Introduction

The increased urbanization is causing pressure on many environmental sectors. How to mitigate these pressures is one of the major nowadays challenges. The rapid expansion of impermeable surfaces in our cities and consequent loss of green spaces is promoting an increase in air temperature, air pollutant emissions, flooding risk, biodiversity loss and overall loss of ecosystem services (Bhatta 2010; Liao et al. 2015).

Green infrastructures (GI) are defined as “a strategically planned network of natural and semi-natural areas with other environmental features, designed and managed to deliver a wide range of ecosystem services, while also enhancing biodiversity” (European Commission 2023). Examples of GI include green parks, green roofs and walls and green corridors. This network can be a sustainable and cost-effective answer to mitigate the consequences of high urbanization and climate change (Andersson et al. 2017). One important characteristic of GI is their ability to provide multiple benefits and co-benefits.

Researchers have studied the GI benefits on the atmospheric environment at different scales and with different methods, and with the diversity of studies comes a large range of results (Kumar et al. 2019; Manso et al. 2021). For instance, at the mesoscale level, some studies show a negative impact of GI on air quality, mainly due to the reduction in temperature and consequent reduction of the boundary layer height (Arghavani et al. 2019; Ascenso et al. 2021a; Fallmann et al. 2016; Rafael et al. 2020, 2021; Taha 2008; Zhang et al. 2020). However, other studies show a positive effect on air quality and temperature (Gunawardena et al. 2017; Tallis et al. 2015). At the local scale, both negative and positive impacts are also reported; the GI act as an obstacle and reduce wind velocity nearby, affecting the local dispersion of pollutants, so in the same domain it is possible to have new air pollution hot spots and areas with improved air quality (Amorim et al. 2013; Borrego et al. 2012; Rafael et al. 2018; Rodrigues et al. 2020). Moreover, the effectiveness of GI is determined by many factors, such as location and type of GI, climate conditions, topography and vegetation type. Thus, more studies are needed towards the harmonization of methods to obtain more robust results and to better advise policymakers and municipalities on the implementation of GI in different urban areas.

This study focuses on two types of GI: Green walls (GW) and green roofs (GR). These GI can be effective greening strategies in urban areas with several direct and indirect potential benefits without restricting air flows or occupying additional space (Ysebaert et al. 2021). When compared to green roofs, green walls can cover more surface area, reaching a difference of up to 20 times in high-rise buildings (Besir & Cuce 2018), and technically the infrastructures show fewer restrictions for application (Koch et al. 2020). Some of the direct benefits of these green infrastructures are the removal of atmospheric pollutants by deposition leading to air quality improvement, carbon sequestration, energy savings by reducing temperature extremes, both in warm and cold situations, attenuation of noise levels and mitigation of the urban heat island effect (Besir and Cuce 2018; Koch et al. 2020; Manso et al. 2021; Tomson et al. 2021; Ysebaert et al. 2021). The indirect benefits are related to health and well-being, urban biodiversity, architectural and recreational value (Besir and Cuce 2018; Manso et al. 2021).

While there are several positive impacts of GI, there are also some constraints associated with their implementation, namely (Van Mechelen et al. 2015; Silva et al. 2015; Shafique et al. 2018; Manso et al. 2021): i) maintenance, GW may require frequent maintenance, and plants on both GR and GW can be prone to die in adverse climate conditions; ii) water consumption, GR may require significant water consumption for irrigation; iii) monetary costs, the initial cost and ongoing maintenance expenses of GR and GW can be higher compared to conventional solutions. Additionally, GW and GR require a great number of materials compared to conventional solutions, which can have potential environmental impacts (Scolaro & Ghisi 2022; Mihalakakou et al. 2023). These GI must evolve to become more sustainable solutions, through the use of materials with less incorporated energy and carbon emissions and the application of climate adapted plant species with less irrigation needs. Some examples (Mihalakakou et al. 2023; Shafique et al. 2018), already show sustainability concerns by using natural or recycled materials and native plants, integrating water recovery systems and sensors for water and nutrients minimization. By integrating efficient components, considering lifecycle costs, and implementing standardized disposal methods, these living installations can deliver environmental benefits over the long term. The integration of these practices supports the overarching goal of creating resilient and sustainable urban environments.

Computational Fluid Dynamics (CFD) models allow consideration of the morphological specificities of the urban environment, and consequently, they are able to simulate the flow dynamics perturbations caused by distinct urban obstacles and to assess the effectiveness of distinct countermeasures to control air pollution in urban areas. Specifically, CFD models have been applied to assess the role of GI in mitigating air pollution, mainly considering the effects of the mechanical drag and/or including pollutant removal by deposition and filtration mechanisms (Amorim et al. 2013; Cui et al. 2023; Gromke et al. 2016; Jeanjean et al. 2017; Rafael et al. 2018; Santiago et al. 2019; Vos et al. 2013; Vranckx et al. 2015).

