Integrated air quality assessment using monitoring data, emission inventory, and aermod dispersion modeling: a case study of Sohar Port and Freezone, Oman

Industrial and port areas require integrated air quality assessment methods combining monitoring observations, meteorological conditions and atmospheric dispersion modeling. In this study, we have investigated the pollution behavior and industrial source contributions in the Sohar Port and Freezone industrial area of Oman for a multi-year period of ambient air quality monitoring data and meteorological observations and AERMOD dispersion modeling results for PM 10, NO 2, SO 2 and CO. All the pollutants were compared according to their dominant source properties, atmospheric behaviour and importance to industrial air quality management. The model performance was evaluated using fractional bias, normalized mean square error, index of agreement, geometric mean bias, geometric variance and predicted to observed concentration ratios. PM 10 concentrations were strongly influenced by regional background dust and resuspension processes, whereas NO 2 and SO 2 were more closely related to combustion and industrial point sources. The validation results for PM 10 and SO 2 were better agreed on than for NO 2 and CO. The results were affected by uncertainty regarding background concentrations, fugitive and traffic emissions, completeness of emission inventory and monitoring site representation. The study provides a practical and regionally relevant approach to assessing industrial air quality in arid coastal environments. The findings indicate the importance of improving emission inventories, characterizing background pollution in the region, accounting for fugitive and traffic pollution and optimization of monitoring infrastructure to help in sustainable industrial development and evidence-based air quality management.

Air pollution is one of the main causes of premature deaths even though clean air is essential for human health. Every year millions of people die of pollution-related diseases. Nitrogen dioxide (NO 2), sulfur dioxide (SO 2), particulate matter (PM 10 and PM 2.5) and ground-level ozone (O 3) are also major contributors to serious health problems including respiratory and cardiovascular diseases. It is a major environmental issue in developed countries where rapid development and the release of unhealthy pollutants are a big problem. To meet this objective, several cities are using AQMN to measure and calculate pollutant concentrations and for other purposes report using the Air Quality Index (AQI).

From health point of view, a good understanding of six criterion pollutants is important (Silva et al., 2003). The increase in the levels of six pollutants NO 2, SO 2, PM 10, PM 2.5 and O 3 at ground level impacts the quality of air and human health (Rupakheti et al., 2021). Several epidemiological reports indicate that certain pollutants are primary risk factors for diseases associated with cardiovascular and respiratory conditions, which are correlated with lung and heart diseases and mortality (Charlson and Rodhe, 1982; Haq, 2022; Karimi et al., 2019; Ravindiran et al., 2023; Taheri Shahraiyni and Sodoudi, 2016).

The AQI is a dimensionless numerical scale designed to communicate air-quality conditions and their potential health implications to the public (Singh and Chaturvedi, 2023: Yousefi et al., 2019). This score aims to indicate the immediate impact of air pollution on public health (Ravindiran et al., 2023). The measurement of AQI is influenced by temperature, wind speed, relative humidity, and environmental variables such as pollutants (PM 2.5, PM 10, SO 2, NO 2, CO, and O 3) (Rao et al., 2024; Ravindiran et al., 2023; Yousefi et al., 2019). The relationship between AQI and future air quality is also important to predict air quality (Sobirin et al., 2023). It is better to keep this air quality at a minimum and implement policies to reduce the pollution of particulate matter. The accuracy of AQI is very important for society and human health. Forecasting air pollution levels also involves deterministic, traditional statistical and computer methods such as machine learning and traditional analysis methods. Deterministic methods are derived from stochastic processes, statistical theory and several general statistical methods based on multiple linear regression and time series analysis. Dust is a major constituent of the air quality in the Gulf Cooperation Council (GCC) region, both natural and anthropogenic. Dust storms are common in the area, and they have a great effect on visibility and air quality. Air pollution is also caused by industrial activities, traffic jams and the use of energy. Oman has been ranked among the worst countries for air quality with PM 2.5 concentrations above WHO recommendations in the 2021 World Air Quality Report. Dry regions such as Al Batinah in Oman can release large amounts of particulate matter into the atmosphere through dust storms or “desert dust.” Some of it is transported to and from the site over long distances, and some of it is brought to and from the site. Understanding the sources and effects of these pollutants is key to clean air and development. Wind erosion causes destruction of 46% of degraded land all over the world (Rashki et al., 2015). The most important and active sources of dust emissions in the world are in Oman, and the GCC region in the dust belt (Darvishi Boloorani et al., 2022; Kalantari et al., 2024; Miri et al., 2021). These breezes transport fine and light sediments in the dry season to affect the provinces of Sistan Plain, Sistan and Baluchestan, Afghanistan, Pakistan and Oman. However, they have limitations (for example, they assume that the variables are linear and therefore they are not valid for non-linear air quality data). Machine learning (ML) techniques are increasingly applied in the field of urban air quality analysis because of their ability to address non-linear problems.

