Ground-level ozone in the Mekong Delta region: precursors, meteorological factors, and regional transport

The Mekong Delta region (MDR), also known as Vietnam’s rice bowl, produced a bountiful harvest of about 23.8 million tons in 2020, accounting for 55.7% of the country’s total production, providing food security for 20% of the world population. With the rapid pace of industrialisation and urbanisation, the concentration of ozone in the lower atmosphere has risen to a level that reduces crop yields, especially rice, and is therefore the subject of research. This study aims to simulate the spatiotemporal distribution of ground-level ozone in the area and evaluate the impact of precursor emissions and meteorological factors on the spatiotemporal distributions of ozone concentrations. The study area was divided into seven zones, including six agro-ecological zones (AEZs) and one low-mountainous area, mainly to clarify the role of emissions in each AEZ. The simulation results showed that ground-level O3 in the MDR ranged from 40.39 to 52.13 µg/m3. In six agro-ecological zones, the average annual ground-level O3 concentration was relatively high and was the highest in zone 6 (CPZ) and zone 3 (LXZ) with values of 96.18 µg/m3 (exceeding 1.60 times the WHO Guidelines 2021) and 94.86 µg/m3 (exceeding 1.58 times the WHO Guidelines 2021), respectively. In each zone, the annual average O3 concentration tended to gradually increase from the inner delta to coastal areas. Two types of precursors, NOx and NMVOCs, are the main contributors to O3 pollution, with the largest contribution coming from zone 1 (FAZ) with 91.5 thousand tons of NOx/year and 455.2 thousand tons of NMVOCs/year. Among the meteorological factors considered, temperature (T), relative humidity (RH), and surface pressure (P) were the three main factors that contributed to the increase in ground-level ozone. The spatio-temporal distribution of ground-level O3 in the MDR was influenced by emission precursors from different zones as well as meteorological factors. The present results can help policy-makers formulate plans for agro-industrial development in the entire region.


Introduction
Although ozone (O 3 ) is essential in the Earth's upper atmosphere, its presence in the lower atmosphere acts as a secondary air pollutant which negatively impacts human health and crops (Wilkinson et al. 2012). The effect of phytotoxic ozone concentrations on crop yields has been extensively studied (Pleijel 2011;Arshad 2021;Ramya et al. 2021;Wang et al. 2021a, b;Shang et al. 2022). Europe's economic loss in 2000 was estimated at €6.7 billion, based on the effects of ground-level ozone on 23 crops (Ene et al., 2015); losses in global crop production for the same year were estimated at US$11-18 billion (Avnery et al. 2011). Recent studies have shown that rice crops are also sensitive to ground-level O 3 , especially in Asian cultivars (Shang et al. 2022). Several pilot studies conducted in Malaysia, Pakistan, and Vietnam have shown that the local rice varieties are affected by O 3 (Ishii et al. 2007;Arshad 2021).
Ground-level ozone has long been recognised as a threat to crop health and human health, and studies on its effects have been conducted (Nouchi et al. 1991;Wahid et al. 1995;Brunekreef and Holgate 2002;Tang et al. 2014). Plants that grow under favourable conditions form stomata on leaf Responsible Editor: Gerhard Lammel * Long Ta Bui longbt62@hcmut.edu.vn surfaces to exchange gases for photosynthesis (Emberson et al. 2009). In this case, ozone is one of the gases that can penetrate plants. The reduction in crop yields due to the cumulative effects of ground-level ozone is evident when ozone "waves" are prolonged; if this period coincides with a significant increase in precursor emissions, the effect is doubled (Feng & Kobayashi 2009). Symptoms of injury can be observed through leaf wilt or necrosis on the leaf surface, which can coalesce to form larger areas of injury; the leaves often dry up and fall prematurely (Emberson et al. 2009). Ozone reduces grain size, weight, and nutritional quality in field crops such as certain varieties of wheat, rice, maize, beans, and soybeans (Biswas et al. 2008;Decock & Six 2012).
Overall, ground-level O 3 has been shown to be one of the most difficult pollutants to control worldwide (Sharma et al. 2016) because surface O 3 formation is considerably influenced by both precursor emissions and meteorological conditions (Cao et al. 2020). Currently, many researchers have analysed, discussed, and demonstrated the effects of precursor emissions on surface O 3 . When there is an increase in the emissions of primary pollutants, the ground-level O 3 concentrations also increase (Fowler et al. 2008). Chen et al. (2019) reported that the increased contribution of NO x emissions and VOCs was responsible for the rise in surface O 3 concentrations in Beijing during the 2006-2016 period, and the results of Li et al. (2019a) clarified the role and contribution of NO x and VOC emissions to the increased-surface O 3 concentration in northern China during the period of 2013-2017.
The Mekong Delta region (MDR) of Vietnam has an area of nearly 40,000 km 2 and a population of nearly 18 million, accounting for approximately 12% of the land area and 22% of the country's population. This is a land with a strategic location in Vietnam, bordering the East Sea and Southwest Sea on three sides, with a coastline of more than 700 km with 360,000 km 2 , and an exclusive economic zone in a strategic location that is very favourable for economic development (Nguyen et al. 2019;Tran et al. 2019). The MDR, also known as the "rice bowl" of Vietnam, produced a bountiful yield of around 23.8 million tons in 2020, accounting for 55.7% of the country's total yields (Wassmann et al. 2019;Tu et al. 2021;Bich Tho & Umetsu 2022;Ferrer et al. 2022), providing food security for 20% of the world's population, particularly in regions where rice is the staple diet. With accelerated industrialisation and urbanisation of the MDRs, the concentration of ground-level ozone in the lower atmosphere has risen to levels that have resulted in reduced agricultural yields, particularly of rice; it is therefore the focus of this research (Danh et al. 2016).
Ground-level ozone is produced by nitrogen oxides (NO x ) and volatile organic compounds (VOCs) through photochemical reactions (Emberson et al. 2009;Guo et al. 2019); therefore ground-level ozone dispersion was simulated through the application of photochemical transport and meteorological models (Hogrefe et al. 2000;Sokhi et al. 2006;Davis et al. 2011;X. Li and Rappenglück 2014;N. Wang et al. 2015a;Astitha et al. 2017;Tao Wang et al. 2017;Trieu et al. 2017;Y. Chen et al. 2021;Mousavinezhad et al. 2021;Wang et al. 2021a, b). The use of meteorological/ photochemical transport models to predict hourly groundlevel ozone concentrations and the modelling results were evaluated based on the measured field data, which were done in a various studies (Sokhi et al. 2006;Davis et al. 2011;Li & Rappenglück 2014;N. Wang et al. 2015b).
Ground-level ozone concentrations often occur on days with strong sunlight and weak winds, creating favourable conditions for the production and accumulation of ozone and its precursors . Wind direction was also important because it affects the transport of pollution, thereby increasing the temperature of windy sites (Epa 2006;Li & Rappenglück 2014;Mitchell et al. 2021). The quantification of ozone production from different precursors, including NO x , VOCs, and individual VOC species, was performed (Garner & Thompson 2013;Simon et al. 2015;Wang et al. 2015a). Thus, with the help of these models, it was possible to propose solutions to reduce ozone pollution by minimising precursor emissions Xue et al. 2014).
Favourable meteorological conditions for photochemical episodes have been extensively studied (Camalier et al. 2007;Simon et al. 2012;McNider & Pour-Biazar 2020). For example, sunny weather and low wind velocity lead to pollution accumulation and O 3 production (Xue et al. 2014;Carter et al. 2017). Several studies have been conducted to determine the relationship between ozone concentration and local meteorological parameters, including solar irradiance, temperature, relative humidity, wind speed and direction, and cloud cover .
Recognising the special role and importance of the Mekong Delta region (MDR) in multiple aspects at the national and regional levels, a weather research and forecasting (WRF) model system combined with the community multiscale air quality (CMAQ) model was used to simulate and evaluate the distribution of surface O 3 pollution in space and time throughout 2018. Simultaneously, the various contributions of physical and chemical factors to surface O 3 formation were explored and quantitatively evaluated through photochemical reactions in the MDR during the simulation period. To solve this requirement, based on the simulation results, this study focused on presenting a systematic statistical analysis to quantify the correlations according to a multivariate linear regression (MLR) model of ground-level O 3 concentrations with variations in meteorological parameters (temperature T, relative humidity RH, and surface pressure P) and a number of emission precursors (CH 4 , CO, NO x , and volatile organic compounds [VOCs]) for the study scope, mainly on six agro-ecological regions of the MDR and for a simulation period to be evaluated throughout 2018. The main objective of this study is to identify the crucial correlation, which can be used to further understand the sensitivity of ground-level O 3 to changes in meteorological conditions, and to assess the change in O 3 precursor emission factors to the photochemical responses to surface O 3 formation simulated in the coupled WRF/CMAQ model system. The results of this study will support local governments and policymakers in developing medium-and long-term strategies to effectively control O 3 pollution, contribute to building an integrated socio-economic-environmentally sustainable relationship in the MDR, and aim to maximise the goals of economic benefits and environmental performance.

