16% of As, and 9% of Zn. While coal combustion is found to account for only 0.3% of total sulfur, it is essential to note that the % contribution of this factor to elemental PM2.5 has remained quite low (less than 1.9%) throughout the study period.
As and Se have been widely used as markers for coal combustion (Gupta et al., 2007;Hien et al., 2001;Lee et al., 2008;Sharma et al., 2007). Zn again has been used as a marker for coal combustion in India, due to relatively higher Zn content in Indian coals (Almeida et al., 2006). While commercially available coal has lower Pb content, Negi et al.
(1967) reported the higher concentrations of Pb and Zn in domestic soft coal. It is also important to note that domestic Indian coals have been found to have low sulfur content (less than 0.6% by mass) with an exception to coal deposits in north-eastern India with high sulfur content (Chandra, A, and Chandra, 2004;Sarkar, 2009). Also, past studies have reported Indian power plants to utilize blends of imported and domestic coals supporting the higher sulfate emissions from power plants (Central Electricity Authority, 2012; Chandra, A, and Chandra, 2004).
Evaluating the temporal variation associated with this factor ( Figure S3(a)), we notice that the lockdown implementation brings about a 95% reduction in the average concentration of the coal combustion source, comparing eLD-1 to PLD conditions. With increasing relaxations, the percentage reduction in average concentration w.r.t the PLD falls to 90% in eLD-2 and LD-3 and finally 85% in LD-4 ( Figure S6(b)).
Since the source profiles and supporting literature indicate domestic soft coal burning, the real-world sources may be connected to small scale industrial setups, eateries or household usage of domestic grade coal, and such sources appear to be drastically affected by the lockdown and display only a marginal increase in emissions even with increasing relaxations. Such a variation could possibly stem from the massive outflow of migrant laborers from the NCT region, resulting in the sudden downfall of domestic coal usage for cooking purposes (Roy and Agarwal, 2020).

Source Apportionment of Organic Aerosols (measured using Q-ACSM)
The organic content of the total PM2.5 mass is subjected to source apportionment using PMF. A six-factor solution was found to fit the input data best. The apportioned factors were identified by the mass spectra signatures, their correlation with tracers, and their diurnal behavior Zhang et al., 2005b). The present study further correlates each apportioned factor to corresponding reference factor profiles from Ng et al. (2011). Figure 3 presents the predicted mass spectra for each profile, along with their temporal variation and correlation with external markers. The detailed description of each predicted source profile is as follows:

Semi-Volatile Oxygenated Organic Aerosol (SVOOA)
The factor profile, as seen in Figure 3(a), is characterized by a significant peak at m/z 43, which is a characteristic of less oxidized secondary organic aerosol (Li et al., 2019;Zhu et al., 2018). The diurnal variation (Figure S5) of this factor presents two peaks: early morning (6:00 -9:00 IST) and a weaker peak at midnight, signaling SVOOA concentrations to be affected by photo-oxidation of fresh emissions (morning peak) along with boundary layer height (midnight peak), similar to observations made by Chakraborty et al. (2018). The SVOOA source profile predicted from the PMF analysis was noted to have a Pearson R correlation of 0.93 with the reference SVOOA profile from Ng et al. (2011) The SVOOA factor time series displayed a strong correlation with NO3 (measured using ACSM) (Pearson R = 0.95) (Figure 3(c)), which is in agreement with the trend reported in past studies (DeCarlo et al., 2010;Dzepina et al., 2009;Volkamer et al., 2006), and is attributed to the analogous semi-volatility of SVOOA and nitrate resulting in similar gasparticle partitioning.
The SVOOA factor drops significantly (86%) after the lockdown is implemented; the emissions increase with increased relaxations from eLD-2 to LD-3, however, there is a small drop (18%) in average concentration again from LD-3 to LD-4 ( Figure S6(c)).

