Trends and Variability of PM2.5 at Different Time Scales over Delhi: Long-term Analysis 2007–2021

Received: April 30, 2022
Revised: July 15, 2022
Accepted: August 6, 2022

 Copyright The Author(s). This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are cited.

Cite this article:

Chetna, Dhaka, S.K., Longiany, G., Panwar, V., Kumar, V., Malik, S., Rao, A.S., Singh, N., Dimri, A.P., Matsumi, Y., Nakayama, T., Hayashida, S. (2023). Trends and Variability of PM_2.5 at Different Time Scales over Delhi: Long-term Analysis 2007–2021. Aerosol Air Qual. Res. 23, 220191. https://doi.org/10.4209/aaqr.220191

HIGHLIGHTS

• PM_2.5 showed an overall small declining trend over Delhi during 2007–2021.
• Planetary boundary layer height (PBLH) and wind speed showed a decreasing trend.
• Temperature and rainfall showed no trend, pressure and RH showed a rising trend.
• Relative humidity displayed an inverted U-shape relationship with PM_2.5.
• Stubble burning contributes significantly to Delhi's PM_2.5 levels during Oct.–Nov.

Keywords: Long-term trend analysis, Seasonal variation, Theil-Sen approach, Particulate matter, Stubble crop burning

1 INTRODUCTION

PM_2.5 are highly respirable fine particles or liquid droplets, with diverse chemical composition, suspended in air with aerodynamic diameter less than or equal to 2.5 microns. Fine particulate pollution is responsible for environmental degradation, poor air quality (Fuzzi et al., 2015), an array of health disorders, mortality, and morbidity (Guttikunda and Goel, 2013; Li et al., 2013; Maji et al., 2017; Nagpure et al., 2014). Presently air pollution is India’s second-biggest public health hazard after child and maternal malnutrition. India was ranked as the third and fifth most polluted country in the world in 2020 and 2021 respectively based on annual average population-weighted PM_2.5 levels (IQAir, 2020, 2021). PM_2.5 concentration is further expected to increase and deteriorate the air quality in India in the coming decades due to its expanding population and rapid urbanization (Conibear et al., 2018). The WHO’s International Agency for Research on Cancer (IARC) has classified outdoor air pollution as a Group 1 human carcinogen (IARC, 2013). In 2012 exposure to outdoor and indoor air pollution caused 7 million premature deaths worldwide (WHO, 2014). The study on Global Burden of Disease reports that air pollution killed 1.7 million Indians in 2019 and the majority of these deaths were due to ambient particulate pollution and household air pollution (Pandey et al., 2021). Currently, more than 55% of the world's population is living in urban areas and it is projected to increase to 68% by 2050 (United Nations, 2019). The vast population residing in the urban centers, especially in developing economies like India, is facing the problem of poor air quality. In terms of particle pollution, Delhi is one of the most polluted capital cities in the world (IQAir, 2021). Particle pollution is a key challenge in Delhi. The major sources of its PM are transport, construction activities, power plants, industries, road dust, and open biomass burning (ARAI and TERI, 2018). In the PM_2.5 concentration, the carbonaceous aerosols contribution is 26%, which comes from fossil fuel combustion, vehicular emissions, and the biomass burning (Sharma et al., 2022). Over the last two decades, there have been good efforts are made by different groups to investigate variability and trends. However, studies carried out by them were confined for a short duration or more inclined to the seasonal changes (chronology is shown in Table 1).

Although the central and state governments have taken numerous initiatives in the last decade to tackle the air pollution, especially in the national capital. The measures taken includes ban on biomass burning, a pilot project of Odd-Even policy for vehicles (2016), launching the National Air Quality Index (NAQI) (2016), de-registering of the diesel vehicles older than 10 years or more (2016), Graded Response Action Plan (GRAP) for Delhi-NCR (2017), shut-down of Badarpur thermal power plant (2018), use of Bharat Stage BS-VI grade auto fuels in Delhi (April 2018), and the National Clean Air Program (NCAP) (2019) (MoEFCC, 2019), but still particle pollution levels in the city frequently exceed the national and WHO standards. The proposed study used non-parametric statistical methods for trend analysis of fine particles at different time scales. Main objectives of the study are to investigate and estimate long-term trends annually, seasonally, and monthly, and understand the temporal variability in PM_2.5 over Delhi during 2007–2021. To understand the relationship between PM_2.5 mass concentration and meteorology, we conducted a Pearson correlation analysis using monthly data. The long-term trends in meteorological parameters were also studied to explore the influence of meteorology on the observed trend in PM_2.5. Understanding the temporal trends in the key pollutants, PM_2.5, and meteorology may be useful in devising new effective strategies for air pollution abatement. The understanding of variability in PM_2.5 at different time scales and its relationship to the local meteorological conditions will help in understanding the drivers of air pollution in a complex urban environment (Dhaka et al., 2020).