Study area
The MDR consists of 13 provinces (Fig. 1). According to data from the General Statistics Office of Vietnam in 2019, the Mekong River Delta has the largest total area in Vietnam, with a higher economic growth rate than the whole Fig. 1 Study area including six agro-ecological zones and one low-mountainous area country. Rice farms alone account for 47% of the area and 56% of the country's rice production, and rice exports from the entire region account for 90% of production (Cramb 2020;GSO 2020).
Based on the characteristics of geography, physics, and water resources, the MDR can be divided into seven different areas, including six important agro-ecological zones along with one area of hills and low mountains (HMZ) (Nguyen et al. 2007). The six agro-ecological zones include the (1) freshwater alluvial zone (FAZ), (2) Dong Thap Muoi region (Plains of Reeds zine, PRZ), (3) Long Xuyen-Ha Tien Quadrangle (LXZ), (4) Trans-Bassac Depression zone (TBZ), (5) coastal zone (CZ), and (6) Ca Mau peninsula (CPZ) (Nguyen et al. 2007). The industry is frequently affected by floods or saline intrusion, as indicated by the demarcation lines in Fig. 1. The FAZ, which is located along the Trans-Bassac and Mekong rivers in the central part of the delta, is characterised by alluvial soils and covers approximately 900,000 ha of available fresh water. In this zone, the residents subsist by diversifying agricultural activities and practising two-or three-crop rice farming combined with the cultivation of fruit trees, vegetables, and aquaculture. The PRZ, located in Dong Thap province and part of Long An province, covers approximately 500,000 ha. This is the upper part of the delta at 0.5 m below the average sea level. The area has alkaline soil and water, and related environmental factors are partially controlled by flooding and acid toxicity. The residents in this area subsist with rice farming and integrated aquaculture. The LXZ, located between the provinces of An Giang and Kien Giang, with an area of 400,000 ha, is characterised by alkaline soil. Water and environmental factors changed from saline alum ecology to freshwater ecology after the year 2000. Rice is the main crop in this area and rice-based farming is the main activity. The TBZ, located to the west of Can Tho Province, covers approximately 600,000 ha. It is a low-lying area on a plain that is only mildly affected by floods and saline intrusion and has favourable conditions for intensive farming and diverse crop production. Residents subsist by cultivating two or three rice crops. The CZ, located along the eastern part of the MDR, covers approximately 600,000 ha. Large areas in this zone have acidic soils. Since 1998, parts of this region have been subjected to changing water and environmental conditions from brackish to permanent freshwater ecology. In this area, residents subsist by farming shrimp and growing rice. The CPZ, located at the southernmost tip of the MDR, covers approximately 800,000 ha. The area is characterised by seasonal salinisation of the soil and various rice-based farming systems in terms of rainwater use. Since 1998, large swaths of the region have undergone changes in water and environmental conditions to a permanent freshwater ecoenvironment, with shrimp farming being the dominant activity here (Nguyen et al. 2007).

Observation of meteorology
In this study, the hourly realistic meteorological data from the observation stations of Can Tho and Tien Giang (Fig. 2) in the freshwater alluvial area and Kien Giang in the lowland area were extracted in March and October 2018 representing the dry and wet seasons of the MDR from the websites of https:// www. wunde rgrou nd. com/ and https:// www. timea nddate. com/, respectively. A time-series dataset of temperature, wind speed, wind direction, relative humidity, and surface pressure was captured from each monitoring site, which was used to verify the results of the WRF simulations in terms of accuracy and error levels using statistical indicators.

