Assessment of Air Quality and Health Impact in Hanoi (Vietnam) Due to Traffic Emission—Seasonal Analysis and Traffic Emission Reduction Scenarios

Abstract

This study assesses air quality and health impact in Hanoi, Vietnam, using the Community Multiscale Air Quality (CMAQ) model and health impact assessment to evaluate the effectiveness of traffic emission reduction strategies under two scenarios. An updated emission inventory was used as the input data for the CMAQ model. The Weather Research and Forecasting (WRF-CMAQ) model (version 5.4), incorporating the CB6 chemical mechanism, was applied alongside a calibrated meteorological model to simulate pollutant dispersion. The model achieved strong performance in PM_2.5 simulation, with a correlation coefficient (R) of 0.78, an index of agreement (IOA) of −0.5, a Normalized Mean Bias (NMB) of 7.11%, and a normalized mean error (NME) of 28.51%. Seasonal analysis revealed higher concentrations of CO, NO₂, O₃, and SO₂ in January compared to July, driven by traffic and industrial emissions. Improved air quality in July was attributed to favorable meteorological conditions, such as increased rainfall and clean airflows from the sea. Spatial distribution highlighted elevated pollutant levels in urban areas, while PM_2.5 was significantly influenced by long-range transport and atmospheric processes. However, fine dust concentrations remained high in suburban areas, driven by secondary emissions and nearby industrial zones. An emission reduction scenario based on the Hanoi city policy decree focusing on traffic sources demonstrated its potential to reduce NO₂, SO₂, and PM_2.5 concentrations, though the impacts varied across time and space. Health impact due to population exposure to PM_2.5 shows that the densely populated suburbs surrounding the urban core have the largest impact in terms of mortality and cardiovascular diseases hospitalization. As PM_2.5 has the largest impact on these two health endpoints, only PM_2.5 impact assessment is performed. Health impact due to air pollution is higher in January (dry season) with estimated 625 deaths and 124 cardiovascular diseases (cvd) hospitalization as compared with estimated 94 deaths and 18 cvd hospitalization in July (wet season). One of the research questions posed by the city authority is whether converting diesel buses to electric buses can yield environmental and health benefits. Our work shows that the scenario based on Hanoi city decree of replacing 50% of fossil fuel combustion buses with electric buses by 2035 does not yield perceptible change in mortality health effect. This is due to emission from buses being small as compared to those from the whole transport sector and other sectors. This study emphasizes the need for integrated, targeted emission control strategies to address spatial and temporal variability in pollution. The findings offer valuable insights for policymakers to develop effective measures in urban planning for improving air quality and protecting the health of people in Hanoi.

1. Introduction

Air pollution is one of the most serious environmental problems in urban areas. The World Health Organization (WHO) has estimated that air pollution causes the death of more than 8 million people per year in the world, and among that, more than 70% the deaths per year occur in developing countries. Millions of people are found to be suffering from various respiratory illnesses related to air pollution in large cities. Therefore, urban air quality management should be urgently considered in order to protect human health. Air quality in developing countries has deteriorated considerably, exposing millions of people to harmful concentrations of pollutants because urban air quality management has not been adopted because of a variety of difficulties.

The sources of air pollution can be categorized by their origins, primary or secondary pollutants, stationary or mobile sources, and point sources or area sources. The United States Environmental Protection Agency (US EPA) commonly classifies sources of air pollution by splitting them into two main categories: anthropogenic and natural sources.

Anthropogenic air pollution originates from large stationary sources (e.g., industries, power plants, and municipal incinerators), small stationary sources (e.g., households and small commercial boilers), and mobile sources (e.g., traffic). In addition, anthropogenic sources can be classified into two main source groups: stationary sources, which are point sources and non-point (Area) sources, and mobile sources, which are on-road and non-road (off-road). Stationary sources include smokestacks of power plants, manufacturing facilities (factories), waste incinerators, furnaces, and other fuel-burning heating devices. Traditional biomass burning is the primary source of air pollutants in developing and poor countries. Traditional biomass includes wood and crop waste, generating NO_x, SO₂, PM, and CO₂. Mobile sources include motor vehicles, marine vessels, and aircraft. In cities, transportation is known as the primary source of air pollution. Along with congestion and large, poorly maintained vehicle fleets pollute air quality in cities.

Air pollution is increasingly problematic in Vietnam due to rising emissions from industrial growth. The country’s reliance on fossil fuels has intensified pollution in recent years, particularly in major cities like Hanoi and Ho Chi Minh City [1,2]. Reports indicate that Hanoi’s annual PM₁₀ and PM_2.5 levels often exceed local air quality standards [3]. Key contributors to this pollution include the combustion of coal, oil, and natural gas [4]. Despite Vietnam’s commitment to reducing greenhouse gas emissions through renewable energy by 2030 as in nationally determined contributions (NDC, 2015), ongoing reliance on fossil fuels is expected to persist, potentially increasing emissions over the next decade [5].

Hanoi, the capital city of Vietnam, has undergone significant urban growth, leading to the conversion of agricultural land into urban areas and an associated rise in pollution from traffic, industry, and residential activities [6]. This rapid development has resulted in increased levels of air pollutants, such as nitrogen dioxide (NO₂) and sulfur dioxide (SO₂), particularly in newly urbanized zones [7]. Pollutant concentration from roadside emission from traffic in Hanoi streets had been modeled using the Operational Street Pollution Model (OSPM) [7]. The predictions from five streets were evaluated against measurements of NO_x, SO₂, CO, and benzene (BNZ). Calibration of the model was used to calculate the average emission factors of the vehicle fleet for various pollutants. Traffic emissions are a significant source of air pollution, with studies showing that particulate matter (PM₁₀) from transportation contributed to thousands of premature deaths in previous years [1]. Additionally, roadside air pollution presents serious health risks for the city’s population [8]. To address these pressing issues, in 2024, the Hanoi city authority promulgated a decree to convert the fleet of buses from combustion engines to 50% electric buses by 2035. Evaluating air quality in Hanoi, focusing on seasonal changes and the potential impact of reducing traffic emissions, is crucial for implementing effective pollution control measures. Air quality dispersion models on the regional scale such as the WRF-Chem (Weather Research Forecast—Chemistry) or WRF-CMAQ (Weather Forecast—Community Multiscale Air Quality) model is used by many authors in various countries to assess the impact of emission from various sources on air quality in the domain under consideration [9,10,11,12].