Several review studies (Abhijith et al. 2017; Koch et al. 2020; Tomson et al. 2021; Tsoka et al. 2018) identified a lack of research investigating the impacts of green walls and roofs on air pollution at the local scale, recommending further work to make solid recommendations and to draw more conclusions regarding green infrastructures implementation for air quality improvement. Aiming to reduce this research gap, the current paper presents the results of a quantitative assessment of how green walls and green roofs can impact the air quality of a selected central area of the municipality of Lisbon, the capital city of Portugal. This area is a vastly urbanized zone with generally tall buildings (over six floors). It is surrounded by residential buildings but also by important infrastructures such as train stations, commercial and academic centres, parks, and offices. Therefore, this location comprises important traffic accesses with multiple lanes and an underground passage.

The main objective of this paper is to evaluate the GW and GR potential benefits and effectiveness in mitigating air pollution by analysing the local flow dynamics and air quality levels in a high-traffic area of the city of Lisbon considering two distinct green scenarios. The manuscript is organized as follows: Section 2 describes the used model and setup; in Section 3 modelling results are presented and discussed; conclusions are drawn in Section 4.

Methodology

To study the impact of GW and GR in terms of air pollution, wind pattern, turbulent kinetic energy, and temperature, the ENVI-met CFD model was applied to a particular hour of a summer day in Lisbon. This section presents the description of the model along with a detailed description of all necessary input and initialization parameters, including information about the simulation domain within Lisbon.

ENVI-met model description

ENVI-met is a Computational Fluid Dynamics (CFD) software based on a three-dimensional model designed to simulate complex surface-vegetation-air interactions in urban microclimate (Bruse and Fleer 1998; ENVI-met 2008). The wind field is calculated using the non-hydrostatic three-dimensional Navier–Stokes equation. Turbulence is parametrized in ENVI-met using a k-ε 1.5 order closure model. ENVI-met uses the Finite Difference Method to solve the partial differential equations. The scheme is partly implicit, partly explicit depending on the subsystem analyzed. The atmospheric advection and diffusion equations are implemented in a fully implicit scheme. ENVI-met uses an orthogonal Arakawa C-grid to represent its environment. Topography is included by marking cells as being filled with soil. Because of this scheme, ENVI-met does only allow straight and rectangular structures. For the ground surface, the exact exposition and inclination is considered for the energy balance calculations. For building walls and roofs, inclined or curvy surfaces must be approximated by grid points. In addition, the model applies an Eulerian approach to simulate the dispersion of pollutants.

The pollutant dispersion module of ENVI-met allows the synchronous release, dispersion and deposition of up to six different pollutants including particles and both passive and reactive gases. The photochemical reaction cycle between NO, NO2 and ozone is considered, as well as the release of biogenic volatile organic compounds (BVOCs) through plants. Among the BVOCs, isoprene plays a central role compared to other BVOCs, it forms higher quantities of reactive oxygen compounds which in turn can directly lead to higher ozone concentrations (Calfapietra et al. 2013). Together with the fact that isoprene is the most abundant volatile organic compound emitted by plants, the emission of isoprene and the isoprene-induced reactions play a key role in the formation of tropospheric ozone (Guenther et al. 2006; Sharkey et al. 2008; Simon 2016). Therefore, the ENVI-met’s chemistry module simulates the formation of tropospheric ozone by photolyzing nitrogen dioxide and its destruction by reacting with nitrogen monoxide. In addition, through an isoprene emission model, ENVI-met’s chemistry model has been extended to isoprene-induced reactions, leading to changes in ozone concentrations.

The deposition velocity is modelled as the inverted sum of the resistances, the aerodynamic resistance, the sublayer resistance, and the surface resistance. The aerodynamic resistance at ground surfaces or walls is derived from turbulent exchange coefficient for heat at the ground surface or at walls (Bruse 2007), while the aerodynamic resistance at plants is calculated using plant specific constants for deciduous, grass and conifers (Bruse 2007). The sublayer resistance is calculated using the friction velocity and the Schmidt number and represents resistances caused by the quasi-laminar layer adjacent to the surface. The transfer velocities depend on the molecular properties of the compound and the surface. The surface resistance accounts for solubility processes of the compound with various surfaces. It is dependent on chemical properties of both the gas/ particles and the surface type.

The sedimentation of particles due to gravitational forces and the deposition of both, particles and gases, to surfaces are complex processes that influence the local concentration to a large extent. Especially the absorption of gases at natural surfaces (leaves, soils) and the capturing and re-releasing of particles at rough surfaces are processes, that are not fully understood and lack usable numerical parameterizations (Bruse 2007). ENVI-met treatment of sedimentation and deposition processes is composed of two components: (i) concentration change due to gravitational settling including deposition at surfaces (composed of downward flux and flux received from grid boxes above); (iii) deposition at leaf surfaces. The concentration change due to gravitational settling only applies to particulate matter whereas the deposition at solid surfaces and plant leaves are calculated both for particles and gases (Bruse 2007).