 
2 METHODS

 
2.1 Acquisition of Data and Analysis

2.1.1 PM_2.5 data

The monitoring of air quality is done by the Central Pollution Control Board (CPCB) of India through the widely spread network of stations across the country but most of the stations in the city were added in 2018. The longest data of PM_2.5, from 2007 except for 2014 to 2016, is available for the ITO station only. The daily average data from CPCB, Delhi Pollution Control Committee (DPCC), and Indian Meteorological Department (IMD) are obtained through the CPCB’s online portal (https://app.cpcbccr.com/ccr/#/caaqm-dashboard-all/caaqm-landing). The instruments at CPCB are calibrated using certified reference standards; hence quality data are provided following a variety of standard operating procedures. The details regarding the calibration of the instruments, quality assurance, and quality control (QA/QC) procedures in air quality monitoring can be found elsewhere (CPCB, 2012) and at https://cpcb.nic.in/quality-assurance-quality-control/. Hourly observations of PM_2.5 have also been made in the U.S. Embassy and Consulates in New Delhi since 2013. PM_2.5 is measured at CPCB stations and the U.S. Embassy using Met One Instruments, Inc. BAM-1022 (MetOne, 2020) which is based on the principle of attenuation of beta rays when they are passed through an the aerosols loaded air sample. The accuracy of the instrument is within 2 µg m^–3. The maintenance and calibration of the instruments are done according to the regulations of the U.S. Environmental Protection Agency (U.S. EPA, 2022). The PM_2.5 data of the U.S. Embassy and consulates is available from the AirNow website (https://www.airnow.gov/).

 
2.1.2 Satellite data

The number of fire events over the Punjab region, in India, were obtained from the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor on NASA’s Suomi NPP satellite.

The area-averaged daily value of Aerosol Optical Depth 550 nm (MYD08_D3_6_1_Deep Blue, Land-only) from the MODIS sensor on board the NASA-Aqua satellite was obtained from the NASA Giovanni platform (https://giovanni.gsfc.nasa.gov/giovanni/) for the study period (2007–2021) to determine the correlation between the PM_2.5 mass concentration and AOD over Delhi.

Monthly mean area-averaged data of planetary boundary layer height (PBLH) were retrieved from the MERRA-2 Model M2TMNXFLX with 0.5° × 0.625° (lat × lon) spatial resolution during 2007–2021 from the NASA Giovanni portal.

2.1.3 Meteorological data

Hourly averaged meteorological data from the IGI T3 terminal of Delhi airport are obtained. The R package accessed the surface meteorological data from the web servers of National Oceanic and Atmospheric Administration (NOAA). The daily averaged data of precipitation were taken from the NASA data viewer (https://power.larc.nasa.gov/data-access-viewer/). NASA Power provides the meteorological and solar datasets for the support of renewable energy and agricultural needs.

 
2.2 Study Area and its General Meteorology

The national capital, located at 28.7°N, 77.1°E, is in the northern of India at an elevation of 216 m above the mean sea level (MSL) covering an area of 1483 km². The city has more than 18 million human population (http://census2011.co.in) and a high population density of 29,259 people per square mile (https://worldpopulationreview.com/world-cities/delhi-population). Delhi is an inland city with a continental type of climate having extreme variations in the meteorological parameters from one season to another. The months were sequentially clubed and divided into the seasons according to the India Meteorological Department (IMD, 2015): winter (Jan.–Feb.), summer (Mar.–May), monsoon (June–Sep.), and post-monsoon (Oct.–Dec.). The dry and hot summer (15–40°C, 41% relative humidity (RH)), moderate rainfall with high humidity during monsoon (571 mm average rainfall, 68% RH), moderate temperature and humidity during post-monsoon (21°C, 62% RH) and low temperature and high humidity in winters (5–25°C, 72% RH) are the characteristics of Delhi’s climate. Due to low temperature, calm winds (< 1 m s^–1), shallow planetary boundary layer height (PBLH), and high relative humidity fog and pollution episodes with surface temperature inversion are quite frequent during the post-monsoon and winter seasons. Fig. 1 shows the map of the study area showing the location of six observing stations. The monitoring sites are selected from different zones of the city therefore average concentration of all stations can be taken as the representative average concentration of the entire city. The location, land-use type, major pollution sources, and period of availability of PM_2.5 data for all the study sites are presented in Table S1.

