Introduction

Near-road traffic-related air pollution (TRAP) is one of the most serious public health concerns facing many countries due to emissions of several toxic ambient air pollutants from on-road vehicles (Liang et al. 2018; Mannucci and Franchini 2017; Munir and Habeebullah 2018). Studies have shown that exposure to TRAP has resulted in the incidence of pulmonary diseases such as abnormal lung function (Bowatte et al. 2018), incident of cardiovascular diseases (Cai et al. 2018), respiratory cancer (Ribeiro et al. 2019), childhood asthma (Khreis et al. 2018), and diabetes mellitus (Pedersen et al. 2017). World Health Organization (WHO) estimated about 12–70% of total ambient air pollution levels in African and Asian (especially Middle Eastern) countries due to road TRAP (WHO 2019). This is mainly due to the increase in motorization rate, lack of adequate green infrastructure, the absence of engine powered electric vehicles, and urban development in nearby roadways, whereas TRAP comes into play (Korfant and Gogola 2019; Omidvarborna et al. 2018; Rosenlieb et al. 2018; Zhou and Gao 2018). Among the gaseous pollutants, nitrogen oxides (NOx) and carbon monoxide (CO) are the most common forms of near-road TRAP (Bodisco et al. 2019; Gokhale and Pandian 2007; Ko et al. 2019; López-Pérez et al. 2019; Triantafyllopoulos et al. 2019; Wallace et al. 2012). Thus, accurate measurements and model estimations are the fundamental factors of reducing the incidents of TRAP morbidity and mortality through policy formulations and mitigation actions (US EPA 2019).

Assessment of TRAP has well explored in parts of the world. For example, Perugu (2019) enhanced the emission rates estimates of CO, NOx, and hydrocarbons (HC) for light-duty vehicles (LDV) in India based on local driving conditions by adapting the United States Environmental Protection Agency (US EPA’s) Motor Vehicle Emission Simulator (MOVES) model. Gibson et al. (2013) found a good model performance between US EPA Regulatory Model (AERMOD) dispersion model output and that of the measured field data for near-road traffic-related particulate matter (PM) and sulfur dioxide (SO2) concentration levels in Nova Scotia, Canada. Also, many European cities have also used similar TRAP modeling approach such as URBan AIR (URBAIR) and Urban Air Quality Model (ADMS-Urban) models in evaluating the exposure concentrations of several air pollutants mostly NOx, PM, HC, and black carbon (BC) due to road traffic activities (Borrego et al., 2016; Mallet et al., 2018).

The literature review showed that the concentration levels of gaseous TRAP (e.g., CO and NOx) near major roadway were poorly investigated in Oman. In such environments with a high level of solar radiation, emission of TRAP is considered as the major challenge as it readily causes the production of ground-level O3 (Baawain and Al-Serihi 2014). The first and only study in Muscat was conducted by Abdul-Wahab and Fadlallah (2014), in which the California Puff model (CALPUFF) model was employed in simulating concentration of NOx and CO in a single road within Sultan Qaboos University. This study could not validate the predicted concentrations with the field data which were important in determining the prediction, performance, and the association of the modeled results with the measured data (Kadiyala and Kumar 2012). In addition, the study was limited to a local area, where the majority of vehicles were listed as LDV. Hence, the model could not predict busy roadways with different vehicular volume levels and vehicle categories (heavy-duty vehicles (HDV) and LDV). The second TRAP modeling study was focused only on the modeling of NOx and CO in Salalah city, Oman, by comparing the predicted pollutant concentration levels between monsoon and non-monsoon seasons with CALPUFF. The estimates were made during only peak and off-peak traffic volumes and did not present average traffic flows and emission in the Salalah city. Also, the reliability of the model in determining the predicted air pollutant concentration levels was not validated due to the lack of measured data (Charabi et al. 2018).

Most developed countries have established their own TRAP emission rate database for detailed road traffic emission modeling (Borrego et al. 2016; Milando and Batterman 2018; Righi et al. 2009; Wallace et al. 2012). However, literature review studies in Oman revealed poor background study due to lack of time-series air quality monitoring (AQM) data, local meteorological data, reliable emission rate database, and well-fitted models. This study aimed to evaluate the performance of two emission rate models (COmputer Programme to calculate Emissions from Road Transport (COPERT IV) and Comprehensive Modal Emission Model (CMEM)) to predict CO and NOx concentration levels from the near-road environment using AERMOD dispersion model. CMEM was considered for the study as it offers a flexible traffic input data requirements, including local meteorological parameters, whereas the acquired COPERT IV data presents emission factors of different vehicle categories similar to vehicle types under the current study. Both COPERT IV and CMEM are the most comprehensive emission rate models in terms of input parameters. Therefore, they can be utilized as the most reliable emission rate models for the countries with no established emission rate database.