Air quality indices (AQIs) aggregate concentrations of specific air contaminants into a single numerical value using appropriate methodologies. These indices are intended to meet public demand for accurate, easily comprehensible air pollution information by standardizing and creating standardized air pollution statistics. For AQI to be developed, there are core assumptions, which are communication of basic public knowledge on a common level: main criteria pollutants and their relationship, ensuring that it is easily scalable by adding more pollutants with different average pollutants and their durations in the population in a way with national ambient air pollution program requirements of AQI guidelines, the AQI cannot become “eclipsing,” to eliminate “ambiguity,” to serve as an alarm system, and to rely on the accuracy of data from location-specific measurement stations, i.e., particular measurement stations.

To aid in the management of air pollution issues at specific locations, the Air Quality Index (AQI) – also known as the Air Pollution Index (API) or the Pollutant Standards Index (PSI) was created. Assembling the AQI, several environmental groups and scientists have worked together to produce a comprehensive, widely used standard for use in most countries and regions. Nevertheless, there is no universally accepted approach to developing AQI systems. This analysis critically evaluates many AQI systems, including the Pollutant Standards Index (PSI) issued by EPA in 1976. The daily PSI is governed by five primary air pollutants: PM 10, O 3, SO 2, CO, and NO 2. But PSI does not factor in exposure to other pollutants, some of which can be fatal for people with respiratory disease. Challenges in analyzing air quality in Oman and at Sohar Port include inadequate monitoring infrastructure, calibration and maintenance issues, and difficulties in identifying pollution sources.

We propose an all-encompassing framework for environmental monitoring to overcome these obstacles. There are several gaps, including limited engineering tools, insufficient data integration, and insufficient temporal and spatial resolution. More data should be integrated with other platforms to better understand the number of stations needed to be deployed in populated and industrial areas. More collaboration is required among government agencies, corporate clients, and other interested parties to improve monitoring infrastructure, integrate widely available datasets, and encourage data availability. Such an approach can reveal pollution hotspots and guide evidence-based decisions to maintain public health and protect the environment.

Therefore, the novelty of this study is based on integrating long-term ambient air quality monitoring data with AERMOD dispersion modeling for a complex industrial coastal area in Oman. The study does not only present concentration levels, but also evaluates how industrial source emissions, meteorological conditions, monitoring station locations, and data limitations can affect the interpretation of air quality results. This is important because air quality assessment in arid coastal industrial areas such as Sohar Port is influenced by both industrial sources and regional background dust. The work therefore provides a practical framework for air quality assessment and pollution management in data-limited industrial regions.

Study area

The study area was located in North Al Batinah in Oman. It is the largest industrial complex in Oman (SOHAR, 2024) as it has the Sohar port and free zone and the city of Sohar Madayen (Figure 1). It has a hot, humid climate with summer temperatures of over 50 °C. Very low rainfall with an average no more than 100 mm only occurs in each year. In Oman this climate that called hyper-arid climate is characterized by two main seasons yearly with on clear present of spring and fall as the winter, starting in November and ending in March, and the summer, in April and October. It experiences a temperature from 3 °C in winter to 48 °C in summer, with heavy rainfall in the northwest. Sohar Wilayat has a hot summer and a moderate winter, with low annual rainfall. The average relative humidity is around 92%, with the dominant wind direction being from the east.

The topography of the project sites is similar to that of undeveloped areas along the Al Batinah coast but has been altered by industrial and economic development and by road network expansion. The Al Batinah North area is composed of alluvium gravel and the piedmont zone, with the piedmont areas consisting of ancient gravel terraces and modern alluvium. The study area is almost 95% covered with sub-recent alluvial fans and lowermost terrace deposits.

