Introduction

Urban air pollution is one of the crucial factors affecting public health for city residents1. As traffic-induced emissions are the dominant contributor to poor air quality in urban areas2,3, exposure to traffic-related air pollutants (TRAP) is linked to increased morbidity and mortality4. The density and morphology of the built environment affect airflow, leading to the accumulation of TRAP in near-road areas5,6. Urban green spaces can regulate air quality7 and provide other environmental quality benefits, such as attenuating noise pollution8, reducing the urban heat island effect9, and supporting biodiversity10. City parks provide a space to support citizens’ physical and mental health by offering infrastructure like sports facilities and seating11, and promoting subjective well-being12,13. Small urban parks (≤ 5 hectares) can be fragmented in cities and are commonly found near busy streets for accessibility by local inhabitants14. Due to their locations, elevated TRAP may put potential users of these green spaces at increased risk of air pollution exposure15.

Particulate Matter (PM) is a critical indicator of local air quality16,17. The World Health Organization (WHO) released new global air quality guidelines in 2021, recommending that annual air quality guideline levels of PM with aerodynamic diameters of ≤ 2.5 microns (PM2.5) and ≤ 10 microns (PM10) should not exceed 5 μg/m3 and 15 μg/m3, respectively18. In the European Union (EU), the Air Quality Directive (2008/50/EC) set limit values at 25 µg/m3 for PM2.5 and at 40 µg/m3 for PM1019, with new WHO guidelines promoting further reductions and an ongoing revision of these limits are underway20. Previous studies have demonstrated that the PM2.5/PM10 ratio offers crucial insights into the source identification of PM pollution21,22. Specifically, lower PM2.5/PM10 ratios often indicate a dominance of coarse particles from natural sources (e.g., dust and sand), while higher ratios are typically associated with fine particles from anthropogenic activities (e.g., combustion sources and industrial processes)23,24.

Urban morphology significantly influences air quality through the dynamic interplay between grey (urban development) and green (maturing parks) elements25,26. High-density urban development tends to trap pollutants, leading to deteriorated air quality due to restricted air circulation and heightened emissions from vehicles and buildings27. Conversely, integrating ample green spaces and maturing parks into urban areas can generally enhance air quality28. These green elements improve ventilation and facilitate pollutant dispersion, thereby mitigating the negative impacts of dense urban development.

Vegetation affects air quality by influencing both pollutant deposition and dispersion29,30. Numerous studies have demonstrated that vegetation can reduce airborne PM concentrations. Vegetation has been shown to remove PM through dry deposition on the leaf surfaces, thereby reducing ambient PM concentrations31,32,33. The magnitude of reduction in PM is influenced by factors such as leaf area index (LAI), leaf area density (LAD), vegetation height, and thickness, etc15,34,35. For instance, vegetation with extensive surface areas and complex structures is particularly efficient in trapping particulate matter32,36. Moreover, vegetation influences the dispersion of pollutants through its impact on local wind patterns. Trees and other plants can create turbulence that helps disperse pollutants, diluting their concentrations and spreading them over a larger area. Vegetative barriers like hedges and green walls act as windbreaks, reducing wind speed and promoting particle deposition, while also blocking air pollution from reaching people, such as street trees between traffic and pedestrian pathways37. A field investigation showed notable reductions in PM at an urban green space. PM10 was reduced by up to 2.9-fold in summer (2.4-fold in winter), and PM2.5 was reduced by up to 2.1-fold in summer (1.3-fold in winter)38.

However, vegetation may sometimes have adverse effects. Changes in turbulence and wind speed due to increased friction can inhibit dispersion and lead to trapped pollutants in some areas, resulting in deteriorated air quality11,39. Zheng et al. observed an accumulation of particulate pollutants on the sidewalk and bikeway near dense roadside vegetation, with particle concentrations significantly increasing by 0–35%40. Moreover, the emission of biogenic volatile organic compounds (BVOCs) from vegetation can also worsen air quality. BVOCs are emitted by numerous plant species and can react with nitrogen oxides (NOx) in sunlight to produce ground-level ozone (O3) and secondary organic aerosols (SOA)41. These chemical reactions generate PM2.5 and other pollutants, further deteriorating urban air quality.Various physical and environmental factors influence the spatiotemporal distribution of air pollutants in urban areas42,43,44,45. Microclimate models, such as ENVI-met, Ansys Fluent, OpenFOAM, and PALM, provide a means to analyse these interactions and evaluate the potential for passive controls, such as nature-based solutions (NBS), to improve air quality46,47,48. ENVI-met is widely used to assess urban greening effects on air quality, with studies showing its effectiveness in simulating vegetation impacts on pollutant deposition and dispersion11,32,49,50. For example, Xing and Brimblecombe examined the influence of different vegetation types and layouts to explore the optimum configuration to minimize TRAP exposure in near road areas11. Despite the sophistication of CFD models like Ansys Fluent, which offers detailed fluid dynamics simulations46, PALM, suitable for large-scale urban studies51, and OpenFOAM, known for its extensive customization and wide range of applications52, ENVI-met remains advantageous. ENVI-met is praised for its user-friendly interface, integrated vegetation modelling, and fine-scale modelling capabilities. However, it may have limitations in handling very large-scale scenarios compared to the other models. Additionally, methods like wind tunnel experiments and scale models provide valuable insights into air pollutant distribution through controlled physical simulations, helping to validate and support modelling results53,54. However, because of ENVI-met’s ability to accurately and efficiently simulate vegetation-pollutant interactions, it stands out as the most suitable tool for this study and is the preferred choice for detailed urban air quality assessments49.