Fig. 1. Map of study area showing the location of six monitoring stations from each zone of the city.

 
3 RESULTS AND DISCUSSION

 
3.1 Inter-annual Variation in PM_2.5

Statistical description of the long-term PM_2.5 data from 2007–2021 is summarized in the whisker-box plot and shown in Fig. 2. The yellow star inside the box represents the average value, and the lower, middle, and upper lines of the box represent 25^th, 50^th (median), and 75^th percentile, respectively. Red dot at the bottom and top of the box represents the 5^th and 95^th percentile, respectively. The black circles are the outliers and dashed green and blue lines show the WHO and Indian National Ambient Air Quality Standards (INAAQS) (CPCB, 2009). It is clear from Fig. 2 and Table S2 that the annual mean and median PM_2.5 levels don’t meet the standards during the entire study period. We found that for the entire period annual mean concentration was greater than the median PM_2.5 concentration suggesting that PM_2.5 distribution is highly skewed towards the right end indicating a few but high PM_2.5 episodic events. The black circles in Fig. 2 indicate that during the episodes, concentration may reach up to hazardous levels (> 750 µg m^–3). The inter-annual variation showed the highest levels in 2014 (154 ± 51 µg m^–3) and the lowest in 2009 (95 ± 60 µg m^–3). The annual mean concentration in 2020 (99 ± 81 µg m^–3) was low, probably due to the reduction in particulate levels due to the COVID-19-related lockdowns imposed in various phases (Dhaka et al., 2020; Mahato et al., 2020). Mean concentration during the entire study period of 15 years was 125 ± 86 µg m^–3 which is 3 times higher than the Indian NAAQS. Year-wise descriptive statistics of PM_2.5 are presented in Table S2.

Fig. 2. Whisker-Box plot showing the yearly statistical overview of PM_2.5 during 2007–2021. The yellow star inside the box represents the average value, the lower, middle, and upper line of the box represents 25^th, 50^th (median), and 75^th percentile. The red dot at the bottom and top of the box represents the 5^th and 95^th percentile. The black circles are the outliers and dashed green and blue lines shows the WHO and Indian NAAQS.

 
3.2 Long-term Pattern on an Annual Scale

Understanding the trends of air quality is very important for the scientific community and policymakers to properly assess the effects of various air quality control programs. Using the Shapiro-Wilk normality test at p < 0.05 we found that the distribution of PM_2.5 is not normal (W = 0.85966 and p-value < 0.05). The non-parametric rank-based Mann-Kendall (MK) test (Gilbert, 1987; Kendall, 1975; Mann, 1945) was used to detect the presence of a monotonic trend (upward or downward) in the time series of the daily average data set of PM_2.5 from 2007 to 2021 using the MannKendall function of the Kendall R package (McLeod, 2011). The MK test is not affected by the presence of outliers and the distribution of data (normal or non-normal). This method tests the null hypothesis H_o of no trend in PM_2.5 time series during the study period against the alternate hypothesis H_a which assumes the presence of a trend. By plotting the auto-correlation and partial auto-correlation plots of monthly PM_2.5 data, we found the presence of a significant auto-correlation in the time series and it is also known that the fine particle concentration shows a strong seasonal variation (Fig. 7(b)). Therefore, we first deseasonalized monthly time series and then used the MK test in conjunction with the block bootstrapped technique. The block bootstrapped is a powerful method if the time series has auto-correlation. MK test results revealed the presence of a statistically significant monotonic downward trend in the long-term PM_2.5 data set. Kendall's statistics tau value is –0.253 with 2-sided p-value < 0.05, confirming the rejection of the null hypothesis. Tau is a measure of the strength of the linear trend. Negative value of tau shows trend is decreasing. To estimate the slope of this linear trend, also known as the magnitude of the trend, we used the smoothTrend and TheilSen functions (Sen, 1968; Theil, 1950) of openair package (Carslaw and Ropkins, 2012) in R software. Monthly mean values were calculated from the daily average data to detect the long-term trend. The smoothTrend function plots the monthly mean concentration and fits a smooth line with 95% confidence limits. mgcv R package was used to determine the smooth line using the Generalized Additive Modelling (GAM) approach. The Bootstrap simulations were used to estimate the uncertainties in the trend. To get a clear view of temporal trend, monthly mean was first deseasonalized using the stl function (Seasonal Trend decomposition using Loess) which is inbuilt in SmoothTrend and TheilSen functions. The auto-correlation was not considered in the trend uncertainty estimates. Theil-Sen estimator is a robust, efficient, and non-parametric statistical method that can be used for non-normal distribution and it is resistant to the missing data and outliers in the time series because this is based on the median, not the mean, of the data set. The TheilSen function calculates the slopes between all pairs of data points, and the median of the slopes is known as the Theil-Sen slope or the trend estimate. This approach has been extensively used worldwide to study the temporal trends in air pollutants and climate variables (Althuwaynee et al., 2020; Jassim et al., 2018; Munir et al., 2013; Sarkar et al., 2019a; Torbatian et al., 2020).