Observation of ground-level O 3
In 2018, field-measured ground-level O 3 concentration datasets were obtained from 11 monitoring sites within the simulation domain, including N, NT1, DT1, DT3, DT4, DT6, GT2, GT3, CN1, CN2, and CN4 (Fig. 2), which had been managed and operated by the Binh Duong Center of Natural Resources and Environment-Technical Monitoring (BREM) (Binh Duong 2019). These were intermittent manual monitoring stations with four measurements in a monitored day/month (namely, 09:00, 11:00, 13:00, and 15:00), and the sample collection time for one measurement was approximately 1 h (Table S1). Each monitoring station is operated to ensure the objectives of the designed monitoring program; specifically, positions DT1, DT3, DT4, and DT6 serve the purpose of monitoring the impact of urban environmental activities; the GT2 and GT3 sites are used to monitor the impact of traffic activities; the CN1, CN2, and CN4 sites are used to monitor the influence of activities from industrial parks and industrial clusters; and the NT1 and N positions are used to monitor the effects of agricultural activities and the background environment. These datasets were used to support the processes of calibration and validation of the ground-level O 3 concentration results simulated by the CMAQ model; specifically, the measured data from six stations, N, DT3, DT4, DT6, CN2, and GT3, were used for calibration, and the observed data of five stations, DT1, GT2, CN1, NT1, and CN4, were used for verification.

Emission inventory datasets
In this study, anthropogenic and natural biogenic emission inventory datasets from ECCAD (Emissions of Atmospheric Compounds and Compilation of Ancillary Data) (https:// eccad3. sedoo. fr/) in 2018 were used; specifically, two anthropogenic emission datasets, CAMS-GLOB-AIR and CAMS-GLOB-ANT and the biogenic emission dataset CAMS-GLOB-BIO developed by Copernicus Atmospheric Monitoring Center (CAMS) (Granier et al., 2019a, b), Joint Research Center of the Ispra Institute for Environment and Sustainability, Italy (Janssens-Maenhout et al. 2015), and CNRS Laboratories -Paris, France (Sindelarova et al. 2014), were included. This was the output from a combination of EDGARv4.3.2 emissions data developed by the Common European Center (Crippa et al. 2018) and CEDS emissions (Hoesly et al. 2018) that provide the emissions for IPCC reporting, and next, AR6. The datasets were captured and used in fluxes (kg.m −2 .s −1 ) with a spatial resolution on a global scale (0.50° × 0.50°, 0.25° × 0.25°, and 0.10° × 0.10°, respectively); we calculated the spatial distribution of grid cells of domain D02 (covering the entire MDR study area) with a spatial resolution of 9.0 km × 9.0 km and distribution over time (hourly) based on the distinction of point source, mobile source, and area source combined with information by industry sectors. The types of emission sources could be attributed to the emission contribution with the method detailed in previous studies (Friedrich et al. 2006;Bui et al. 2021).
The CAMS-GLOB-ANT dataset of different species was estimated from 14 sectors of anthropogenic emissions, including power generation 1 , road transportation 2 , off-road transportation 3 , fugitives 4 , industrial process 5 , solvents 6 , ships 7 , solid waste and wastewater 8 , residential and other sectors 9 , agriculture livestock 10 , agriculture soils 11 , agricultural waste burning 12 , general agriculture activities 13 , and the total contribution of sectors 14 (Granier et al. 2019a, b).
The emission inventory datasets, estimated emission levels of 19 primary substances (EC, OC, PMC, PMOTHR, PNCOM, H 2 O, K + , Na + , Ca 2+ , Al 3+ , Fe 2+/3+ , Mg 2+ , Mn 2+ , Si 4+ , Ti 2+ , NH 4 + , Cl − , NO 3 − , and SO 4 2− ) and 34 substances involved in secondary reactions (CH 4 , CO, NO x , SO 2 , NH 3 , non-methane volatile organic compounds [NMVOCs], and other VOCs) contributing to ground-level O 3 generation were coded according to the 3rd updated CB6 Carbon Bond Chemical Mechanism (CB6r3) (Yarwood et al. 2010;Emery et al. 2015). The emission loads of different types of VOCs were also included in the global emission datasets of CAMS-GLOB-AIR, CAMS-GLOB-ANT, and CAMS-GLOB-BIO by sector contributing to VOC emissions (Granier et al. 2019a, b). In this study, the global emission datasets  (Granier et al. 2019a, b) were used to uncover the anthropogenic and biologically induced emission contributions in the MDR. The results of processing the emission load data distribution are shown in Figs. S1 to S36.