CFD simulations offer controlled environments, cost efficiency, and flexibility, but their accuracy depends on the quality of the models applied and assumptions made in relation to inputs. On-site measurements provide real-world data but involve higher costs and temporal and spatial limitations. Consequently, it remains crucial to measure localized data to accurately support simulation results at a hyperlocal level20,55. Combining both methods enhance accuracy and reliability by leveraging the broad scope of simulations while validating with on-site data56.This case study uniquely investigates how environmental parameters in and around a historical city square park affects upon fine and coarse particulate matter (PM) concentrations predominantly from TRAP sources due to evolving green and grey urban infrastructure evolves over time. Both building development (grey infrastructure) and vegetation maturation (green infrastructure) were evaluated in this study to support a better understanding of how green and grey infrastructure interact in urban settings. The study conducted an initial air quality monitoring campaign to investigate the distribution and influencing factors of PM2.5 and PM10 in the park, Fitzwilliam Square Park in Dublin, Ireland, with this data also used to support model validation. An initial analysis of the spatiotemporal distribution of TRAP pollution provides insights into the monitoring campaign in the city square park. Simulations using the modelling software ENVI-met were then employed to examine the dynamic relationship between local air quality and the dimensions of vegetation around the perimeter of the park with surrounding buildings. The simulations examine the evolving nature of buildings and vegetation to reflect the dynamic morphology typical of an urban setting. The impact of these changes in the city square park quantifies their effects on local PM concentrations for park users and local inhabitants. It identifies which conditions will present the greatest deteriorations in local air quality. The findings of this study can provide valuable guidance for urban planning and development, whilst informing best-practice in the future design of urban parks to enhance the value of NBS for improving local air quality.

Methods

Case study

Dublin is a compact city in Ireland with dense buildings and limited land resources. Fragmented parks are scattered throughout the city, providing leisure space for nearby residents. Fitzwilliam Square Park (coordinates: lat. 53.33°N, long. 6.25°W) in Dublin was selected for this investigation (Fig. 1).

Figure 1
figure 1

Graphical representation of the basic information about the case study area and the installation of air quality monitors. (a) Location of case study area in Dublin; (b) Location of the case study area, local weather station, and background air quality monitoring station in Dublin city centre; (c) Illustration of the relative location of Fitzwilliam Square Park, two major roads, and the four fixed monitors inside the study area; (d) Photos showing the vegetation barriers around Fitzwilliam Square Park; (e) & (f) Installation of eLos air quality stations outside and inside the park, respectively.

Full size image

Fitzwilliam Square Park was chosen to represent a typical urban park with a significant amount of vegetation, surrounded by main city roads and multi-story buildings. The two main roads near the park are Fitzwilliam Street East on the east side and Fitzwilliam Street West on the west side. In addition, its regular shape and small size make it ideal for advanced computational modelling of air quality. The characteristics of Fitzwilliam Square Park and the surrounding neighborhoods are presented in Supplementary Material (S.M.) Table S1.

Data collection

Air quality monitoring campaign

Stationary monitoring started on 25th March 2022 and ended on 1st July 2022, since it’s a period of peak usage of parks and green spaces due to favourable weather. Four autonomous, solar-powered monitors (eLos air quality station, Elichens)57 were installed on posts inside and around the park. These monitors, using light-scattering, continuously measured the concentration of different size fractions of Particulate Matter (PM1, PM2.5, PM10) in 1-h intervals during the monitoring period. Two stations were fixed inside the park at agreed points on west and east side behind the vegetation barrier (I-W & I-E), and two stations were fixed on road-side lamp poles on Fitzwilliam Street East (O-E) and Fitzwilliam Street West (O-W) (Fig. 1c). The installation of monitors for inside and outside of the park is shown in Fig. 1e,f, respectively. All the stations were installed at the height of 2.5 m to avoid interference from pedestrians.