Air quality data collection

There are many ways of collecting air quality data, and they are carefully designed to get a good measurement of different aspects of the air pollutants. Different pollutants can be measured with standard methods. These are PM 10, PM 2.5, NOx, SO 2, O 3 and volatile organic compounds (VOCs). The methods are passive and active sampling, remote sensing and continuous monitoring. The reference methodology was used in this study because it is the most accurate technical method to measure global air quality for regulatory and research purposes. Analyzers used in reference stations of air quality monitoring as per the reference method of USEPA. This is considered the most effective and accurate method of measuring air quality in the environment. They are constructed and put together in a certain way to measure air pollutants with standard reference methods. These analysers are also called reference-method analyzers. “EPA (United States EPA) says these analyzers are the best way to measure air pollution. The exterior configuration of the monitoring station and its internal analyzer and data-acquisition systems are shown in Figures 3A,B. They ensure that air quality data from various monitoring stations and agencies are available in same format and can be compared with each other. The continuous ambient air monitor (CAAMS) is a permanent ambient monitoring station located in a city, industrial and rural area to measure air pollution. The CAAMS system, therefore, provides access to analysers for PMs, NO 2, SO 2, O 3, CO and volatile organic compounds (VOCs). These analyzers are based on well-known measures like chemiluminescence, ultraviolet absorption and laser-based spectroscopy to provide accurate real-time information about the amount of pollutants (Verghese and Nema, 2022). Reference method analyzers for PM detection use beta attenuation, tapered element oscillating microbalance (TEOM) and filter-based gravimetric methods to get very accurate PM measurements and chemiluminescence and UV fluorescence for NO 2 and SO 2 measurements (Galán-Madruga et al., 2022). Reference method analysers are calibrated and maintained according to USEPA guidelines and manufacturer’s manual. Engineers make regular check-ups to ensure the highest data are measured by the stations (U.S. Environmental Protection Agency, 2017).

Air emissions monitoring

Air emission monitoring is often conducted using a variety of methods, ranging from routine sampling (e.g., manual sampling) to automated sampling systems for continuous air emissions monitoring, including traditional and next-generation automated systems. Manual sampling methods, which are the grab sampling methods (e.g., grab sampling, isokinetic sampling) or isokinetic sampling techniques, are prevalent for emission monitoring such as grab sampling and isokinetic sampling for emissions monitoring from stationary sources, such as industrial stacks, are the tools used to obtain short-term data, which is used for high-quality high accuracy data, and reliable for monitoring the long-term performance. Nevertheless, these techniques are usually quite labor-intensive and may poorly cover temporal variation. An automated systems can be used to continuously monitor pollutants, e.g., continuous emission monitoring systems (CEMS), predictive emission monitoring systems (PEMS), etc. Pollutants include SO 2, NOx and particulate matter. The Continuous Systems (CEMS) are widely used to monitor emissions and are also used by the regulatory authorities, and this in turn leads to a rising demand for high-resolution data (U.S. Environmental Protection Agency, 2026). Oman has been using (CEMS) since 2004 to meet national industry emissions standards set in MD118/2004.

Most emission facilities use this system to monitor facility-level and in-stack concentrations in real time. The CEMS-generated data are highly valuable for the development of emission factors, the construction and accounting of fine industrial pollutant emission inventories, and the study of changes in pollution emissions, etc. Scholars have widely used CEMS data for estimating emissions from industrial sources. However, CEMS is generally deemed too costly for small and mid-scale companies. Consequently, these companies usually select alternative emissions-monitoring systems that align with their operational workflows.

Validation and analysis

This study’s dataset includes multiple years of data collected from stations located around the study area. In addition to the main criteria pollutants, meteorological data was also considered in this assessment. The stations were located in sensitive locations based on the selection guidelines developed by Ambu-Saidi, 2019. The coordinates of the stations are presented in Table 1.

Date sets were extracted from each station through the central data center. Data extraction is hourly for 3 years, 2019 to 2021. The total number of data samples in the study from each station is approximately 26,280 lines for all parameters. That’s about 105,120 lines of data. The data were then filtered and checked. The key to high confidence monitoring results is data analysis. A strict data validation procedure is mandatory prior to analysis. This means verifying the accuracy of high-resolution data so that specialists can use validated data for both short and long term. The validation protocol for the measured data and the pollutant trends is presented in this section. The following steps are outlined in this protocol, and this protocol may be changed as more data and time become available. The raw data is subjected to a multi-step process of validation starting from the raw data, and involves several validation stages:

Short-term means

Download short term data from data center (15-min intervals).

Do a post-process of the data with external media/software.

Do not modify or destroy data in the data center or database.

Change measurements

To convert pollutant readings from parts per million (ppm) to micrograms per cubic meter (µg/m 3).

Data validation

Use the validation protocol on short-term data.

Mean short term data.

Determine whether the data sets are reportable and whether the values are reasonable and understandable.

Keep valid data and delete invalid data from the analysis sheet but save it in the raw data server.

Exclude any calibration readings

Do not round or substitute “zero” for negative values in the analysis.

Accept rolling-average data as valid if at least 65% of the data for the 24-h and 8-h averages are valid. Do not accept any averages of less than 8 h in length.