Model configuration
In this study, the offline WRF model version v.3.8 (Skamarock et al. 2008) was used to simulate the meteorological conditions. The NCEP (National Center for Environmental Prediction) Final (FNL) Operational Global Analysis data from every 6 h have a spatial resolution of 1.0° × 1.0° from the US National Center for Atmospheric Research (https:// rda. ucar. edu/ datas ets/ ds083.2/) and were used as the initial and boundary conditions and the heuristic analysis for the WRF model. These NCEP FNL data were generated from the Global Data Assimilation System (GDAS) (https:// rda. ucar. edu/ datas ets/ ds083.2/) (NCEP 2000) based on continuously collected monitoring data sources, which are meteorological parameters such as ground pressure, sea level pressure, geopotential temperature, sea surface temperature, soil temperature, ice cover, relative humidity, U wind, and V wind. The FNL data have been widely used in many studies to simulate meteorological conditions and air quality in various regions of the world (Wang et al. 2021a, b).
The simulation started on 15 December 2017 and continued for all 12 months in 2018 (from 00:00 local standard time (LST) of 1 January 2018 to 23:00 LST on 31 December 2018). Of these, the first 5 days of the simulation was used to generate deep soil temperature and humidity because soil effects are often used to optimise surface temperature and moisture parameters (Pleim & Xiu 2003;Pleim & Gilliam 2009;Qin et al. 2019). The WRF model was set up with two nested computational domains (D01 and D02) (Fig. 2), with the number of grid cells for D01 and D02 being 76 × 94 and 55 × 43 grid cells, respectively. The outer largest domain (D01) has an area of approximately 5.41 × 10 6 km 2 and a horizontal grid spatial resolution of 27.0 km covering the entire country of Vietnam; the inner domain (D02) nested in the D01 calculation domain has an area of approximately 2.11 × 10 5 km 2 and has a horizontal grid spatial resolution of 9.0 km covering most of the South Central and Central Highlands provinces of Vietnam. The specific description of the setting parameters for the D01 and D02 calculation domains is shown in detail in Table 1. Vertically, there are all 31 sigma layer levels for all grid cells of the two domains D01 and D02 in the WRF model, from ground level to a fixed peak sigma layer at 100 hPa pressure. The output from the WRF model was processed by the MCIP tool version v.4.5.3 (Meteorology-Chemistry Interface Processor) to generate the input format required by the CMAQ model. Next, the CMAQ model version v.5.2.1 (http:// cmasc entre. org/ cmaq/) that was updated and published in June 2017 by the US Environmental Protection Agency (US EPA) (Borge et al. 2014;Hu et al. 2015;Lang et al. 2017) was applied to simulate the surface O 3 concentration distribution in the MDR from 1 January 2018 to 31 December 2018.
To ensure the accuracy of the boundary conditions of the meteorological fields, the horizontal domains of the WRF model are normally slightly larger than those of the CMAQ model (Li et al. 2022). The CMAQ model in this study was configured with the same nested domains as the WRF model, but three grid cells in each direction of the computed domains were removed from the D01 and D02 domains of the WRF model, so the number of grid cells of the groundlevel O 3 simulation domains in the CMAQ were 73 × 91 and 52 × 40 grids, respectively. For the CMAQ model, there were a total of 29 layers in the sigma coordinate system; specifically, the sigma values (σ) for the CMAQ computing domains at the layer boundaries were 1. 000, 0.997, 0.990, 0.983, 0.976, 0.970, 0.962, 0.954, 0.944, 0.932, 0.917, 0.898, 0.874, 0.844, 0.806, 0.760, 0.707, 0.647, 0.582, 0.513, 0.444, 0.375, 0.308, 0.243, 0.183, 0.126, 0.073, 0.023, and 0.000. The third updated and extended edition  (Granier et al. 2019b) of NO x and VOCs (2018), the intensity spatial resolution was 0.25° × 0.25° grids. All these emissions were interpolated linearly (Jiang et al. 2010;Liu et al. 2013;Wang et al. 2016) into the spatially resolved interior domain of 9.0 × 9.0 km and were used to simulate the ground-level O 3 concentration on the D02 computing domain. Initial conditions for domain D01 for each run of the CMAQ model were obtained by running the model for 17 days before (from 15 December 2017 to 31 December 2017) the modelling period throughout 2018 for the MDR and began with clean air conditions on the calculated domain. For the chemical boundary conditions of the calculated domain, D01 was mainly built based on the monthly average concentrations of ground-level O 3 , nitrogen dioxide (NO 2 ), nitric oxide (NO), carbon monoxide (CO), peroxyacetyl nitrates (PANs), methane (CH 4 ), ethane (C 2 H 6 ), formaldehyde (HCHO), and nitric acid (HNO 3 ), which were obtained from the Global Model of Ozone and Related Tracers, version 4 (MOZART-4) (Emmons et al. 2010). The boundary conditions of other chemicals according to CB6r3 (Yarwood et al. 2010;Emery et al. 2015) in the CMAQ model for photochemical processes were set to zero and varied during the simulation period based on the input emission data developed for the CMAQ model. At the same time, the boundary conditions of domain D02 were also obtained from running the model for the large domain D01. The initial conditions for the calculation domain D02 were created on the basis of the available monitoring data sources and the outputs of the simulation system for outer domain D01. Simultaneously, a spin-up period of 4 days was used to minimise the influence of initial chemical conditions when emulating empirical studies (Yu et al. 2014;Qin et al. 2019;Wang et al. 2021a, b).

WRF/CMAQ modelling performance protocol
In this study, the fusion data method between the measured and simulated ground-level O 3 results was used to correct the initial outcomes estimated by the coupled WRF/CMAQ model, based on the study outcomes of Friberg et al. (2016) and Senthilkumar et al. (2019). Equation (1) illustrates the calculation of two regression parameters, α and β, reflecting the correlation relationship applied to correct the computed ground-level O 3 concentration results by the coupled WRF/ CMAQ model, as follows: where CMAQ x is the exported initial ground-level O 3 concentration based on the computing results of the coupled WRF/CMAQ model at the observation station x at the given time t (h); CMAQ x, Corrected is the verified ground-level O 3 concentration close to the field-measured data at the monitoring site x at the specific time t (h); and α and β are the coefficients of the correlation equation used for calibration.
In general, statistical metrics are frequently used to verify the performance of WRF and CMAQ models by determining the relationship between the measured and simulated values (Nghiem and Oanh 2008;Wang et al. 2015aWang et al. , 2016. Actual hourly data on meteorological parameters and surface O 3 concentrations were used for these assessments. To verify the performance of the WRF model, hourly 2-m temperature (T), hourly 2-m relative humidity (RH), hourly 10-m wind speed (W s ), hourly 10-m wind direction (W d ), and hourly 2-m surface pressure (P) were evaluated using the following statistical metrics, which were calculated for each meteorological location: mean bias (MB), mean gross error (MAGE), root mean square error (RMSE), and index of agreement (IOA) (Huang et al. 2005;Wang et al. 2013). To assess the performance of the CMAQ model, the actual hourly data of groundlevel O 3 concentrations from 5 ambient air quality monitoring stations (DT1, GT2, CN1, NT1, and CN4) in Binh Duong Province (Fig. 2) were used. Similar to the WRF model, MB, ME, and RMSE, and other metrics such as normalised mean bias (NMB), normalised mean error (NME), unpaired peak accuracy (UPA), and the Pearson correlation coefficient (R) were also calculated to assess the efficiency of the chemical conversion simulations (Yu et al. 2006;Nghiem & Oanh 2008;N. Wang et al. 2016;Li et al. 2022). The formulae for calculating the statistical metrics of MB (2), MAGE (3), RMSE (4), IOA (5), NMB (6), NME (7), UPA (8), and the correlation coefficient R (9) are as follows:  (Fig. 2). Coefficients α and β were determined based on the difference between the actual measured and modelled datasets (Table 2).
Moreover, the corresponding benchmarks (criteria) applied to each statistical metric (from Eqs. (2) to (9)) to verify the performance of the WRF and CMAQ models based on previous evaluations from meteorological simulation studies (Emery & Tai 2001;Yu et al. 2006;Jiang et al. 2010;Wang et al. 2013) and air quality (US EPA 1991;Wang et al. 2015a;Emery et al. 2017) are presented in Table 3.  Table 4; x k and y are the temporal means of x k and y, respectively; s y and s k are the standard deviation values of y and x k , respectively; k is the normalised (dimensionless) regression coefficient; and t is the time. Moreover, the k * original linear regression coefficients have the unit µg.m −3 .D −1 , where D is the dimension of the x k variables including meteorological and O 3 precursor emission parameters as described in Table 4, determined as Eq. (23):  The MLR model is applied to each individual zone of MDR, including FAZ, PRZ, LXZ, TBZ, CZ, and CPZ (only HMZ is not calculated). All data ( x k and y) were deseasonalised and detrended by subtracting the 30-day average from the original data, such that x k = y = 0 (A. P.K. Tai et al. 2012). This process allows analyses to focus on changes at a synoptic-scale variability and avoids aliasing from common seasonal variations (Tai et al. 2012). The estimated normalised regression coefficients ( k ) allow direct comparison of the correlation between ground-level O 3 concentrations and analysed meteorological variables, as well as precursor emissions (Kutner et al. 2004). Nevertheless, the MLR models examined in this study did not consider interaction terms because the influence of these variables was insignificant and had been demonstrated in previous studies (Tai et al. 2012;Li et al. 2019aLi et al. , b, 2020).

