Adaptation to Climate Change at Local Scale: A CFD Study in Porto Urban Area Adaptation to Climate Change at Local Scale: A CFD Study in Porto Urban Area

Green infrastructures play an essential role in urban planning, namely with their potential to reduce the impact from air pollution episodes together with extreme weather events. This chapter focuses on the assessment of green infrastructures’ benefits on current and future microclimate and air quality patterns in Porto’s urban area (Portugal). The effects of green infrastructures on flow dynamics are evaluated for the baseline scenarios by means of numer - ical and physical simulations, using the computational fluid dynamics (CFD) model VADIS and the wind tunnel of the University of Aveiro. The baseline morphological (BM) scenario focuses on the current morphological characteristics of Porto’s urban area, while a baseline green (BG) scenario comprises the replacement of built-up areas by green areas and parks. In addition, the benefits of green infrastructures on air quality are assessed for the baseline and under future climate scenarios. The air quality simulations focus on particulate matter, one of the most critical air pollutants with severe impacts on human health. For the BM scenario, the simulated concentrations are compared with hourly averaged PM10 concentrations mea - sured during a weekday at the air quality station located within the study domain.


Introduction
Despite the advances over the last decades on mitigation of greenhouse gas emissions, climate change still remains a major concern, threatening ecosystems and human systems [1][2][3]. Currently, climate change (CC) impacts are recorded worldwide, through extreme weather events, such as heat waves, droughts, floods, cyclones and wildfires [1][2][3]. In addition, urban areas are facing a continuous unsustainable growth of population [4,5], associated with extreme weather events and critical air pollution episodes, highly affected by these climate-related events, among other environmental pressures, again threatening human health and lifestyle standards [1][2][3][4][5].
Given the expected increase of extreme weather events, the design and implementation of countermeasures in densely populated urban areas are an important goal in adaptation of societies to climate change in coming decades. Adaptation policies and strategies have the potential to increase resilience of urban areas to climate change, i.e., to improve the ability of a city and the surroundings to readily recover from a disturbance and return to the original functions of the urban metabolism [6]. Recently, several adaptation strategies turn to innovative solutions supported by nature, the so-called nature-based solutions, such as green infrastructures and urban forests, white roofs and highly reflective facades, urban water streams and wetlands, etc. [7][8][9][10][11][12][13][14]. These nature-based solutions provide sustainable, cost-effective, multipurpose and flexible alternatives for tackling societal and environmental challenges, such as climate change, air pollution and human health protection. In particular, green infrastructures (e.g., green roofs, walls and corridors) and urban forests play an essential role in urban planning, leading to several benefits to the environment as solutions of not only resilience to climate change but also to improve air quality and thermal comfort [7][8][9][10][11].
An improved knowledge about the highly complex urban microclimate and consequent air pollutant dispersion patterns is therefore of utmost importance to circumvent climate change impact at local scale (e.g., at street canyon level). A better understanding of the overall urban microclimate, and all the physical-and chemical-related processes, requires not only accurate measurements but also numerical and physical modeling of the exchanges within the urban atmospheric boundary layer [15,16]. In recent years, several studies have been performed using physical models, such as wind tunnel simulations, experimental campaigns and numerical models [17][18][19][20][21][22][23][24][25][26][27][28][29]. In particular, computational fluid dynamics (CFD) models have been used to compute turbulent flow dynamics and atmospheric pollutant dispersion within the urban surface layer. These models usually allow taking into account the morphological specificities of the urban environment, and consequently, they are able to simulate the flow dynamics perturbations caused by distinct urban obstacles [23][24][25][26][27][28][29]. Specifically, CFD models have been applied to assess the role of urban vegetation in mitigating air pollution, mainly considering the effects of trees induced by mechanical drag of trees and/or including the pollutant removal capacity of trees by deposition and filtration mechanisms [30][31][32][33][34][35][36][37][38]. Still, most of the numerical, physical and experimental studies have been performed over urban-like geometries, through idealized configurations and hypothetical scenarios [33]. Although the effects of green infrastructures on urban atmospheric dynamics have been widely studied, a deeper knowledge is still required regarding the overall perturbations induced by trees on turbulent flow dynamics and, consequently, air pollutant dispersion. Up to now, despite the several contributions from outdoor measurements, wind tunnel measurements, extensive database on idealized street canyons and several CFD studies, the current understanding of turbulent flow dynamics within and around vegetation is not sufficient, denoting local increases and decreases of wind speed and turbulence, depending on several parameters (e.g., characteristics of vegetation, urban morphology and meteorological conditions). Therefore, the first identified knowledge gap, which motivates this work, refers to the need to contribute to improve the knowledge of the effects of green infrastructures on turbulent flow dynamics at a local scale and, consequently, on air pollutant dispersion patterns.
Furthermore, in literature, the study of climate change impacts ranges from global scale down to the regional and urban scale [39,40] (e.g., in [40] the authors have performed a 10-year air quality projection under climate and city-level emission changes at urban scale applying a horizontal resolution of 4 km × 4 km). Nevertheless, the full understanding of the impact on air quality at a local scale is a challenge and the effects are still unknown. Another identified knowledge gap is the lack of studies focused on local-scale effects, with an insufficient understanding and accurate predictions of the changes in urban microclimate patterns and in pollutant concentrations under climate changes. Air pollutant dispersion is mainly driven by meteorological forces, and thus it is natural that changes in future climate will strongly affect air quality patterns, resulting, potentially, in an increase of the magnitude and frequency of air pollution episodes. Therefore, several issues arise concerning how the changes on global and regional circulations will affect local-scale urban microclimate, how to use GI as an adaptation measure to increase resilience to CC at local scale, and what is the link between the impact of GI under future, past and current climate.
This chapter aims to foster urban microclimate knowledge and to assess CC effects at a local scale, evaluating strategies of adaptation to CC toward an urban sustainable development. The chapter's main objective is to evaluate the impact of distinct resilience measures on the urban microclimate and air quality, in recent past, current and future climate scenarios for Porto's urban area. These resilience measures will be based on green infrastructures.