Ignore data affected by microenvironmental peaks from known activities, such as open fires or vehicle movement nearby.

For high stations other than all high stations, high stations indicate a sandstorm, highlight hourly PM 10 averages above 600 μg/m 3 or below 20 μg/m 3 for further review by specialists.

Highlight hourly PM 2.5 averages greater than 100 μg/m 3 or less than 10 μg/m 3 for specialists to review, except when all stations are high indicating a sandstorm, or during and after port rainfalls.

Be aware of the ozone analyzer data at times of high humidity particularly during summer 16:00–18:00.

Please note that nitrogen dioxide and sulfur dioxide are also sensitive to high humidity.

Compute long-term averages

Obtain long-term averages from validated short-term data.

Derive 1-h, 3-h, 8-h or 24-h averages using the rolling average formula. However, 1-h averages were used in this evaluation.

Calculate fixed equations to find daily, monthly and yearly averages.

The monitoring data sets were validated prior to the analysis, but some uncertainty may be associated with the measured concentrations. This uncertainty may be because of analyzer calibration, frequency of maintenance, high humidity conditions, local activities in the vicinity of the monitoring stations and representativeness of the station locations. The coastal and industrial nature of the study area may also influence the monitoring results, as some stations may be affected by the nearby roads, port activities or local dust movement. The monitoring data were therefore considered adequate for model comparison, but the interpretation of the results should consider these possible sources of uncertainty.

In this study, the selected pollutants were categorized based on their source characteristics, atmospheric behavior, and relevance to air quality management in Sohar Port and Freezone. PM 10 was considered as the main particulate pollutant because it is affected by both industrial activities and regional dust conditions. NO 2 and SO 2 were considered as combustion-related and industrial gaseous pollutants, mainly linked to fuel combustion, stack emissions, and industrial processes. CO was considered as a combustion tracer because it is relatively stable in the atmosphere and can be used to support the interpretation of combustion-related emissions. This grouping was applied to support the interpretation of the monitoring and modeling results and to link each pollutant group with its likely source and environmental behavior.

For the modeled data, The sources input was as the following for each pollutants (coordinate, stack height, flow velocity, base elevation, diameter, exit temperature, and pollutant emission rates) a total of 87 fixed sources were used in the model divided into three main groups as per the clusters in the Sohar Port and Freezone (petrochemical, metals and utilities)—all of those consideration as input for the AERMOD. The AERMOD modeling system is a comprehensive modeling tool that includes meteorological data (surface, upper air, precipitation, and wind rose) processors, geophysical data processors, a diagnostic meteorological model, a dispersion model, and post-processing modules. The modeling system is a multilayer, multi-species, non-steady-state model for plumes from sources to the air, and for settling velocity at each concentration, based on the meteorological data at the same time. It is capable of modeling instant or continuous releases over distances starting from tens of meters to many kilometers. The AERMOD modeling system consists of three main components: AERMOD, ISCST3, and ISC-PRIME. Meteorological data is a key input required for running the modeling system. The AERMET data, which is hourly from 2019 to 2021, is a diagnostic model that requires gridded meteorological data as input. The data were was purchased from the AERMOD developer based on the location of Sohar Port and Freezone Area (Figure 1). The data were then processed in AERMET using MM5, as per the user-provided location coordinates (460121mE, 2,708,265mN). After finalizing the met data, the AERMOD study area was defined as the following (451,000mE, 2695000m N), then extended 20km to the east and north, and completed with a square of 20km × 20km that covers the Port and the Freezone. In addition to the affected area due to the industrial activities, as shown in Figure 2, with a grid spacing of 500 m. The terrain elevation and land use data for the modeling domain was obtained from AERMOD as elevated/flat.

The modeling was used to model PM 10, SO 2, NOx, and CO pollutants. Some calculations have been made for the standing vessels in the berths, as included in the modeling per the berth timing; however, those numbers are very limited compared to the industrial stacks. The modeling was done for each day to obtain a 24-h average of the output. Then the data were plotted for the monitored date, which was validated and collected using the stations. Emission rates for industrial point sources were derived from CEMS datasets when available; otherwise, USEPA AP-42 emission factors were applied based on stack type, fuel type, and available process information. However, background concentrations, fugitive emissions, road traffic emissions, and other low-level diffuse sources were not fully included in the base modeling scenario because reliable and continuous data were not available for the full study period. This limitation is important for PM 10 and NO 2, because PM 10 can be strongly affected by regional dust, resuspended dust, and port handling activities, while NO 2 can be influenced by road traffic and low-level combustion sources. Therefore, the modeled concentrations should be interpreted mainly as the contribution from selected industrial point sources and not as the total ambient concentration in the area. Future work should include background monitoring, traffic emissions, fugitive dust inventories, and sensitivity analysis to improve the reliability of source attribution. In future assessment, AERMOD setup should be done by adding line and area sources to consider other emissions like traffic, fugitive dust, and hydrocarbon emissions, as it is important for initiating reduction and mitigation efforts, starts with understanding the impact of each source.