Meanwhile, the Rathmines air quality station in Wynnefield Road in the southern suburb of Dublin (53.32° N, 6.27° W) was selected as a pollutant background station, providing hourly concentrations of PM2.5 and PM10 during the monitoring periods. The location of Rathmines air quality station is shown in Fig. 1b.

Traffic data

Manual traffic survey was conducted on 5 Wednesdays from April to June (27 April, 11 May, 25 May, 8 June, and 15 June) to show the traffic condition on a typical weekday. All the vehicles passing both Fitzwilliam Street East and Fitzwilliam Street West during three time periods (9:00–10:00, 13:00–14:00, 17:00–18:00) were recorded and counted. The vehicles were divided into six categories when counted according to pollution emission intensity: Passenger cars (PC), Light duty vehicles (LDV), Heavy duty vehicles (HDV), Buses (B), Coaches (C), and Motorcycles (MC). The average manual traffic volume data of Fitzwilliam Street East and Fitzwilliam Street West obtained from the five-day traffic surveys was presented in Fig. S1.

Meteorological conditions

The meteorological data during the monitoring period, including air temperature, relative humidity, wind direction, and wind speed, were obtained from Weather Underground58. The data was collected from a nearby weather station (Merrion Hotel AWS-IDUBLI27, 53.34° N, 6.25° W), located 430 m north of Fitzwilliam Square Park (Fig. 1b).

Analysis of air quality data

Initial calibration of PM data

PM2.5 and PM10 were chosen as target pollutants to investigate their spatio-temporal distribution in and around the urban park because both PM2.5 and PM10 are target pollutants in WHO and EU guidelines. In addition, both PM2.5 and PM10 were collected in field monitoring and at the Rathmines background station, providing a comprehensive understanding of the combined effect of different factors on local pollutant concentrations.

To overcome any potential impact of drift in PM measurements, and to quantify correction factors if required, all fixed monitors were co-located together at the start and end of the monitoring campaign. The instruments were placed on the rooftop of the Civil Engineering Building in Trinity College Dublin for 9 days before the field measurements (Phase 1) and 15 days after that (Phase 2). Pearson correlation coefficients of hourly PM data between the four monitors and the background stations are listed in Table S2. The PM data measured by the four monitors exhibited a strong correlation with each other and between Phase 1 and Phase 2. Although the correlation among the monitors in Phase 2 was slightly weaker than in Phase 1, it remained within an acceptable range.

Calibration factors of PM concentrations for each monitor were obtained by calculating the deviation of its readings from the average of the four monitors. Respective calibration factors for April, May, and June were obtained through linear interpolation. The calibration factors, which range from − 0.04 to 0.28 for PM2.5 and from − 0.01 to 0.22 for PM10, are presented in Fig. S2.

Statistical analysis

Descriptive statistics, scatterplots, histograms, and boxplots were used to characterize distributions of PM2.5 and PM10 concentrations, traffic volume, and meteorological data (air temperature, relative humidity, wind speed and direction). Pearson correlation analysis and Stepwise Regression model were utilized to examine the relative contribution of each influence factor (background concentrations, traffic volume, and meteorological parameters) to local PM2.5 and PM10 concentrations.

Random forest model

The Random Forest model has the ability to evaluate feature importance, providing insights into which features are the most influential in predicting the target outcome3. In this study, the Random Forest model was used to evaluate the relative importance of each influence factor on local PM concentrations based on the average decrease in impurity (the percentage increase in mean square error, %IncMSE) that a feature contributes across all the trees in the forest.

Microclimate modelling

Model domain

ENVI-met 5.0 was used to simulate the dispersion characteristics of different traffic-related air pollutants from both Fitzwilliam Street East and Fitzwilliam Street West in our study area. The initial model was developed based on Google maps and on-site investigation. To enhance stability of the simulation and minimize boundary effects, 150 m extension of buildings and roads and 5 nesting grids were added on each lateral boundary, ensuring enough distance from the area of interest to the boundary, as shown in Fig. 2a. The model domain covers an area of 530 m × 458 m, of which the area of interest had dimensions of 230 m × 158 m. The size of the horizontal grid was set to 2 m by 2 m, taking into account the computational costs while achieving as accurate results as possible. Varying grid sizes were used in vertical direction. Equidistant grids of 1 m (dz) were used below 20 m, and a telescoping factor of 5% was applied for the space above 20 m, resulting in 89.76 m of total height with fifty vertical grids. As the highest building in the domain is 14 m, the total model height was defined as six times higher than the building height, adhering to best practice for built environment simulations48. At this grid size, each simulation runs for approximately 55 h. A finer grid resolution would result in excessive computational cost, and this resolution has been applied and validated in previous scale models in Dublin6. The vertical grid with a maximum of 1 m height was split into five sub-grids to improve the accuracy of the simulation of near-surface airflow.