Hydrocarbon-like Organic Aerosol (HOA)
Alkyl fragment signatures distinctly mark this factor profile (Figure 3(a)) with prominent contributions from m/z 55 and 57 (Aiken et al., 2009;Ng et al., 2011). The resultant HOA profile has a strong correlation (Pearson R = 0.95) with reference HOA spectra from Ng et al. (2011).
Past studies have found HOA to correlate well with black carbon (BC) (Mohr et al., 2009;Sun et al., 2016). In the present study, we note an excellent correlation between BC and HOA (Pearson R = 0.96) (Figure 3(c)), during the PLD phase; however, post-lockdown the trend of BC and HOA become completely disparate, resulting in a negligible correlation between the two. The significant correlation with BC often is taken as support for the vehicular origin of HOA (DeWitt et al., 2015).
However, from section 3.1.1, we note the vehicular emissions to drop significantly post-lockdown (96%) while, HOA concentrations lower only by 14% post-lockdown. At the same time, it departs from the trend followed by BC ( Figure   2(d)), indicating that during the lockdown, HOA originates from a source other than vehicular emissions.
The initial studies that looked into the deconvolution of HOA from POA (Zhang et al., 2005a) suggested its connection to vehicular origin based on the significant correlation with vehicular markers like NOx and BC and the fine mode of particulate matter corresponding m/z 55 and 57 as compared to m/z 44 which grows larger while aging. Zhang et al. (2005a) also discussed car-chaser and lab-based studies, wherein both diesel emissions and lubricant combustion resulted in HOA like spectra, as heavy oils, lubricants, cooking oils are known to correspond to m/z 55, while mass spectra associated with gasoline and diesel-like fuels displayed a more definite m/z 57 peaks. Hao et al. (2014) also observed appreciable HOA contributions in a low-traffic village setting. They attributed the source to be a combination of industrial, cooking, and biomass burning along with the low contribution from traffic.
Thus, it may be hypothesized that either HOA during the lockdown originates from diesel/lubricant based emissions from sources other than vehicles, like diesel-based generators in industries or cooking-related activities. However, similar to SVOOA, even HOA experiences a sharp reduction towards the end of LD-4.

Biomass Burning Organic Aerosol (BBOA-1 and BBOA-2)
In the present study, we resolve two separate biomass burning related factors (Figure 3(a)). However, BBOA-1 can be categorized as a fresh/ primary emission, with its mass spectra highly correlated to the BBOA profile from Ng et al. and m/z 44, indicating that BBOA-2 is relatively aged. Also, the diurnal variation ( Figure S5) associated with BBOA-1 displays primary emission like behavior, with early morning peaks, while BBOA-2 also displays peaks around noon, which is a characteristic of m/z 44 or CO2 + , indicating the possibility of regionally transported emissions responsible for BBOA-2. BBOA-1 and BBOA-2 are marked by intensified peaks corresponding to m/z 60. Levoglucosan is known to be proportional to C2H4O2 + (a fragment at m/z 60) is extensively used as a marker for biomass burning in AMS based studies (Aiken et al., 2009).
Similar to the biomass burning factor in section 3.1, it is not intuitive to presume an effect of the lockdown on biomass burning. However, comparing the two BBOA factors, we see that BBOA-1 displays higher concentrations in prelockdown, while reduced order (76%) values following lockdown, steadily rising from eLD-1 to LD-3. BBOA-2 displays lower concentrations in the PLD period and steadily rise to a maxima (5.1g/m 3 ) up till LD-3 ( Figure   3(b)).Thus, indicating that while the lockdown leads to a decrease in the primary BBOA emissions (BBOA-1), it in some way enhances the regional transported/aged fraction of BBOA (BBOA-2). However, similar to both SVOOA and HOA, both BBOA factors significantly reduce both in absolute concentration and percentage contribution in LD-4 ( Figure S3(b) & Figure 3(b)).
Interestingly none of the BBOA factors display any positive correlation or similar trend with the satellite-based fire counts (LANCE FIRMS, 2020) (Figure 3(c)), which was seen in the potassium dominated biomass burning factor in section 3.1.
