Introduction

Over the decades, high population growth has triggered rapid industrialization, urbanization, and increase in vehicular traffic that has led to deterioration of air quality. Furthermore, the lack of strict implementation of environmental regulations has only added to the pollution woes in developing countries like India. According to the World Health Organization (WHO) [1] in 2016, half of the top 20 polluted cities are in India. In order to monitor the air pollution trends, Central Pollution Control Board (CPCB) has set up 591 monitoring stations around the country covering 248 cities/towns in 28 states and 5 union territories [2]. In 2015, for example, averaged concentrations of SO2 and NO2 in major cities did not exceed National Ambient Air Quality Standards (NAAQS) in any of the major cities in India. In contrary, the annual average concentration of PM2.5 in major cities in north, east, west, and south Indian cities were 3.3, 3.7, 2.3, and 1.6 times of NAAQS standard (40 μg/m3). Even the 8-h concentrations of O3 and CO in major cities in north (47.8 and 1.26 mg/m3), east (48.1 and 1.73 mg/m3), west (58.6 and 1.27 mg/m3), and south India (58.6 and 0.94 mg/m3) were less than the NAAQS standards (100 and 2 mg/m3) [3]. Exceedance of pollutant concentrations can primarily be attributed to dust emissions, vehicular emissions, biomass burning etc. [4]. Aggravated concentrations of criteria pollutants pose constant threat to human health and environment. India accounted for 25.7% of global premature deaths due to PM2.5 exposure in the year 2015 [5]. PM2.5 concentrations exceeding WHO set limits resulted in 6.5% excess mortality in the Indian capital [6].

Many of the high- and middle-income group countries have extensive network of monitoring stations covering most of rural and urban areas, which provide continuous measurements of pollutants (for example, see [7, 8]). Setting up of a network of monitoring stations in developing countries like India is not economically viable. Since continuous monitoring of air pollutants is of prime importance in India where air pollution levels are among one of the world’s highest [5], other methods such as data from satellites [9, 10] and air quality models are used to estimate pollutant concentration in such cases. Moreover, sparse spatial coverage of monitoring stations limits their use and hence air quality models are used to obtain the spatial and temporal variations of pollutant concentrations. On the basis of spatial resolution, air quality models can be divided into (i) local, (ii) regional, (iii) global/meso scale models. Models with a domain size of few meters to few km are termed as local/urban scale (for example, see [11]), tens of km to hundreds of km are treated as regional models (for example, see [12, 13]). All the models coarser than regional models are treated as global/meso scale models [14]. A sensitivity study using a regional photochemical model of ozone in local/urban scale to regional ozone episodes indicated that increase in ozone levels in local/urban zones were related to increase in ozone in regional scale [15]. Thus, pollutants released locally can affect regional air quality.

Local air quality modeling studies provide critical information regarding the status of air quality in our vicinity. A three-dimensional Eulerian model [16] was used to simulate dispersion of near road gaseous air pollutants using on road emissions [11]. Source contributions of volatile organic compound fluxes in a city were carried out using a receptor model [17]. Regional models have a greater spatial coverage and help in predicting concentrations or identifying sources at a much wider scale. A 7-year air quality simulation in eastern USA [7] revealed that while O3 and PM10 were modeled satisfactorily, PM2.5 and its components were slightly over-predicted. Similarly, a yearlong simulation of ozone and particulate matter in China [18] revealed over prediction of O3 at low concentration ranges, and under prediction of PM2.5 during summer. Source apportionment studies using source oriented regional air quality models enable one to obtain contributions from sources without being limited by the frequency and spatial coverage of observations. These studies are often used to validate emission inventory of pollutants [19]. A study [13] indicated significant contribution of isoprene to secondary organic aerosol (SOA) in eastern USA during summer. A source apportionment study of particulate matter in China [20] using source oriented air quality model revealed residential and industrial emissions to be chief contributors to primary PM2.5, and industries, agriculture, power plants, and transportation were important sources to secondary aerosols. However, only one such study was carried out in India [21].

Many air quality-modeling studies, around the world, in the past have been carried out at both local and regional scale for various applications. It is important to combine and evaluate the results from the studies of different scales to achieve better understanding of the model performance. For example, an overview of modeling practices in European Union (EU) revealed that no single model is currently capable to accurately predict air quality at all spatial scales, with emissions carrying the most uncertainty among inputs [22]. Such reviews can also aid in identifying possible mitigation strategies. For example, a study reviewing ozone modeling studies around the world have revealed over-estimation of night time ground level ozone (GLO), greater influence of NOx over VOCs in export of ozone from urban region, better performance of fine resolution inputs, and greater contribution of temperature rise and biogenic VOC to GLO concentration [23].

Moreover, a review of modeling studies carried out in a particular scale in a country/region can also aid in better understanding of the sources in the entire region. For example, a recent review [24] on source apportionment (SA) studies carried out in China shows that while South China is categorized by higher contributions of vehicle exhaust and secondary sources, dust, coal combustion, and biomass burning are the leading sources in North China. A European study [25] which reviewed the published literature using meta-analysis found that sea salt, dust, traffic, point sources, and biomass burning are the constantly reported sources in different European cities. While in India, SA of airborne particulates was thoroughly reviewed [26] and concluded that dust emissions as the principal source followed by traffic emissions, industrial emissions, coal combustion, sea salt, and biomass burning are the most commonly apportioned sources.