The choice of ENVI-met for this study is based on its ability to simulate simultaneously microclimate with highly detailed 3D vegetation, green facades/roofs, and pollutant dispersion and deposition as well as vegetation-surface-pollutant interaction with a high spatial and temporal resolution. The software is designed for microscale with a typical horizontal resolution from 0.5 to 10 m and a typical time frame of 24 to 48 h with a time step of 1 to 5 s (ENVI-met 2019). All the simulations performed for this work used the ENVI-met version 4.4.6. The detailed description of ENVI-met with all the mathematical equations can be found in Bruse and Fleer (Bruse and Fleer 1998) and Huttner (2012).

ENVI-met has been widely applied to assess the impact of greening strategies in urban microclimates (Lee et al. 2016; Peron et al. 2015; Tsoka et al. 2018; Vos et al. 2013), although the impact of green walls, in particular, using ENVI-met is still scarce. Multiple validation exercises were carried to assess the model performance where overall acceptable/good agreements were obtained (Crank et al. 2018; Detommaso et al. 2021; Di Giuseppe et al. 2021; Lee et al. 2016; Simon et al. 2018). However, the model presents some limitations, namely: does not have a separate mechanism to account for re-suspension of particles that occurs concurrently with deposition and consequently either influences the removal rate or the amount deposited (Morakinyo et al. 2016). The results from Crank et al. (2018) suggest that ENVI-met is not grid independent, contrary to the best practices in the CFD modelling community. The surface energy balance is limited in its ability to represent the urban environment, as ENVI-met accounts for latent and sensible heat fluxes but does not consider their dependency on the thermal mass of the urban form, i.e., only considering the radiation budget and ignoring the impact of fluid flow through the urban canyon. An overall drawback is the reduced research studying the accuracy of the model for other atmospheric variables than air temperature.

Input parameters

The input and initialization parameters necessary to run the ENVI-met model are described in this subsection.

Domain configuration setup

ENVI-met was applied to a simulation domain centred in a main avenue in the city of Lisbon in Portugal – the Avenida da República (38°44′33″N 9°08′49″W, elevation 70 m a.s.l.). Figure 1 shows the location of the simulation domain, the location of an air quality monitoring station and the location of the traffic count points used to estimate the traffic emission used in this study.

Fig. 1
figure 1

Location of the case study domain (618 m × 594 m), the air quality monitoring station and points of traffic counts

Full size image

The simulation domain for this study covers a horizontal area of 618 m × 594 m with a uniform Cartesian grid and a vertical height of 143 m with a variable grid setup. Due to the orientation of the main avenue, a rotation of -9˚ (anti-clockwise rotation) was applied to the domain to improve the quality of the overall mesh. The horizontal resolution of 2 m × 2 m (L x W) was kept the same all over the domain. For the vertical resolution, due to the importance of the pedestrian height, the first cell was divided into five grid cells with a finer resolution of 0.4 m; then a 2 m resolution was set from z = 2 m to z = 60 m, and for the remaining cells above 60 m a telescoping factor of 25% was applied. To avoid numerical problems caused by model boundary interference with internal model dynamics and guarantee a minimum spacing between the obstacles and the model boundaries, 30 nested grids were added to each side of the computational domain. Thus, totalling 309 × 297 × 43 grid cells. The domain contains 184 buildings, with the tallest building being 56 m.

To quantify the impact of the green scenarios, three computational domains were created: a baseline (Fig. 2(a)), a green walls scenario (Fig. 2(b)) and an integrated scenario (Fig. 2(c)). These scenarios were co-created with local stakeholders and the location of the green infrastructures was decided based on a set of practical constraints. The green walls scenario consists of the baseline set-up with the implementation of green walls in 13 specific buildings near the main avenue (identified in green in Fig. 2(b)). Hedera helix, with a Leaf Area Index (LAI) of 1.5 and a leaf angle distribution of 0.5, was used for greening the buildings walls, because it is the most common species found in direct green facades around the world (Ysebaert et al. 2021). For the integrated scenario, in addition to the green walls, green roofs at 8 of the buildings were considered together with a green corridor (GC) in three distinct areas over the road tunnel of the main avenue. The GC consists of bushes with a height of 1 m, as well as trees with a trunk height of 2 m, and a total height of 5 m. For reference, the existing trees in the baseline scenario have a height of 15 m.

Fig. 2
figure 2

ENVI-met geometric modelling setups: (a) the baseline scenario, (b) the green walls scenario, and (c) the integrated scenario, which includes green walls and roofs, together with a green corridor in the main Avenue

Full size image

For the soil and surface parameters, since the area is highly urbanized, a default pavement (Pavement (Concrete), database-ID [0100PP]), with an albedo of 0.3 and an emissivity of 0.9, was applied over the domain, except for the main green areas with unsealed soil (Default Unsealed Soil (Sandy Loam), database-ID [010000]), which has an albedo of 0.2 and an emissivity of 0.9.