A study by Brown et al. (2016) can aid in the reason behind such discrepancy. They studied the comparison between different biomass burning markers like K + , BC, and Levoglucosan or in turn m/z 60. It was noted that K + and BC are more prominent products in flaming combustion (which is usually captured as fire counts). However, Levoglucosan is a more prominent emission in the case of smoldering combustion (Lee et al., 2010).

Low Volatile Oxygenated Organic Aerosol (LVOOA-1 and LVOOA-2)
LVOOA is addressed as an aged or oxidized aerosol and is majorly marked by a distinct peak of m/z 44 or CO2 + . In this study, we resolve two LVOOA factors (Figure 3 Observing the temporal variation, we see that LVOOA-2 is at a high concentration (9.1 g/m 3 ), which reduces after the lockdown (2g/m 3 ) and steadily rises to a noticeably high concentration in LD-4 (12.3g/m 3 ). On the other hand, LVOOA-1 mostly remains at a lower concentration from PLD to LD-3 ( from 2 g/m 3 to 4 g/m 3 ); however, it rises to For the initial base runs, we examine the solution space, with the number of factors ranging from 3 to 10 with 10 seeds each (number of PMF runs with different pseudorandom starts) for both ACSM and Xact based source apportionment.
In the vicinity of the possible optimal solution (lowest Q-value with no significant change with further increment in the number of factors), the process is repeated with 50 seeds each.
The optimal solution in both cases, was tested for rotational ambiguity by subjecting it to DISP analysis available within the EPA PMF 5.0 module. The DISP assess the largest range of solution factor profile values without an appreciable change in the optimal Q-value. In the DISP method, each species in the factor profiles obtained from the base-run are perturbed about the base value, one species at a time and after each adjustment the PMF run is repeated to calculate a new converged solution such that the change in the Q-value, w.r.t the base run remains less than a predetermined maximum change i.e. dQmax (dQmax = 4,8,15,25). With these changes in the factor profiles, it is possible that the modified factor profile may switch identity when compared to the base run factors i.e. a particular factor from the base run after displacement of certain species may be better correlated to some other base factor than itself, such a case is noted as a factor swap. The EPA PMF module accounts for factor swaps using uncenterd cross correlations between the displaced solution and the base run. In the present study no factor swaps were reported for both the Xact-based and ACSM-based PMF solutions. The largest decrease in Q-value for the Xact-based PMF was found to be 0.0668 with % dQ change as However, the optimal base run solutions was further subjected to variation in f-peak values from -1 to +1 at an increment of 0.1 to explore the rotations of plausible solutions, to evaluate the effect of rotations on the fraction of variance explained by each factor, correlation of factors' mass spectra (MS) with reference mass spectra (for ACSM), correlation of factor time series with external tracers and the mutual correlation between the time series of resolved factors (Bhandari et al., 2020;Rai et al., 2020b;Ulbrich et al., 2009). The range of f-peak rotations was limited to ensure that the rotated solution remains in the vicinity (similar total Q-value) of the initial solution (except the case of a rotationally ambiguous solution) in the solution space.
For the Xact-based source apportionment, no appreciable change in the factor source profiles, and external correlation was observed by varying f-peak; thus, the base solution was chosen as optimal. However, for the ACSM-based source, apportionment rotations were found to improve correlations between factor MS of all factors and reference MS from Ng et al. (2011) and also decrease mutual correlations between the resolved factors. Thus the rotated solution with the best correlations with reference MS and minimal mutual correlations among factors was chosen to be optimal (Bhandari et al., 2020;Rai et al., 2020b).
Further, the effect of random errors were evaluated using the bootstrap (BS) randomized resampling strategy (Brown et al., 2015;Efron, 1979). The bootstrap is implemented using US EPA PMF 5.0 by a random selection of non-overlapping blocks of species measurements from the input data set and creates a new PMF input matrix with the total number of samples equal to the original input matrix, where the user specifies the block size. The PMF code then runs over the new input matrix, and the BS factors are then mapped to the primary factors. The BS factors are assigned to corresponding base factors with which have the highest uncentered correlation values (above a user-specified threshold).