Implementation steps
The information/data model system and processing steps used in this study are shown in Fig. 3. The steps for preparing and processing the emission data are described in the "Emission inventory datasets" section. The steps of running, processing, and verifying the results and creating a meteorological dataset using the WRF model and a ground-level O 3 concentration dataset using the CMAQ model are described in the "Model configuration" section. The field-measured data of the meteorological and surface O 3 concentration measurements are presented in the "Observation of meteorology" and "Observation of ground-level O3" sections. The steps for calibrating and validating the coupled WRF/CMAQ model are presented in the "WRF/CMAQ modelling performance protocol" section. The MLR models determined the relationship between ground-level O 3 concentrations and meteorological variables, as well as precursor emission variables, as described in the "Model of multiple linear regression" section.

Distribution of weather conditions and WRF modelling performance
To evaluate the simulation performance of the WRF model, the results of the meteorological parameter fields 2-m T, 2-m RH, 10-m W s , 10-m W d , and 2-m P hourly height output from the WRF model was compared with real data measured at 3 surface meteorological stations including Can Tho, Kien Giang, and Tien Giang (both the dry and wet seasons). Table S2 presented the results of the calculated statistical metrics (MB, MAGE, RSME, and IOA) for each During the dry season, the wind direction at all three meteorological observation sites followed the same patterns in Can Tho, Kien Giang, and Tien Giang (Fig. 2); the dominant direction of the wind was southeast, and the wind speed ranged from 3.3 to 5.5 m/s (considered a light breeze). In the wet season, the wind speed ranged between 5.5 and 7.9 m/s (considered a moderate breeze), and the dominant direction of the wind was also southwest. Furthermore, in the wet season at the two measuring stations in Kien Giang and Tien Giang, the northeast wind direction was observed with a high frequency and the wind speed ranged from 3.3 to 5.5 m/s (considered the light breeze level).
The simulated meteorological parameter results also showed that the relative humidity in the wet season was higher than that in the dry season at all times. At the Can Tho station, the relative humidity in the dry season was approximately 68.3%, whereas in the wet season, this value increased by 15.0%, reaching approximately 83.5%. Among the three measuring stations, the Can Tho station observed a larger difference than the other two stations; the humidity difference of the Can Tho, Kien Giang, and Tien Giang stations obtained was 15.2%, 12.9%, and 12.4%, respectively. In contrast to the relative humidity factor, the surface pressure during the wet season was higher during the wet season, but the difference was not high, ranging from 0.2 to 0.3 mbar. The monthly mean surface pressure in both seasons showed a declining trend at the measuring stations of Can Tho, Kien Giang, and Tien Giang (Fig. 2). The estimated meteorological factor results consisting of the temperature, relative humidity, and surface pressure used to assess the study area were also applied in the CMAQ model shown in Figs. S37, S38, and S39.

CMAQ performance evaluation
Similar to meteorological parameters, the simulated groundlevel O 3 concentration from CMAQ corresponds to the monitoring times at the nearest grid nodes for the location of 5 observing stations (Fig. 2). The D02 domain was used to compare with the ground-level O 3 concentration values observed by measured hourly hours (09:00, 11:00, 13:00, and 15:00), respectively, at these 5 monitoring stations, including DT1, GT2, CN1, NT1, and CN4 in Binh Duong province (Fig. S40). Statistical metrics (MB, MAGE, RMSE, NMB, NME, UPA, and the correlation coefficient R) were estimated for each monitoring sites, respectively, most of which met the recommended benchmarks corresponding for each statistical metric by US EPA (1991), Wang et al. (2015a), and Emery et al. (2017) and were detailed in Table S3.
Thus, from the results of the estimated statistical metrics, it could be seen that the simulation results from the CMAQ model had errors, which were mainly caused by the uncertainty in the emission data, meteorological data, boundary conditions, and chemical transport. On the other hand, based on the results from the calculated statistical metrics and the corresponding recommended benchmarks, it had also been reported that the CMAQ model was capable of simulating the spatio-temporal distribution of ground-level O 3 concentration in the study area.

Spatiotemporal distributions of ground-level ozone concentration
The .02 µg/m 3 ; in the fifth zone (CZ), the concentration ranged from 42.65 to 50.22 µg/m 3 ; and in the sixth zone (CPZ), this value ranged from 42.85 to 52.13 µg/m 3 . Simultaneously, on the basis of the simulation results, the highest annual mean ground-level O 3 concentrations were in the fifth zone (CZ) and sixth zone (CPZ), with average concentrations of 46.11 µg/m 3 and 46.41 µg/m 3 , respectively. The third zone (LXZ) had an average concentration of 44.70 µg/m 3 (1.04 times lower than that of CPZ); followed by zones 4 (HMZ) and 5 (TBZ) with average concentrations of 43.35 µg/m 3 and 43.19 µg/m 3 , respectively (1.07 and 1.08 times lower than that if CPZ, respectively); and zones 1 (FAZ) and 2 (PRZ) had the lowest average concentrations across the entire MDR with 42.70 µg/m 3 and 41.83 µg/m 3 , respectively (1.09 and 1.11 times lower than that of CPZ, respectively). In each zone, the average ground-level O 3 concentration tends to increase gradually from the inner delta to the coastal areas of the east and west coasts (especially the east coast and the southernmost at the tip of the South China Sea of Ca Mau province), from north to south, typically localities with concentrations > 48.5 µg/m 3 concentrated mainly in parts of Tien Giang, Ben Tre, Tra Vinh, Soc Trang, and Bac Lieu provinces (about the east, southeast, and south) of zone 6 (CZ); the eastern and south-eastern parts of Ca Mau and Bac Lieu provinces in zone 7 (CPZ); and the western and southwestern parts of Kien Giang province in zone 1 FAZ (west coastal area). The map of ground-level O 3 concentration by year and month is shown in Figs. 4 and 5, respectively.