Most of the reviews of modeling studies in India are either on a single pollutant (for example, see [27]) or constrained to a single scale (for example, see [28]). Therefore, the goal of the present study is to provide a comprehensive review of all local and regional scale modeling studies carried out in different parts on various pollutants in India. For this purpose, the studies have been divided into regional scale and local scale, based on the domain size. Therefore, this study will not only aid in understanding the current knowledge gaps, but also help in providing proper control strategies in India, which is currently facing severe air pollution-related issues.

Study Design

Metadata reported by the air quality modeling studies carried out throughout India on PM, VOCs, PAHs, O3, SO2, NO2, NOx, and Black Carbon (BC) were reported in this study. For the present study, each article published in a peer-reviewed journal was logically searched using database from Scopus, Web of Science, and Google searches. Table 1 depicts the number of studies covered in different regions using different models in India. It shows that Northern India has been extensively studied with Central India being the least studied region. Overall, 130 research publications during past 5–7 years were thoroughly inspected. Of the studies collected, 60 local and 35 regional studies conducted in five different regions, i.e., north, south, east, west, and central parts of India were used.

Table 1 Total number of studies with respect to different models in regional and local scale distributed all over India
Full size table

Regional Scale

The current review attempted to include studies which used regional models for predicting pollutant concentrations. Studies with models predicting surface concentrations of pollutants were only included for the analysis. Moreover, studies which explicitly mention model performance or at least had a plot of observed and predicted concentrations were only considered. The following are some of the most generally used regional models in India: Weather Research and Forecasting Model coupled with Chemistry (WRF-Chem), Model for Ozone and Related chemical Tracers (MOZART), Community Multiscale Air Quality Model (CMAQ), and Comprehensive Air Quality Model with extensions (CAMx). Since model specific work in India has only gained momentum in recent years, to include diverse set of models, all available studies were included. The regions were further grouped into urban and semi-urban areas based on their population as provided in http://www.newgeography.com/content/002537-urbanizing-india-the-2011-census-shows-slowing-growth.

The search keywords used for regional studies were as follows: air quality model in India, air quality model in Delhi, Patna, and likewise in different cities; air quality model in north, south, east, and central India. Also, the model name followed by city name were used as keywords, i.e., CMAQ Delhi, CMAQ Mumbai, CMAQ Kolkata, CMAQ Kanpur, CMAQ Bangalore, CMAQ Chennai, and likewise, similar searches were done using WRF-Chem, CMAx, and MOZART.

As majority of the studies reported predicted and observed values in the form of plots, to extract the data, Webplotdigitizer software (https://automeris.io/WebPlotDigitizer/) was used. The software is easy to use and provides both manual and automatic point selection modes to easily identify points on the plot, which can be downloaded in the form of a csv file.

Model performance is generally estimated using mean fractional bias (MFB), mean normalized bias (MNB), mean fractional error (MFE), mean normalized error (MNE), factor of 2 (FAC2), correlation coefficient (R), etc. [29]. However, as different studies use only one or a few of the above mentioned statistics, it is tough to compare model performance of each individual study. This study used a simpler approach by evaluating the ratio of predicted and observed concentrations. Ratio between 0.80 and 1.20 were considered as satisfactory model performance.

Local Scale

In this study, based on model formulation, the local studies have been divided into two categories viz. statistical and numerical models.

Local Statistical Modeling Studies

Receptor oriented source apportionment (RSA) models are commonly used statistical models to obtain the source contributions of the measured pollutant in a region. Keywords used for SA studies were source apportionment, receptor models, PM10 and PM2.5, VOCs and PAHs.

Some of the commonly used RSA techniques in different Indian studies were as follows: enrichment factor (EF) (34%), principal component analysis (PCA) (36%), positive matrix factorization (PMF) (15%), UNMIX (3%), and chemical mass balance (CMB) (10%) [27].

Most of the RSA studies in India are carried out to identify the sources and their contributions to measured PM, VOCs, and PAHs. Identification of key tracer species for each source is one of the chief steps in source allocation. Although there is ambiguity, as reported by earlier reviews [27, 30,31,32], in finding the tracers, based on the commonly reported sources in different RSA studies, seven common source categories were considered in this study: traffic, biomass burning, coal combustion, industry, sea salt, dust, and other sources.

  1. (i)

    Traffic: This includes on-road emissions related to motor vehicles, i.e., tail pipe, non-tail pipe, brake wear, and tire wear.

  2. (ii)

    Biomass burning: Emissions from burning of biomass, wood, vegetative burning, etc., which are used for cooking and heating.

  3. (iii)

    Coal combustion: Emissions from mining activities and burning of coal.

  4. (iv)

    Industry: This group contains the mixture of emissions mainly from power plants and all the various types of industries.

  5. (v)

    Sea salt: Sea salt and ship emissions.

  6. (vi)

    Dust: This category includes emissions from re-suspended dust, earth’s crust, and other soil emissions.

  7. (vii)

    Other sources: All other anthropogenic emissions, including contributions of secondary organic aerosols.

Local Numerical Modeling Studies

Study design follows the similar procedure described in the regional modeling section. The search keywords used for the studies were similar to regional studies but with the inclusion of different model names such as AERMOD, CALINE, and ISCST.