The long-term (2007–2021) inter-annual temporal trend in PM_2.5 across the city is shown in Fig. 3. Deseasonalized smooth trend in PM_2.5 is shown by a red solid line (Fig. 3(a)). Smooth trend line shows the long-term variation in the monthly mean data. Red circles denote monthly mean concentration and the shaded area represents the estimated 95% confidence interval (CI). Fig. 3(b) depicts the Theil-Sen linear trend in PM_2.5 which is calculated from the deseasonalized monthly mean concentration shown by blue circles. Red line indicates the trend estimate, the Theil-Sen slope, and the dashed red line shows the 95% CI of the trend (slope) based on resampling methods. Overall trend is shown at the top-left and the square bracket shows the 95% CI in the slope. Symbols ***, **, *, and + indicate that trend is significant at 0.001, 0.01, 0.05, and 0.1 levels, respectively. Numerical value of the slope shows an overall decreasing trend at the rate of –1.35 µg m^–3 year^–1 over the entire period with a 95% CI [–2.3, –0.47]. A very high significance level (p < 0.001) provides strong evidence that concentration decreased over the period; however, the reductions are only marginal, and the annual mean PM_2.5 levels still exceed the national standards over the entire study period of 15 years.

Fig. 3. (a) Deseasonalised smooth trend in PM_2.5 shown by red solid line from 2007–2021. The red circles shows monthly mean concentration and shaded area represents estimated 95% confidence interval (CI). The smooth trend line shows the long-term variation in the monthly mean of the data. (b) Inter-annual Theil-Sen linear trend in PM_2.5 calculated from the deseasonalised monthly mean concentration shown by blue circles. The red line indicate the trend estimate and dashed red line shows the 95% CI of the trend based on resampling methods. The overall trend is shown at the top-left (–1.35 µg m^–3 year^–1) and square bracket shows the 95% CI in the slope or trend. The ***, **, *, + indicate that the trend is significant to 0.001, 0.01, 0.05, and 0.1 levels respectively.

An attempt has been made to estimate the trend in PM_2.5 along with different wind directions. The Theil-Sen function splits the wind direction into eight different sectors viz. NW, N, NE, W, E, SW, S, and SE. The Theil-Sen slope was calculated for each different wind direction and presented in Fig. S1. The trend of PM_2.5 along the different wind sectors showed that the maximum decrease was along the south (–4.01 µg m^–3 year^–1) followed by North (–3.29 µg m^–3 year^–1) and northwest (–3.23 µg m^–3 year^–1). No trend was observed in the North-East direction.

 
3.3 Long-term Pattern on a Seasonal Scale

We also explored the trend in PM_2.5 levels across the different seasons. Fig. 4 shows the Smooth trend and Theil-Sen trend estimate for all the seasons. The maximum decrease, –3.05 µg m^–3 year^–1, was observed in the summer season with a 10% level of significance (p < 0.1) over the period. 95% CI in the trend, shown at the top of each subplot in a square bracket, ranged between –6.08 to 0.33 µg m^–3 year^–1. No significant trend was observed in the post-monsoon and winter season. Also, monsoon season showed an insignificant declining trend of –1.95 µg m^–3 year^–1 with 95% CI [–4.08, 1.12] µg m^–3 year^–1. Results of the MK test (uptrend or downtrend) for each season are similar to the Theil-Sen estimator and it is presented in Table S3. The impact of action control programs over the city are extremely small.