This study aims to assess air quality in Hanoi (Vietnam) using the Weather Research Forecast-Community Multiscale Air Quality (WRF-CMAQ) model and evaluate air pollution reduction for two traffic emission scenarios. Health impact due to population exposure of PM_2.5 air pollutants as simulated from the model will be assessed for the emission scenarios. This study uses the updated emission inventory for thermal power plants, analyzing a variety of pollutants and changes over recent years in Hanoi. This study sets up a meteorological model and simulates air pollution dispersion for Hanoi, calibrating and validating the model. Then, we simulate the dispersion of air pollutants using the photochemical model based on the current and reduced emission scenarios, thereby assessing the impact of road traffic sources on air pollution in Hanoi.

2. Data and Methods

2.1. Emission Inventory in Hanoi

In this study, air emission inventory results were the input of the CMAQ model. We calculated air emission sources based on three main categories: line source, area source, and point source.

2.1.1. Line Source

The EMISENS model is used to calculate emissions from road transportation activities. This study classified streets into five categories: highways, rural roads, urban streets, suburban streets, and industrial streets (located in the industrial zone) [13]. Meanwhile, vehicles were divided into five types: buses/coaches, heavy-duty vehicles (HDVs) with a total gross weight of over 3.5 tonnes, light-duty vehicles (LDVs) which weigh less than 3.5 tonnes, cars (<15 seats), and motorcycles.

This study referenced the number of vehicles in 2022 and surveyed 200 extra questionnaires on 15 streets. We directly interviewed and collected information directly from vehicle owners for each type of vehicle on each specific road.

The sources of road traffic emission factors (Efs) were considered in this study, including the local studies [14,15], the EFs of China [16], and the EFs of other countries in CORINAIR for CO, NO_x, SO₂, NMVOC (non-methane volatile organic compounds), CH₄, and TSP (Total Suspended Particles) [17].

2.1.2. Area and Point Sources

We used emission factors and activity data as the equation below

E = EFs × A × (1 − ER/100)

where

E is emissions (typically in tons/year),

A is activities rate (amount of fuel use, capacity, or number of products)

EF is emission factors (related to A),

ER is emission reduction efficiency (only if abatement devices are used).

For point sources, the key industrial products in Hanoi in 2022 include six products from the mechanical and manufacturing industry (accounting for 18.2%), four products from the electrical and electronic industry (accounting for 12.1%), five products from the information technology industry (accounting for 15.2%); eight products of the textile, garment, and footwear industry (accounting for 24.2%); four products of the agricultural and food processing industry (accounting for 12.1%); three products of the chemical, rubber, plastic and pharmaceutical industries (accounting for 9.1%); two products of the construction materials industry (accounting for 6.1%) and one product of the handicraft industry (accounting for 3%) [18]. CNG was the most significant fuel used in industry, especially in chemical chemistry. The other fuels including oil types (DO, FO, and cashew oil), gasoline, wood and wood products, and coal are used for industry.

Area sources, such as households, restaurants, construction material stores, pagodas, and biogenic sources (straw burning), are calculated.

The sources of EFs considered were referenced from the emission inventory guidebook of the European Environment Agency (EEA) [19,20,21].

2.2. WRF-CMAQ Model

The WRF (Weather Research and Forecasting)-CMAQ model (version 5.3.1, US EPA, 2019) was employed to simulate atmospheric processes using the CB6 (Carbon Bond 6) chemical mechanism. The CB6 mechanism provides a detailed representation of atmospheric chemical reactions, particularly those involving secondary organic aerosol formation [22,23]. The WRF model was first executed to generate meteorological input data for CMAQ using the Meteorology—Chemistry Interface Processor (MCIP).

Domain configuration and physics/chemistry options for WRF are detailed in Figure 1 and

Table 1. Model setup and physical parameterization used for WRF simulation.

WRF Model Setup
WRF (ARW): Version 4.2.1
Mercator_34S_150E
Surface Classification: MODIS data.
Model Configuration
Initial and Boundary Conditions: The initial and boundary conditions for grid nesting are derived from the NCEP Global Forecast System (GFS), with a horizontal resolution of 0.5° × 0.5° from 2007 to the present. Simulation Setup: The simulations are initiated at 00:00 UTC daily. Model output is generated hourly. The time step for most simulations is approximately 60 s.
Grid Nesting: One-dimensional grid nesting with a grid adjustment zone of 5 grid points.
Physical Parametrization
mp_physics	3
ra_lw_physics	1
ra_sw_physics	1
Radt	30
sf_sfclay_physics	1
sf_surface_physics	2
bl_pbl_physics	1
Bldt	0
cu_physics	1
Cudt	5
Isfflx	1
Ifsnow	1
Icloud	1
surface_input_source	1
num_soil_layers	4
num_land_cat	24
sf_urban_physics	0
Maxiens	1
Maxens	3
maxens2	3
maxens3	16
Ensdim	144

. Meteorological data were obtained from the Global Forecast System (GFS) by the National Centers for Environmental Prediction (NCEP). GFS integrates four sub-models (atmosphere, ocean, land/grid, and sea ice) to provide high-resolution forecasts of variables such as temperature, wind, precipitation, soil moisture, and atmospheric ozone.