Results and Discussions

Regional Scale

Studies were divided into urban and semi-urban regions. Figure 1a shows the ratio of predicted to observed concentrations. The results of those studies are described in detail below:

Fig. 1
figure 1

Ratio of predicted to observed concentrations of a regional and b local models. While the point shape indicates pollutant types, the color indicates the region where the study has been undertaken

Full size image

Northern India

Urban

A study simulated PM10 concentrations using WRF-Chem in Delhi with 3.3-km resolution domain using Emission Database for Global Atmospheric Research (EDGAR) emission inventory [33]. Results indicated that the predicted to observed ratio (P/O) increased with the domain size, as the domain size was progressively increased from Delhi region (P/O = 0.12) and so on to North India (P/O = 0.57) and then to whole Asia region (P/O = 0.74). Similarly, another study predicted PM2.5 concentrations using WRF-Chem with 20-km resolution and emissions from EDGAR emission inventory [34]. The model performance at two cities, i.e., Agra (P/O = 0.83) and Delhi (P/O = 0.94), was satisfactory. CHIMERE and Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) models were used to simulate aerosol over the Indian subcontinent at a resolution of 111 km [28] with the former using EDGAR emission inventory and the later using a pre-compiled emission inventory [35]. The predicted concentrations for both models were lower than the observations in Delhi with P/O = 0.64 and 0.22 using CHIMERE and GOCART models, respectively. In another study, BC was simulated using WRF-Chem at 12-km horizontal resolution and Spectral Radiation Transport Model for Aerosol Species (SPRINTARS) at 1.125° resolution with emission inputs from an emission inventory compiled using reanalysis of tropospheric chemical composition (RETRO) and for the former, and using RCP8.5 emission inventory for the later [36]. Both models under predicted BC concentration at Delhi (P/O = 0.409 and P/O = 0.505) and Varanasi (P/O = 0.39 and P/O = 0.363). Regional climate model was used by some studies to determine BC concentrations using optical properties of the model at a horizontal resolution of 30 km [37]. Observations from two urban city centers when compared with predictions were found to be good in Kanpur (P/O = 1.02) with under predictions in Delhi (P/O = 0.68). In an another study, the simulated NO2 concentrations, from Model for Ozone and Related chemical Tracers (MOZART) at 2.8° horizontal resolution with emission inputs from Precursors of Ozone and their Effect on the Troposphere (POET) [38], in a northern urban center were higher than observations (P/O = 1.8). Another study in Delhi was conducted using WRF-Chem model with EDGAR emission inventory at 10-km resolution [39]. Two chemical mechanisms were used for the analysis of O3: carbon bond mechanism (CBMZ) and regional atmospheric chemical model (RACM), results indicated that simulation using CBMZ (P/O = 1.29) performed better than RACM (P/O = 1.42).

Semi-Urban

A study used CHIMERE and GOCART for simulating aerosol over the Indian subcontinent at a resolution of 111 km [28], using EDGAR and pre-compiled emission inventory, respectively. Comparison of model predictions with observations in a semi-urban area revealed under prediction, i.e., P/O = 0.76 and 0.13 by both CHIMERE and GOCART, respectively. WRF-CAMx predicted non-methane VOCs (NMVOC) concentrations, using an in situ emission inventory with 25-km horizontal grid resolution [40], at two semi-urban areas were closer to observed concentrations except in areas with limited observation data. An in situ emission inventory at 36-km resolution was input to WRF-CMAQ to simulate concentration over India [41]. The model performance was good at Nainital (P/O = 1.04) and indicated slightly over predictions in Gandaki (P/O = 1.38). Similarly, CMAQ was used to simulate tropospheric ozone over south and East Asia with horizontal resolution of 60 km and in situ emission data. Comparison with observed data revealed that the model performed satisfactorily (P/O = 1.04). In another study, WRF-Chem and MOZART were used for simulating tropospheric ozone over India with resolutions of 45 km and 2.8°, respectively, using INTEX-B and RETRO emission inventories [42]. The model performance of MOZART and WRF-Chem were found to be good in one location (P/O = 1.18 for both models), and over predicting at another (P/O = 1.7 and 1.6). A study simulated O3 over India using WRF-Chem with a spatial resolution of 30 km [43]. The performance of the model was found to be satisfactory at a semi-urban area, P/O = 0.87 and 1.08 using INTEX-B and HTAP-v2 emission inventories, respectively.

Eastern India

Urban

Simulated BC using WRF-Chem at 12-km horizontal resolution and emission inventory compiled using RETRO and EDGAR, and SPRINTARS at 1.125° resolution with emissions from RCP8.5 [36] resulted in satisfactorily performance (P/O = 0.884 and P/O = 0.826, WRF-Chem and SPRINTARS, respectively).