Due to the difficulty and complexity inherent to the measurement of the vertical distribution of urban vegetation, especially in a full-scale analysis, few data are available, especially for trees in European urban areas. In this study, the same tree characteristics (deciduous trees, with a Leaf Area Density (LAD) of 1 m2/m3) were used for the surface green vegetation, considering the analysis of Lalic and Mihailovic’s (2004), which shows an average LAD inside the crown of a large size tree of approximately 1 m2/m3, with no significant variation with height.

Emissions

In Lisbon, air quality problems are strongly associated with road-traffic emissions, industrial activities, shipping, and aviation emissions. These emission sources make a homogeneous contribution throughout the year. During winter days, nights are longer and strong surface atmospheric thermal inversions are more common. This phenomenon combined with the need for residential heating, such as the use of wood burning, or other fuels, can lead to episodes of high levels of particulate matter and NO2 concentrations. While O3 concentrations are of particular concern during the spring and summer months.

For this work, only the transport sector was quantified in terms of emissions. Traffic of motorcycles, light and heavy vehicles as well as buses was counted onsite for two periods in two strategically different locations (see Fig. 1). One traffic count occurred between 7:30 and 10:30 and between 17:00 and 20:00, in 15-min intervals, in an intersection between the main avenue (Avenida da República), another important avenue (Avenida António Serpa) and the Campo Pequeno square, for a total of 12 lanes. Counts were also done at the southern entrance of the road tunnel, between 7:30 and 10:30 and between 17:00 and 20:00, in 15-min intervals, for a total of 2 lanes. Both counts occurred during a weekday. It was not possible to do traffic counts covering all the roads located in the domain, however, all the roads on the northern side of the domain were accounted for. The obtained short-term traffic counts were then used to estimate the annual average daily traffic (AADT), following the methodology proposed by Brito (2012), by multiplying with hourly, daily, and monthly adjustment factors. Hourly distribution was then extrapolated using ENVI-met’s traffic model profiles. For the unaccounted lanes, the values were inferred based on similar lanes in the domain.

It is advisable to use emission factors calculated from local empirical data. However, such data was not available, and the emission factors used in the ENVI-met model were sourced from the European Monitoring and Evaluation Programme (EMEP) (EMEP/EEA 2020). Table 1 shows the selected emission factors, where PC is passenger cars, LDV is Light Duty Vehicles, HDV is Heavy Duty Vehicles, Bus is Urban Buses, and MC is Motorcycles. The fraction of NO2 present in NOx was considered 90%.

Table 1 Emission factors in g/km (EMEP/EEA 2020)
Full size table

In total, 23 different emission sources from road traffic were added to the model (Fig. 3(a)). From those, 21 correspond to the sections of road (line sources), and 2 were considered as area sources, corresponding to the openings of the road tunnel, and their calculation was based on a box model (Eq. 1).

$$q=\frac{CuH}{L}$$
(1)

where:

\(q\) = emission rate (µgs−1 m−2).

\(c\) = concentration inside the tunnel (μgm−3).

\(u\) = wind velocity in x (ms−1).

\(H\) = mixing height (m).

\(L\) = dimensions with one side parallel to the wind direction (m).

Fig. 3
figure 3

Location of the traffic emission sources within the computational domain (a). Emission rate for NO2 (a) and PM10 (b) used in the simulation

Full size image

The concentrations inside the tunnel were set based on a priori simulation considering only the tunnel configuration and the traffic lanes inside the tunnel with the corresponding emission rates. The resulting concentrations were equal to 30 µgm−3 for NO2, 15 µgm−3 for PM10 and 10 µgm−3 for PM2.5. The mixing height was the same as the height of the tunnel (10 m), the dimension (\(L\)) was the length of the tunnel (35 m) and the wind velocity was set to 2 ms−1, based on inflow wind data.

Figure 3 shows the emission rates of NOx (Fig. 3(b)) and PM10 (Fig. 3(c)) estimated by the model.

The contribution of the lanes in the main Avenue, accessing the road tunnel, is quite evident in terms of emissions. In addition, the lanes crossing the main Avenue on the Top and Bottom of the computational domain have also significant emission rates.