Numerical modeling approach
A cascade of numerical models is applied in this work, from global scale down to local scale (e.g., street canyon and surroundings) to assess the urban microclimate and air quality patterns.

From global to urban scale
A cascade of numerical models, from global to urban scale, was applied to the Greater Porto area. The numerical model Weather Research and Forecasting (WRF) Model version 3.7.1 [41] was applied to perform regional climate simulations for Porto city and surrounding areas through dynamical downscaling. WRF setup has included four domains on line nested with increasing resolution at a downscaling ratio of three: domain 1 at 27 km resolution covering West Europe, Atlantic Ocean and North Africa; domain 2 at 9 km resolution over the Iberian Peninsula; domain 3 at 3 km resolution over Portugal; and domain 4 with 1 km resolution over North West Portugal. All these domains were used to perform a representative simulation of the recent past climate  that was used as the reference simulation and a representative simulation of the future medium-term climate (2041-2070). For the future simulation, the Representative Concentration Pathway Scenario RCP8.5 has been adopted, from the new generation of climate change future scenarios included in the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [42]. The MPI-ESM-LR model was used to provide initial and boundary conditions to WRF model, with a spatial resolution of 1.9° and 47 hybrid sigma-pressure levels (Max Planck Institute for Meteorology Earth System Model) [43].

CFD numerical formulation
Pollutant DISpersion in the atmosphere under VAriable wind conditions (VADIS) is the CFD model applied to Porto urban area to assess local-scale flow dynamics and air quality. This CFD model was developed by the University of Aveiro to numerically simulate air pollutant dispersion in complex urban areas and under unfavorable wind conditions. VADIS consists of two distinct modules: the FLOW and DISPER.