Semi-Urban

WRF-Chem was used to simulate BC concentrations at a spatial resolution of 0.25° with emissions from EDGAR [44]. Evaluation of model performance revealed under prediction at Agartala (P/O = 0.61) and Dibrugarh (P/O = 0.75) and satisfactory in Imphal (P/O = 0.87) and Shillong (P/O = 0.86). Similarly, MOZART with emission inputs from HTAP-v2 was used to model ozone and related tracers at a resolution of 1.9° × 2.5° [45]. Apart from CO, other species were over predicted. Predicted BC concentrations using CHIMERE with EDGAR emissions and GOCART with precompiled emissions at a horizontal resolution of 111 km [28, 35] indicated under predictions (P/O = 0.68 (CHIMERE) and P/O = 0.75 (GOCART)). The concentrations were over predicting in a semi-urban site (P/O = 3.07). In another study, WRF-Chem simulated PM2.5 concentration over the Indian subcontinent at 20-km horizontal resolution and emission inputs from EDGAR emission inventory [34]. The predicted concentrations were satisfactory when compared to observations (P/O = 0.92). A study simulated NMVOC concentrations at 25-km horizontal grid resolution using WRF-CMAx with in situ emission inputs [40]. The model over predicted O3 concentrations at a semi-urban site (P/O = 1.36). WRF-Chem was used to simulate O3 concentration in South Asia using INTEX-B emission inventory at 0.5° grid resolution [46]. The model without fire emissions showed under predictions for CO (P/O = 0.67) and NO2 (P/O = 0.74) and over prediction of NO2 (P/O = 2.42) and good performance of CO (P/O = 0.98) when fire emissions were included.

Western India

Urban

PM2.5 simulation was performed using WRF-Chem with EDGAR emission inventory at 20-km horizontal resolution [34]. Model was found to satisfactorily predict PM2.5 concentrations (P/O = 0.89). Predicted BC concentrations at 30-km horizontal resolution using a regional climate model were lower than observations at Ahmedabad (P/O = 0.54) and Pune (P/O = 0.55) [37]. In another study, MOZART predicted NO2 concentrations, at a horizontal resolution of 2.8° resolution with emissions from POET, were higher than observations at Pune (P/O = 1.5) and closer to observation at Ahmedabad (P/O = 0.93) [38]. NMVOC simulation over India using WRF-CMAx with an in situ emission inventory at 25-km horizontal grid resolution [40], slightly over predicted O3 concentration at Ahmedabad (P/O = 1.3). A study simulated O3 and its precursors using CMAQ and MOZART at 60-km resolution with emission inputs from an in situ emission inventory [47]. The predicted concentrations by both models were satisfactory. In an another study, MOZART was used to simulate O3 and its precursors with a resolution of 1.9° × 2.5° using HTAPv2 emission inventory [45]. The predicted CO values were found to be slightly lower than observations in Pune (P/O = 0.76), while they were closer to observations in Udaipur (P/O = 0.81). Similarly, MOZART was used to model ozone and its precursors using POET emission inventory at a resolution of 1.8° [48]. The model showed satisfactory performance for O3 (P/O = 1.16), under predictions for CO (P/O = 0.45) and NOx (P/O = 0.35).

Semi-Urban

Simulated O3 using WRF-CMAx and an in situ emission inventory with 25-km horizontal grid resolution [40] showed satisfactory model performance (P/O = 1.08).

Southern India

Urban

WRF-Chem simulated PM2.5 using EDGAR emission inventory at 20-km horizontal resolution [34] showed under predictions at Chennai (P/O = 0.73) and Hyderabad (P/O = 0.72). CHIMERE and GOCART were used to simulate BC concentration at 111-km horizontal resolution [28] using EDGAR and precompiled emission inventories, respectively [35]. Both these models under predict concentrations at Hyderabad (P/O = 0.30 and 0.15) and Thiruvananthapuram (P/O = 0.68 and 0.50). Another study used WRF-Chem at 12-km horizontal resolution and SPRINTARS at 1.125° resolution with emissions from a precompiled emission inventory and RCP8.5, respectively [36]. The model performance indicated under predictions at four cities. MOZART predicted NO2 concentration at a horizontal resolution of 2.8° resolution with emission inputs from POET [38] were higher than observations at Bangalore (P/O = 1.5). In another study, MOZART was used to simulate ozone and its precursors with a horizontal resolution of 1.9° × 2.5° using HTAP-v2 emissions [45]. The model under predicted CO at Chennai (P/O = 0.75).

Semi-Urban

WRF-Chem was used to simulate PM2.5 concentrations at 20-km horizontal resolution with EDGAR emission inventory [34]. Model slightly over predicted concentrations (P/O = 1.21). Predicted BC using CHIMERE and GOCART [28] at 111-km resolution with emissions from EDGAR and precompiled BC emission inventory [35] were significantly lower than observed concentrations at two semi-urban locations. Similarly, both WRF-Chem and SPRINTARS simulated BC, which use emissions from EDGAR and a precompiled emission inventory, respectively, under predicted concentrations (P/O = 0.27 and P/O = 0.39) in this region [36]. WRF-Chem predicted CO with resolution of 30 km showed satisfactory model performance (P/O = 0.93) using HTAP-v2 emissions and over predictions (P/O = 1.33) using INTEX-B emissions [43]. Ozone was satisfactorily predicted by MOZART and WRF-Chem which used INTEX-B and RETRO emissions in two of three observation sites. Similarly, good model performance of predicted ozone was observed from WRF-CMAx simulations with in situ emissions at two sites (P/O = 1.14 and 0.94) [40]. However, in an another study, CMAQ predicted that ozone concentrations using in situ emissions showed satisfactory model (P/O = 0.9) performance at one location but over prediction at another (P/O = 1.52) [47].

Central India

Compared to other parts of the country, regional modeling studies in central India are rare. In one study, 20-km resolution WRF-Chem predicted that PM2.5 concentrations with EDGAR emissions inventory showed under predictions at an urban location (P/O = 0.42) [34].