Fig. 4. Same as Fig. 3 but for Seasonal trend in PM_2.5 during 2007–2021.

3.4 Long-term Monthly Trends and Variations

3.4.1 Monthly trends

The MK test and Theil-Sen slope estimator were calculated individually for each month. Fig. 5 shows the results of the trend analysis of PM_2.5 concentration on a monthly time scale during 2007–2021. Except Dec. and Jan. all the months showed a statistically significant decreasing trend over the study region. A strong negative trend was found in Oct. (–3.78 µg m^–3 year^–1) followed by Mar. (–3.52 µg m^–3 year^–1), Apr. (–2.81 µg m^–3 year^–1), and Jun. (–2.64 µg m^–3 year^–1) at a 0.001 level of significance. It reveals that the control strategies to curb particle pollution in the city are not effective during Dec. and Jan. months. A statistical significant decreasing trend was observed during the crop burning months viz. Oct. (–3.78 µg m^–3 year^–1) and Nov. (–1.97 µg m^–3 year^–1 with p < 0.01). An overall small average decrement is not enough to keep the PM_2.5 concentrations within the threshold limit. The results of the MK test (uptrend or downtrend) for each month are similar to the Theil-Sen estimator and are presented in Table S3.

Fig. 5. Same as Fig. 3(b) but for Monthly trend in PM_2.5 during 2007–2021.

 
3.4.2 Monthly variations

Variation in monthly averaged concentration was studied and shown in Fig. 6(a). Fine particle levels were continuously very high from Oct. to Feb., which includes both the post-monsoon and winter seasons. Nov. and Aug months show the maximum (227 ± 156 µg m^–3) and the minimum (39 ± 16 µg m^–3) concentrations, respectively. The highest concentration in Nov. may be attributed to the contribution of Crop Residue Burning (CRB) taking place in the neighboring states of Delhi. Minimum concentration in Aug. may be due to the wash-out mechanism of rain. After Sep., concentration starts rising due to increased local emissions, long-range transport, and unfavorable meteorology including a decrease in temperature, PBLH, wind speed (WS), and rainfall (RF), and increase in relative humidity (RH) and surface pressure (SP). Unfavourable meteorology hampers the horizontal and vertical dispersion of pollutants, leading to high levels of aerosols in the post-monsoon and winter seasons. Because of favorable meteorology and decreased local emissions, monthly averaged concentration starts decreasing from Feb onwards. During Apr.–Jun. strong surface winds (> 3 m s^–1) from westerly, and north-westerly directions (Fig. S2: wind rose) brings frequent dust storms from the Indian Thar Desert and Arabian Peninsula (Kumar et al., 2014; Sarkar et al., 2019b) in the Delhi-NCR (Fig. S3 shows HYSPLIT trajectory for one dust storm event on 08 May 2018).

Fig. 6. (a) Monthly mean concentration of PM_2.5 and (b) Diurnal variation of PM_2.5 during different seasons prepared using the hourly average data of U.S. Embassy and Consulates from 2013 to 2021.