The WRF-CMAQ simulation employed three nested domains to capture spatial variability across scales. The outermost domain (D1) consisted of a 42 × 42 grid with a horizontal resolution of 27 × 27 km, covering northern and north-central Vietnam, the East Sea, and parts of Laos, Thailand, and China. The second domain (D2) featured a 40 × 43 grid with a resolution of 9 × 9 km, focusing on northern and north-central Vietnam and adjacent regions in Laos and China. The innermost domain (D3) included a 43 × 43 grid with a 3 × 3 km resolution, encompassing Hanoi city and nearby provinces such as Thai Nguyen. This setup ensured high spatial resolution for the areas of interest while accounting for regional-scale meteorological dynamics.

For domain 1 and domain 2, outside the Hanoi emission inventory area, we use the Emissions Database for Global Atmospheric Research (EDGAR) version 5 (v50_AP) and the biogenic emission MEGAN v2.10, while for domain 3 (Hanoi domain), we use the Hanoi emission inventory

Initial and boundary conditions for chemical species were sourced from the global CAM_Chem model (Community Atmosphere Model—Chemistry), available at NCAR (https://www.acom.ucar.edu/cam-chem/cam-chem.shtml, accessed on 14 October 2025). Anthropogenic emission data were provided by the research team’s 2022 emission inventory for Hanoi. The simulation period covered January and July 2022, representing the winter and summer seasons. Emission data from various sources and formats were processed into CMAQ-compatible inputs using a set of custom-developed Python 3.1 scripts.

Post-processing tools, including Combine 1.1, Sitecmp 1.4, Writesite, Panoply 5.7.1, and QGIS 3.40, were utilized to analyze and validate the WRF-CMAQ simulation outputs. These tools enabled efficient extraction, organization, and comparison of modeled data with observational datasets, ensuring robust evaluation of the model’s performance in predicting air quality and atmospheric chemical composition.

2.3. Calibration and Validation for the WRF-CMAQ Model

The model performance was evaluated using statistical indices such as the Mean Bias (MB), Root Mean Square Error (RMSE), Mean Absolute Gross Error (MAGE), Index of Agreement (IOA), and Pearson Correlation Coefficient (R). In addition, the Factor 2 coefficient and Standard Deviation (SD) were also used [24].

The Mean Bias (MB) is calculated as follows:

$$\mathrm{MB} = \overline{\mathrm{P}} - \overline{\mathrm{O}}$$

(1)

where $\overline{\mathrm{P}}$ is the average value of the model results and $\overline{\mathrm{O}}$ is the average value of the observed data. An MB value greater than 0 indicates an overestimation of the simulated data, while an MB value less than 0 indicates an underestimation.

The Normalized Mean Bias (NMB) is used as a normalization method to facilitate the analysis across a range of concentration levels. This statistic calculates the average difference between the model and observations relative to the total observed values. NMB is a useful model performance index as it prevents excessive inflation of the observed value range, especially at lower concentration levels. The Normalized Mean Bias is defined as follows:

$$\mathrm{NMB} = {\displaystyle \frac{{\sum }_1^n(P_i-O_i)}{{\sum }_1^n\left(O_i\right)}}\times 100$$

(2)

where P_i represents the model values, O_i represents the observed values, and N is the total number of samples.

The Normalized Mean Error (NME) is similar to the NMB, in which the performance statistic is used to normalize the mean error. NME calculates the absolute value of the difference between the model and the observations relative to the total observed values. The Normalized Mean Error is defined as follows:

$$\mathrm{NME} = {\displaystyle \frac{{\sum }_1^n|P_i-O_i|}{{\sum }_1^n\left(O_i\right)}}\times 100$$

(3)

The RMSE (Root Mean Square Error) represents the total error of the model and is equal to zero in ideal cases.

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}{\sum }_{i=1}^n{(P_i-O_i)}^2}$$

(4)

MAGE (Mean Absolute Gross Error) calculates the mean absolute error between simulated values and observed measurements, reflecting the average similarity of simulation errors. Similar to RMSE, lower MAGE values indicate better agreement between the observed data and the model results:

$$\mathrm{MAGE} = {\displaystyle \frac{1}{n}}{\sum }_{i=1}^n|P_i-O_i|$$

(5)

The Index of Agreement (IOA) provides a sophisticated measure of model performance. IOA is a dimensionless parameter with values ranging from −1 to 1, where 1 indicates an ideal model. A value of 0 signifies that the total error of the model is equal to the sum of the deviations of the observations from their mean value. A value of 0.5 indicates that the total error of the model accounts for half of the deviations between observed values and their mean. The constant variable c is related to the frequency of model output and is assigned a value of 2 The variable P_i represents the predicted value at time point (i), while O_i represents the observed value at the same time point (i), and $\overline{\mathrm{O}}$ represents the mean of the observed values.

$$\mathrm{IOA}=1-{\displaystyle \frac{{\sum }_{i=1}^n|P_i-O_i|}{c{\sum }_{i=1}^n|P_i-{\overline{o}}_i|}} \mathrm{when} {\sum }_{i=1}^n|P_i-O_i| \le c{\sum }_{i=1}^n|P_i-{\overline{o}}_i|$$

(6)

$$\mathrm{IOA}={\displaystyle \frac{{\sum }_{i=1}^n|P_i-O_i|}{c{\sum }_{i=1}^n|P_i-{\overline{o}}_i|}}-1 \mathrm{when} {\sum }_{i=1}^n|P_i-O_i| > c{\sum }_{i=1}^n|P_i-{\overline{o}}_i|$$

(7)

Pearson’s correlation coefficient is a measure of the linear relationship between simulated and observed data. Its value is zero when there is no correlation and increases as the coefficient approaches −1 or +1. Values close to +1 indicate a strong positive correlation between the two variables.

$$R={\displaystyle \frac{{\sum }_{i=1}^n\left[\left(M_i-\overline{M}\right)\left(O_i-\overline{O}\right)\right]}{\sqrt{{\sum }_{i=1}^n{\left(M_i-\overline{M}\right)}^2{\sum }_{i=1}^n{\left(O_i-\overline{O}\right)}^2}}}$$

(8)

The Factor 2 coefficient calculates the ratio between the simulated data (P_i) and the observed data (O_i), indicating the percentage of data that falls within the range of 0.5 ≤ P_i/O_i ≤ 2. Clearly, its optimal value is 1.