In addition, AERMOD dispersion modeling has not been used widely for investigating traffic emissions from multiple road networks of varied vehicle volumes and types. This study monitored the gaseous TRAP at six major roadways by using two mobile ambient AQM stations for the first time in Muscat (see Fig. 1), and the developed dataset was used to validate the emission rate models, while it attempted to predict urban air quality at major hotspots.

Fig. 1
figure 1

A map showing road traffic air quality sampling locations in Muscat

Full size image

Background

Description of the study area

This study was carried out in Muscat Governorate (23° 36′ 51.5808″ N, 58° 32′ 43.0224″ E) which is the capital city of Oman located in the northern region of the country with a total area of 1500 km2 (Fig. 1) (Kwarteng et al. 2009). According to Directorate General of Meteorology (2019), mean annual precipitations, ambient temperature, and humidity levels of the Muscat are 150–750 mm, 23.1–33.6 °C, and 29–67%, respectively. Muscat is the most populated city of the country with a current estimated resident population of 1.5 million out of the total 4.6 million national population. Currently, the annual population growth rate is about 3.3–5.9% (National Centre for Statistics and Information 2018).

The rapid increase in road traffic population has been one of the major challenges by the Government of Oman, which is attributed to limited public transportation systems, subsidized vehicle importation taxes, and fuel prices. For example, an average of one vehicle per every household is estimated in Muscat (Amoatey and Sulaiman 2017). In addition, the number of newly registered vehicles in 2017 rose to as high as 175,266 compared to 131,700 vehicles in 2013, indicating a percentage increase of about 25% (Islam and Al-Hadrami, 2012). These have caused a high number of several LDV and HDV attributing to high air pollution levels in the major cities. The details of traffic volumes in these study are presented in “Vehicle population and composition.”

Road traffic condition in Muscat

On-road vehicle activities differ from one city to another in Oman. Currently, National Centre for Statistics and Information (NCSI) and Ministry of Transport and Communication do not have a comprehensive dataset on vehicle emission factors and vehicle kilometer traveled for different vehicle categories in Oman. But vehicle volume, distribution, and road inventories data have been a part of the 2011 national traffic survey study (Muscat Municipality 2019).

Our estimation showed that Muscat has the highest vehicle volumes compared to other cities in Oman. This is because the city served as a hub for most economic activities in the country. Majority of manufacturing companies, oil industries, government agencies, schools, hospitals, and hospitality industries are located in major industrial (Ruwi (RWI) and Rusayl (RSL)) and business (Al Khuwair (ALK), Al Amerat (AMT), A Seeb (ASB), and Muttrah (MRA)) Wilayats/town, thereby, leading to high volumes of HDV and LDV in the city, which varies in terms of vehicles types and population.

Furthermore, the strategic location of Oman sharing a border with the United Arab Emirates (UAE) and Yemen makes it efficient for HDV to transport goods and services through Muscat. Also, low vehicle occupancy rates in addition to the high temperatures in summertime (maximum of 40 °C) make personal vehicles the convenient means of transportation for individuals in Muscat (Amoatey and Sulaiman 2017). In addition, tourism is one of the emerging sources of revenue to Oman. For example, it is estimated that tourism activities contributed to the total vehicle population of 431,105 during monsoon seasons in Salalah city, a known tourist area in the southern part of the country (Muscat Municipality 2019). However, most of these visitors move back to Muscat, thereby increasing vehicle populations and emissions. All these factors have contributed to high emission levels in Muscat. Hence, to better understand the situation, this study examined on-road vehicle populations in various Wilayats in Oman and found that the highest on-road vehicle numbers and compositions are concentrated within the six main Wilayats/towns (ASB, ALK, MRA, AMT, RWI, and RSL) of Muscat.

Data and methods

Here, the study was conducted in three stages. For the first and second steps, road traffic data and location-specific meteorological data were employed to estimate traffic emission rates in each zone. Next, the ambient air pollution dispersion model was utilized to estimate the near-road pollutants concentration levels. Finally, the modeled concentrations were validated with field data through statistical analysis.