 
3.4.3 Why November shows peak levels: stubble crop burning

An attempt has been made to show that high fine particle levels during Nov. are closely related to the CRB in the neighboring states of Delhi, mainly the Punjab region (Cusworth et al., 2018; Jethva et al., 2018; Montes et al., 2022; Rahman et al., 2022; Singh et al., 2020; Takigawa et al., 2020). Using satellite data we tried to correlate the total number of active fire events in Punjab with the surface PM_2.5 levels in Delhi. The monthly mean number of total fire counts for the Punjab region and monthly average PM_2.5 for four stations viz. ITO, Anand Vihar, DTU, and the U.S. Embassy in New Delhi during 2017–2021 are presented in Fig. 7. It depicts that every year the crop stubble is burnt in Punjab from May–Jun. (wheat crop) and Oct.–Nov. (paddy crop). Fire activity during rice crop harvesting is much greater than during the wheat crop. As the number of fire counts increases, the fine particles loading over the Delhi-NCR region also increases. The CRB in Nov. is a little higher than in Oct. In 2021 the fire counts during Oct. are much smaller, mainly due to the prolonged lingering monsoon rains. Therefore the major residue was burnt during the Nov. month as shown in the Fig. 7. It is clear that the high levels of PM_2.5 in Delhi exactly coincide with the CRB period. The wind rose in Fig. S2 shows that during Oct.–Nov., northwest (NW) is the dominant wind direction for Delhi and the city is located in the downwind direction of the Punjab. The smoke plumes from CRB are loaded with aerosols and precursor gases (SO₂, NO_x, NH₃, and VOCs) which get transported to the Delhi-NCR via the long-range transport (Jethva et al., 2018; Singh et al., 2020; Takigawa et al., 2020) resulting in excess fine particle loading over the city. We studied the Nov. and Dec. 2021 months in more detail by taking the daily average data of PM_2.5 from all 37 monitoring stations in Delhi. The average concentration in Nov. was 237 ± 156 µg m^–3 and the daily average minimum concentration ranged between 40 to 150 µg m^–3 (Fig. S4). All the monitoring stations showed a similar range of particulate matter (PM) concentrations with little local variations due to their specific location, clearly indicating that the entire city is engulfed with high PM pollution. Occasionally hourly PM_2.5 concentrations may reach up to the hazardous level of 750 µg m^–3 resulting in severe pollution episodes (Kanawade et al., 2020). In 2021 the concentration of Nov. was even greater than Dec. for all 37 stations except Anand Vihar, R K Puram, and the Nehru Nagar. The high concentration of PM_2.5 in Nov. over Delhi was due to contribution of CRB. The difference between the monthly average concentration of Nov. and Dec. ranged from 20 to 60 µg m^–3 as shown in Fig. S5. We can conclude that due to the significant contribution of CRB to Delhi’s air pollution, the fine particle concentration during the mid-Oct. and Nov. months is even greater than the concentration during cold Dec. month which has comparatively adverse meteorology and increased emissions due to the burning of solid fuel for heating. The fine particle levels also remains high during the winter season mainly due to the combined effect of increased local emissions and unfavorable meteorology. By comparing monthly mean data during Oct., Nov., and Dec., it is estimated that 50 to 100 µg m^–3 contribution is coming from CRB along with the northerly, and north-westerly winds during the crop burning period.

Fig. 7. (a) Monthly mean total number of hotspots in Punjab using data of VIIRS sensor on NASA’s Suomi NPP satellite. (b) Monthly average concentration of PM_2.5 for ITO, Anand Vihar, DTU, and U.S. Embassy stations of Delhi during 2017–2021.

3.5 Diurnal Variation

We used the hourly observations of PM_2.5 from the U.S. Embassy and Consulates during 2013–2021 to study the distinct seasonal variation in the diurnal profile of PM_2.5, which is presented in Fig. 6(b). The concentration of PM_2.5 was maximum in post-monsoon. The diurnal concentration during monsoon ranged from 43 to 55 µg m^–3. The bi-modal peaks were almost absent during the monsoon season. Due to the washout mechanism of rainfall and favorable meteorology, the concentration of PM_2.5 does not show much variation throughout the day resulting in no clear diurnal pattern. Other studies also reported no pronounced diurnal variation during the monsoon season (Chen et al., 2020; Sreekanth et al., 2018). Throughout the day a small difference (5–15 µg m^–3) in hourly average PM_2.5 concentration was observed during winter and post-monsoon seasons. Daytime peak of PM_2.5 during summer was observed at 0800 IST but in winter it was shifted to1000 IST. This may be due to the shift in morning rush hours (0800–0900 IST) from summer to winter (0900–1000 IST). Chen et al. (2020) also reported a similar pattern. Interestingly, during the post-monsoon and winter seasons, the afternoon minima are well defined but in summer the afternoon minima are flat. PM_2.5 levels during the post-monsoon and winter season start increasing at 0800 IST but in summer it starts at 0600 IST, probably due to the change in traffic timings according to the season. The diurnal profile of PM is strongly dependent on the local meteorological parameters and local emissions due to human activities. During the winter season, very often it shows high values (> 190–200 µg m^–3) until 1100 IST then it declines with increasing solar insolation and strong wind during the day, and pollution again rises in the evening and reaches a peak (170–190 µg m^–3) during the late night. Semi diurnal peak around 1000–1100 IST in local surface pressure coincides with a peak in PM_2.5, which is also followed in the late night. During the day, both wind speed and solar insolation play a significant role in the dispersion and dilution of pollutants. Critical values of wind speed (> 2 m s^–1) and solar radiation (> 150–200 W m^–2) were identified to be effective to disperse pollutants during the peak pollution period (Nov.–Jan.). The concentration of fine particles is quite high throughout the day except for a small afternoon period from 1200 IST to 1800 IST which may be attributed to high wind speed, high solar radiation, and deep PBLH during the afternoon time (Sathyanadh et al., 2017).