2.4. Population Exposure and Health Impact

The population exposure to air pollutants, especially PM_2.5, and its impact on population health, is an active area of epidemiological research in public health. To assess the risk, the relative risk (RR) factor is usually used to determine air pollution impact on health. The World Health Organization’s (WHO) Health Risks of Air Pollution in Europe project (HRAPIE) has recommended concentration—response functions for different health endpoints such as mortality (all cases) and cardiovascular (cvd) diseases for PM_2.5 [25] (WHO, 2012).

These RR of each of the health endpoints for a 10 μg/m³ increase of PM_2.5 on mortality (all cases) and cardiovascular diseases hospitalization are

RR (mortality) = 1.0123 (CI: 1.0045, 1.0201)

RR (cvd) = 1.0091 (CI: 1.0017, 1.0201)

where CI is the confidence interval at a 95% level.

The HRAPIE relative risk for mortality value above means that a 10 μg/m³ increase of PM_2.5 would increase the mortality of the exposed population by 1.23% (CI: 0.45%, 2.01%). Similarly, it would mean an increase of 0.9% (CI: 0.17%, 2.01%) for cvd hospitalization for each increase of 10 μg/m³ in PM_2.5 concentration.

As in the previous studies, such as [26], in this study, these recommended short-term concentration—response coefficients for mortality and cvd disease hospitalization are used to assess the population health impact due to PM_2.5 under different emission scenarios.

From the RR, the impact factor (IF) is then calculated to determine the impact associated with a specific change in air pollution concentration ΔX which is the PM_2.5 concentration increase due to an emission change, and IF is calculated as

IF = RR^ΔX/10

(9)

Equation (9) is a log-linear model which states that the logarithm of the impact factor (or relative risk function) is linearly changed with a change in ambient PM concentration.

Finally, the attributable number of the impact on health endpoint due to an increase in ΔX concentration in a population with a specific incidence rate for this endpoint is

AN = (IF−1) × Population × incidence rate

(10)

The above Equations (9) and (10) are used by the US EPA to estimate the health impact due to changes in air quality in the United States [27] and implemented in the BenMAP v1.5 software tool. They are also used by many authors to assess the health impact of air pollution in various countries [28,29,30].

The population data used in this study is the 2019 high-resolution population 1 km by 1 km dataset from the UN Humanitarian Data Exchange in TIFF format. The incidence rate is based on The Global Health Data Exchange (GHDx) from the Institute for Health Metrics and Evaluation (IHME) available from 1990 to 2019 (http://ghdx.healthdata.org, accessed 25 August 2025). According to GHDx data for 2019, the total number of deaths in Vietnam is 631,817 (CI: 538,099, 714,078). The corresponding population and mortality incidence rates Vietnam is derived as (96,372,928; 0.65%).

For the incidence rate of cvd hospitalization in Vietnam, there is currently no data publicly available that we can find. We use the figure similar to the figure of the heart failure (HF) hospitalization rate in Thailand of 168 per 100,000 in 2013 as provided by [31].

The population data at 1 km × 1 km resolution is converted to Netcdf file and then interpolated to the PM_2.5 modeling domain. Population exposure and attributable number of people affected are then calculated from Equations (2) and (3).

The relative risk as a percentage of increase in risk implies that a reference value of PM_2.5 is required. This reference value could be determined using the lowest value of PM_2.5 where there is no effect on population health. The Global Exposure Mortality Model (GEMM), developed by [32], defined the minimum observed PM_2.5 level in the cohort data as 2.4 µg/m³, below which there was no effect on mortality. This study is frequently used in global burden assessments. The other value that can be used is the so-called Theoretical Minimum-Risk Exposure Distribution (TMRED) that was proposed by [33] in the Global Burden of Disease GBD 2010. This value is defined based on the lowest and 5th percentile of PM_2.5 concentrations observed in major cohort studies from 5.8 to 8.8 µg/m³. In this study, we use the WHO value of 5 µg/m³ recommended as a safe level or a reference value below which there is no impact on population health. There are other studies which showed that there is no safe level of PM_2.5. However, a zero PM_2.5 level is not real and a practical approach is to use the WHO guideline value.

To calculate health impact due to different scenarios, such as the Hanoi city authority recent decree on vehicle emission fleet having a 50% percentage of electric vehicles (EV) by 2035 as compared to business as usual (BAU) scenario with no EVs, we can determine the relative impact between these scenarios without having to use the reference value such as WHO value. Simply use Equations (1) and (2) where ΔX is the change in PM_2.5 concentration between the base case emission scenario without EVs and the scenario with EVs in the fleet. In our study, the daily mean of PM_2.5 as predicted by WRF-CMAQ at each of the cells in the domain and the population within the same cell represents the exposure and hence the effect on health (mortality) will be calculated. We calculate the effect of PM_2.5 on mortality in Hanoi, modelling domains in January (representing dry season) and July (representing wet season) and the corresponding changes in mortality effects when compared between the baseline emission and the scenario when 50% of fossil-combustion buses are replaced with electric buses.

3. Results and Discussion

3.1. Development of Input Emission Datasets for the CMAQ Model

This study used emission inventory results as the input data for the CMAQ model. The summary of Hanoi’s emission sources in 2022 is presented in

Table 2. Emissions in Hanoi in 2022 (ton/year).