Vehicle population and composition

Annual vehicle population from Muscat Municipality was obtained based on the national traffic survey study conducted across the major Wilayats in Muscat (Muscat Municipality 2019). In order to understand the trends of vehicle volumes in Muscat, the total number of daily traffic flows from Monday to Sunday is presented in Fig. 2. There was a sharp decrease in vehicle volumes on Thursdays and Fridays as these days are the last day of the working days and first days of weekends compared to the main business days. Interestingly, though Saturdays are the weekend, yet there were high traffic flows as people visit places within the city thereby increasing the vehicle populations. As shown in Fig. 3, vehicle flows increased from 7:00 to 11:00 AM and showed a slight reduction from 17:00 to 21:00 PM during weekdays. Analysis of inter-city road traffic volumes showed that high traffic was found in ALK (16.5%), AMT (16.5%), and RSL (38.5%) compared to other Wilayats such as MRA (15%), RWI (7.1%), and ASB (6.4%) during morning time. On the other hand, in the evening time, ALK and RSL showed a higher number of vehicles as these are commercial and industrial centers with a high number of the population compared to other places (Fig. 3).

Fig. 2
figure 2

The average daily traffic volumes, the total number of cars, in Muscat (March–May 2011)

Full size image
Fig. 3
figure 3

The average daily traffic and peak volumes (AM-PM) of the six modeling locations on weekdays

Full size image

Road transport vehicles in the six major Wilayats were categorized into eight types: car, taxi, van/light goods, 2-axle truck, 3-axle truck, private and public transports, and others (i.e., those that could not fit into aforementioned classifications) as shown in Fig. 4. Despite the variations in vehicle types, the figure clearly shows that the population of cars was distributed across most of the locations. These were found mostly in business and residential areas (ALK, AMT, ASB, and MRA) where most of the vehicles are composed of cars, taxis, van/light goods, and private/public transports. Also, it could be observed that most industrial areas such as RWI (35–65%) and RSL (18–25%) had a higher proportion of 2- and 3-axle truck vehicles, respectively, than all the remaining locations where these HDV populations were found low (8–18%). Thus, data on types and numbers of vehicles within the study locations are important in order to assess their emission levels in different Wilayats according to the number of HDV and LDV (Wen et al. 2017; Zhang et al. 2019). Since the study aimed to model emission factors of CO and NOx from these locations, the various vehicle categories were broadly classified into HDV and LDV, where the proportion of these vehicle types was used as an input parameter for the CMEM as indicated in Table 1. In addition, location-specific meteorological variables such as temperature, humidity levels, wind speed, and ambient pressure, which were obtained through ambient AQM campaigns, were incorporated into the emission rate models. The detailed description of acquiring the field data is discussed in “Emission modeling.”

Fig. 4
figure 4

Average vehicle classification by volumes in six modeling locations in Muscat

Full size image
Table 1 Fleet composition and input environmental data for CMEM model
Full size table

Meteorological data

In this study, the surface and upper air meteorological data for January–December 2018 were obtained from Seeb International Airport (23° 35′ 43” N, 58° 17′ 54″ E) at an elevation of 8 m. These data have been analyzed through the Meteorological Assimilation Data Ingest System (MADIS), and then they were processed further by Breeze (2019) according to US EPA (2018a) meteorological processor (AEMET) guidelines. Here, the two meteorological datasets (upper and air-surface data) processed by AERMET were used as ready input data for the models. The detailed description of the meteorological factors including land use features of Muscat acquired from Seeb International Airport is indicated in Table 2.

Table 2 Surface and upper air meteorological information used for the air dispersion modeling acquired from Muscat International Airport
Full size table

Air pollution data

Field data

To collect traffic air pollution data and validate the model results, a series of field measurement campaigns were conducted from January to December 2018 across the six study locations. At each location, continuous AQM data for CO and NOx were collected with AirPointer AQM station (MLU/Recordum-GmbH, Austria) and an ETL3000 station (UniTec, Italy) based on European Union (EU) and US EPA directives (Pelletier et al. 2017; Weichenthal et al. 2015). CO was measured through non-dispersive infrared (NDIR) filter gas correlation procedure with a lower detection limit of 0.04 ppm and a precision of ±0.1 ppm. NOx data were collected via chemiluminescence method. The NOx detection limits and precision levels were 0.4 ppb and < 500 ppb ± 1% of reading > 100 ppm, respectively. In addition, hourly ambient temperature, pressure, and wind speed were measured at each location with an integrated meteorological station installed on the instruments. These meteorological data were applied for the simulation of the vehicle emission rates at each study location.

Quality assurance

Measurements were taken for 24 h for each of the six monitoring locations within January–December 2018 for both summer and winter seasons. The instruments were placed at a distance of 6.6–11 m away from the roads during the entire monitoring period. The maximum optimal temperature (+ 40 °C) for operations of the instruments was maintained. Calibrations including regular changes/cleaning of adsorbents were performed while ensuring zero in-built span gas in the instrument. After spending 1440 min of sampling from each location, the measured traffic data across the six locations were visually assessed for possible outliers and other data gaps (< 1% of the total measured data).