Impacts of meteorological variables on ground-level O 3
The outcomes in Fig. 6 show the relationship between ground-level O 3 concentrations and meteorological variables (case [i]), with the normalised regression coefficients (β k ) calculated based on the original regression coefficients (β k * ), as in Eq. (22), corresponding to each meteorological variable (as shown in Fig. 6-a and -b). In the six divided zones of the MDR, the coefficients of determination (or the R correlation coefficients) for the MLR models ranged from 0.319 in LXZ to 0.436 in CZ, while the determination coefficients of the other zones were as follows: TBZ < CPZ < FAZ < PRZ with 0.365 < 0.368 < 0.417 < 0.425, respectively (Table S4). From the MLR models, it was found that all meteorological variables of T, RH, and P were positively correlated across the entire MDR (β k > 0, Fig. 6-a) and contributed to the increase in ground-level O 3 concentration with the contribution explained as LXZ < TBZ < CPZ < FAZ < PRZ < CZ with explained values of 10.16% < 13.34% < 13.55% < 17. 39% < 18.09% < 19.04%, respectively. When separating the effects of RH and P for the sensitivity analysis between T and surface O 3 , it was performed by increasing T by 1 K and the RH and P variables remained unchanged. The results showed that the ground-level O 3 concentrations in the FAZ, PRZ, LXZ, TBZ, CZ, and CPZ increased by 3.041 µg/m 3 , 3.278 µg/m 3 , 3.249 µg/m 3 , 2.716 µg/m 3 , 4.208 µg/m 3 , and 4.849 µg/m 3 , respectively. Similar to T, the sensitivities of the RH and P variables to ground-level O 3 were also calculated. The results also reflected the increase in O 3 concentration in the FAZ, PRZ, LXZ, TBZ, CZ, and CPZ; however, the sensitivity between P and the ground-level O 3 was markedly greater than that of RH. When the variable P increased by 1 mbar, the values of ground-level O 3 concentration increased by 7.404 µg/m 3 , 7.193 µg/m 3 , 5.987 µg/ m 3 , 6.647 µg/m 3 , 8.675 µg/m 3 , and 7.712 µg/m 3 , respectively, compared to 0.648 µg/m 3 , 0.489 µg/m 3 , 0.661 µg/m 3 , 0.883 µg/m 3 , 0.632 µg/m 3 , and 0.779 µg/m 3 , respectively, when the RH variable was increased by 1%, and the T and P variables remained unchanged.
Hot climates, high temperatures in the summer, and reduced monsoonal winds to ventilate inland areas with marine air were considered to be the main meteorological drivers causing the increase in ground-level O 3 concentrations in the troposphere, as demonstrated by Li et al. (2020). In addition, in the chemical transport model sensitivity study by Dawson et al. (2007), high temperatures commonly caused a negative effect on the total PM 2.5 concentration during the summer because of the high-temperature volatilisation of ammonium nitrate; simultaneously,  Li et al. (2019a, b) showed that decreased PM 2.5 concentrations were also a major driver of the increase in groundlevel O 3 concentration because of the role of PM 2.5 as a scavenger of hydroperoxy (HO 2 ) radicals and NO x , which are factors involved in the photochemical reactions forming surface O 3 .
Alternatively, simple linear regression models of the ground-level O 3 concentration with each meteorological variable (T, RH, and P) were performed to evaluate the differences compared with the MLR models (as shown in Fig. 6-b  and -c). The results showed that P had a positive correlation with ground-level O 3 in the six zones; for RH, the positive correlation occurred only in three zones, namely LXZ, TBZ, and CPZ. A negative correlation was also observed across the entire MDR for T (Table S5). The R correlation coefficients between P and ground-level O 3 were the best, with a range varying from 0.286 (in LXZ) to 0.410 (in CZ). In all zones, RH and P were statistically significant (p < 0.05), and T was only statistically significant in PRZ, CZ, and CPZ. This also reflected that there were at least two meteorological factors that were independent parameters for predicting ground-level O 3 concentrations in the MDR.
Thus, in general, it could be concluded that low T, RH, and P were unfavourable meteorological conditions for the formation of surface O 3 pollution in the study area. According to the results of the correlation analysis, P was the most significant meteorological factor for ground-level O 3 in the six zones, whereas the RH meteorological factor was mainly in FAZ, LXZ, TBZ, and CPZ, and the T meteorological factor was only in CZ and CPZ.