	NOx	CO	SOx	NMVOC	TSP	CH4	PM2.5	PM10
Line sources	69,907	2,033,569	4025	189,225	33,271	30,826	29,757	31,388
Area sources (including biomass burning)	1560	21,639	672	2266	14,581	2178	6100	8995
Point sources	3788	21,351	7038	13,195	3859	165	4044	5312
Total	75,255	2,076,559	11,735	204,686	51,711	33,169	39,901	45,695

and Figure 2. Overall, traffic line sources accounted for the most significant emissions of all pollutants. The line source accounted for 92.9%, 97.9%, 34.2%, 92.4%, and 92.9% of NO_x, CO, SO₂, NMVOC, and CH4. Industrial activities contributed 59.9%, 10.1%, and 11.6% of total SO₂, PM_2.5, and PM₁₀ emissions in Hanoi. The area sources accounted for 28.1% of TSP, 15.2% of PM_2.5, and 19.6% of PM₁₀ in Hanoi, whereas others were negligible sources.

3.2. Development of Emission Reduction Scenarios

In this study, alongside simulating air quality for two representative periods representing the rainy and dry seasons in Hanoi for the current state (2022), an emission reduction scenario was also developed to assess the potential for improving air quality by reducing emissions from one type of traffic source. Specifically, in the reduction scenario (Scenario 2), it is assumed that 50% of the fossil fuel-powered public buses operating in the city are replaced by electric buses. Currently, there are 2094 public buses operating in Hanoi. By 2030, around 1047 electric buses are projected to replace traditional buses, representing approximately 2% of the total number of buses and coaches in the city. The emission data was adjusted according to Scenario 2 and then integrated into the CMAQ air quality model. The simulation results from Scenario 2 are compared with the baseline scenario (Scenario 1), as presented in the previous sections. The differences in the concentrations of key pollutants between the two scenarios will provide detailed information on the effectiveness of the traffic emission reduction measures in improving urban air quality.

Table 3. Emission results of pollutants from bus sources in 2022 for two scenarios. (Unit: tons/year).

Pollutant	Emissions from Buses in Scenario 1	Emissions from Buses in Scenario 2	Emission Reduction	Reduction Rate (%)
NOx	14,613.3	12,365.7	2247.6	15.38
CO	1154.2	954.9	199.3	17.27
SO2	58.8	50.6	8.2	14.02
VOC	347.8	290.7	57.0	16.40
CH4	83.4	71.4	12.0	14.43
NMVOC	267.1	222.1	45.0	16.85
PM2.5	4727.5	4192.3	535.3	11.32

and

Table 4. Emission results of pollutants from traffic sources in 2022 for two scenarios. (Unit: tons/year).

Pollutant	Emissions from Traffic Sources in 2022
	Scenario 1	Scenario 2	Reduction Rate (%)
NOx	69,907.4	67,600.2	3.30
CO	2,033,569.2	2,033,365.1	0.01
SO2	4025.0	4016.5	0.21
VOC	150,081.8	150,023.3	0.04
CH4	30,825.6	30,813.2	0.04
NMVOC	189,224.8	189,178.7	0.02
PM2.5	29,756.8	29,202.2	1.86

present detailed data on pollutant emissions for both simulation scenarios. In Scenario 2, the substitution of 50% of fossil fuel-powered buses with electric buses leads to a reduction in emissions ranging from 11.32% to 17.27% for key pollutants. Among these, carbon monoxide (CO) demonstrates the most significant reduction, while particulate matter (PM_2.5) shows the most minor decrease. However, when evaluating total emissions from all traffic sources, the reduction in Scenario 2 relative to Scenario 1 is relatively modest, with a range of 0.01% to 3.3%. Notably, NO_x exhibits the most considerable reduction, at 3.3%. These findings suggest that while introducing electric buses effectively lowers local emissions from this specific source, its impact on the overall reduction in emissions across all traffic sources is limited. It highlights the necessity for supplementary strategies to substantially reduce emissions within urban traffic sectors.

3.3. Model Calibration and Validation of CMAQ

Temperature validation involved extracting values at a height of 2 m from the Time Series (TS) output files corresponding to the measurement location at Ha Dong station. Since the simulated temperature was provided in Kelvin, these values were converted to Celsius for comparison with observed data. Table 2 presents the comparison results between simulated and observed temperature values. A strong correlation (R = 0.89, Table 2) was observed between the simulated and observed temperature data. The test results indicated that the simulated temperatures met the MAGE and Factor 2 criteria (

Table 5. The valuation parameters comparing the temperature and wind speed results from the WRF simulation with the observed data at Ha Dong meteorological station.

Index	Temperature	Wind Speed	Reference Value
MAGE	1.18	0.61	≤2
RMSE	1.38	0.70	-
Factor 2	0.96	1.21	1.00
R	0.89	0.59	-

), suggesting that the model accurately reflected surface temperature changes during the validation period.

For wind validation, wind speed components in the x and y directions were extracted from the TS output files at a height of 10 m. These components were then used to calculate the total wind speed and direction. The MAGE indices for wind speed at Ha Dong station met the standard values. The correlation coefficient (R) was 0.59, indicating that the WRF model more accurately reflected surface wind conditions. However, observed data at the station was limited to rounded wind speed and direction values recorded at specific times (1:00, 7:00, 13:00, and 19:00 LCT), restricting statistical comparison. Therefore, wind validation was further conducted by comparing wind rose diagrams and analyzing prevailing wind direction from WRF model output. Figure 3 shows the wind rose diagrams for observed and simulated data at Ha Dong station. A significant similarity in both wind speed and direction distributions was found. Both the WRF model and observations indicated a dominant southwest wind direction during the study period.

For CMAQ result validation, PM_2.5 values were extracted from the CCTM-AELMO output file at the U.S. Embassy station in Hanoi. The CMAQ model demonstrated strong performance in simulating PM_2.5 concentrations, with an R-value of 0.78 (Figure 4), MB of 0.64, and RMSE of 8.99. Additional performance metrics, including the IOA = −0.5, NMB = 7.11%, and NME = 28.51%, all met the U.S. EPA performance criteria (NMB ≤ ±30%, NME ≤ ±70%). These results indicate that the CMAQ model configuration used in this study is reliable and suitable for simulating air quality in northern Vietnam.