Emission modeling

US EPA (2018b) recommends that receptors should be placed at least 5 m or 25 m closer to the road source and later wider spacing of at least 100 m further away from the road sources. Following Liu et al. (2019) approach, high-resolution receptors from 5 to 100 m at 15 m intervals closer to the road source and later 50 m interval from 100 to 200 m further away from the road source were used for each of the modeling locations. Finally, the hourly (HH:MM) concentration of CO and NOx were predicted by the AERMOD from 00:00 to 23:00 for each road across the six locations.

CMEM model

CMEM model was used to calculate emission factors (g/veh km) for CO and NOx emitted from on-road vehicles in Muscat. CMEM is a microscopic traffic model developed by Centre for Environmental Research and Technology (https://www.cert.ucr.edu/cmem/) at the University of California-Riverside, USA. The microscopic nature of CMEM enables simulation of second-by-second tailpipe pollutants and fuel consumption rates for a wide range of vehicle types/technologies for operating conditions such as properly functioning, deteriorating, and malfunctioning states of the vehicles (Zhang and Ioannou 2016). CMEM was selected for the study because it is less data-driven and incorporate local meteorological data compared to other models such as US EPA’s MOVES model and Mobile Source Emission Factor Model (MOBILE6). In order to run the CMEM model, traffic microsimulation model (VISSIM) output of second-by-second velocity and acceleration rate data are required (Abou-Senna et al. 2013). Due to the lack of such data in Muscat road network, the default VISSIM data for this study was employed. Also, default data on model calibration parameters such as fuel, soak time, hot catalyst, and engine out–pollutant emission–coefficient values were used. Additionally, the proportion of fleet volumes defined as LDV and HDV from each of the study location was used as input parameter into the model. Further, location-specific meteorological data on ambient temperature, pressure, and wind speed were incorporated. Table 1 indicates the fleet data and environmental parameters used for the CMEM model. Employing local meteorological data in this study is crucial in predicting locally based emission factors due to harsh environmental conditions in Muscat. The CMEM emission factors were converted into emission rates as required by the dispersion model based on total vehicle population (Fig. 2). The roadway coordinates and the total area of each location (Table 3) were assumed according to the method used by Charabi et al. (2018) and Abdul-Wahab and Fadlallah (2014). Table 4 shows the final emission rate values for the dispersion model for all the six modeling locations.

Table 3 The road and terrain parameters of the selected locations
Full size table
Table 4 Emission rates for CO and NOx based on CMEM model
Full size table

COPERT IV emission data

The study also aimed at comparing emissions from CMEM modeling outputs with COPERT IV emission factor data to understand which of these emission factors perform well with traffic emissions in Muscat. Therefore, COPERT IV emission factor data (g/veh km) of CO and NOx were obtained from National Atmospheric Emission Inventory data available at http://naei.beis.gov.uk/data/ef-transport. Here, the emission factors were developed based on fleet composition, weight emission standards, vehicle specifications, abatement technologies, fuel types, temperature conditions, and driving behaviors. Table 5 provides a summary of emission factors and their associated process for CO and NOx. Air dispersion emission rates (g/m2/s) were estimated by considering the COPERT IV emission factors and average speed of 30 km/h (Table 5), total vehicle volumes (Fig. 2), and the area of the road (Table 3) for each of the six roads in Muscat (Abdul-Wahab and Fadlallah 2014). Table 6 lists the calculated emission rates for each location for air dispersion modeling study.

Table 5 COPERT IV road transport emission factors (NAEI 2019)
Full size table
Table 6 Modeled location-specific emission rates estimations based on COPERT IV data
Full size table

Air dispersion modeling

The individual roads in Muscat were modeled with US EPA Line source tool of the AERMOD dispersion model. The detailed description of the AERMOD model is available at Amoatey et al. (2018a). The emission rates (g/m2/s) from CMEM and COPERT IV data were applied to the model to estimate the near-road dispersion of CO and NOx. The start and end UTM coordinates (m) of each road, release height, and the initial vertical dimension (σzo) were calculated by dividing the effective height by 2.15 (Thus, the point of release of the pollutants should be 2.15 times the vertical dimension, σzo) as shown in Table 3 and Table 7, respectively (US EPA 2018b). All these data were used as an input parameter of the US EPA Line source in AERMOD (Abdul-Wahab and Fadlallah 2014; US EPA 2018b). Here, the line source uses the area source algorithm with the assumption that the traffic emissions (g/m2/s) are uniformly dispersed across the dimensions of the line sources. Also, the model does not simulate the horizontal meander component of the pollutant dispersion compared to point and volume sources. Thus, the line source is used to predict concentrations of the pollutants located within the dimension of the source (US EPA 2018b). There is no reliable background concentration data available for these sites for years preceding this study, thereby the lowest recorded concentration value was utilized as background concentration in each site.