Summary of Regional Modeling Studies in India

MOZART and WRF-Chem were the most frequently used models in the studies across the country. Performance of studies using CMAQ (north P/O-1.21, west P/O-0.98, and south P/O-1.21) and WRF-Chem (north P/O-1.01, east P/O-0.805, west P/O-0.89, south P/O-0.63, and central P/O-0.42) were better. The performance of all models in semi-urban areas of north (P/O-1.06) and south (P/O-1.04) was better as compared to the studies in the corresponding urban areas. However, the performance of models in semi-urban and urban regions in east (P/O = 0.86) and west (P/O = 1.08) was comparable. HTAP-v2 (P/O-0.785) and EDGAR (P/O-0.74) emission inventory were commonly used models for predicting CO, O3, and PM2.5, respectively. Studies with in situ emission inventories performed satisfactorily (P/O-1.04), indicating the requirement of better representative finer scale emission inventories in India.

Local Scale

Local Statistical Models

Diverse Sources of PM over India

Figure 2 shows the source contributions to PM in different cities over India. PM concentrations in northern and eastern India were higher than cities in other parts of the country. These studies are further divided into residential, city center, and industrial areas in different regions of the country as described below.

Fig. 2
figure 2

Relative source contributions (%) of PM in major cities in India

Full size image

Northern India

Around 19 RSA studies were carried out at Delhi in last decade with 37% in the last 5 years. RSA studies in other cities were fewer in number.

Delhi

Delhi, the capital city of India, often features in the world’s top polluted cities list. Around 30% of its population was diagnosed with respiratory disorders caused due to air pollution [49] which led to 18,600 premature deaths every year [50].

Residential Area

In Delhi, most commonly found sources were traffic (23%), industrial emissions (29%), biomass burning (15.02%), and dust emissions (35%) [28, 29, 31, 32].

City center

Traffic (20.3%), biomass burning (13.95%), industrial emissions (10.04%), dust (25.89%), coal combustion (12.94%), and sea salt (4.4%) were the major sources near city centers in Delhi [28, 29, 33,34,35].

Industrial Area

Most important sources were industrial emissions (46.2%), dust sources (22.8%), and traffic-related emissions (10.2%) [28].

Agra

Residential Area

RSA studies conducted in residential areas of Agra [51, 52], one of the most polluted cities in Uttar Pradesh, identified traffic emissions (28%) and dust emissions (48%) as the major sources contributing to PM10. While to PM2.5, industrial emissions (30%), traffic emissions (17%), dust (27%), and other anthropogenic activities (18%) were the main sources. Dominant sources to water soluble ions in TSP were biomass burning and local soil [53]. Apportionment of PM sizes and trace metals in this region found the following sources [52]: traffic emission and soil dust (38.1%), fossil fuel combustion (25.8%), garbage burning, and other activities (25.7%).

City Center

RSA study of PM10 identified that traffic (27.45%), biomass burning (13.67%), industrial emissions (13.52%), and dust emissions (30%) as the dominant sources contributing to PM10 [36, 37]. Concentrations of PM sizes and trace metals identified the following sources [52]: traffic emission and soil dust (47.6%), vehicular wear and tear (33.8%), and biomass burning (17.9%).

Kanpur

Kanpur, which is located in central IGP region, poses constant threat of varying emission sources thereby leading to higher concentrations of PM. One of the studies conducted in a residential area [54] in this region found out traffic emissions (38.78%), industries (32.52%), and dust (17.07%) as major sources of PM1. Another study in the same location found that anthropogenic sources were crucial during winter, while crustal emissions are high in summer [55].

Eastern India

Kolkata

Population in Kolkata, one of the megacities in India, is suffering from 41.3 to 47.8% of upper and lower respiratory issues, respectively [56]. Number of RSA studies carried out in Kolkata are 57% lesser than Delhi.

City Center

In order to identify sources of total suspended PM, road side locations in Kolkata were studied and found that traffic emissions (42%), dust (17%), and industrial emissions (7%) were the major sources [57]. Extensive sampling conducted in this megacity to understand the sources of PM2.5 using PCA identified traffic emissions (38%), biomass burning (27%), dust (18%), and other secondary anthropogenic (11%) as the dominant sources [58].

Residential and Industrial

One of the studies conducted in 16 different locations of Kolkata (i.e., residential areas close to traffic junctions, coal fired power plants, industrial belts, waste incineration plants, cement factories, and brick kilns) identified major sources as traffic, industries, and coal combustion using PCA [59].

RSA studies in northeast India were mainly concentrated in the mid Brahmaputra valley region. One of the studies in suburban area, using enrichment factor (EF) found out that traffic induced emissions (38%), coal combustion (26%), industrial emissions (19%), and biomass burning (9%) were the dominant sources [60].

A study at a residential region [61], using PCA-multiple linear regression (MLR), marked three major sources: biomass burning (23%), dust (26%), and traffic emissions (22%).

Western India

Mumbai

Mumbai, the industrial capital of India, which is growing in all the commercial activities leading to the deterioration of air quality day-by-day. To understand these increasing sources, several RSA studies have been conducted in the past which are well discussed in the following section.

Residential Area

Sources to PM10 were dust (35%), sea salt (17%), coal combustion (12%), and traffic emissions (20%) and for PM2.5 were dust (26%), industrial emissions (4%), sea salt (14%), and coal combustion (7%) [62].