3.4. Air Quality Simulation Results in Hanoi

3.4.1. Temporal Distribution of Air Pollutants in Hanoi

Table 6. Monthly average of air pollutants in 2022 for domain 3 (unit: µg/m³).

Air Pollutant	January			July
	Average	Min *	Max *	Average	Min *	Max *
PM2.5	100.1	77.3	128.9	17.0	9.3	23.2
CO	1861.8	505.3	11766.6	965.3	108.6	8583.4
NO2	30.3	3.5	222.2	16.9	0.2	162.5
O3	59.2	25.0	82.2	45.5	30.7	52.3
SO2	11.8	5.4	104.9	5.2	0.3	85.0

* Min and max values across the entire domain D3.

summarizes the average concentrations of air pollutants derived from the CMAQ model outputs for January and July 2022, specifically focusing on the northern region, particularly domain D3.

The temporal air quality analysis in Hanoi reveals substantial seasonal variations influenced by meteorological conditions and emission dynamics. In January 2022, pollutant concentrations were significantly higher compared to July 2022, as shown in Table 6. During winter, meteorological conditions such as low temperatures, stable atmospheric layers, and reduced vertical mixing limit pollutant dispersion. This leads to the accumulation of pollutants, particularly PM₂.₅, which exhibited an average concentration nearly six times higher in January than in July. Similarly, CO, NO₂, O₃, and SO₂ concentrations increased by factors of 1.9, 1.8, 1.3, and 2.3 during the winter months. These elevated levels reflect the impact of local and regional emission sources and unfavorable weather conditions.

Despite the seasonal increase in pollution, the average concentrations of some pollutants in January remained within the regulatory limits of QCVN 05:2023/BTNMT. For instance, the average CO concentration was recorded at 1861.8 ± 1535.1 µg/m³, approximately 5.4 times below the standard, while NO₂ and SO₂ concentrations were about 3.3 and 10 times below their respective limits. The average O₃ concentration, at 59.2 ± 10.6 µg/m³, was approximately half of the regulatory threshold. These findings suggest that while pollution levels are higher in winter, they generally comply with Vietnamese air quality standards.

In contrast, the summer season presented a markedly different scenario. Air quality in July 2022 was classified as good, primarily due to favorable meteorological conditions. Prevailing air masses from the ocean, high humidity, and enhanced precipitation facilitated the washout of atmospheric pollutants. This washout effect, combined with cleaner marine-origin air masses, significantly reduced pollutant concentrations. Specifically, PM_2.5 concentrations decreased substantially, highlighting the influence of precipitation on fine particulate matter removal. CO, NO₂, O₃, and SO₂ concentrations also declined, reflecting reductions in both local emissions and regional contributions. The seasonal variation underscores the critical interaction between emission sources and meteorological factors, providing insights for developing targeted air quality management strategies

3.4.2. Spatial Distribution of Air Pollutants in Hanoi

The spatial distribution of air pollutants in Hanoi highlights distinct patterns across different regions, shaped by the interplay between emission sources, topography, and atmospheric processes. PM_2.5 concentrations, as illustrated in Figure 5, were significantly higher in suburban areas compared to central urban regions. These elevated levels in the outskirts of Hanoi are linked to contributions from industrial activities and cross-boundary pollutant transport. We also study the backward trajectory analysis (BWT) which indicates air masses moving from the northern part of Vietnam carried pollutants from industrial zones into Hanoi, exacerbating PM_2.5 pollution in suburban areas. This highlights the role of regional emissions and long-range transport in fine particulate matter accumulation.

In contrast, urban areas exhibited lower PM_2.5 concentrations, which could be attributed to enhanced dispersion facilitated by urban heat island effects and lower contributions from distant sources. However, urban regions showed higher concentrations of gaseous pollutants such as CO, NO₂, and SO₂ (Figure 6, Figure 7 and Figure 8). These pollutants are closely associated with local emission sources, including traffic, industrial operations, and residential combustion activities. High traffic density, particularly during peak hours, contributes significantly to CO and NO₂ levels, while fossil fuel combustion in industries adds to SO₂ emissions. The concentration of O₃, a secondary pollutant, showed a more complex spatial pattern (Figure 9), influenced by photochemical reactions and precursor emissions from local and regional sources.

The spatial differentiation between PM_2.5 and gaseous pollutants underscores the distinct mechanisms driving their formation and distribution. While PM_2.5 pollution is primarily influenced by distant sources, chemical transformations, and accumulation under stable atmospheric conditions, gaseous pollutants are more strongly dependent on localized emission dynamics. Additionally, the presence of fine and coarse particulate matter in suburban areas suggests the contribution of secondary atmospheric reactions, further complicating the pollution landscape.

These spatial and temporal findings emphasize the importance of a comprehensive air quality management strategy for Hanoi. Addressing local emission sources in urban areas is critical to controlling CO, NO₂, and SO₂ concentrations, while mitigating PM_2.5 pollution requires regional cooperation to reduce industrial emissions and manage transboundary pollution. By integrating meteorological considerations and emission control policies, Hanoi can achieve sustainable improvements in air quality and reduce the health risks associated with air pollution.

3.5. Evaluation of Air Quality Simulation Results with Emission Reduction Scenarios

Table 7. Difference ratio of average air pollutant concentrations for January and July between the two scenarios for domain 3.