The diurnal variations of pollutants are affected by atmospheric stability, temperature and solar radiation. During the day the surface is heated by the sun which increases turbulence in the atmosphere. This is combined with convective mixing which increases the dispersion of atmospheric pollutants. The increased mixing reduces pollutant concentrations near emission sites by distributing pollutants over a much larger volume of air (Jacob and Winner, 2009). For example, as shown in Figure 4 pollutants such as nitrogen oxides (NOx) and ozone (O 3) are more prone to spread better during daytime, as solar radiation also promotes photochemical reactions and secondary pollutant formation (Seinfeld and Pandis, 2016). The surface of the earth cools at night resulting in the formation of a stable boundary layer. Turbulence is suppressed and mixing heights are reduced. A stable atmosphere such as this traps pollutants near the surface often resulting in high concentrations at the surface. Their release is due to traffic, industry and home heating. One of the major reasons for the phenomenon of contaminated air moving away from an outlet and into a stagnant body of air to be increased is low wind speeds at night (Stull, 1988). The reduction of breeze enhances this phenomenon. In urban areas, this pressure is compounded further as the pollutants that build up overnight reach their peak concentrations in the morning as airborne pollutants get trapped in stagnant air. Knowledge of these diurnal patterns is important to direct effective air quality management activities, particularly to restrict human exposure to pollutants at their maximum concentrations (Wallace and Corr, 2015).

Seasonal and diurnal variation

The mixing height is lower in winter than in other seasons in the early morning hours. This limits the dispersion of air and reduces the extent of contamination. This is due to the winter season having a lower average mixing height. This process occurs also during summer evenings, where the role of humidity is relevant to reducing the mixing height and consequently the concentration of pollutants is higher. Conversely, the average mixing height often attains its maximum during the summer months. This causes more mixing over a larger area of the troposphere and therefore lower concentrations of pollutants. Meteorological stability, in particular low boundary-layer height, favors the accumulation of pollutants near the surface, while deeper mixing in warmer periods dilutes the pollutants and lowers their concentrations (Salvador et al., 2020).

During the rainy season, substantial precipitation helps decrease air pollution via wet deposition. This is often the case with rainfall, which reduces the concentration of most pollutants, particularly the amount of dust in the air. Alternatively, the increased intensity of sunlight reaching the mixing layer due to the decreased scattering of light by dust particles may result in increased production of ozone from the interaction of sunlight with nitrogen dioxide. Continued rainfall contributes to reducing the quantity of airborne particles and other pollutants generated and dispersed by a variety of sources, ultimately improving air quality during these episodes. According to the research done by Abed Al Ahad et al., 2020, lower uncertainty is observed during winter months and summer evenings as compared to the rest of the year. The weekly concentration patterns for CO, NO 2, SO 2, and PM 10 are presented in Figure 5–8, respectively for study area, and that clearly showing the results of the effects of seasonal variation.

Air quality modeling (AERMOD) results

AERMOD modeling was performed for all selected parameters, and the table below summarizes all runs for the four locations at each station, which were selected as sensitive receptors (Table 2).

The following Table 3 summarizes the major pollutants that were assessed in this study, along with its concentration ranges in study area against the local standards, it also highlights the regulatory standards for ambient air quality at some continues in the GCC.

In Table, we see that the Omani ambient air quality standards under MD 41/2017 are largely similar to the standards in other GCC countries. PM 10 and CO standards of Oman are in line with the UAE, Kuwait, and Qatar, but there are differences in the limits set by Saudi Arabia as Saudi stander is the most relaxed one in the area.

Gulf limits on gaseous pollutants such as NO 2 and SO 2 are known in various countries and over various average periods. Each country is responsible for its own regulation based on its environment, industrial conditions and policy. The Omani standard has a major disadvantage: there are no annual average thresholds for the major air contaminants. Annual limits are important for long-term monitoring, for pollution levels to be set and for authorities to know long-term trends and whether mitigation works.

Thus, although MD 41/2017 establishes a strong regulatory framework for ambient air quality in Oman, next revisions should include annual average limitations. This would be more conducive for long-term air quality management and lead to a more definitive way to assess chronic exposure and future pollution control measures. The comparisons between the modeled and monitored pollutant concentrations are presented in Figures 9–12.