Turbulent flow dynamics
FLOW is a Eulerian module able to simulate the turbulent flow dynamics under stationary conditions within the atmospheric boundary layer. This module numerically solves by means of finite differences (SIMPLE solver), over a Cartesian tridimensional grid, the Navier-Stokes (NS) equations for the wind velocity components, the turbulent viscosity, pressure and the turbulent kinetic energy, applying Reynolds averages (the so-called Reynolds-Averaged Navier-Stokes [RANS] approach). The equation of mass conservation is written as Eq. (1), while the NS equations for the momentum conservation are given by Eq. (2).
where ui represents each velocity component (u, v and w) and x i the spatial coordinates (x, y and z), P denotes the pressure, ρ is the air density, ν e is the effective viscosity and S u indicates the source and/or sink term for u i .
The computation of the NS equations is based on the mechanics of fluid fundamental laws, simplified through a set of considerations and completed with closure formulations, namely in the turbulence simulation. The equation system closure is made through the turbulent isotropic viscosity approach, calculated from the kinetic turbulent energy transport equations and their dissipation rate. The first-order k-ε closure scheme [44] is used to model the turbulent viscosity (ν t ), solving two additional equations for the turbulent kinetic energy (k), Eq. (3), and the dissipation rate (ε), Eq. (4).
where σ k , σ ε , c ε1 and c ε2 are closure constants set to the standard values of 1, 1.3, 1.44 and 1.92, respectively. S k and S ε are the additional source terms of k and ε.
VADIS model is suited for the simulation of the wind field and the turbulent viscosity affected by a set of obstacles (e.g., buildings) localized over the Cartesian grid. The CFD model VADIS was updated by [31] to consider the aerodynamic effect of urban vegetation, adding source terms to momentum (Eq. (5)), turbulent kinetic energy (Eq. (6)) and its dissipation rate (Eq. (7)) equations, calculated based on tree-airflow interactions, neglecting viscous drag relative to form (or pressure) drag [45,46].
where C d is the mechanical drag coefficient, LAD denotes the leaf area density and |U| is the magnitude of the wind speed vector.
S k denotes the production of turbulence by the action of vegetation elements.
where β p is set to 1 and indicates the fraction of the mean flow kinetic energy converted to wake-generated turbulent energy by canopy drag. β d is set to 4 and denotes the fraction of turbulent energy dissipated by "short-circuiting" of the Kolmogorov cascade [47,48] and β d |U|k points out the dissipation of the generated wakes [49]. Then, S k represents the sum of the source and sink of turbulent kinetic energy due to the effect of vegetation elements.
The numerical formulation of S ε is similar to the formulation of S k .
where c ε4 and c ε5 are both equal to 1.5. Therefore, VADIS model is suited for the calculation of the perturbations induced by vegetation elements on the flow dynamics and dispersion patterns.

Pollutant dispersion over complex geometries
The DISPER module uses the data provided by the previously mentioned module, namely the wind field, and estimates the tridimensional concentration field of the air pollutant dispersion, based on the Lagrangian Stochastic approach. This methodology assumes that the pollutant spatial and temporal dispersion is conveniently represented by a large number of numerical particles randomly released in the flow. Langevin equation (Eq. (9)) computes particle displacement at each time step by the sum of the deterministic component obtained from the wind velocity (u i ), the aleatory component (u i ') related with the local turbulence and the influence of the fluctuation forces. T L is the Lagrangian timescale.
The fundamental Lagrangian equation assumes that each particle is represented by a constant mass quantity. The mean concentration of the pollutant at point X and time t due to the contribution of point X 0 is represented by the probability of occurrence (the probability P(X 0 ,X,t) of a particle emitted at X 0 to be at point X and at time t) and the concentration at the original point. Then, the sum of the contributions of all points within the computational domain represents the mean concentration at point X and instant t. C(s) is written as [50]: The performance of VADIS model has been evaluated by comparison with wind tunnel data, other CFD models and in-situ measurements [31,[50][51][52].