Air Pollutant	January			July
	Average	Min *	Max *	Average	Min *	Max *
PM2.5	0.017	−0.064	0.088	0.3387	−5.2564	5.1372
CO	0.007	0.002	0.010	0.0166	0.0107	0.0298
NO2	2.964	1.592	3.420	3.1496	1.9106	3.7838
O3	−1.959	−5.558	−0.831	−0.2714	−1.0228	0.0658
SO2	0.068	−0.075	0.158	0.7998	−3.5126	10.3743

* Min and max values across the entire domain D3.

presents the difference ratio of average air pollutant concentrations for January and July between the two scenarios for domain 3, as calculated from the CMAQ simulation outputs.

Implementing emission reduction measures in Hanoi in January 2022 demonstrated varying levels of effectiveness across different pollutants, as shown by CMAQ model simulations. The concentration differences between the scenarios are summarized in Table 4. The results indicate that bus emission reductions led to concentration changes ranging from approximately −1.9% to 2.96%. However, the overall effectiveness in January was limited, primarily due to unfavorable meteorological conditions, including low temperatures and a shallow boundary layer, facilitating the accumulation of pollutants at the surface. From the site record, January is characterized by markedly cooler and more stagnant conditions than summer. The mean air temperature in January is 18.3 °C, versus 29.3 °C in August, and winds are generally weaker in January (1.7 m/s, versus 1.9 m/s in August). Such a combination suppresses turbulent convection and maintains a shallow planetary boundary layer, which enhances near-surface accumulation of pollutants. Precipitation in January is also limited (0.29 mm vs. 2.51 mm), providing little wet scavenging. Together, these factors are consistent with the reduced apparent effectiveness observed in January despite the same emission-control measures. Additionally, long-range pollutant transport from industrial zones in the northern region and beyond significantly influenced air quality in Hanoi, reducing the impact of local mitigation measures.

PM_2.5 concentration differences showed minimal changes, with an average difference of 0.017% ± 0.041%, ranging from −0.064% to 0.088%. Spatial analysis (Figure 10a) revealed higher reductions in central Hanoi, where traffic emissions are most concentrated. It indicates that traffic-focused measures moderately reduced PM_2.5 levels in urban centers. However, the reduction effect was less evident in suburban areas, highlighting the necessity for more comprehensive strategies to manage PM_2.5 across the entire region.

For NO₂, emission reduction measures achieved a more pronounced effect, with an average concentration reduction of 2.964% ± 0.332%, ranging from 1.592% to 3.420%. It underscores the critical role of controlling traffic emissions in mitigating NO₂ pollution. In contrast, SO₂ exhibited a modest average reduction of 0.068% ± 0.028%, ranging from −0.075% to 0.158%. The limited decrease in SO₂ likely reflects the influence of secondary emissions and atmospheric transport processes. Spatial distribution patterns of NO₂ and SO₂ reductions (Figure 11a and Figure 12a) suggest shared sources, primarily from traffic emissions, in urban and suburban areas.

CO concentrations showed minor reductions, with an average decrease of 0.007% ± 0.002%, ranging from 0.002% to 0.010%. This indicates a consistent but limited impact of emission reductions on CO levels (Figure 13). Conversely, O₃ concentrations experienced a modest increase under Scenario 2, with an average difference of −1.959% ± 0.664%. The increase from −5.558% to −0.831% can be attributed to reduced NO_x emissions. In urban environments with high NO_x levels, NO acts as an O₃ scavenger. Reducing NO_x emissions diminishes this effect, causing localized increases in O₃ concentrations, particularly in areas previously dominated by high NO_x levels. Figure 14a illustrates these spatial differences, showing significant O₃ increases in central Hanoi.

In July 2022, the application of emission reduction measures led to more substantial reductions compared to January. Concentration difference rates ranged from −0.27% to 3.1%, reflecting the influence of favorable summer meteorological conditions. PM_2.5 concentrations decreased by an average of 0.3387% ± 2.8198%, ranging from −5.2564% to 5.1372%. The reduction was more pronounced in urban areas, as shown in Figure 10b, where traffic-related emissions dominate.

NO₂ reductions in July were notably higher, with an average rate of 3.1496% ± 0.2045%, ranging from 1.9106% to 3.7838%. Enhanced dispersion and washout effects during summer amplified the effectiveness of emission reductions. CO and SO₂ reductions also increased, with average rates of 0.0166 ± 0.0044% and 0.7998 ± 3.1284%, respectively. These results suggest that summer meteorological conditions, such as higher temperatures and solar radiation, improve the efficacy of control measures by facilitating oxidation and dispersion processes.

O₃ concentrations in July exhibited slight increases in the emission reduction scenario, with an average difference of −0.2714% ± 0.2921%, ranging from −1.0228% to 0.0658%. Like January, this trend reflects the complex interplay between reduced NO_x emissions and O₃ formation dynamics. Figure 10b, Figure 11b, Figure 12b, Figure 13b and Figure 14b show the spatial distribution of concentration differences for July, emphasizing the variability in pollutant responses across the study area.

In general, emission reduction measures demonstrate varying effectiveness depending on the pollutant and season. NO₂, SO₂, and PM_2.5 showed significant reductions, especially in summer, while O₃ concentrations exhibited localized increases. These findings underscore the importance of integrating seasonal meteorological conditions and chemical interactions into air

3.6. Health Impact Results with Emission Reduction Scenarios

From the results above on the PM_2.5 concentration over the domains, the health impact as indicated by the mortality rate and cvd hospitalization due to emission of anthropogenic sources on the population exposure are calculated and summarized in

Table 8. Mortality and cvd hospitalization number with a 95% confidence interval due to population exposure to PM_2.5 air pollution in Hanoi. The PM_2.5 ambient concentration is simulated using WRF-CMAQ air quality model. (*) The WHO PM_2.5 safe guideline of 5 µg/m³ is used as the reference concentration value below which there is no effect on health.