Figure 2
figure 2

The geometric configuration of ENVI-met models. (a) 3D view of the model (14 m buildings & 7 m vegetation scenario) and the two wind directions input into the simulations. The grey blocks indicate the buildings in the study area, the green blocks represent vegetation in the park area, the red areas indicate the emission sources of PM2.5. (b) Location of the seven PM2.5 receptors set in the ENVI-met model and the four buildings (referred to as E, S, W, and N in this study) around the park.

Full size image

Model inputs

Simple forced meteorological boundary conditions were employed in ENVI-met simulations. The meteorological input parameters (air temperature, relative humidity, wind speed, and wind direction) were obtained from the Merrion Hotel AWS weather station. For the simulations with varying urban morphology, two wind directions were selected: one parallel to the road (~ 210°), which is the predominant wind for the area, and the other perpendicular to the road (~ 300°). Other meteorological parameters were set to the average values observed during the monitoring period.

Traffic emissions were employed as line sources at a height of 0.15 m along the two roads, Fitzwilliam Street East and Fitzwilliam Street West, and the number of vehicles in the five categories from the manual traffic survey was averaged over 1-h periods. The emission factors for each vehicle category were derived from the EMEP/EEA air pollutant emission inventory guidebook 2019, using the traffic emission factors (g/km) for both Euro 5 and Euro 6 vehicle standards59. Among the vehicles, 54% PC and LDV, 75% HDV, and 73% Bus/Coach were calculated as Euro-6 emission standard, while the others were Euro 5 standard60.

Seven points located along the east–west axis of the park were selected as receptor sites, including the positions of four fixed monitoring stations (O-W1, I-W, I-E, and O-E1), footpaths on both sides of the park (O-W2, O-E2), and the central point within the park (I-M), to get the values of PM2.5 concentrations at these points (Fig. 2b). The configuration and parameter values of the ENVI-met models are listed in Table S3.

Model validation

The manual traffic surveys took place during the fixed monitoring period to facilitate the model validation process. Therefore, the ENVI-met model was validated using data from six time periods over two days, 27 April 2022 and 8 June 2022. The locations of the four fixed monitors were selected for model validation by comparing the modelled and measured PM2.5 concentrations at the four places. The PM2.5 emission rates on both roads were calculated from the average traffic volume obtained from the manual traffic survey at each time period. Real-time meteorological parameters from Merrion Hotel AWS weather station and the background of air pollutants from the air quality station Rathmines were applied in the model input.

Models with changing urban morphology

To investigate the influence of urban morphology on the air quality in urban microenvironment, the distribution of PM2.5 concentrations around Fitzwilliam Square Park was simulated using ENVI-met under scenarios with different vegetation heights and building heights. The wind direction was set as perpendicular and parallel wind to the main roads.

Since the vegetation surrounding the park is composed of multiple layers of trees, hedges, and shrubs forming a complex vegetation barrier, this diversity in the model inputs was simplified. The vegetation was represented as block vegetation using simple-tree models with a higher Leaf Area Density (LAD) at the bottom and a lower LAD at the top. The specific LAD data are provided in Table S3. The heights of vegetation used in the models were set to 7 m in the initial model, then it was changed to 0 m (no vegetation), 14 m, and 21 m to evaluate the impact of the vegetation height on local air quality.

The buildings around the park were named West (W), North (N), East (E), and South (S) according to their locations. All buildings had an initial height of 14 m (Fig. 2a). The heights of these buildings were then increased or decreased by 50% (21 m & 7 m) and input into the model.