If a particular BS factor doesn't have an uncentered-correlation higher than the threshold with any base factor, then it is considered to be "unmapped." This analysis provides us with a proxy to understand the uncertainty associated with the solution and the apportionment of each species in the resolved factors.
In the present study, the optimal solutions were subjected to 800 BS runs each, with a threshold correlation of 0.8, and more than 764 BS runs were classified as good solutions, having no unmapped factors.

S3. Supplementary Results
The COVID-induced lockdown in India, lasted over a period of 70 days starting 25th March 2020 up to 31st May 2020.
The lockdown progressed in a phased manner, during the first phase of the lockdown (LD-1) starting 25th March un till 14th April marked the strictest phase of the lockdown, with nearly all services and commercial activities completely suspended, with an exception to providers of essential goods and services like hospitals, grocery stores, and pharmacies.
The lockdown was extended with phase-2 (LD-2) lasting up to 3rd April 2020. However, the first set of relaxations were implemented starting 20th April 2020 with allowances to agricultural industries, farming supplies, cargo services. In order to capture the impact of these relaxations in the present study, we focus on eLD-1 and eLD-2, as defined in section 2.1, instead of LD-1 and LD-2. With the start of LD-3 from 4th May 2020, most regions were subdivided into green, orange, and red-zones based on the intensity of the spread of the virus in the region. In green zones, normal movement was restored with public buses running at 50% capacity, only movement with private vehicles was allowed in orange zones, and no relaxations in red zones. LD-4 started on 18th May with further increase in relaxations, the interstate movement was permitted, all categories of small-scale shops were allowed to open, and all industries allowed to restart, private offices were allowed to reopen with 33% staff.
The following sections draw further support to the main manuscript, evaluating the impact of the aforementioned phased relaxations in the lockdown on the variation of sources contributing to ambient PM2.5:

S3.1.1 PM2.5 and its constituents
Analyzing the daily average time variation of the instrument total PM2.5 and the fractional contribution of each constituent (Figure 1(a)), as discussed earlier in section 3, we note that the average PM2.5 values fall by 53.6% from prelockdown to lockdown phase-1. We can also observe that each of the PM2.5 constituents, i.e., elements, sulfate, nitrate and ammonium (SNA), black carbon (BC), and organics, is found to decrease by 31%, 67.7%, 77.2%, and 47.8% respectively, between PLD and lockdown eLD-1 (Figure 1(b)). However, chloride remains almost unaffected or minimally affected, with only a 12% decrease due to the lockdown.
From Figure 1 (a) and Figure 1 (b), we can also note that total PM2.5 along with the elements, SNA, BC, and organics steadily trend back towards their initial concentrations with increasing relaxations in subsequent phases of the lockdown.
However, even in the final phase of the lockdown, total PM2.5 was 33% lower when compared to the PLD values, while SNA, BC, and organics remained 59.8%, 59.5%, and 19.2% lower than their PLD concentrations.
While average chloride concentration was 46% lower in LD-4 as compared to the pre-lockdown phase, it seems to be affected more by metrological conditions rather than the lockdown. Firstly, chloride concentration was only marginally affected in eLD-1; also, it reached its minimum average concentration (1.8 g/m 3 ) in eLD-2 rather than eLD-1( Figure   1 (b)). This coincides with the observation that a significant portion of eLD-2 and beginning of LD-3 is affected by disturbances from the southeast (from the wind direction/wind speed, Figure S2(a)). This dependence of chloride on wind direction or other metrological conditions is concomitant with the observations made for the apportioned secondary chloride factor in section 3.1.3. Also, it is interesting to note that the elemental concentration increased by 65% in LD-4 w.r.t the PLD concentrations. This could possibly be due to the increase in gust/dust storm events during LD-4 leading to the rise in apportioned dust factor, as discussed in section 3.1.5.