Case study
Porto urban area is located in coastal northern Portugal bounded by the Atlantic Ocean and the Douro River. Porto is the second largest city in Portugal with a population of 237,591 inhabitants and the city center of Porto metropolitan area covering 1900 km 2 with more than 1.5 million inhabitants [53]. Porto metropolitan area is characterized by a fast urban expansion with dense residential and commercial areas and a few and sparse green areas. Porto city features a temperate Atlantic climate with warm and dry summers and mild rainy winters [54]. The urban area is surrounded by two metropolitan rings, each one with important links to Porto metropolitan area, in terms of mobility, with high rates of emissions from road traffic. Porto urban area has been selected as the case study in the framework of CLICURB project-"Urban atmospheric quality, climate change and resilience." The main objective of CLICURB project was to improve the scientific knowledge about urban microclimate and to bridge the gap between global climate change trends and urban development considering the inclusion of adaptation strategies on urban planning and decision-making processes. This research project is aimed to identify a set of measures suited for the increase of urban resilience, including the quantification of their effectiveness to mitigate climate change impact. CLICURB project intended to assess the impact of future climate on urban areas at different levels, e.g., meteorology, energy fluxes and air quality. The most relevant outcome of CLICURB was the production of an urban atlas, both for present and mid-twenty-first century, consisting of a series of layers for urban climate, thermal comfort, air pollutant emissions and air quality.

Computational domain
The modeling system was applied following three distinct steps, in order to assess the impact of green infrastructures on flow dynamics and air quality levels, evaluating their effectiveness as green urban planning strategies, under recent past, current and future medium-term climate. The first step focuses on the comparison of flow dynamics between wind tunnel and CFD simulation results, for both a baseline and a green scenario. The second step includes the comparison of CFD simulation results with PM10 concentrations measured at the air quality station located within the computational domain, for 24th September 2010 (this specific day was selected due to the data availability of traffic counting and meteorological data). The third step comprises the simulation of meteorological fields, from global down to urban scale (following the procedure described in section 2.1), as well as the CFD simulation at local scale of the urban microclimate and air quality under recent past climate and future medium-term climate. Figure 1 shows the selected study area within Porto city center. D1 points out the selected study area for step 1, while D2 denotes the selected study area used in step 2 and step 3. Figure 2 shows the simulation domain used in the wind tunnel and in the CFD simulations (step 1). The study area is centered in part of Rua da Constituição, a street canyon located in the city center of Porto. The baseline domain (Figure 2a) corresponds to the current morphological conditions of the study area, while the hypothetical green scenario (Figure 2b) corresponds to the replacement of two blocks of buildings for parks flanked by groups of trees.
A set of physical simulations were carried out using the wind tunnel facilities of the University of Aveiro, for the baseline and the corresponding green scenario (Figure 2), to assess the impact    Figure 2a and b. The atmospheric boundary layer was simulated using a specific setup of turbulence generators and floor roughness elements, located upstream the test section. The flow analysis was performed by measurements of the wind speed and turbulence, with a time resolution sampling rate of 0.5 s, using a hot-wire anemometer (TSI IFA-300).
Both the mock-up of the baseline and the green scenario were tested in the wind tunnel. The wind speed was measured at different locations, from points 1 to 8 indicated in  location of the traffic count devices (1-4 and 7), the air quality station, as well as the location of buildings, trees and roads, corresponding to the current morphological data of the study area. The green scenario is shown in Figure 3b and comprises the implementation of green areas in 35% of the current built-up area located in the Southeast part of the domain (pointed out by the red rectangles). Within the newly implemented green areas, the height of trees varies between 3 and 30 m. The overall roads keep the same hourly emission rate as for the baseline scenario.
In the third step, the CFD model is also applied to the area defined in the city center of Porto included in the computational domain D2.