Result and discussion

Spatio-temporal distribution of PM

The four fixed stations inside and outside the park performed well from April to June and collected about 33,000 readings of PM2.5 and PM10 mass concentrations. Figure 3 presents the mean PM concentrations recorded at the four fixed monitor locations during the campaign. Despite lower traffic volumes observed on Fitzwilliam Street West, the highest PM2.5 and PM10 concentrations were observed at the O-W fixed air quality station on this road. The average monthly PM2.5 and PM10 concentrations collected were 15.0–18.8% (6.64–7.81 μg/m3 in April, 2.46–2.90 μg/m3 in May, and 2.18–2.69 μg/m3 in June) and 8.9–14.9% (9.46–11.11 μg/m3 in April, 6.23–7.32 μg/m3 in May, 6.73–7.39 μg/m3 in June) higher respectively, on Fitzwilliam Street West (O-W) compared to Fitzwilliam Street East (O-E). For the two stations inside the park, the monthly concentrations of PM2.5 collected by the station at the west side of the park (I-W) were up to 7.8% (2.00–2.17 μg/m3) higher than the one at the east side of the park (I-E), while PM10 concentrations were found up to 8.1% (6.22–6.77 μg/m3) higher on the west side (I-W).

Figure 3
figure 3

The distribution of the measured (a) PM2.5 and (b) PM10 concentrations in and out of the park from the four fixed monitors during the monitoring period.

Full size image

Figure 4 shows the diurnal variations of calibrated PM2.5 and PM10 concentrations for different months in the study area from the four fixed stations (O-W, I-W, I-E, and O-E). The diurnal variations of PM2.5 concentrations presented a bimodal distribution pattern (Fig. 4a–c). The first peak in PM2.5 occurred between 04:00 and 08:00 (9.95 μg/m3 in April, 3.78 μg/m3 in May, and 3.17 μg/m3 in June), then it reduced and fluctuated until 17:00 (4.16 μg/m3 in April, 1.45 μg/m3 in May, and 1.04 μg/m3 in June). A second peak occurred between 20:00 and 00:00 (1.27 μg/m3 in April, 3.84 μg/m3 in May, and 3.24 μg/m3 in June). This pattern aligns with the diurnal cycle of PM2.5 observed in previous studies43,61. The increase in anthropogenic activities (e.g., vehicle exhaust emissions) and more intensive photochemical reactions in the morning lead to an increase in PM2.5 concentrations. Afterward, upward airflow attributed to the rising atmospheric boundary layer (ABL) and increased atmospheric instability promoted the upward dispersion of air pollutants, leading to a rapid decrease in PM2.5 concentrations in the afternoon42. In the evening, the stable boundary layer, increasing vehicle exhaust emissions and human activity (e.g., restaurants and bars) cause the gradual increase of PM2.5 concentrations61.

Figure 4
figure 4

Diurnal variations of PM concentrations in the study area from the four fixed monitoring stations (O-W, I-W, I-E, and O-E) at Fitzwilliam Square Park in April, May and June 2022. (a–c) PM2.5 concentrations, (d–f) PM10 concentrations.

Full size image

In contrast, the diurnal variations of PM10 concentrations presented significant unimodal distribution patterns (Fig. 4d–f). The PM10 concentration in the morning presented a decreasing trend and reached its lowest point at around 05:00 (6.67 μg/m3 in April, 3.84 μg/m3 in May, and 2.95 μg/m3 in June), then rapidly increased to its peak at around 12:00 (12.58 μg/m3 in April, 8.83 μg/m3 in May, and 9.48 μg/m3 in June). After that, the PM10 concentration decreased gradually. Similar to the study conducted in Korea, the peak of PM10 occurred with a 2-h delay after the peak hour45. Compared to PM2.5 concentrations, which are substantially influenced by atmospheric diffusion conditions and chemical transformations, coarser particles like PM10 exhibit shorter atmospheric residence times due to their inherent physical properties. Consequently, concentrations of PM10 are predominantly ascribed to local direct sources (e.g., urban road dust and sand), and generally remain elevated throughout the daylight hours, decrease during the night, and reach a minimum in the early morning24. It is notable that both PM2.5 and PM10 concentrations produced a peak at midnight during April, which may be linked to ongoing home heating at that time, causing higher emissions from fossil fuel combustion at night.

Figure 5 shows the overall diurnal changes of PM2.5, PM10 and the corresponding PM2.5/PM10 ratios during the monitoring period. The hourly mean PM2.5/PM10 ratios ranged from 0.28 to 0.66 during the day, with an average value of 0.45. In Europe, PM2.5/PM10 ratios have been observed to range from 0.39 to 0.74 in cities with high urbanization, which is consistent with the results of this study17. The ratios demonstrated an increase during the morning hours (prior to 6:00) and in the evening hours (after 18:00), with lower concentrations observed during mid-day. The peak PM2.5/PM10 ratio (Mean = 0.66, SD = 0.36) values was observed at 5:00, whereas the lowest value (Mean = 0.28, SD = 0.32) was typically observed at 17:00. Similar diurnal patterns of PM2.5/PM10 ratios were found in studies conducted in the UK and China22,43.