Table 7 AERMOD EPA Line source input parameters for the modeled locations
Full size table

Model evaluation

The performance of the AERMOD (US EPA Line source) model was determined by comparing the predicted values of hourly CO and NOx concentrations with the observed values from 00:00 to 23:00 for January–December 2018 for each of the six locations. These evaluations were conducted by using US EPA recommended air quality statistical models (Mohan et al. 2011). These statistical performance measures have been previously applied as common framework for European Union Initiative on “Harmonizing within the Atmospheric Dispersion Modeling for Regulatory Purposes” (Cai and Xie 2011). To achieve the model evaluation, three linear statistical measures including fractional bias (FB), normalized mean square error (NMSE), and correlation coefficient (R) were considered. The FB and NMSE are both quantitative statistical models used to determine over-/underprediction and accuracy of the model performance, respectively, whereas qualitative statistical measures (R) determine the association between the modeled and the observed results. Geometric mean bias (MG) and geometric variance (VG) performance measures were also employed to account for a more balanced treatment of extremely high and low predicted values based on logarithm transformation, of which FB and NMSE cannot do. In addition, bootstrap resampling statistical technique was used to estimate the confidence interval (CI) at 95% using NMSE, FB, and R performance measures for COPERT and CMEM models. This is because they are deemed as the most basic statistical measures, and they are more suitable for mid-range model output pollutant concentrations to evaluate air quality models (Kadiyala and Kumar 2012). This analysis will help to establish whether the performance of COPERT and CMEM model outputs are statistically different from 0 at 95% CI compared with observed data or not. In this study, all the primary and bootstrap resampling analysis were performed through Microsoft Excel® (Version 2016) and BOOT® Software (version 2.01), respectively. All the statistical performance measures are presented by Eqs. 1–5 below.

$$ \mathrm{FB}=\frac{\overline{\mathrm{Co}}-\overline{\mathrm{Cp}}}{0.5\left(\overline{\mathrm{Co}}+\overline{\mathrm{Cp}}\right)} $$
(1)
$$ \mathrm{NMSE}=\frac{\Big(\overline{\mathrm{Co}-\mathrm{Cp}\Big){}^2}}{\Big(\overline{\mathrm{Co}\mathrm{Cp}\Big)}} $$
(2)
$$ R=\frac{\Big(\overline{\mathrm{Co}-\overline{\mathrm{Co}}\Big)\ \left(\mathrm{Cp}-\overline{\mathrm{Cp}}\right)}}{\sigma \mathrm{Co}\ \sigma \mathrm{Cp}} $$
(3)
$$ \mathrm{MG}=\exp \left(\overline{\mathrm{lnCo}}-\overline{\mathrm{lnCp}}\right) $$
(4)
$$ \mathrm{VG}=\exp \left[{\left(\overline{\mathrm{lnCo}}-\overline{\mathrm{lnCp}}\right)}^2\right] $$
(5)

The symbols Cp, Cο, \( \overline{\mathrm{C}} \), σ and ln denote the predicted concentration, the observed concentration, the average of the concentrations, the standard deviation, and the natural logarithm, respectively. Figure 5 illustrates a flowchart describing the steps of the entire modeling study in “Data and methods.”

Fig. 5
figure 5

Full-chain road traffic modeling processes of Muscat

Full size image

Results and discussion

The COPERT IV emission data and the estimated CMEM emission rates were employed as input for EPA Line source of the AERMOD dispersion modeling system in prediction of CO and NOx. Also, the measured concentrations and allowable threshold limits of the pollutants are presented and compared with the predicted concentrations for the six roads located in ALK, AMT, ASB, MRA, RWI, and RSL. The analysis of the results consists of three components, including CO concentrations, NOx concentrations, and model evaluation.

CO concentrations

By selecting CO as a case study, distinctive predicted concentrations for COPERT IV and CMEM emission rates input data were obtained. Table 8 presents the top five maximum hourly concentrations of CO for the three main traffic rush hours in Muscat defined as morning (6:00 and 8:00), afternoon (13:00), and evening (20:00 and 22:00) peak hours along with the near-road locations. It can be seen that although the model predicted higher concentrations of CO for COPERT IV input data compared to CMEM across all the six roads, most of the predicted concentrations were within the US EPA standard (40,096.1 μg/m3). However, MRA experienced two highest predicted concentrations during the time 20:00 (83,038.83 μg/m3) and 22:00 (55,695.54 μg/m3), thereby exceeding the US EPA’s allowable limit. Similarly, slightly higher prediction of 41,455.64 μg/m3 than the US EPA limit also occurred in AMT during the 22:00 period.