Initial and boundary conditions

Based on data from a 10-year period, between 2010 and 2019, from the meteorological station (Gago Coutinho) located near the Lisbon Airport and distancing approximately 2.9 km from the domain and taking into consideration the air quality data from a traffic monitoring station (Entrecampos, see Fig. 1), an hour was selected for the simulations. This simulation period was selected taking into consideration data availability (e.g., emission data, meteorology, air quality) and the criteria of particular interest for the involved stakeholders, such as a period of rush hour (peak of road traffic emissions) and high temperatures. Seasonality is also a key factor when assessing the impacts of green infrastructures on flow dynamics and air quality in urban areas. Spring and summer months are of particular interest considering vegetation growth and greening levels. Thus, the analysis was focused on 16:00, which corresponds to a traffic pollution peak and to a high temperature period. The dominant wind blew from NW, with a wind speed of 2 ms−1. The relative humidity was set to 51% and the temperature to 25˚C. Therefore, initial meteorological boundary conditions in the simple forcing option were set for temperature, humidity, mean wind speed and direction. For the meteorological measurement site, the surface roughness length value was defined as 1 m, which is considered typical for a build-up area (Wieringa 1992). Based on these user-defined data, ENVI-met created one-dimensional profiles that were consequently used as inflow boundary conditions for the model.

Background concentrations were also added to the simulation, as initial data, using observations from an urban background monitoring station located in Olivais, approximately 4.4 km east from the domain. For the simulation period, the background concentrations were equal to 68 µgm−3 (NO2), 12 µgm−3 (NO), 87 µgm−3 (O3), 20 µgm−3 (PM10) and 14 µgm−3 (PM2.5).

Results and discussion

Simulation results were analyzed comparing the baseline with the green walls and the integrated scenarios. Section 3.1 shows the simulation results for the baseline, while Section 3.2 shows the impact of the scenarios using green infrastructures.

Baseline

Figures 4, 5, 6 show the horizontal maps of the simulation results for the baseline simulation, considering the estimated values for the magnitude of the wind speed, together with the three components of the wind speed (u-, v- and w-), and followed by the turbulent kinetic energy, temperature, and solar radiation (Fig. 4), together with the NO2, O3, PM10 and PM2.5 concentrations (Fig. 5) and deposition levels of pollutants (Fig. 6).

Fig. 4
figure 4

Horizontal colour map of the magnitude of the wind speed (a), and the three components of the wind speed u- (b), v- (c), w- (d), turbulent kinetic energy (e), temperature (f), and solar radiation (g) obtained at pedestrian level (z = 1.4 m) for the baseline simulation

Full size image
Fig. 5
figure 5

Horizontal colour map of NO2 (a), O3 (b), PM10 (c) and PM2.5 (d) concentrations, obtained at pedestrian level (z = 1.4 m) for the baseline simulation

Full size image
Fig. 6
figure 6

Horizontal colour map of NO2 (a), O3 (b), PM10 (c) and PM2.5 (d) total amount of mass deposed to ground or leaf surface unit area obtained at ground level (z = 0 m to 2 m) for the baseline simulation

Full size image

The colour map in Fig. 4(a) shows the magnitude of the wind speed, revealing slight increases near the complex corners of buildings compared to the initial inflow wind speed of 2.0 ms−1, reaching a maximum of 2.4 ms−1. Overall, lower wind speeds are observed in passages between buildings, and downwind the buildings. The wind vectors, also shown in Fig. 4(a), highlight turbulent vortices generated by the buildings’ corners. In Fig. 4(b), the wind u-component indicates negative values on the west- and east-facing buildings’ facades (minimum of 0.6 ms−1), while the highest positive values are simulated in a few buildings’ corners (maximum of 1.9 ms−1). Figure 4(c) displays positive values of the wind v-component near the north and south buildings’ facades (maximum of 0.6 ms−1), with some spots recording significantly high negative values in specific buildings’ corners (ranging from 1.5 to 2.0 ms−1). Figure 4(d) shows the wind w-component with positive values around the dense vegetation surrounding building 5, as well as in the region of buildings’ facades.

The vortexes generated by the presence of buildings’ corners align with existing knowledge in literature. Nevertheless, these structures may also be linked to numerical artifacts due to the meshing of building corners, a topic requiring further investigation.

Simulation results denote turbulent kinetic energy values (Fig. 4(e)) over the entire domain in line with what is found in literature for urban areas (Kastner-Klein and Rotach 2004). Near complex building corners higher turbulent kinetic energy values are estimated, due to vortices generation (e.g., building 5).

Temperature patterns indicate higher temperatures close to western and south-western buildings facades– the sun-exposed facades at this period of the day, while lower values are registered downwind the buildings and in the expected shadowed areas.

The simulated total solar radiation over the entire computational domain is equal to 1080 Wm−2, while lower values of solar radiation (from 120 to 240 Wm−2) are registered in the zones of buildings facades non exposed to the sun at this period of the day, as well as in the shadowed areas created by the presence of buildings and vegetation.

Figure 5 shows higher values of NO2, PM10 and PM2.5 concentrations in the main traffic lanes, where the emission sources with higher emission rates are allocated. There are specific hotspots of concentrations in the main Avenue (north and south lanes), directly linked with the highest emission rates at the entrances of the road tunnel, in line with the colour of emissions in Fig. 3. The low wind speed seen in the traffic lanes also contribute to the formation of hotspots of air pollution. For instance, the accumulation of pollutants near building 8, is linked with a low wind speed area on the north facade of the building.