The mean concentration results show the current conditions of the air quality at the study area, and as expected that PM 10 is high at 119.9 ± 91.8 μg/m 3 as it represents the effects of natural dust, industrial activities and port operations and current data shows that there are elevated levels of dust above the limits sets in the Omani Air quality standards. However, for the gaseous pollutants NO 2, SO 2 and CO the means were much lower at 21.3 ± 11.2 μg/m 3, 31.1 ± 12.61 μg/m 3 and 310.5 ± 239.22 mg/m 3, respectively. This indicates that on average the pollutants were within the Omani short-term regulation limits, but the high standard deviation, especially for PM 10 and SO 2, indicates that the pollution was not short-term and that the Omani standard should be updated to add the annual limits.

As previously reported, the impact of PM 10 on residential areas may be very harmful, especially when background values are not included in the present results. Ambient dust concentrations in urban Sohar over the past years have been capture in the above table with mean is naturally high at 119.9 ± 91.8 μg/m 3. The modelling results indicate that there is a worst-case scenario (41.4 μg/m 3 of PM 10 in the air over 24 h). Overall, this figure is around the Omani standard of 150 μg/m 3 over 24 h. The NO 2 modeling results show that the industrial NO 2 concentration had minimal impact on the residential area. NOx was reportedly completely converted from sources to NO 2 in the ambient air. The analysis did not consider sunlight, which causes NO 2 to be transformed into NO and the production of ozone and NO 2. Our assumption is violated, and we get a result that does not comply with the criteria and is 15% below the permissible limit (250 μg/m 3) for a 1-h average, measured amount of MGL 212.6 μg/m 3. A 24-h mean suggests an acceptable window of operation, with an average ground-level concentration (MGL) of 76.4 μg/m 3, <130 μg/m 3. Based on historical data from local inquiries, the mean values for measurements not at the port were 20 μg/m 3 (24-h rolling average). The model does not consider traffic sources, which are also likely to be major contributors to NO 2 emissions and measurements, because emissions occur at lower elevations and thus the station could be affected by road emissions. Carbon monoxide is chemically stable and oxidizes only for its changes, which makes it useful as a tracer of combustion. The remaining CO concentration in the air is adequate for future analysis. The total content in the total concentration of CO from the model results is significantly lower than the limit, with only 10% of the limit at maximum ground-level concentrations of 3,182 μg/m 3 and between 1.9% and 4.1% of the receptor concentration at the station sites. SO 2 is an environmentally harmful pollutant emitted by gases from sulfur-burning fuels. In the majority of cases, SO 2 stays down in the lower atmospheric layers (Omidvarborna et al., 2018), and relatively high levels are limited mainly to the port area, and very little of the emissions are spreading to outside places. This indicates a low impact of emissions on station sites, with the highest ground concentrations beyond the port area being slightly over 34.7 μg/m 3 for 1h and 5.5 μg/m 3 for 24 h. Fugitive emissions were removed from this simulation owing to data loss and uncertainty. Most of the SO 2 emissions arose from coke consumption in metallurgical and crude refining processes. The power plant in the utilities cluster operated only on natural gas, with diesel as a redundant fuel source. The significant diesel emissions are from car traffic in the area, with more than 15,000 vehicles entering the port and the free zone every day, though this factor is not captured in the model.

Model validation results

Data from the monitoring stations were used to evaluate the modeling performance and to test the AERMOD prediction against reference station data. We used the data from 2019 to 2021 and ran the model to evaluate the results. As mentioned above, the modeling results were without background concentrations; therefore, the validation results interpretations considers to be for selected points sources not total concentrations for CO, PM, SO 2, and NO 2. Direct standards: statistical measure methods used to validate the models by running comparisons of the monitoring data. Fractional bias (FB), index of agreement (IOA), normalized mean square error (NMSE), geometric mean bias (MG), and geometric mean variance (VG) are some statistical equations used to validate data and used in this study, as shown in Equations 1–5.

A graph of the modeling results and the monitoring numbers used in the graphs, for more straightforward presentation of the data, evaluates the matching percentage using standard statistical analysis. The available monitoring data from the stations we have dates set. In the model validation, we used a sensitive receptor at the same location for the monitoring stations in AERMOD. The modeling data were validated against a real monitoring station installed at the site in the receptor areas. The average of the monitoring data was used to evaluate and match the modeling results. The monitoring data available during the study. The calculated statistical performance indices used to evaluate agreement between the modeled and monitored concentrations are summarized in Table 4. Therefore, an additional run of the model was conducted for the selected period. The results were analyzed using the monitored data.