Emissions from road traffic
The traffic roads are defined within the computational domain as line sources (see Figures 3  and 4). All these roads were considered as emission sources with an associated emission rate.
Hourly averaged PM10 emission rates from on-road transport vehicles were calculated with high temporal and spatial resolution using the Transport Emission Model for Line Sources (TREM). This model was developed at the University of Aveiro, based on MEET/COST319 methodology [50]. The hourly emission rates are estimated for each road segment, considering the local information on traffic counting data. Vehicle counting data was acquired using automatic devices installed in seven distinct locations (Figure 3a). Empirical rates expressing the relation with the traffic in the surrounding roads were applied in case of roads without available data.
In addition, the calculation of the emission rates are also based on emission factors following the European guidelines [55] and considering the fleet composition and the characteristics of the vehicles, such as age of vehicles, average speed, engine type, capacity and technology, vehicle weight and fuel consumption (e.g., diesel, gasoline or LPG). TREM algorithm applies an aggregation of vehicles by categories (e.g., passengers, light-duty vehicles, heavy-duty vehicles, busses, motorcycles and new-technology vehicles, such as hybrid and electric cars). For all vehicle categories, 350 classes were considered including EURO1 to EURO5 vehicles and depending on the characteristics of vehicles and their emission standards. Fleet data of the percentage of vehicles in each class were obtained from national databases [56,57].   Figure 4 shows the hourly averaged PM10 emission rates calculated for each road segment, in 68 segments. The figure points out the variation of PM10 emission rates depending on the road segment for two distinct hours characterized by low emission rates, at 4 am (Figure 4a), and high emission rates, at 8 am (Figure 4b).
The results correspond to a weekday (Friday, 24th September 2010) typical behavior of downtown traffic flow.
The main road in the domain is Avenida de Fernão Magalhães, constituted by five lanes, i.e., three on the right side and two on the left, from South to North direction, separated by a central row of trees. This main avenue is divided into four road segments, for the computation of the emission rates. The two parallel road segments with high emission rates are located in the intersections with the roundabout where the Air Quality Station is located. On the contrary, the low emission rates are registered in the road segments located in the Southeast neighborhood of the computational domain. The maximum emission rates registered at 4 and 8 am are equal to 9.2 and 60.2 g km −1 h −1 , respectively, while the minimum rates are equal to 0.12 g km −1 h −1 at 4 am and to 14.9 g km −1 h −1 at 8 am. The high emission levels are registered during the morning, after 8 am, until evening, while the low emission levels are estimated during the night.

Meteorological inflow boundary conditions
Several CFD simulations were performed over the computational domain D2, in the second step, for 24th September 2010, for both the baseline morphological and the baseline green scenarios. The meteorological inflow conditions for these CFD simulations were obtained from mesoscale simulations using WRF model [41]. The mesoscale simulations were conducted over Continental Portugal with a horizontal grid resolution of 5 km. The meteorological data for 24th September 2010 are presented in Figure 6, i.e., the hourly average wind speed (Figure 6a) and wind direction (Figure 6b). The hourly wind speed (Figure 6a) ranges from low to moderate wind speed conditions, between 0.1 m s −1 , during the night at 5 am, and 6.5 m s −1 , during the afternoon at 3 pm, while the wind direction (Figure 6b) varies from South-Southwest to North-Northeast, blowing predominantly from the 4th quadrant. The high wind speed is registered during daylight time, between 8 am and 6 pm. In addition, the friction velocity ranges from a minimum near zero to a maximum of 1 m s −1 . The aerodynamic roughness length was calculated in an hourly basis from the vertical wind profile and ranges from 0.8 to 2.2 m.

Meteorological data for recent past and future medium-term climate
The meteorological results in Figure 7 show the greatest reduction of wind speed in the first autumn months in the mid-term future climate, compared to the recent past climate. For the remaining seasons, the simulation results only show negligible variations with any clear trend. The simulation results show a slight decrease in the average number of days with moderate to strong wind speed or higher and a slight increase in the average number of days recording low wind speed conditions, in the future medium-term climate.
Temperature data show an annual average increase of about 1.1°C in the mid-term future climate. Figure 8 shows the minimum and maximum monthly values of temperature, together with the monthly average values. Monthly maximum temperatures increase almost every month. The highest anomaly records are projected for summer and autumn, e.g., in July (one of the hottest months), the anomaly points out an increase of 5°C. The monthly minimum temperatures also increase, again with the largest anomalies occurring in the summer and autumn time.
The number of heat waves in the mid-term future climate increases about seven times compared to recent past climate. The average of duration of the heat waves also increases and the days with the maximum temperature greater than 35°C increase 4 times. The results also show an increase in the number of summer days and tropical nights. Although for the recent past climate the greatest number of heat waves occurs during spring and summer time, in the mid-term future climate, these extreme events will occur in summer and autumn time.