Figure 5
figure 5

Diurnal variations of mean PM2.5 and PM10 concentrations and their ratio for the whole monitoring period.

Full size image

PM concentrations and corresponding influencing factors

Comparison of the influencing factors

This study employed three approaches including the Pearson correlation coefficient, Stepwise Regression, and Random Forest to explore the coupling effects of background concentrations, traffic volume, and meteorological parameters on ambient PM concentrations.

Table S4 presents the Pearson coefficients between PM2.5, PM10 concentrations, and their influencing factors. For PM concentrations, the most significant positive correlations were obtained with background concentrations, as background concentrations contribute significantly to local air quality42. In this case, PM2.5 and PM10 were strongly correlated (r = 0.825, p < 0.01) since the emission sources of PM are to some extent similar at this time of year i.e., vehicle exhausted sources, fuel burning and re-suspended road dust. Air temperature shows negative correlations with PM2.5 concentrations but positive correlations with PM10, which is consistent with results from Kassomenos et al. in winter conditions16. The Pearson correlation result shows negative correlations between PM10 and humidity while positive correlations between PM2.5 and humidity. This is due to increased PM deposition velocities attributed to high relative humidity, which is more significant for larger particles, while high relative humidity promotes the formation of secondary aerosols from gaseous precursors (e.g., SO2), found in PM2.5. A positive correlation between pressure and PM concentrations was found due to increased vertical air currents at ground level during low-pressure weather conditions, resulting in higher winds which accelerate the dispersion of pollutants44.

Stepwise Regression and machine learning model Random Forest were conducted to disaggregate the individual contributions of background concentrations, traffic volume, and meteorological parameters to local PM concentrations. The results of Stepwise Regression showed that PM concentrations were primarily affected by background concentrations, wind speed and traffic volume (Table S5). The contribution of background concentration to the measured PM2.5 and PM10 concentrations was 80.9% and 68.2%, respectively, making it the most dominant factor, consistent with the results of correlation analysis. In addition, Stepwise Regression indicated that surface pressure had a negligible effect on both PM2.5 and PM10 concentrations.

Random Forest was introduced to address the non-linear relationship between PM concentrations and other influential factors, and to further explore the contributions of background concentrations, local traffic volume, and meteorological parameters to local PM2.5 concentrations. The R2 value of the Random Forest model on training and test sets for PM2.5 and PM10 was 88.96% and 83.25%, respectively. The results demonstrated the reliability of the Random Forest model in understanding the influence of different factors on local PM2.5 concentrations.

As presented in Fig. 6, Random Forest model used the percentage increase in mean square error (%IncMSE) to quantify the relative importance of the background concentrations, traffic volume, and meteorological parameters to local PM concentrations. The results further confirmed the findings from stepwise regression that background concentrations were the most dominant factor in determining local PM concentrations, accounting for 93.2% and 90.4% of local PM2.5 and PM10, respectively. Following it were relative humidity and pressure, which contributed 46.4% and 44.6% to the PM2.5 concentration and 64.4% and 53.8% to the PM10 concentration, respectively. Different from the stepwise regression, the results of Random Forest indicated a low contribution rate of wind speed to local PM concentrations, explaining only about 28.4% of PM2.5 and 31.9% of PM10.

Figure 6
figure 6

The relative importance of influence factors (meteorological parameter, traffic volume and background concentration) to measured PM2.5 and PM10 concentrations from Random Forest model.

Full size image

Background concentrations were identified as the most dominant factor by all three approaches—Pearson correlation coefficient, Stepwise Regression, and Random Forest—consistent with previous research findings4,42,62. Wind speed ranked as the second contributor by Pearson correlation coefficient and Stepwise Regression, but showed the lowest importance by Random Forest. Meanwhile, Pearson correlation and Random Forest indicated a high contribution of pressure, which ranked last in Stepwise Regression. The differences observed among the three approaches may be attributed to the fact that both Pearson correlation and Stepwise Regression assume a linear relationship between the dependent and independent variables when assessing variable importance63. Consequently, these models may exhibit bias and inaccuracy if non-linear relationships exist among the variables. In contrast, the Random Forest model can capture the complex interplay of multiple influencing factors on local PM2.5 concentrations, resulting in a more accurate and informative representation of their contributions.