City Center

In this city, major sources observed were dust emissions (11%) and sea salt (11%) for PM2.5, and for PM10, traffic emissions (22%), biomass burning (18%), sea salt (19%), and dust emissions (25%) [62].

Industrial Area

In this region, dominant sources found were dust emissions (10%) and sea salt (8%) for PM2.5, and for PM10, traffic emissions (25%), biomass burning (15%), and dust emissions (27%) [62].

Ahmedabad

Ahmedabad, a semi-arid and well-populated area in western India, has all types of settings like residential, small- and large-scale industries, commercial activities etc. Identification of the major sources to TSP using PMF revealed that dust (57%), biomass burning (10%), vehicular emissions (17%), and sea salt (5%) are the resolved factors [63]. Study using PMF, in a city center, reported the dominant sources of PM10 as dust (37%) and biomass burning (33%), and industrial emissions (11%), sea salt (6%), and vehicular emissions (31%) for PM2.5 [64].

Nagpur

Nagpur, the centrally located fast growing metropolis city of India, is also victimized of poor air quality. A recent RSA study in which PM2.5 sampling was carried out in Nagpur using CMB found the following major sources: vehicular emissions (57, 62, and 65%), biomass burning (15, 11, and 9%), and dust emissions (6, 10, and 7%) in residential, commercial, and industrial regions, respectively [65].

Southern India

RSA studies over South India have immensely increased during the last decade and is extensively concentrated in Chennai, Hyderabad, and Visakhapatnam.

Chennai

Sources to PM in a city center was studied and identified as dust emissions (74%), sea salt (16%), and traffic emissions (10%) [66]. Another study conducted in an urban site using PMF reported that sea salt (40.4% in PM10 and 21.5% in PM2.5), traffic emissions (20% in PM10 and 11% in PM2.5), biomass burning (0.7% in PM10 and 14% in PM2.5), and dust emissions (3.4% in PM10 and 4.3% in PM2.5) are the major sources [67].

Hyderabad

Hyderabad, an emerging metropolitan 400-year-old city, has an increasing trend of urbanization since 1960 due to which a significant degree of air quality decline is being witnessed. RSA in the city center identified dust emissions (40%), traffic emissions (22%), coal combustion (12%), industrial emissions (9%), and biomass burning (7%) as dominant sources for PM10 and traffic emissions (31%), dust (26%), coal combustion (9%), industrial emissions (7%), and biomass burning (6%) as the source contributors to PM2.5 [68]. In a residential area, the common sources found were dust emissions (36 and 20%), traffic emissions (41 and 38%), biomass burning (6 and 9%), and coal combustion (6 and 12%) for both PM10 and PM2.5, respectively.

Visakhapatnam

Visakhapatnam, also known as the financial capital of state of Andhra Pradesh, had gradual decline of air quality and thereby turning out to be air pollution hotspot. RSA of PM10 showed dust emissions (22.5%), sea salt (9.7%), coal combustion (15.5%), industrial emissions (5.1%), and biomass burning (35%) in residential sites, and dust emissions (22.5%), sea salt (5.5%), coal combustion (26.1%), industrial emissions (7.8%), and traffic (14%) in industrial sites [69].

Central India

Durg

RSA studies in other cities of the Central India are few. At a city center of Durg in Chhattisgarh, source identification study [70] using PCA revealed two principal components which explain 76.6 and 65.9% of the variance for PM2.5 and PM1, respectively. One component had coal combustion, traffic emissions, and biomass burning (52 and 45%) and another was dust emissions (25 and 21%).

Raipur

Study near city center, using PCA [71], revealed that while coal burning (33.2%), dust (31.6%), and traffic emissions (18.4%) were prominent sources contributing to PM2.5, coal burning (50.7%), dust (25.3%), and traffic emissions (18%) were sources to PM10.

Summary of Sources in Different Regions of India

In residential areas in north, south, east, and west, the dust emissions (42%), traffic (41%), traffic (30%), and dust emissions (43%) were the dominant sources. While in city centers in north, south, east, west, and central India, dust emissions (28%), dust emissions (47%), traffic (40%), traffic (42%), and coal combustion (45%) were the main sources. In industrial regions in north, south, and west, industrial emissions (46%), coal combustion (26%), and traffic emissions (44%) were important sources, respectively. Overall, in India, dust (30%) and traffic emissions (31%) are the main contributors to PM.

Diverse Sources of Components of Polycyclic Aromatic Hydrocarbons over India

In the recent years, studies on RSA of PAHs are gradually increasing every year with majority of them in cities. Sources of PAHs in Kolkata, identified using suitable biomarkers, were coal combustion, vehicle exhaust, and wood burning [72]. RSA study of PM1 bound PAHs carried out using PCA reported that diesel vehicles and coal combustion were the predominant sources in Kanpur city [73]. In Delhi, a RSA study using PCA revealed that diesel, natural gas and lubricating oil combustion (49.5%), wood combustion (25.4%), gasoline (15.5%), and coal combustion (9.6%) as sources [74]. RSA results of PM10 bound PAHs found that vehicular exhaust, wood combustion, and coal combustion as most important sources in Mumbai city [75]. RSA of PAHs using PCA in Agra identified industries (44%), coal and wood combustion (22%), and gasoline driven vehicles (14%) as the major sources [76]. Another study of fine particulate phase PAHs in Chennai using PCA model identified that vehicular emissions, off-road combustion activities such as wood, solid waste, and other garbage as the dominant sources [77]. PM2.5 bound PAHs in Tiruchirappalli identified the sources using PCA as diesel engine emissions (31.71%), vehicle emissions (28.2%), and gasoline emissions (14.5%) as dominant sources [78].