The background concentration of O3 considered in the simulations is equal to 89 µgm−3. The spatial distribution colour map (Fig. 5(b)) shows areas of higher O3 concentrations than the background value (e.g., bottom of the main Avenue, top and bottom lanes crossing the main Avenue), mainly due to high levels of NO2 concentrations available to be used in the O3 formation. Overall, O3 concentrations are lower near the east-side buildings façades, where low solar radiation was simulated (see Fig. 4(g)). The lower O3 concentrations near building 5 can be explained by the higher deposition levels (Fig. 6(b)), where the vegetation is located. Typically, O3 is not a pollutant analyzed at microscale, taking on the hypothesis that there are no relevant chemical reactions at these spatial and temporal scales. However, ENVI-met accounts for the photochemical reaction between NO, NO2 and O3. Figure 5(b) patterns indicate that chemical reactions leading to O3 formation may occur at this very local spatial scale.

Despite being located outside the computational domain, the Entrecampos air quality monitoring station (an urban traffic station located approximately 300 m northern from the top of the domain), could be representative of the order of magnitude of the air pollutant concentrations over the main Avenue, and the monitored concentrations during the simulation period can be used as indicative values for verification of the simulation results over the Avenue. The monitored concentrations are compared with the minimum and maximum concentrations simulated within the computational domain (Table 2).

Table 2 Overview of the monitored concentrations at the air quality station (Cmeasurements) and the simulations results, including the minimum (Cminimum) and the maximum (Cmaximum) concentrations simulated over the entire domain
Full size table

Thus, the simulated concentrations fit well within the order of magnitude of the measured levels. Supposing that the air quality station has sufficient spatial representativeness to cover the study area, we can argue that simulated concentrations show agreement with the measurements. It is noteworthy that the large variability in the simulated values along the main Avenue emphasizes the relevance of CFD results in providing comprehensive spatial coverage, especially when contrasting with monitored concentrations at specific locations.

Figure 6 results indicate that NO2 mass is deposed mostly on the green infrastructures, but also, in a much lower quantities, on surfaces corresponding to the road lanes with higher emissions. The bottom right set of green infrastructures surrounding building 5 led to lower deposition levels, which could be associated with the lower values of NO2 concentrations being dispersed to this area. O3 deposition patterns are related with the NO2 deposition patterns. O3 deposition values on surfaces range from 0 to 40 µgm−2, while on vegetation range from 1650 to 2941 µgm−2. Deposition of PM10 occurs over the entire domain with a minimum value of 32 µgm−2. The higher values of PM10 deposition occur on the green infrastructures surrounding building 5, but also on surfaces in the main avenues, where the emissions are located. Inside ENVI-met, the leaf surface PM10 deposition and sedimentation is directly driven by the atmospheric PM10 concentration (Hofman and Samson 2014). In the main Avenue higher concentrations of PM10 led to higher deposition values. High deposition values near building 4 are related to the presence of green elements. Deposition values of PM2.5 follow a similar pattern as the PM10 deposition values, with lower values (in line with the lower values of PM2.5 concentrations).

The magnitude of the Total PM10 deposed (≈ 30 g in an hour) is of the same magnitude of similar studies using ENVI-met in an urban area (Viecco et al. 2021), and the deposition velocities estimated for NO2 and O3 (≈ 0.4 cms−1) also fall within the range of other studies (0.1–0.55 cms−1) (Ainsworth et al. 2012; Holland et al. 2005; Lovett 1994), however it is not possible to directly compare such results due to the high dependency on variables like size and density of the particles, type of vegetation (including factor like stomatal opening, LAI, canopy closure, vegetation height), area of vegetation in the computational domain, vapor pressure deficit and wind velocity and air pollutant concentration. In addition, deposition velocities are proved to significantly vary between hours of the day and month of the year (Ascenso et al. 2021b).

Green walls, green roofs and green corridor scenarios

When comparing results from the green walls and the baseline scenarios we found a negligible impact for all the pollutants, while the integrated scenario denotes the capability to change pollutants concentrations. Figures 7, 8 show the horizontal colour of the differences between the integrated and the baseline scenarios, considering the estimated values for wind speed, turbulent kinetic energy, and temperature (Fig. 7), together with the NO2, O3, PM10 and PM2.5 concentrations (Fig. 8).