S3.1.2 Gaseous pollutants
In Figure 1 Figure S6(a), a significant drop in NO2 and NO concentrations 56% and 90%, respectively, after the lockdown, while NO2 seems to rise back with increasing relaxations, NO concentrations do not rise at the same pace. During LD-4, the average NO2 levels were 22% lower than its pre-lockdown concentration, while for NO, the average concentration was 85% lower than its PLD average. Transportation is believed to be the primary source of both NO and NO2 emissions in Delhi (Tyagi et al., 2016). Similar anomalies have been reported by Nagpure et al. (2013)and Badarinath et al. (2009 in Delhi. However, the disparate behavior could stem from chain reactions initiated by the attack of hydroxyl radical on VOCs and CO catalyzing the conversion of NO to NO2 (Tiwari et al., 2015), outside the photo stationary state of reactions between NO, NO2, and O3 as described by Leighton (1961).
Similarly, CO also follows the intuitive trend of a sudden reduction (by 32%) in LD-4, followed by a gradual increase with increasing relaxations in the lockdown. It is again interesting to note that SO2 remains mostly unaffected by the lockdown based on measurements for the RK Puram station; this is in line with a recent study by Kumari and Toshniwal (2020), where only a slight decrease in SO2 was reported for Delhi, Kumari and co-workers supported this observation with the hypothesis that SO2 in Delhi mainly stems from power plant emissions which have remained unaffected by the lockdown. The negligible effect of transportation on SO2 is also supported by the fact that BS VI fuel with low sulfur content is already in use in Delhi since 2019 (Confederation of Indian Industries (CII) and NITI Aayog, 2018).
In Figure S2(b), we further analyze the effect of the lockdown on ambient air quality based on aerosol neutralization ratio (ANR), as defined by Zhang et al. (2007) The ANR was found to be weakly affected by the lockdown for the PLD phase; the average ANR was 0.83, followed by 0.73, 0.77, 0.78, 0.78 for each of eLD-1 to LD-4. The ANR indicated weak acidity associated with the aerosols, which is slightly intensified by the lockdown, followed by a gradual trend back towards the PLD value.

S3.1.3 Markers for organic sources
While section 3.2 elaborates on the source apportionment of the organic fraction of the particulate matter, in Figure   S2(b), we plot some significant m/z ratios as tracers to known apportioned factors like m/z 43 for semi-volatile OOA (SVOOA) and m/z44 for low volatility OOA (LVOOA), m/z 60 & m/z 73 for Biomass Burning Organic Aerosol (BBOA) and m/z 55 & m/z 57 for Hydrocarbon like Organic Aerosol (HOA). Interestingly, we observe that most of these m/z's decrease significantly during lockdown phase-4, while m/z 44 increases significantly for the same period. Usually, m/z 55 and m/z 57 and the apportioned factor marked by them, i.e., HOA, is found to correlate well with vehicular markers like BC (DeWitt et al., 2015), however, in the present case, the reduction in gaseous pollutants like NO2 and NO, also associated with vehicular origin (Tyagi et al., 2016), due to the lockdown is much more intense than observed in m/z 55 or 57, supporting possible alternate sources for the same during the lockdown as discussed in section 3.2 with the source apportionment of the organic PM2.5.

S3.2 Diurnal Variation of Factors apportioned using PMF
The details on the factors resolved from source apportionment of Xact and ACSM derived species are discussed in section 3.1 and 3.2, respectively. In this section, we present the diurnality associated with each resolved factor and the implications associated with the same. Figure S4 displays the diurnal variation of each factor constituting elemental PM2.5, for each phase from PLD to LD-4.
The vehicular emissions factor shows appreciable diurnal variation during PLD and LD-4, with significant peaks around 6:00 IST to 8:00 IST and also a relatively weaker peak around 18:00 IST to 21:00 IST, both of which correspond to traffic rush hours in the vicinity of the sampling location; also some peaks after midnight may signal toward the heavyduty vehicular movement (for logistics and transportation) during the night. However, it is essential to note that the lockdown severely impacts this typical diurnality associated with this factor, and no characteristic peaks are observed during eLD-1 to LD-4, it is essential to consider that the average factor concentration was noted to drop by 96% as discussed in section 3.1. Also, the restricted vehicular movement during the lockdown is not expected to follow a significant trend, justifying the atypical nature of the diurnal pattern.