 
3.6 Correlation among the Study Sites: Spatial Variability

Using daily average data of PM_2.5 for all locations and the area-averaged value of Aerosol Optical Depth (AOD) for the entire city, we studied the Pearson linear correlation among all the study sites to understand the spatial variability of fine particles in the city. The correlation plot is shown in Fig. S6. Pairwise correlation among all the study sites showed a strong positive linear correlation (0.62–0.92) indicating a weak spatial variability in PM_2.5. Thus it can be concluded that by and large the pollution sources for all the sites are similar and the entire city is facing the problem of critical levels of particle pollution. Hama et al. (2020) study also observed high Pearson correlation among the six study sites of the Delhi region and the coefficient of divergence values for sites was mostly less than the threshold value of 0.2 indicating high homogeneity and low spatial divergence in PM_2.5 among the study sites. PM_2.5 value of each station showed a very strong positive correlation (0.81–0.95) with the average PM_2.5 concentration of all the stations indicating that averaged concentration is the best representative value for the entire city. Therefore, we used all stations' average value for all the analyses done in this paper. We also observed that during the study period the daily average PM_2.5 concentration showed a weak positive linear correlation (0.22–0.40) with the daily area-averaged AOD over Delhi.

 
3.7 Influence of Meteorology on the Long-term Trend of PM_2.5

The concentration of PM is not only dependent on primary emissions and precursor gases but it is also influenced by meteorological factors (Tella et al., 2021). It plays an important role in the observed seasonal and diurnal variations in PM_2.5 by governing its transport, dilution, dispersion, and plenty of photochemical reactions among the primary pollutants in the atmosphere. It may affect the short- and long-term variability and trend in PM concentration. The distinct inter-annual, seasonal, and diurnal variations are the combined results of anthropogenic activities and meteorology. To understand the correlation between PM and meteorology the Pearson correlation and regression analysis was conducted on the monthly average data set from 2007–2021 and it is shown in Fig. 8. The correlation coefficient and p-value are shown on each subplot. Linear regression line (Black solid line) is added for all subplots except (c) for which loess, local regression fitting was used because relative humidity has a nonlinear association with PM_2.5. The shaded area shows a 95% CI of the best fit curve. We found a strong and negative correlation between PM_2.5 and wind speed (–0.62), PBLH (–0.40), air temperature (–0.75), and rainfall (–0.60) whereas surface pressure (0.78) showed a strong positive correlation. Fig. 8(c) shows an inverted U-shaped curve illustrating that initially as RH increases, PM_2.5 mass concentration increases at first but as RH reaches 70% the concentration of fine particles starts decreasing. Similar scientific findings were reported by Lou et al. (2017) in their study of the Yangtze River delta in China. To assess the possible meteorological influences on the observed decreasing trend of PM_2.5, we estimated the long-term trends in major meteorological parameters viz. wind speed (WS), relative humidity (RH), temperature (temp), rainfall (RF), PBLH, and surface pressure (SP). The smooth and linear trends of WS and RH are shown in Fig. 9. The WS showed a very small (–0.03 m s^–1 year^–1, 95% CI: –0.04, –0.02) but statistically significant (p < 0.001) decreasing trend over time on the other hand RH showed an increasing trend of 0.42 units year^–1 (95% CI: 0.16, 0.65) at 0.001 level of significance. Surface pressure showed a very small but significant uptrend, 0.08 mm Hg year^–1 (p < 0.001), whereas PBLH showed a significant downtrend with an average decrease of –12.38 m year^–1 with p < 0.001 (95% CI: –20.97, –4.34) as shown in Fig. 10. No significant trends were observed for temperature and rainfall as shown in Fig. S7. Singh et al. (2021) reported no significant change in the meteorological parameters namely in temperature, wind speed, precipitation, and PBLH over Delhi during 2014–2019. The monthly mean area-averaged time series of PBLH during 2007–2021 over Delhi was obtained from MERRA-2 Model M2TMNXFLX from the NASA Giovanni platform and it is presented in Fig. S8. Our results of an uptrend for surface pressure and downtrend for wind speed and PBLH suggests that the mass concentration of fine particles should have increased over the study region during the study period. On the contrary, we found a weak decreasing trend in PM_2.5 levels. It reveals that the observed long-term negative trend in PM_2.5 might be the result of various pollution control programs of Government agencies. The observed long-term trend of meteorological parameters over Delhi tends to negate the effects of pollution abatement initiatives during the study period. It may be one of the reasons why we observed only a marginal decline in PM_2.5 levels during the long period of 15 years. The other probable reasons may be the weak enforcement of the norms (Bhave and Kulkarni, 2015) and the increasing human and vehicular population in the city. Therefore, more strict and comprehensive regional action plans are required to curb fine particle pollution and comply with the recommended air quality guidelines. More than a Decade-long exposure to the highly polluted air, as shown in the analysis, raises very serious concerns for human health including vegetation.