	January (Dry Season)		July (Wet Season)
	Mortality	CVD Hospitalization	Mortality	CVD Hospitalization
Overall baseline impact due to PM2.5 pollution (*).	626 [245, 963]	124 [−1, 232]	94 [35, 152]	18 [0, 36]
Overall impact due to PM2.5 pollution with 50% EV buses in fleet (*).	625 [245, 962]	124 [−1, 232]	94 [35, 152]	18 [0, 36]
Change in impact between baseline and 50% EV buses scenarios.	0.25 [−717, 717)	0.05 [−233, 234]	0 [−117, 117]	0 [−36, 36]

and in Figure 15 (for January 2022). Here, we consider health impact due to one pollutant PM_2.5 as it has the most impact compared to other gaseous pollutants such as NO₂, SO₂, or O₃. The densely populated area in and around Hanoi Old Quarter has the highest population exposure and high mortality and cvd hospitalization as compared with those in the surrounding areas.

From Table 8, the wet season as in July has less impact on mortality and cvd hospitalization in the Hanoi domain due to lower levels of PM_2.5. There were about 625 mortalities in Hanoi and 124 hospitalizations due to air pollution in January 2022 as compared with estimated 94 mortalities and 18 cvd hospitalizations in July. There is no improvement in terms of health impact when 50% of the fossil-fuel combustion engines in the public bus fleet are replaced with EV buses. This is due to the fact that emission from public buses is only a part of the emission in the transport sector and emission from other sectors. Higher proportions of EV buses in public buses and more EVs in the no-buses motor vehicle fleet can yield measurable benefits in population health. The overwhelming vehicle type that contributes most to the air pollution in Hanoi is motorcycles. This should be the focus of the effort of reducing the health impact and the air pollution such as PM_2.5.

In this study, we only consider the effect of PM_2.5 as it has the major impact on mortality and cardiac diseases hospitalization health endpoints but other pollutants such as NO₂, SO₂, or O₃ can have the effect on different health endpoints such as respiratory diseases hospitalization such as asthma and Chronic Obstructive Pulmonary Disease (COPD). Muti-pollutant effect is much more complex to consider and beyond the scope of this study. Future studies will focus on other important traffic emission sources beside buses and more comprehensive health impacts on various health endpoints including respiratory diseases from a number of pollutants including NO₂ and O₃.

4. Conclusions

This study utilizes the WRF-CMAQ version 5.4 model with the CB6 chemical mechanism to simulate air quality in Hanoi. The WRF model provides meteorological data to CMAQ through MCIP, using reanalysis data from the US NCEP. Emission data is derived from sources, including the 2022 emission inventory in Hanoi, the EDGAR v5 database, biogenic emissions from MEGANv2.10, and other natural sources. The simulation domain consists of three nested levels: D1 (27 km × 27 km), D2 (9 km × 9 km), and D3 (3 km × 3 km), with a focus on the Hanoi area in D3. Boundary conditions and initial inputs are provided by the global model CAM-Chem. This study offers a comprehensive assessment of air quality in Hanoi city and provides a scientific basis for future air pollution management strategies.

Air quality simulations were conducted for two representative periods: the dry season (January) and the wet season (July) of 2022, while also developing an emission reduction scenario where 50% of fossil fuel-powered buses are replaced by electric buses. The simulation results under the current scenario indicate that air quality in Hanoi during January is polluted at certain times and certain locations, with increasing PM_2.5 concentrations due to adverse meteorological conditions, limiting the dispersion of pollutants. The spatial distribution reveals higher pollution levels in suburban areas, mainly due to terrain effects and local emission sources. Backward trajectory analysis shows transboundary air pollution, with air masses from the northern part of the northern region of Vietnam significantly contributing to fine dust concentrations, highlighting the importance of regional cooperation in air pollution management. Concentrations of gaseous pollutants (CO, NO₂, O₃, and SO₂) in January were higher than in July, reflecting seasonal variability, although they remained within the regulatory limits of QCVN 05:2023/BTNMT. Spatial analysis shows elevated CO, NO₂, and SO₂ concentrations in urban areas, primarily due to traffic and industrial emissions, while PM_2.5 is influenced by long-range transport and atmospheric processes. In July, air quality improved significantly due to favorable meteorological conditions, including clean airflows from the sea and increased rainfall, which helped remove pollutants. Pollutant concentrations decreased by 1.3–5.9 times compared to January, with PM_2.5 showing different spatial distributions. Higher fine dust concentrations in suburban areas reflect contributions from secondary emission sources and nearby industrial zones.

Simulation data from Scenario 2 were compared with Scenario 1, showing changes in pollutant concentrations. In January, bus emission reductions resulted in concentration changes ranging from −1.9% to 2.96%, with an overall low reduction due to unfavorable meteorological conditions and long-range pollutant transport from northern Hanoi city and the northern part of the north region of Vietnam. The reduction in PM_2.5 was not significant, ranging from −0.064% to 0.088%, with the highest reductions in urban areas, where traffic emissions are concentrated. For NO₂, the reduction was more significant at 2.964% ± 0.332%, while SO₂ and CO showed only slight reductions. O₃ showed a slight increase due to reduced NO_x emissions. In July, the results showed more distinct air quality changes, with pollutant concentration differences ranging from −1.6% to 3.1%. The PM_2.5 reduction was 0.3387%, with substantial spatial variability. NO₂ decreased by 3.1496% ± 0.2045%, higher than in January, and CO and SO₂ also showed more significant reductions due to favorable meteorological conditions. The results demonstrate that emission reduction measures are effective in mitigating concentrations of NO₂, SO₂, and PM_2.5, although their impact exhibits temporal and spatial variability.

Health impact assessment based on impact of simulated PM_2.5 air pollution in Hanoi on mortality shows that the estimated mortality during January dry season is much higher than that during the July wet season. The replacement of 50% of fossil-fuel combustion with electric vehicles in the public buses in Hanoi yields no measurable improvement in mortality for both seasons. Higher targets of conversion of combustion buses to electric buses and more focusing on the reduction in emission from motorcycles which dominates the traffic emission will yield more tangible benefits.