The biomass burning factor is found to peak in the early morning around 6:00 IST, indicating the influence of gasparticle partitioning and ambient temperature on the time variation of this factor, which is in line with the findings of past studies (Rai et al., 2020a). Also, the nature of the diurnal profile has remained very similar during all phases of the lockdown, indicating a less pronounced effect of the lockdown on the nature of the source.
The diurnality associated with the secondary chloride factor is discussed in section 3.1.3 and is indicative of dependence of the concentration on gas-particle due to increment in volatility and consequent evaporation after sunrise supporting the significant peak around 6:00 IST followed by the rapid decline in concentration. The Zn-K-Br rich factor is found to have diurnal behavior similar to secondary chloride and could partially contribute to the same as a potential source of HCl and HBr, as discussed in section 3.1.3 and 3.1.4. However, it is vital to take into account that secondary chloride remained mostly unaffected by the lockdown, while the Zn-K-Br rich source was found to drop by 42% in eLD-1.
In the present study, while the lockdown is found to have little or no impact on the dust-related factor in terms of the concentration, however, there is significant variation in the diurnal pattern of the same across different phases of the lockdown. But since this factor is expected to be influenced by a variety of metrological and transport problems, it is difficult to attribute the variation to a single cause.
The power plants factor is found to have a relatively weak diurnal variation; the S-rich factor was noted to correlate well with SOR in section 3.1.5, indicating the factor to be driven by sulfate emissions, the weak diurnality can be attributed to the low volatility associated with sulfate; also the peaks around 8:00 IST indicate the rise in factor concentration due to photo-oxidation from SO2 to SO4 2-, along with gas-particle partitioning of compounds like (NH4)2SO4.
The coal combustion diurnal profile displays a peak around 6:00 IST, similar to secondary chloride signaling towards the role of gas-particle partitioning in giving rise to the observed diurnality.
The diurnal variation profiles of the factors resolved from the source apportionments of organic fraction of NR-PM2.5 are collated in Figure S5. The diurnal variation of SVOOA has been discussed in section 3.2.1 and is found to be dependent on photo-oxidation and boundary layer height. The diurnal profile of HOA is typical of a primary organic aerosol with seeming high dependence on boundary layer height, leading to a rise in concentration when the boundary layer height is lower, i.e., during the night and following a decrease in the morning when the boundary layer rises.
Overall the diurnal profile has no discernable rush hour peaks to attribute the factor to have a vehicular origin.
Similar to HOA, both BBOA-1 and BBOA-2 display diurnal behavior characteristic of primary organic emissions and their dependence on boundary layer height or ambient temperature (Bhandari et al., 2020), however BBOA-2 in addition to the increased concentration during the night, also presents a distinct peak around 6:00 IST -9:00 IST, similar to biomass burning factor from Xact source apportionment, indicating the role of gas-particle partitioning or a primary emission associated with that time.
LVOOA-2 shows a weak diurnal behavior which is characteristic of LVOOA as noted by previous studies (Aiken et al., 2009;Zhu et al., 2018), and is attributed to the low-volatility associated with this factor; the factor observes a small peak around noon due to enhanced photo-oxidation/ photochemical aging of other organic aerosols, in turn, is converted into LVOOA. However, as discussed in section 3.2.4, LVOOA-1 display diurnal behavior similar to primary organic aerosols, and this behavior is further intensified during LD-4 when the highest average concentration corresponding to this factor is recorded. This anomalous behavior indicates the local formation/emission of LVOOA rather than the typical aging route citing the atypical diurnal as well as other factors discussed in section 3.2.4 Figure S1: Linear regression analysis between sum of total PM2.5 as measured by Xact, ACSM and Aethalometer and a co-located Beta Attenuation Monitor (BAM)