Impacts of precursor emission variables on ground-level O 3
The effects of ground-level O 3 precursor emission variables were examined using MLRs and simple linear regression models. The results in Fig. 7 show the relationship between the surface O 3 concentration and five precursor emission variables of CH 4 , CO, NO x , NMVOCs, and other VOCs (case [ii]), with the original regression coefficients (β k * ), as in Eq. (11), corresponding to each variable of O 3 precursor emissions (as described in Fig. 7-a and -b) and the calculated R correlation coefficients of the simple linear regression models (as shown in Fig. 7-c). In general, the R correlation coefficients for the MDR of the MLR models ranged from 0.258 (in CPZ) to 0.541 (in PRZ), and the R correlation coefficient levels in the six zones were in the order of CPZ < FAZ < TBZ < LXZ < CZ < PRZ, with values  (Table S6), respectively. When analysing the separate sensitivity of each precursor emission, for the NO x precursor when increasing the emission value by 1 kg.day −1 and keeping the remaining precursor variables unchanged, the results showed that the ground-level O 3 concentrations in PRZ, LXZ, TBZ, and CZ increased by 15.773 µg/m 3 , 4.237 µg/m 3 , 0.851 µg/m 3 , and 0.597 µg/m 3 , respectively; while in FAZ and CPZ, the surface O 3 concentrations decreased by 0.021 µg/m 3 and 0.025 µg/m 3 , respectively. The results of the sensitivity analysis also tended to be similar for the precursor emission variables CH 4 and VOCs; particularly, when increasing the CH 4 emissions by 1 kg.day −1 , the ground-level O 3 concentration increased in PRZ, LXZ, TBZ, CZ, FAZ, and CPZ with values of 0.716 µg/m 3 , 0.287 µg/m 3 , 0.058 µg/m 3 , 0.071 µg/m 3 , 0.003 µg/m 3 , and 0.029 µg/m 3 , respectively; compared with increasing the VOC emissions by 1 kg.day −1 , which resulted in surface O 3 concentrations of 9.253 µg/m 3 , 4.075 µg/m 3 , 0.501 µg/m 3 , 0.575 µg/m 3 , 0.1681 µg/m 3 , and 1.501 µg/m 3 , respectively. A similar analysis of the sensitivity of the CO precursor emission with surface O 3 , when increasing the emission value by 1 kg.day −1 and keeping the remaining precursor variables unchanged, showed decreasing ground-level O 3 concentrations in the PRZ, LXZ, TBZ, CZ, FAZ, and CPZ with values of 0.756 µg/m 3 , 0.247 µg/m 3 , 0.051 µg/m 3 , 0.052 µg/m 3 , 0.001 µg/m 3 , and 0.014 µg/m 3 , respectively. From the MLR models, it can be seen that most of the precursor emission variables of CH 4 , NO x (except CPZ and FAZ), and VOCs (except FAZ) were positively correlated with the surface O 3 concentrations in the study area (β k > 0, Fig. 7-a), whereas the CO precursor emission variable alone was negatively correlated for all zones in the MDR (β k < 0, Fig. 7-a). Emission precursors such as CH 4 , NO x , and VOCs contributed to the increase in ground-level O 3 concentrations, with the levels of contribution explained as CPZ < FA Z < TBZ < LXZ < CZ < PRZ, with values of 6.64% < 7.39% < 7.44% < 12.84% < 14.73% < 29.26%, respectively.
Additionally, an analysis based on simple linear regression models of ground-level O 3 concentrations with each O 3 precursor emission variable was performed to independently evaluate the predictive precursor emission factors for ground-level O 3 formation (Figs. 7-b and 8-c). The results showed that the precursors of CH 4 , CO, NO x , and NMVOCs were statistically significant independent factors (p < 0.05) (Table S7) and had a remarkable impact on the ground-level O 3 concentration levels in the MDR, whereas the precursors Overall, it could be concluded that the low emissions of CH 4 , NMVOCs, and NO x , and the high emissions of CO were the unfavourable precursor emission conditions for surface O 3 pollution formation in this study area. In addition, according to the results of the correlation analysis, precursor emissions of NO x and NMVOCs were among the most remarkable emission factors for surface O 3 concentrations in the PRZ, LXZ, TBZ, and CZ, whereas the CH 4 emission factor was mainly in the PRZ and LXZ, and the CO emission factor was only in the PRZ, LXZ, and TBZ.

Impacts of both meteorological and precursor emission variables on surface O 3
A multivariable experimental linear regression model was built to evaluate the overall relationship between O 3 concentration (C O3 , y) and meteorological factors, including temperature value x 1 (T, K), relative humidity x 2 (RH, %), and surface pressure x 3 (P, mbar), and the types of CH 4 precursor emissions were E CH4 , x 4 ; CO was E CO , x 5 ; NO x was E NOx , x 6 ; NMVOCs were E NMVOCs , x 7 ; and other VOCs were EOther-VOCs , x 8 . The results are shown in Table 5, along with Eqs. (24)-(29). Table 5 shows that the common correlation coefficients at zones 1-6 were R 1 = 0.47549, R 2 = 0.61360, R 3 = 0.53909, Fig. 8 Emission distribution of two precursors NO x and NMVOCs with major impacts on ground-level ozone pollution in the MDR Table 5 Multivariable correlation equation of ground-level O 3 with meteorology and precursors (C O3 = F (T, RH, P, E CH4 , E CO, E NOx , E NMVOCs , E Other-VOCs ))

Uncertainty analysis
These limitations could lead to uncertainty in the research results and create errors in the simulation evaluation of ground-level O 3 concentration distribution. This could be explained in detail, first, by the limitation in the ground-level O 3 concentration monitoring data and the field data measured at meteorological observation stations in the MDR. For ground-level O 3 concentration measurement data in the entire MDR, there was no monitoring station to measure the ground-level O 3 concentration (it only monitored some basic pollution indicators, such as TSP, SO 2 , NO x , CO, and PM 10 ). The dataset in this study was collected from monitoring stations in the vicinity of Binh Duong Province, which was also an area within simulation domain D02 covering the entire study area. The ground-level O 3 concentration measurement dataset was created by the observation method, which was performed manually and measured only four times at 9:00, 11:00, 13:00, and 15:00 on an important day unique monitoring at each station from January, 2018, to December, 2018. Second, the contribution of the emission of two main precursors of NO x and NMVOCs in each group of industries/ activities was different among zones in the Mekong Delta, especially in zones 1 (FAZ), 2 (PRZ), 5 (TBZ), 6 (CZ), and 7 (CPZ), with a of 520-1530 thousand hectares compared to zone 3 (LXZ), and zone 4 (HMZ) with an area of only 95-225 thousand hectares. Significant differences in socioeconomic activities, management levels of each locality, and the difference in soil problems by geographical zone (between freshwater alluvial soils and acid and saline soils) led to an assessment of six zones (not considering the four HMZ zones), which may not fully reflect the level of detail according to the administrative management level in each zone. Moreover, the 2018 emission inventory dataset built for the entire region (range of calculation D02) had a low resolution and a relatively large grid size (approximately 9.5 km) and natural emissions from forest fires in zones 5 (TBZ) and 7 (CPZ) were not considered due to lack of detailed inventory data during the simulation.
Third, the correlations were limited to evaluation in the form of a multivariable linear function between O 3 concentration and meteorological factors; the main precursor emissions were NO x and NMVOCs. In this study, correlation of some forms of VOC precursor components according to the chemical mechanism of the CMAQ model was not performed. Moreover, the three selected meteorological factors, including temperature, relative air humidity, and surface pressure, cannot fully account for the impact of meteorology on the formation of O 3 . As a result, the correlation level is not high enough (R only reaches a maximum of 0.61 in the LXZ).
Fourth, in the MLR multivariable linear function models between total ground-level O 3 concentration and meteorological factors combined with emission precursors, the interaction between variables (combination of quadratic or ternary variables) increased the confidence level of the analytical model.