The difference between the modeled and monitored concentrations ranged from 16.13% to 35.07%. The trend of monitoring and modeling data has some relative shapes. The city is very active in industrial, transportation, and residential areas. Therefore, this variation between the data and building and locals’ lives may result in differences in the predicted results. There is a possibility of an error in data recording at the monitoring station due to lack of maintenance, data drifting, or local atmospheric conditions at the station. The station’s location might be a limiting factor in the model’s estimation and observed results. The PM 10 model performance is generally strong, with low bias (FB) of 0.13, minimal logarithmic error of 1.14, and excellent variability representation (1.02). While NMSE is moderate, it remains within an acceptable range and the IOA value (0.53) that resulted in good to very good agreement. If the additional value (0.88) refers to correlation, it further confirms good agreement.

In the air, the NOx group is very active in reacting with other parameters, such as PM, water, and Oxygen. Sunlight availability plays a significant role in activities. Those reactions result in the formation of secondary pollutants such as HNO 3 and Ozone (Al-Salem and Khan, 2010). From the above chart and table, there was a 40.62% - 60.46% variation between the modeled and monitored concentrations, with a similar trend. This high variation is expected, as the model setup was for full conversion of NOx to NO 2, which is not the case in real life, as NO 2 reacts rapidly with O 2 in the presence of sunlight via the following reaction.

On the other hand, the monitoring station was located close to a main highway, which meant some NOx emissions from vehicles were captured by the station. However, traffic emission was not considered as line-sources in this assessment, if was, it might have a better alignment between the monitoring and modeling data. Also the Port is located in a coastal area where humidity is high and this can also cause some variation in the data. The NO 2 limit is generally lower than the limits for the area and no exceedances were reported in the past. Most of the monitored data in the area was fluctuating around 40 μg/m 3. The model shows high overprediction of NO 2 concentrations with a high Fractional Bias (0.75) and Geometric Mean Bias (2.20). For the NMSE and VG values this indicates a moderate performance in terms of error and variability. However, the IOA value seems to be very inconsistent with its expected range and has to be re-checked. As a general rule the model needs to be recalibrated or refined in input data or assumptions to better match actual monitoring results.

The result was a clear matching trend from a monitoring and modelling based SO 2 -concentration data point of view. Both datasets are in the same shape with some discrepancies. Drifting is observed only at higher concentrations, which is always observed in the modeling case. The very good predictive relationship between monitored and modelled SO 2 concentrations in this survey shows that the air quality model has a very good predictive power for the future behaviour of sulfur dioxide in the atmosphere. These two data sets show very similar temporal patterns as the shape of the observed data is very similar to the modeled data along the observed time course, indicating that the model is able to reproduce the main emission sources, spread patterns and meteorological effects that influence SO 2 concentrations. The model does fairly well on this run. The fractional bias and MG suggest a bias to underestimate SO 2 concentrations, whereas NMSE is slightly deviated from the observed values. Given that the reported value of IOA was above the expected range, this should be reviewed.

The model is fair in general; but its accuracy is not good, and it needs calibration to optimize the predictions and reduce bias. To solve this problem all the statistical indicators were recalculated by means of Python (SciPy, NumPy). IOA values were recalculated using the standards to the accepted theoretical range (0–1) to conform to current best-practice validation standards. This revision confirms that PM 10 and SO 2 were more consistent with the available data than NO 2, which exhibited an overprediction bias (probably due to unaccounted road traffic emissions).

Numerical modeling validation

Fractional bias (FB), index of agreement (IOA), normalized mean square error (NMSE), geometric mean bias (MG), and geometric mean-variance (VG) are statistical metrics used to validate data and employed in this study. A graph of the modeling results and the monitoring numbers is used to present the data more easily and to evaluate the matching percentage using standard statistical analysis. The available monitoring data from the station are based on monitored and modeled data.

The validation results indicate that the model performance varied between pollutants. PM 10 showed the closest agreement with the monitoring data compared with the other pollutants, mainly because the fractional bias was closer to zero and the variability indicators were more stable. SO 2 showed moderate agreement, which may be related to the fact that most SO 2 emissions are linked to identifiable industrial sources. In contrast, NO 2 showed higher bias, which may be due to the assumption of full NOx conversion to NO 2, in addition to the exclusion of traffic emissions and other low-level combustion sources. CO also showed uncertainty in some statistical indicators, which suggests that further refinement of the emission inventory may be required. Therefore, the model can be used as a useful screening tool for industrial source contribution, but it should not be interpreted as a complete representation of total ambient air quality without including background and fugitive sources.