Diverse Sources of Components of Volatile Organic Compounds over India

VOCs are the group of species containing non-methane oxygenated and halogenated hydrocarbons released from anthropogenic and natural sources. Toxicity of VOCs is severe where it ranges from respiratory irritation to carcinogenic effects such as lung, blood, and kidney. However, in Indian context, the focus on VOCs is far less compared to the importance given to PM. A study in city center of Delhi revealed following major sources using CMB: diesel engine exhaust (26–58%), vehicle exhaust (14–23%), evaporative exhaust (10–18%), auto repair (4–16%), degreasing (2–4%), and natural gas (2–12%) [79]. A recent study using PCA in a residential area of Delhi found the dominant sources contributing to VOCs as traffic emissions (37%), solvent usage and degreasing solvents (19%), and industrial emissions (13%) [80]. Studies carried out in Mumbai in a city center, using factor analysis, found that evaporative emissions (60–70%), vehicle exhaust composite (11–33%), degreasing (2–20%), diesel internal combustion (4%), and refinery (21%) as the important sources [81, 82]. In Kolkata, SA study using CMB in a city center observed vehicle exhaust (38.8–44.8%), coal combustion (up to 37.9%), and other source (pesticides, wood combustion, printing) as major sources [83]. Studies in Raipur (in an industrial site), Ahmedabad (at a city center), and Dehradun (at a city center) concluded that apart from contributions from vehicular emissions, emissions from natural gas and leakage of liquefied petroleum gas (LPG) can be major [84,85,86].

Local Numerical Models

Northern India

Urban

Industrial Source Complex (ISCST3) models were used to predict PM10 concentrations at 2-km resolution in an industrial area using an in situ emission inventory [87]. The model under predicted concentrations at seven locations (P/O 0.45–0.61). In another study, AERMOD was used to determine the total suspended particulate matter in an urban area using in situ emission inventory at 2-km resolution [88]. The model was found to satisfactorily predict concentrations in Delhi (P/O = 0.86). In another study using AERMOD [89] with in situ emissions revealed that while NOx and PM10 were predicted satisfactory (P/O = 1.22 and 0.85), SO2 was under predicted (P/O = 0.68). While AERMOD satisfactorily predicted total suspended PM (TSPM) concentrations in Delhi (P/O = 0.84), atmospheric dispersion modeling system (ADMS) under predicted the concentrations (P/O = 0.77) [90]. SPM model with in situ emission inventory was used to determine the roles of various sources affecting TSPM concentrations [91]. While the predictions when compared with observations using emissions from only vehicles were lower (P/O = 0.78), predictions with vehicle and natural emissions (P/O = 0.86), vehicle and domestic emissions (P/O = 0.87) and vehicle, natural and domestic emissions (P/O = 0.92) were found to be closer to observations. California Line source (CALINE4) model was used to predict concentrations in flat and hilly locations [92] using in situ emission inventory calculated using vehicle emission factors from ARAI [93]. While the model slightly under predicted concentrations in flat terrain (P/O = 0.72), it severely over predicted concentrations in hilly terrain (P/O = 3.14). Similarly, impact of diesel vehicles on air quality was studied using CALINE and Indian Institute of Technology Line Source (IITLS) models with in situ emissions [94]. Both models under predict concentrations in Delhi, with IITLS (P/O = 0.61) having a slightly better performance as compared to CALINE (P/O = 0.51). In another study, IITLS and ISCST3 were used with an in situ emissions [95]. Results showed that while IITLS model slightly over predicted Pb (P/O = 1.28), ISCST3 under predicted (P/O = 0.63) the same.

Semi-Urban

An assessment of air quality in an industrial area using GFLSM and ISCST3 with in situ emissions revealed that industries were the main source followed by vehicles [96]. Both the models under predicted NO2 concentrations (P/O = 0.23 and 0.73 using GFLSM and ISCST3, respectively).

Eastern India

Urban

Roadside air quality was evaluated at a busy traffic intersection using CALINE3, General Finite Line Source Model (GFLSM), and CAL3QHC [97] using vehicular emissions computed from emission factors provided by national environmental research institute [98]. Models were able to predict PM10 and PM2.5 concentrations satisfactorily. P/O for GFLSM, CALINE3, and CAL3QHC were 0.78, 1.21, 0.93 for PM10 and 1, 1.13, and 1.39 for PM2.5.

Semi-Urban

Similarly, assessment of contribution of SO2 and NO2 from different sources in an industrial area [99] using in situ emissions [100]. While NO2 was under predicted (P/O = 0.41), SO2 predictions were satisfactory (P/O = 0.98).