Fig. 7
figure 7

Horizontal colour map of the differences between the integrated scenario and base scenario for wind speed (a), turbulent kinetic energy (b) and temperature (c), in percentage, obtained at pedestrian level (z = 1.4 m)

Full size image
Fig. 8
figure 8

Horizontal colour map of the differences between the integrated scenario and base scenario for NO2 (a), O3 (b), PM10 (c) and PM2.5 (d), in percentage, obtained at pedestrian level (z = 1.4 m)

Full size image

The integrated scenario will lead to a reduction of wind speed up to a maximum of 22% due to the presence of the green corridor in the main Avenue (Fig. 7(a)), followed by a reduction of TKE up to 37% (Fig. 7(b)). However, the results also point to increased levels of wind speed and TKE around the green corridor vegetation. Temperature decreases (maximum of 1.6%, which corresponds to a maximum decrease in temperature of 0.3ºC) due to the presence of the green corridor, but also due to the green walls in the buildings (e.g., the west façade of building 8). The green corridor was implemented in an area of 1500 m2. A decrease of 0.3ºC is in line with data available in literature, e.g., Aram et al. (2019) shows a cooling effect intensity ranging from 0.32ºC to 0.57ºC for green areas of 0.3 ha and 0.8 ha. Furthermore, in our study, this cooling effect is only slightly advected by the turbulent flow, which is also in line with the findings of Aram et al. (2019), which found a cooling effect distance of 1 m (green area of 0.3 ha) and of 22 to 44 m (green area of 0.8 ha).

The implementation of the integrated scenario will promote distinct impacts on air quality, but following similar spatial patterns, with the exception of O3. It will lead to a maximum increase of 9%, 4% and 3% of NO2, PM10 and PM2.5 concentrations (Fig. 8(a), (c) and (d)), at specific grid cells, while those concentrations will decrease up to a maximum of 9%, 4% and 3%, at other specific cells, as compared to the baseline. The major impacts are linked with the presence of the green corridor, with an increase of concentration at the Avenue where the trees are planted. This happens as expected, as it was already demonstrated that vegetation is capable of reducing wind speed by attenuating turbulent kinetic energy and acting as a diffuser, breaking down turbulent eddies, but also may increase shear stress around vegetation and therefore induce mechanical turbulence to the flow (De Maerschalck et al. 2010).

The impact of the integrated scenario on O3 concentrations (Fig. 8(b)) is distinct compared to its influence on NO2 concentrations, because this scenario will result in a no increase of O3 concentrations with a potential decrease of 7%. The locations of higher NO2 concentrations occur where we have reductions of O3 concentrations due to depleted chemical reactions in these regions, aligned with the findings of Moradpour et al. (2017).

To better understand the impact of the GI, vertical profiles of the studied variables were analised for a plan along the main avenue (Figs. 9, 10). These vertical profiles also confirm these findings. The green corridor of the integrated scenario will lead to an overall decrease of TKE (up to 43%), wind speed (up to 39%) and temperature (up to 2.3%). The results also show an increase of TKE at the height between 5 and 10 m, near buildings 1 and 2 (where the tallest trees were planted).

Fig. 9
figure 9

Vertical colour map of the differences between the integrated scenario and base scenario for temperature (a), wind speed (b) and TKE (c) in percentage, obtained at X = 250 m

Full size image
Fig. 10
figure 10

Vertical colour map of the differences between the integrated scenario and base scenario for NO2 (a), O3 (b), PM10 (c) and PM2.5 (d) in percentage, obtained at X = 250 m

Full size image

At this specific X-cut, the green scenario will promote a maximum decrease of 10%, 4% and 3% of NO2, PM10 and PM2.5 concentrations, while those concentrations will increase up to a maximum of 8%, 3% and 2%, following similar spatial patterns, registering the highest differences in the northern part of the green corridor. The highest increase will occur at heights between 5 and 10 m. O3 concentrations will decrease up to a maximum of 7% due to the presence of the green corridor.

The reduction in wind speed inside the street canyon/ main Avenue (Fig. 7(a)), due to the aerodynamic effects of vegetation, leads to an increase of NO2 concentrations and a decrease of O3 concentrations. These results are aligned with the findings of Hang et al. (2023), who discuss the aerodynamic effects of trees as the process with the greatest influence on the O3-NOx-BVOCs chemistry and pollutant dispersion process inside the street canyon, followed by the BVOCs emissions, and the dry deposition.

Therefore, these results point out potential local benefits of green infrastructures on O3 concentrations, followed by either positive or negative impacts on particulate matter and NO2 concentrations, highlighting the relevance of further research focused on the interlinks between NO2 and O3 concentrations and the potential impacts of green infrastructures on the concentration of those pollutants at microscale.

The simulation results show that pollutant concentrations of NO2, O3, PM10 and PM2.5 are not reduced in general with the implementation of the integrated scenario, but only redistributed in the computational domain. While the pollutants concentrations are typically reduced in the green areas of the scenario, they tend to increase in upwind areas. These effects can be of large negative impact since pedestrians do not walk, bike, or drive the car in the green corridor, but on the street, where the pollutant concentrations are higher.