Evaluating an evolving urban morphology on local air quality

Model validation

The PM2.5 and PM10 concentrations from the four fixed stations were collected during the period of manual traffic counts (morning peak, off-peak, and evening peak) on 27 April and 8 June. The modelled PM concentrations at the four locations fitted well with the fixed monitored data, with coefficient of determination (R2) of Pearson’s correlation of 0.93 and 0.80 for PM2.5 and PM10, respectively (Fig. S3). This study selected PM2.5 as the focus for ENVI-met modelling to investigate the impact of urban landscape on local air quality.

Changes in vegetation height on local PM2.5

Figure 7 presents the differences in ENVI-met results of PM2.5 concentrations between the no-vegetation scenario (No-VEG) and various vegetation heights of 7, 14 and 21 m, under the wind direction separately parallel and perpendicular to the main roads. The modelled distribution of PM2.5 concentrations, as well as the vertical plane visualization can be found in Fig. S4.

Figure 7
figure 7

Comparison of the modelled distribution of PM2.5 concentrations at the seven receptors along the central axis of the park with different vegetation heights (7 m, 14 m, 21 m) to the no-vegetation model. The models were run under two wind directions: (a) perpendicular and (b) parallel to the main roads. The height at which data was collected is 1.5 m.

Full size image

For a wind direction perpendicular to the roads, particularly with the west wind, the existing vegetation contributed to an increase in PM2.5 concentrations (7.4–14.4%) on the leeward road (West) outside the park, compared to the no-vegetation scenario. Conversely, this led to a decrease of 65.4–74.6% on the windward road (East), with the lowest vegetation (7 m in this case) presenting the best air quality conditions. With the perpendicular wind direction, the lowest vegetation (7 m in this case) most effectively improved air quality in the study area, with a reduction of 30.4% of the average PM2.5 concentrations (0.17–0.24 μg/m3).

For the wind direction parallel to the roads, the existing vegetation contributed to a decrease in average PM2.5 concentrations (1.52–10.45%) in the study areas. Specifically, the highest vegetation (21 m in this case) presented the greatest improvement in air quality over the study area in parallel wind conditions, decreasing average PM2.5 concentrations by 10.5% (0.19–0.21 μg/m3). For the footpaths on both roads, increased PM2.5 concentration occurred on the footpath near the buildings as the vegetation height increased. Meanwhile, the concentrations of PM2.5 on the park-side footpaths decreased with the height of the vegetation, as well as at locations within the park.

Previous studies showed that vegetation improves air quality more effectively when the wind direction is perpendicular to the road rather than parallel, consistent with the findings of this study6. Li et al. found that the optimal vegetation barrier heights for street canyons under perpendicular wind conditions are 1.1 m and 2.0 m, which are lower than the optimal heights determined in this study37. However, optimal vegetation heights are closely related to the morphology of buildings and vegetation, and CFD simulations can determine these optimal heights for specific locations15,37.

Changes in building heights on local PM2.5 concentrations

The differences in modelled PM2.5 concentrations at seven receptors resulting from changing the height of one building are shown in Fig. 8, and the concentration data with varying building heights, as well as the vertical plane visualization, can be found in Fig. S5.

Figure 8
figure 8

Comparison of the modelled distribution of PM2.5 concentrations at the seven receptors along the central axis of the park for varying building heights (7 m and 21 m) to the initial 14 m model. (a1–a4) present the changes in the heights of the buildings W, N, E, and S under perpendicular wind, respectively. (b1–b4) present the changes in the heights of the buildings W, N, E, and S under parallel wind, respectively. The height at which data was collected is 1.5 m.

Full size image

For wind direction perpendicular to the road, changes in building height on the leeward side (W) had the most significant impact on the PM2.5 concentrations on the leeward side (west) road. Decreasing the height of the building W by 50% (7 m) resulted in a decrease and increase in the traffic-emitted PM2.5 concentrations at the building (O-W1) and park (O-W2) sides of the footpath by 105.1% and 38.3%, respectively. Conversely, increasing the height of building W by 50% (21 m) resulted in an increase and decrease in the PM2.5 concentrations at O-W1 and O-W2, respectively, by 20.0% and 17.1% (Fig. 8a1).

Similarly, changes in building height on the windward side (E) had the impact on the PM2.5 concentrations mostly on the windward side (east) of the park. Decreasing the height of building E by 50% (7 m) resulted in an increase and decrease in the traffic-exhausted PM2.5 concentrations at the building (O-E1) and park (O-E2) sides of the footpath by 25.0% and 96.8%, respectively. And increasing the height of building E by 50% (21 m) resulted in a decrease and increase in the PM2.5 concentrations at O-E1 and O-E2, respectively, by 11.3% and 86.4% (Fig. 8a3). These changes in building height affected ground-level air pressure, which in turn modified the vortices above the road25. This mechanism either facilitates or inhibits the dispersion of traffic-emitted PM2.5 from the road onto the footpaths (Fig. S6). This morphology resembles wide street canyons, where previous studies have observed similar variations in pollutant concentrations with changes in building heights27,64.