Table 8 Comparison of top 5 peaks modeled AERMOD values from COPERT IV and CMEM input values with observed concentrations of CO among the six modeled locations in Muscat from 00:00 to 23:00
Full size table

Moving to the measured concentrations of CO, they were also found below the US EPA standards similar to most of the model concentrations. However, higher concentrations were measured for all the three peak hours in MRA and AMT exceeding an average of 10,000 μg/m3 compared to the remaining locations. Figure 6 shows a time-series concentrations levels of CO for each location measured over 24 h. Interestingly, the results from this study showed consistency in trends for higher concentration levels of CO for both the predicted and the measured values in these two locations (MRA and AMT) (Table 8). The similarities in the concentrations of these two locations may be attributed to the influence of meteorological factors due to its proximity to the sea (Sultan Qaboos Port) and mountainous terrain features compared to the other roads (Abdul-Wahab and Fadlallah 2014). Also, the measured and predicted concentrations may not be due to poor dispersal of the CO emissions as these locations are characterized by several mountains (Zhang et al. 2019). In general, visual comparison of all the predicted and the measured concentrations showed increased in CO levels during the morning where traffic was at the peak and decreased when traffic volumes reduce and later increased again during evening hours. Thus, these variations in CO concentrations may be highly dependent on traffic emissions.

Fig. 6
figure 6

Average measured hourly concentrations of CO (top) and NOx (bottom) from each modeling location

Full size image

NOx concentrations

For the emissions of NOx, Table 9 shows the AERMOD predicted concentrations of COPERT IV and CMEM emission rate input parameters for the six modeled locations in Muscat. The predicted concentrations were visually compared with the measured and the allowable limit established by the US EPA threshold limit of 188.2 μg/m3 for NO2 since, currently, there is no criterion set for NOx emissions (US EPA 2018c). It could be observed from Table 9 that the two emission rate data of the model predicted maximum hourly NOx concentrations of about 10–100 folds higher than the observed concentrations. The model prediction based on CMEM emission data for AMT and ASB was found very low, thereby meeting the US EPA standard of 188.2 μg/m3 for NO2. Despite the high predictions made by the model in certain hours of the day, there were a similar pattern in terms of increase, decrease, and increase in NOx concentration levels with respect to morning (6:00 and 8:00), afternoon (13:00), and evening (20:00 and 22:00) peak hours across all the model locations, respectively. For example, in ALK, the model predicted high NOx levels at 6:00 to be 2087 μg/m3 which later decreased drastically to 275 μg/m3 at 13:00 increased again to 902 μg/m3 during the peak hour of 22:00 based on the CMEM emission data. Similar to COPERT IV data, the model predicted concentrations of 89,432 μg/m3, 11,801 μg/m3, and 38,646 μg/m3 during the same three peak hours of the day as described above. The high predictions of COPERT compared to CMEM could be due to the utilization of local or location-specific traffic and meteorological data during CMEM emission rates estimation, whereas such local data was lacking in COPERT. These variations in the predicted concentrations attest that the NOx levels were primarily attributed to traffic emissions. Also, the road traffic NOx concentration levels were found to be very low with average hourly concentrations (22 to 160 μg/m3) among the six roads. Figure 6 also shows detailed time-series NOx concentrations measured from each location over 24-h durations. These measured values were found to be very low compared to the predicted concentrations (Table 9), which could be due to rapid transformation of the NOx to O3 through photochemical reactions as solar radiation levels are generally high in Muscat (Agudelo-Castaneda et al. 2014; Directorate General of Meteorology 2019). This low model prediction levels were also evidenced by the lower measured NOx levels (Table 9). According to Amoatey et al. (2019), nitric oxide (NO) is very unstable gas and is the major component of NOx (95% vol. NOx) when emitted from vehicular tailpipes. These imply that most of the NOx might have been converted to an unstable NO thereby decreasing the measured and the predicted concentration levels across the modeled locations. Thus, it is imported to incorporate NOx (NO2–NO–O3) chemical transformation algorithms into future traffic modeling studies in order to improve the accuracy of the predicted concentrations levels (Halonen et al. 2016).

Table 9 Comparison of top 5 peaks modeled AERMOD values from COPERT IV and CMEM input values with observed concentrations of NOx among the six modeled locations in Muscat from 00:00 to 23:00
Full size table

Model evaluation

The statistical performance measure outputs for the measured and predicted concentrations across the six locations are summarized in Table 10. In overall, the statistical models predicted the concentrations of CO and NOx moderately well and poor for certain locations. In ASB, FB model revealed moderate propensity of AERMOD to underestimate NOx (FB = 0.39) and CO (FB = 0.47) concentration levels for CMEM and COPERT IV input data, respectively. However, FB showed a low tendency for the COPERT IV model to over-predict CO (FB = − 0.23) in AMT. For the performance measures, there was satisfactory model performance based on NMSE statistics in ASB for CO = 0.44 and NOx = 0.5 for COPERT IV and CMEM input parameters, respectively. However, as shown in Table 10, NMSE results for CO in AMT (0.95) and MRA (0.69) were found to be moderate. The above model performances of FB and NMSE were based on the criteria used by the study conducted by Mohan et al. (2011); thus, a model prediction is deemed acceptable if − 0.5 < FB < + 0.5 and NMSE < 0.5.