The best model usually produces FB around zero. The table above shows FB values from 0.13 to 0.75 where PM 10 is closest to the monitored values. The IOA ranges from 0.01 to 0.69, with 1 being a perfect model. MG and VG are normalizing FB and P/O and best matching in PM 10. NMSE across all runs was not well predicted for all pollutants. AERMOD has lower NMSE predictions and assessments. To test the robustness of the model, a sensitivity analysis was performed by varying NO 2 and SO 2 emission rates ±30% and changing wind speed inputs ±1 m/s. Results showed that PM 10 predictions were highly sensitive to wind speed and mixing height assumptions, whereas NO 2 outputs varied significantly due to emission rate uncertainty and the lack of NO–NO 2 –O 3 chemical transformation modeling.

In the air, the NOx group is very active in reacting with other pollutants, such as PM, water, and Oxygen. Sunlight availability plays a significant role in activities. Those reactions resulted in the formation of secondary pollutants such as HNO 3 and Ozone. Three modeling scenarios were employed: NOx to NO 2, Ambient Ratio method, and Ozone Limiting Method. The three modeling methods did not differ significantly, and in all cases the models underestimated the concentrations of NO 2 in the air. The high NO 2 reading is to be expected as the vehicle’s emissions from the main highway near the station were not accounted for in the model set-up.

In addition, the Sohar Port is located in coastal areas with high humidity, which can also cause some variation in the data. Overall, the NO 2 limit is usually below the area’s limits, and no exceedances have been reported in the past, and most of the monitored data in the area fluctuated around 40 μg/m 3. All concentration charts clearly show that all observations match each other. It is also reflected in the index’s SO 2 performance across all indices, which is close to the perfect model. A sensitivity analysis was carried out to determine the impact of varying emission rates (±30%) and meteorological inputs on modeled concentrations. It became apparent that PM 10 estimates were most sensitive to simple wind assumptions and their speed/mixing height, and that NO 2 values are mostly affected by traffic emissions, mainly due to the highly variable nature of chemical transformation assumptions.

Limitations and uncertainty

This study presents an integrated air quality assessment of Sohar Port and Freezone using monitoring data and AERMOD modeling. However, the results should be interpreted within the limitations of the available data and model assumptions. Background concentrations were not included in the base AERMOD results because representative background data were not available for all pollutants and all periods. This is especially important for PM 10, because the study area is located in an arid coastal region where natural dust, resuspension, and regional transport can contribute significantly to ambient particulate levels.

Fugitive emissions, road traffic, material handling, and other diffuse sources were also not fully included because of limited activity data and uncertainty in emission factors. These sources can influence measured concentrations, especially near roads, port operation areas and industrial handling areas. As a result, the model outputs should be considered as an estimate of selected industrial point-source contributions rather than a full source-apportionment assessment.

Monitoring uncertainty may also affect the comparison between observed and modeled concentrations. This uncertainty can be related to calibration, maintenance, analyzer drift, high humidity, local activities, and station siting. Therefore, the validation results are useful for understanding the general performance of the model, but they should not be used as the only basis for final source attribution. Future studies should include additional monitoring stations, background concentration analysis, fugitive emission estimation, traffic emission inventories, and sensitivity analysis to improve the robustness of the model.

AQI modeling and health implications

At 24-hour-averaged modeled concentrations, AQI values were calculated using USEPA breakpoints for PM 10, SO 2, CO, and NO 2. AQI was mainly PM 10 -dominated across all receptor sites, with ∼60% of the days modeled falling under “Moderate” and ∼25% under “Unhealthy for Sensitive Groups.” These results confirm the observed AQI reports from the Sohar area. Increased PM background content, without industrial input, also further supports the need for natural dust and traffic control strategies in addition to industrial emission regulation.

This study uses both high-resolution monitoring and dispersion modeling to evaluate the industrial air quality impacts involved in the Sohar Port and Freezone Area. Even PM 10, well below regulatory thresholds, is strongly controlled by its natural and industrial sources. NO 2 levels were close to hourly limits in worst-case model designs, but actual values were rarely above these limits. The validation results show an acceptable performance of the model, especially for PM 10 and SO 2, although traffic and fugitive emissions are major unmodelled sources.

For improved future evaluations, emission inventories should include current-time traffic and fugitive emissions, and the model calibration must account for photochemical reactions. Infrastructures for monitoring information would need to be expanded to enable more accurate measurement of near-source changes and capture near-source variation. This work provides basis for sustainable industrial planning and stresses the importance of data-informed decision-making in plications for an application for a port or city experiencing.