Fig. 8. Pearson linear correlation between monthly average PM_2.5 and different meteorological parameters (a) wind speed (b) air temperature (c) relative humidity (d) surface pressure (e) rainfall and (f) planetary boundary layer height for p < 0.05. Correlation coefficient and p-value is shown on each subplot. Linear regression line is added for all subplot except (c) for which loess, local regression fitting is used because relative humidity has non-linear association with PM_2.5.

 Fig. 9. Same as Fig. 3 but for wind speed and relative humidity during 2007–2021 in Delhi.

 Fig. 10. Same as Fig. 3 but for PBLH and surface pressure during 2007–2021 in Delhi.

4 CONCLUSIONS 

Satellite fire count data and daily averaged surface data of PM_2.5 from six stations of Delhi during 2007–2021 were analyzed to understand the temporal trend and variability in fine particles inter-annually, seasonally, monthly, and along with the different wind directions using the non-parametric Mann-Kendall and Theil-Sen approach. The main highlights of the study are:

• PM₅ mass concentration showed an overall average decrease at the rate of –1.35 µg m^–3 year^–1 (p < 0.001).
  
  
• The seasonal temporal variation of PM₅ showed a significant decreasing trend in summer (–3.05 µg m^–3 year^–1, p < 0.1), an insignificant downtrend during monsoon, and no trend in post-monsoon and winter season.
  
  
• Trend of PM₅ in the different wind sectors showed a maximum decrease along the South (–4.01 µg m^–3 year^–1) followed by North (–3.29 µg m^–3 year^–1) and northwest (–3.23 µg m^–3 year^–1) directions. No trend was observed in the North-East direction.
  
  
• Except for Dec. and Jan., all the months showed a statistically significant decreasing trend in PM₅.
  
  
• Using monthly data, PM₅ showed a negative linear correlation with wind speed, rainfall, PBLH, and air temperature; whereas a positive correlation with surface pressure and an inverted U-shaped relationship with relative humidity was prominent.
  
  
• Relative humidity and surface pressure showed a significant long-term increasing trend whereas wind speed and PBLH suggest decreasing trend. Air temperature and rainfall showed no trend.

 
ADDITIONAL INFORMATION

Author's Contributions

SKD and Chetna conceptualized the idea and analysis was carried out of the particulate matter. G L and V P analyzed meteorological data; V K, S M, and ASR have contributed on diurnal variability. NS and A.P.D examined long term changes in pressure and relative humidity. Y. M, T. N., and S H have contributed to writing and editing of the manuscript.

ACKNOWLEDGMENTS

We acknowledge the Central Pollution Control Board (CPCB), Delhi Pollution Control Committee (DPCC), and U.S. Embassy and Consulates in New Delhi for providing surface data of PM_2.5 online. The Authors also thank NASA for providing fire counts data from the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor on the Suomi NPP satellite and NASA Giovanni online Platform; we thank to Prof. R. Imasu and A A Adhari for preparing fire count data. The authors would also acknowledge the NOAA Air Resources Laboratory (ARL) for the provision of the HYSPLIT transport and dispersion model and the READY website. This study was supported by the Research Institute for Humanity and Nature (RIHN: a constituent member of NIHU), Project No. 14200133.