Discussion
The seasonal variation of ground-level O 3 was noted in many regions of the world (Simon et al. 2012;Tang et al. 2013;Maji et al. 2019) in MDR, the change in ground-level O 3 concentration is quite complicated according to the season; in the dry season, the O 3 concentration is generally higher than that in the wet season with O 3 values. In the month with highest concentration of the dry season, February, the 1-h average concentration ranged from 17.2 to 102.1 µg/ m 3 , which was still significantly lower than the allowable limit of NAAQS (QCVN 05:2013/BTNMT-hourly average with 200 μg/m 3 ), but different effects on crops should be considered. During the wet season, the highest O 3 value in the transitional period at the end of the season (October) ranged from 8.6 to 117.4 µg/m 3 . Considering the monthly average, the O 3 concentration was high (> 60 µg/m 3 ), occurring in February, June, October, and December. In January, March-May, and July-September, the concentration level was much lower than in August (wet season), with the lowest concentration in the range of 16.7-25.1 µg/m 3 . October (end of the wet season) had the highest monthly average concentration, ranging from 55.4 to 84.5 µg/m 3 . The spatial distribution trend of terrestrial O 3 concentration shows that there is a clear and gradual shift from the inner delta to the eastern coastal areas in zone 6 (CZ), Cape Ca Mau in zone 7 (CPZ), and the western coastal areas of zone 1 (FAZ) in Kien Giang Province.
The role of precursors in the distribution of ozone concentrations has been discussed in similar studies (Xue et al. 2014;Sharma et al. 2016;Han et al. 2019;Chen et al. 2021;Mitchell et al. 2021;Mousavinezhad et al. 2021;). The modelling results showed that the two precursors of NO x and NMVOCs were the main contributors to ground-level O 3 pollution. These two precursors originate from anthropogenic activities as well as natural biological sources, with the largest contribution coming from zone one (FAZ) with 91.5 thousand tons of NO x /year and 455.2 thousand tons of NMVOCs/year, followed by zone 2 (PRZ), zone 7 (CPZ), zone 5 (TBZ), zone 6 CZ, and zone 3 (LXZ); the lowest was in zone 4 (HMZ) (Fig. 8). In particular, the NMVOC precursors in each zone had a significant impact and were much more sensitive than the NO x precursor emissions with the NMVOCs/NO x ratio in zone 1 (FAZ) of 4.97, zone 2 (PRZ) was 8.03 times, zone 3 (LXZ) was 7.84 times, zone 4 (HMZ) was 6.82 times, zone 5 (TBZ) was 6.55 times, zone 6 (CZ) was 5.50 times, and zone 7 (CPZ) was 5.55 times. The pollution contribution was predominantly from anthropogenic activities with three main groups of industries/activities in the following order: (1) industrial production activities, (2) activities from residential zones and people's livelihood, and (3) agricultural production activities such as rice and crop cultivation, land used, livestock, etc.
Meteorology is one of the factors that directly affect the distribution of ground-level ozone concentration (Camalier et al. 2007;Davis et al. 2011;Godowitch et al. 2015;Mousavinezhad et al. 2021). The results showed that relative humidity (RH) and surface pressure (P) were the two main factors contributing to the increase in ground-level O 3 concentration, specifically, the increase in ground-level O 3 concentration from 5.9 to 8.7 µg/m 3 /Pa with each pressure unit increase (when temperature and relative humidity did not change), respectively, for zone 3 (LXZ), zone 5 (TBZ), zone 2 (PRZ), zone 1 (FAZ), zone 7 (CPZ), and zone 6 (CZ); and from 0.5 to 0.9 µg/m 3 /% with each unit of humidity increase (when surface temperature and pressure remain unchanged) for zones 2 (PRZ), zone 6 (CZ), zone 1 (FAZ), zone 3 (LXZ), zone 7 (CPZ), and zone 5 (TBZ), respectively. Moreover, the correlations between the ground-level O 3 concentration values and NO x and NMVOCs precursor emissions were statistically significant (p < 0.05), showing that they contributed to the increase in ground-level O 3 concentration in zones 2 (PRZ), 3 (LXZ), 5 (TBZ), and 6 (CZ), whereas this is the opposite in zones 1 (FAZ) and 7 (CPZ). At the same time, when evaluating the correlations between the ground-level O 3 concentration and meteorological factors combined with precursor emissions, the correlation index (R) gives results that are at a satisfactory level with values fluctuating from 0.42 to 0.61 (R > 0.4).

Conclusion
The MDR has a strategic location in Vietnam, bordering the East Sea and the Southwest Sea, the exclusive economic zone on the east, west, and south. It has a very favourable environment for growing rice, an important food crop globally. Rapid industrialisation in the MDR has increased the trend of ground-level zone concentrations, leading to a decline in crop yields. The distribution of ozone concentration for a year was modelled to determine the dependence of ozone concentration on emission factors and meteorology. Using a combination of WRF/CMAQ and multiple linear regression models, the following results were obtained. First, the ground-level O 3 concentration during the dry season was higher than that in the wet season; the month with the highest concentration reaches an average concentration of 1 h, ranging from 17.2 to 102.1 µg/ m 3 . In the wet season, the O 3 value was highest at the end of October, reaching 8.6-117.4 µg/m 3 . Within the Mekong Delta, ground-level ozone concentrations ranged from 40.39 to 52.13 µg/m 3 . Second, the MDR was divided into six agro-ecological zones corresponding to geographical features, agriculture, and water resources: the FAZ, PRZ, LXZ, TBZ, CZ, and CPZ. The two precursors of NO x emissions and NMVOCs are major contributors to terrestrial O 3 pollution. The largest contribution of emissions comes from FAZ with 91.5 thousand tons of NO x /year and 455.2 thousand tons of NMVOCs/year. Next, is zone 2 (PRZ), zone 7 (CPZ), zone 5 (TBZ), zone 6 (CZ), and the lowest was zone 4 (mountainous). Third, in each zone, the NMVOC precursors have a significant impact and are much more sensitive than the NO x precursor emissions with the NMVOCs/NOx ratio in the first zone (FAZ) of 4.97, respectively. Zone 2 (PRZ) is 8.03 times, zone 3 (LXZ) is 7.84 times, zone 4 (HMZ) is 6.82 times, zone 5 (TBZ) is 6.55 times, zone 6 (CZ) is 5.50 times, and zone 7 (CPZ) is 5.55 times. Anthropogenic activities were the main sources of pollution, from three main groups of industries/activities in the following order: (1) industrial production activities, (2) residential zones and people's livelihood activities, and (3) from agricultural production activities such as rice and crop cultivation, using land, livestock, etc. Fourth, in the sixth zone, the emissions of precursors E CH4 , E CO , E NOx , and E NMVOCs were independent factors mainly affecting the ground-level O 3 concentration in these six zones. Other-VOC precursor emissions had little or no influence on the O 3 concentration values in each area of the MDR. Fifth, humidity (RH) and surface pressure (P) were the two main meteorological factors that contributed to the increase in O 3 concentration, among the three meteorological factors considered in this study, the humidity (RH), surface pressure (P), and temperature (T).