Western India

Urban

An unsteady and a steady-state dispersion model was used to understand the effect of meteorology on ambient concentrations using in situ emission inventory in an urban area [101]. While irrespective of mixing layer heights, the steady-state model had similar performance, the unsteady state model over predicted 24-h concentrations at 50-m mixing layer height (P/O = 1.5), and satisfactorily predicted concentrations at 100 m (P/O = 0.90). In an another study, NOx and PM concentrations were modeled using Operational Street Pollution Model (OSPM) [102] with ARAI provided emission inputs [93]. The model prediction in Mumbai indicates under prediction of PM (P/O = 0.56) and slight over prediction of NOx (P/O = 1.28). A study used chemical transport model to simulate PM10 concentration using in situ emission inventory at 1-km resolution [103]. The predictions were found to be satisfactory for Ahmedabad (P/O = 1.02) while Surat (P/O = 0.78) and Rajkot (P/O = 0.66) were found to be under predicted. Another similar study was carried out in an urban area using AERMOD to evaluate PM10 concentrations with WRF meteorological inputs [104]. Model under predicted concentrations at commercial (P/O = 0.63), background (P/O = 0.38), residential (P/O = 0.72), and sensitive (P/O = 0.31) regions.

Southern India

Urban

Effect of stone crushing units on PM concentration was evaluated in an urban area [105]. Three models FDM, ISCST3, and AERMOD were used for modeling dust emissions. Model results indicated dust concentrations to exceed INAAQS. Out of the three models, AERMOD was the better in predicting dust concentrations (P/O = 0.80) as compared to ISCST3 (P/O = 0.53) and FDM (P/O = 1.84). Furthermore, model simulation at other nearby areas was also carried out to determine the effect of quarry using AERMOD, ISCST3, and FDM: Saranagar (P/O = 0.54, 0.72, and 1.27), Zamin Pallavaram (P/O = 0.45, 1.02, and 1.12), Everedy nagar (P/O = 0.27, P/O = 6.09, and P/O = 0.16), Krushna nagar (P/O = 1.36, P/O = 5.37, and P/O = 0.19), and Kennady nagar (P/O = 0.52, P/O = 11.24, and P/O = 0.10). As observed, barring AERMOD, the other two models either severely over or under predicted concentrations. Another study used ATMOS to simulate PM10, SO2, and NO2 concentrations [106]. The PM10 concentration was found to be satisfactory in two urban locations (P/O = 1.20 and 0.91). Similarly, another study using SIM-air quality modeling tools in Chennai predicted PM10 concentrations satisfactorily (P/O = 1.03) [103]. In another study, ISCST3 model was used to study the impact of industries on air quality [107]. The model performance for SO2 at Hyderabad in the months of May and April indicated over predictions (P/O = 1.27 and 1.48, respectively).

Semi-Urban

AERMOD satisfactorily predicted PM10 concentrations at three semi-urban sites (P/O = 0.78–0.88) [108]. The study concluded that transportation, small-scale industries, and combustion of conventional fuels were the sources of particulate pollution. ISCST3 model performed satisfactorily (P/O = 0.89), while Gaussian Plume Model (GPM) under predicted (P/O = 0.67) NOx and SO2 [109].

Central India

SIM-air using ATMOS chemical transport model was used for predicting PM10 concentrations with in situ emission inventory at 1-km resolution in an urban area [110]. The simulated concentration of PM10 was found to be satisfactory (P/O = 0.82) when compared with observed values in Indore.

Summary of Local Numerical Modeling Studies in India

Similar to regional modeling studies, emissions resulted in huge uncertainties in local numerical models. ATMOS (west P/O-0.9, south P/O-1.03, and central India P/O-0.82) and AERMOD (north P/O-0.85, west P/O-0.50, and south P/O-0.78) models with improved treatment of emission inputs performed better compared to other models. AERMOD and ISCST3 were the most frequently used models.

Conclusion

Till date, no critical review which combined both local and regional air quality modeling studies, which is extremely essential for the proper formulation of air quality abatement strategies, was carried out in India. A total of 130 publications available in citation databases were intently reviewed and the principal findings are well comprised below:

  1. (i)

    Very few local and regional studies are carried out in central India. Moreover, owing to higher concentrations in Indo Gangetic plain, more studies are concentrated in northern India. For example, 74% RSA studies are documented from Northern and eastern India.

  2. (ii)

    Most of the PM studies in the country are either on PM2.5 or PM10; thus, more studies are required to identify sources and contributions of size resolved PM.

  3. (iii)

    RSA studies on PM are numerous when compared to VOCs and PAHs which are highly carcinogenic and need greater attention.

  4. (iv)

    Inconsistent selection of the source tracers holds the point of ambiguity in RSA studies.

  5. (v)

    Heterogeneous sources at different regions of India lead to varied spatio-temporal, physical, and chemical characteristics of PM. In northern region, while dust emissions are dominant source in Delhi, industry followed by traffic emissions are major sources in Kanpur and Agra. Traffic and dust emissions are commonly reported dominant sources followed by coal combustion in southern and eastern cities of India. Almost all the cities in coastal regions have profound sea salt contributions.

  6. (vi)

    Performance of regional models with finer resolution was better than coarser models.

  7. (vii)

    Irrespective of the emission inventory and the model used, most of the regional studies under predict PM concentrations in Delhi. While the same model set up is doing decent job elsewhere. This could be due to missing out a dominant source in the region. One way to approach the problem could be by modifying the existing emission inventories based on results from the local RSA studies.

  8. (viii)

    The problem of lack of reliable emission inventory is also evident from the analysis of local numerical models. For example, unavailability of load dependent emission factors for vehicles led to either under or over prediction of target pollutants in most of the studies.