Figure 11 presents the total mass of pollutants removed by the GI for the integrated scenario (in comparison to the baseline). Only the ground-level values are shown (0 to 2 m) because the results indicate that the majority of deposition differences happen at this height, and they were promoted by the trees and bushes of the GC located above the tunnel.

Fig. 11
figure 11

Horizontal colour map of the differences between the integrated scenario and baseline scenario for the total amount of NO2 (a), O3 (b), PM10 (c) and PM2.5 (d) mass deposed on ground or leaf surface unit area, obtained at ground level (z = 0 m to 2 m)

Full size image

The ground-level deposition differences due to the GC represent 91%, 87%, 86% and 77% of the impact of the integrated scenario for O3, NO2, PM2.5 and PM10, respectively. The most impact on deposition is therefore seen at the level of the bushes (0 to 2 m), followed by deposition at tree level (4 to 6 m), on the same GC, which is similar to the results obtained by Viecco et al. (2021). In total, the GC was responsible for the removal of 12,7 g of NO2, 7.8 g of O3,, 0.25 g of PM10 and 0.01 g of PM2.5.

Deposition values over the buildings with GR were also analysed. Results indicate the GR assessed in this study had a negligible effect on the air pollutants deposition. This could be related to the low deposition flux at the top of the buildings (10 times lower than the values at ground level), associated with the higher wind speeds and lower concentrations. Moreover, the similar values between baseline and GR suggest that stomatal resistance was very low, probably due to the high temperatures and the type of vegetation of the GR.

ENVI-met challenges and limitations

The results of this case study could be very useful to establish a set of guidelines to be disseminated to stakeholders and decision makers for regulatory purposes, promoting important social, environmental and economic benefits. However, there are some challenges and limitations of this study that can be worth to explore as future research work, to further support end-users.

ENVI-met is a Windows software, and thus depending on the RAM and CPU cores available, the user easily faces computational resources constraints. This characteristic relies on challenges in terms of period of simulation, computational domain size and grid resolution, grid independence and sensitivity analysis. For example, the period of simulation of 1 h is not representative of the daily, monthly and seasonal variation of atmospheric emissions, air pollution episodes, green infrastructures characteristics, which are critical variables when analysing the impacts of GI on local air quality. Furthermore, the model has only available a k-ε turbulence closure model, which is a limitation considering the options now available in literature. As consequence, this study is also affected by these limitations.

In addition, the background concentrations of O3, NO and NO2 were obtained from a background air quality station located 4.4 km away from the study area. The actual background concentrations can be different from those measurements, and they have a key influence on the atmospheric chemistry and, specifically, on the concentration of O3 at pedestrian level.

In this study, the air quality monitoring station was located outside the computational domain leading to difficulties to perform verification of the simulation results. The choice of excluding the monitoring point of the study area is due to the trade-off between the requirements from the stakeholders (interested in the road tunnel and the main Avenue included in the simulation domain) and the computational resources constraints.

Despite those challenges and limitations, ENVI-met presents unique capabilities regarding its database consisting of several parameters for a wide range of vegetated species, as well as its attractiveness to end users, mainly due to its user-friendly interface.

Therefore, as future research we would highly recommend the integration of the database of vegetation into more complex CFD models, the development of a model version to run in other operating systems, or to explore more the abilities of running the model into the cloud.

Conclusions

This study aimed to foster green infrastructures impacts on air quality, evaluating the effectiveness of green urban planning strategies. No disturbances on the turbulent flow dynamics and on the air quality levels were obtained by the green walls’ scenario, when compared to the baseline scenario. The overall results, for the integrated green scenario, point out potential local benefits of green infrastructures on O3 concentrations, followed by either positive or negative impacts on NO2, PM10 and PM2.5 concentrations, mainly associated with the implementation of a green corridor along the main Avenue.

In literature, ozone is not often considered in CFD simulations at microscale and the prevailing hypothesis suggests the absence of significant chemical reactions occurring at these spatial and temporal scales. However, due to the capabilities of ENVI-met, the simulation results indicate that chemical reactions leading to O3 formation may occur at this very local spatial scale. Therefore, the main findings of this study highlight the relevance of further research focused on the interlinks between the NO2 and O3 concentrations, followed by the relationship between the ozone formation and the temperature patterns at microscale, the deposition patterns of both NO2 and O3, together with the impacts of vegetation (e.g., location, characteristics).

The deposition results showed a strong impact of the integrated scenario on levels of NO2, O3, PM10 and PM2.5 but only in the main Avenue, where the trees and bushes were considered. This highlights the stronger impact of trees and medium vegetation on the deposition levels of pollutants, at the pedestrian level (0 to 2 m) and tree level (4 to 6 m), when compared to the impact of green walls and roofs. We do not intend to directly apply the results of our study to other cases due to the local characteristics. However, we consider the main conclusions to be valid for similar conditions based on the results and previous studies available.