Table 10 Linear performance measures of AERMOD predictions based on COPERT IV and CMEM input values and the measured concentrations across the six locations
Full size table

It is very clear that very few locations had moderate prediction and performance while the majority showed very poor over-prediction and performances with the measured results (Table 10). Regarding the relationship between the modeled and measured pollutants, the association (R) were found to be moderate in all the locations for CO (0.51–0.64), with an exception of MRA (R = 0.25), which showed a weak correlation. For NOx, very good correlation was found in ALK (R = 0.70), while moderate association were found in AMT (R = 0.50) and RSL (R = 0.49). The detailed statistical associations between the predicted model results based on COPERT IV and CMEM input data and the observed concentrations in each location are shown in Table 10. In addition, due to the varying (extremely low and high values) predicted concentration levels of both COPERT and CMEM outputs, MG and VG performance measures were utilized to evaluate these predicted concentrations based on a logarithmic scale (Table 11). According to the standard model acceptance criteria range for MG (0.70–1.30) and VG (1.50–4.00) suggested by Chang and Hanna (2004), the model performance for the current study seems moderate; however, several locations could not meet the above performance. For example, COPRET IV showed good performance for CO (MG = 1.07) in AMT. For the same model, the VG statistical measure for several roads including ASB (2.33), MRA (2.75), and AMT (2.00) had predicted CO levels within the above model acceptance criteria. In the case of CMEM model, a similar situation was also found in RSL where the estimated VG = 1.91 satisfied the acceptable limits for CO. As indicated in Table 11, the model evaluation showed weak performance towards NOx. For example among all the location, only ASB (VG = 2.32) satisfied the performance measure for CMEM model. It should be noted that, in a more stringent model acceptability criteria, both COPERT IV and CMEM models will show poor performance (Amoatey et al. 2018a; Kumar et al.1992).

Table 11 Logarithmic performance measures of AERMOD predictions based on COPERT IV and CMEM input values and the measured concentrations across the six locations
Full size table

Tables 12 and 13 indicate detailed 95% CI estimates over NMSE, FB, and R obtained from bootstrap resampling analysis for both individual models (COPERT and CMEM) and between models for CO and NOx (Kumar et al. 2012). In ALK, the bootstrap 95% CI estimates over NMSE and FB for CO were statistically different from 0 (the signs for lower and upper limits did not change) for the individual model comparison except for R (due to change in signs for the CI limits) in CMEM model (Table 12). Intra-model (combined) comparison also indicated statistical significance for all the performance measures for both COPERT and CMEM models as listed in Table 13. A similar situation was also recorded for NOx where individual model 95% CI estimates over NMSE and FB were found not to be statistically different from 0 with the exception of R for both models. However, all the performance measures (NMSE, FB, and R) were statistically different from 0 between model comparisons. Also in AMT, the individual comparison for CO showed that all the measures were not statistically different from 0 except NMSE for NOx; however, both NMSE and FB were found to be statistically different from 0 (Table 12). Interestingly, comparison of CO and NOx was significantly different from 0 between COPERT and CMEM models (Table 13).

Table 12 Inter-model (individual models) comparison using bootstrap CI 95% estimates over NMSE, FB, and R for CO and NOx
Full size table
Table 13 Intra-model (between models) comparison using bootstrap CI estimates over NMSE, FB, and R for CO and NOx
Full size table

Interestingly, in ASB, both CO and NOx bootstrap 95% CI estimates for all the measures were statistically different from 0 for both individual models and between model performances with the exception of R for NOx which was not statistically different from 0 for individual model performance. The study found an interesting model performance evaluations in RWI, where all the pollutants (CO and NOx) showed statistical significance (at 95% CI) for both the individual models (COPERT and CMEM) and between model comparison (Tables 12 and 13). Also for individual model performance comparison, the bootstrap 95% CI estimates over NMSE and FB were statistically different from 0 for CO and NOx for all the models in MRA, whereas R was not found significantly different from 0. Similar performance was also observed in RSL for CO and NOx for the individual COPERT and CMEM model comparison except for NMSE (in CMEM model), which was not statistically different from 0 (Table 13). Finally, the bootstrap 95% CI estimates for the combined model comparison of CO and NOx for both MRA and RSL were found statistically significant from 0 (Table 13).