In terms of the air quality on footpaths outside the park, increasing the leeward building’s height from 7 to 21 m resulted in a 25.3% increase in average PM2.5 concentrations (0.24–0.30 μg/m3). Increasing the height of the windward buildings from 7 to 21 m resulted in a 37.0% increase in average PM2.5 concentrations (0.27–0.37 μg/m3). Thus, reducing the heights of both the windward and leeward buildings is beneficial for improving the air quality on the footpaths and throughout the entire study area. However, reducing the height of leeward buildings can increase PM2.5 concentration inside the park. The impact of changing the windward building’s height had a negligible effect on air quality inside the park.

For wind conditions parallel to the road, changes in the height of the leeward buildings (S) had the greatest impact on overall PM2.5 concentrations in the study area, especially on the footpaths near the park, i.e., O-W2 and O-E2 (Fig. 8b). Setting the height of the leeward building (S) to 7 m and 21 m resulted in the lowest and highest average PM2.5 concentrations across all scenarios, at 0.19 μg/m3 and 0.27 μg/m3, respectively, representing a 30.9% difference. Taller buildings on the leeward side impede airflow into the study area, subsequently inhibiting PM2.5 dispersion. Adjusting the heights of the other three buildings has minimal effect on PM2.5 concentrations in the study area. Previous study found demonstrated that ground-level pollutant with wind direction parallel to the street axis are comparable to, or even higher than, those in perpendicular wind models47. For Fitzwilliam Park, where parallel winds are predominant, the height of leeward buildings is particularly important in local air quality.

This finding demonstrated that increasing building heights on specific sides of an urban square park can amplify air quality deterioration, depending on local meteorological conditions. This is potentially relevant to similar urban settings. In urban forms characterized by green spaces surrounded by buildings, as observed in the study area, if the predominant wind direction is perpendicular to the road, lowering the height of windward buildings is the optimal choice for improving air quality. Reducing the height of leeward buildings is also viable, but may slightly increase pollutant concentrations inside the park. Measures to optimize air quality inside the park, such as implementing taller vegetation, should be considered in this case. If the predominant wind direction is parallel to the road, minimizing the height of leeward buildings yields the best air quality in the city square park. For both wind directions, if high-rise buildings are necessary, it is recommended to position them along the side aligned with the wind direction to minimize air quality deterioration.

Conclusions

The integrated environmental measurement and modelling study presented in this paper examined the evolving urban morphology and the interconnected relationship between grey and green infrastructure, and their impact on urban PM concentrations. The study considered a classically proportioned city square park surrounded by buildings and vegetation.

The measured data from the fixed monitoring campaign revealed the distribution of PM in the study area. This highlighted the importance of various influencing factors on local PM levels. The diurnal variations of PM2.5 and PM10 presented different patterns with PM2.5 showing a bimodal distribution throughout the day, peaking at 4:00–8:00 and 20:00–00:00, while PM10 presented a unimodal distribution with a 12:00 peak. The hourly mean PM2.5/PM10 ratios ranged from 0.28 to 0.66 during the monitoring period. Background concentrations were identified as the most dominant factor affecting PM concentrations in the study area by Pearson correlation stepwise regression, and random forest.

The environmental modelling simulations examined the effects of changes in urban morphology on air quality, focusing on the height of the vegetation and buildings around the city square park. The study found that the optimal vegetation height for improving air quality in the park varied with wind direction: vegetation 21 m tall was most effective for parallel winds, while vegetation 7 m tall was better for perpendicular winds. Additionally, taller buildings increased PM levels by up to 37.0% on footpaths with perpendicular wind, and raising leeward buildings led to a rise in PM by up to 30.9% with parallel winds. The findings highlight the complex morphological relationship between the buildings and vegetation in air quality assessment. They emphasize the importance of considering urban form dynamics in planning and decision-making for city square parks and modifications to buildings, as both green and grey infrastructure represent evolving features in the built environment that can affect air quality.

The limitation of the study is the evaluation of only simplified vegetation as the NBS, without investigating a range of different vegetation specie. Future studies should focus on the impacts of various vegetation species and other NBS solutions on air quality, and field measurements should also be conducted in different seasons with varying meteorological conditions. This work supports best-practice design guidance and aids decision-makers in considering the characteristics of NBS and the impacts of urban development to optimize air quality conditions both inside and outside city square parks.