In terms of modeling study, the study faced several limitations including the inability of the EPA line source to simulate vertical dispersion of the pollutants but rather considers the horizontal distribution of the pollutants long the road, which could lead to over prediction and underestimation of the predicted pollutants. Also, the lack of background data for CO and NOx in close proximity to the modeled locations could be a major factor in the overall performance of the model predictions. Khreis et al. (2018) found that incorporating background NOx concentrations (ranging from 8.5 to 71 μg/m3) other than road sources could enhance the performance of the model. In addition, most roads have planted green vegetation (mainly turf grass) and these turf grasses near the roads are frequently managed through mowing activities (Amoatey et al. 2018b). Thus, the emission from the mowers could be the major contributor to background CO and NOx concentrations.

Although the study estimated emission rates of CO and NOx with CMEM model by using road-specific meteorological factors and the proportion of LDV and HDV for each road, there was a lack of second-by-second vehicle velocity and acceleration data, vehicle engine emission data, as well as the most recent traffic data. All these factors affected the CMEM emission rate data and the performance of the model with respect to measured values. Thus, in order to ensure accurate prediction of traffic emissions from roads in Oman, it is crucial to develop a local emission rate data and air dispersion modeling system since Oman has different meteorological and terrain features compared to developed countries where these models were adapted from.

Despite the limitations, this study provides very useful datasets about the road traffic emission modeling in Muscat and many arid countries compared to previous studies (Abdul-Wahab and Fadlallah 2014). This is the first study where US EPA Line source of AERMOD model has been used in modeling of different on-road vehicles with varied vehicle volumes and compositions under different meteorological factors and terrain features. Furthermore, additional modeling study will be required to incorporate the effects of complex terrain features of Muscat on the pollutants concentration levels.

Conclusions

In this study, the concentration levels of CO and NOx were predicted across six different major roads with various traffic vehicle categories and volumes in Muscat using US EPA Line source tool of AERMOD dispersion modeling system. Two different emission factor models (COPERT IV and CMEM) were established based on fleet composition, weight emission standards, vehicle specifications, abatement technologies, fuel types, temperature conditions, and driving behaviors. Studies have proved that the near-road TRAP estimations were attributed to street level environment and specific conditions including low winds, LDV vehicles only, and limited receptors only. Hence, the field monitoring campaign of CO and NOx along the roads under normal traffic, road, and meteorological conditions were conducted to help in the validation of the modeled results. The results proved that the predicted and the measured CO and NOx were within the acceptable limit set by US EPA. However, certain locations were found to have exceeded the standard, especially from high NOx levels. Also, the overall performance of the predicted concentrations with the measured data was moderate, except some locations that showed significant over/under prediction as well as the poor performance with the measured values. Taking into consideration the overall primary model performance measures, the study found COPERT emission factors showing moderate prediction compared to the CMEM; this may be due to the limited local (arid) vehicle second-by-second velocity and acceleration rate input data of the CMEM model. However, the bootstrap 95% CI estimates over all the performance measures (NMSE, FB, and R) of between model (COPERT–CMEM) comparison for both CO and NOx were statistically significant from 0 across all the model locations. For the individual model comparison, bootstrap 95% CI estimates over NMSE and FB were the most dominant statistical measures that were significantly different from 0 for both models. The contribution of the complex and varied geographical landscape features in Muscat on traffic emission levels was not considered under this study, except road geometry (road length and width), as the model could not account for the effect of complex and hilly terrain features. The major limitation of the model validation was due to the lack of reliable background CO and NOx data in addition to the emissions from other roads which were not a part of the study.

This is the most and the only comprehensive study conducted in Muscat urban and industrial areas, which was supported by the Omani government. As a result of this study, the government aimed to map air pollution hot spots and set up a permanent network of air pollution monitoring stations around the city. The reported results, which are supported by high-quality dataset, would be an essential infrastructure for emission forecasting, which provides a basis for future urban and city-scale air pollution studies and it can be expanded into future health and epidemiologic impact assessments. In the case of arid environments in developing countries, the methodology followed in this study can be utilized in other countries with similar environmental climate, such as developing countries in the Middle East; Belize and El Salvador in Central America; Peru (Southern America); and Arizona, Texas, and Oklahoma in the USA (Checkley et al. 2000; Heusinger and Sailor 2019; Song et al. 2017). This study would contribute to enriching the existing body knowledge and in future studies to filling this particular gap in knowledge. Future traffic modeling study may have the high potential of improving the prediction of traffic emissions for accurate exposure assessment in residential areas in Muscat and Oman at large. Finally, a dynamic model evaluation approach involving detailed meteorological information should be considered for future study.