Assessment of GI effects on flow dynamics using CFD and wind tunnel simulations
The effects of green infrastructures on flow dynamics are assessed by comparison between the baseline morphological scenario and the corresponding green scenario, in the first step of this study. Then, the accuracy of the numerical and physical results is evaluated through the comparison between both results and the estimation of statistical parameters. Figure 9 shows the measured and simulated data for the baseline and the green scenarios and for an inflow wind blowing from North with wind speeds of 3 and 9 m s −1 , as example.  In conclusion, both numerical and physical results point out an overall increase of wind speed and a good agreement between CFD simulations and wind tunnel measurements to predict this trend. Exceptions are found in point 3, in the wind tunnel results, and in point 7, in CFD simulations, where a decrease of wind speed is registered.   Figure 10 presents the measured and simulated data for an inflow wind blowing from West. The effects of GI are over again studied in terms of increase or decrease of wind speed registered in the green scenario, when compared to the baseline.

Impact of GI on flow dynamics
In the specific case of an inflow wind from West direction, point 1 is located at the inlet of the domain and upstream the newly implemented green areas. At this location, the wind tunnel results present negligible differences of wind speed for the green scenario compared to the baseline. However, at the same location, the CFD results show a more significant impact of the green scenario pointing out a decrease of 0.  In conclusion, at a location nearby the implemented green areas the wind speed tends to increase, both in wind tunnel and CFD simulations. CFD and wind tunnel results do not agree at point 8, with relevant increase of wind speed predicted by the measurements, and a slight decrease of wind speed simulated by the CFD model.

Assessment of CFD performance
The CFD performance is evaluated by comparison with the wind tunnel measurements, using the validation metrics, such as the normalized mean square error (NMSE) and the fraction of simulated values within a factor of two of the measured values (FAC2), based on the metrics formula proposed by [58]. The normalized mean square error obtained for each set of wind speeds and directions ranges between 0.3 and 1 (inflow wind from Southeast and West, respectively) in the baseline scenario, while in the green scenario, the obtained NMSE ranges from 0.1 to 0.7 (inflow wind from North and Southeast, respectively). The factor FAC2 ranges between 0.9, for an inflow wind from Southeast, and 1.7, for an inflow wind from North, in the baseline scenario, and between 0.6 and 1.2 in the green scenario (over again for an inflow blowing from Southeast and North, respectively).
The acceptance criteria establish a value less than 1.5 for the NMSE. The fraction of the simulated values within a factor of two of the measured values in the wind tunnel should comprise between 0.5 and 2, following the acceptance criteria. Therefore, we can conclude that the CFD simulation results and the measured data in the wind tunnel are in good agreement, based on the acceptance criteria, for each set of wind speeds and directions, as well as for both the baseline and the green scenarios. Consequently, the obtained validation metrics confirm the ability of the CFD model to simulate the perturbation exerted by the green infrastructures on the turbulent flow dynamics.
Finally, both numerical and physical results denote the effects of green infrastructures on flow dynamics. Consequently, the demonstrated effects on flow dynamics should affect the air quality patterns.

Assessment of GI effects on particulate matter dispersion
Several CFD simulations were performed to assess the impact of green infrastructures on flow dynamics and, consequently, on PM10 dispersion, corresponding to step 2 of this study. Furthermore, the comparison between the CFD simulation results and the PM10 concentrations measured at the air quality station will allow assessing the performance of the CFD model.  points out a disagreement between the PM10 concentrations, simulated by the CFD model and measured at the AQS, and the traffic dynamics, mainly during night period. In addition, PM10 concentrations point out an overall underestimation by the CFD model.

Assessment of CFD performance
The CFD performance is also evaluated using the validation metrics, i.e., the normalized mean square error and the fraction of simulated values within a factor of two of the measured values.
The normalized mean square error is equal to 2.1, and the factor FAC2 is equal to 0.3. Thus, the NMSE is above the maximum value, defined equal to 1.5 as acceptance criteria, and the FAC2 is lower than the minimum accepted value of 0.5. Therefore, the CFD simulation results denote a weakness of the CFD model to simulate the impact of green infrastructures on particulate matter dispersion, when only considering the particulate matter emissions from road traffic.

Impact of GI
The impact of green infrastructures on PM10 dispersion is analyzed in an hourly basis on 24th September 2010. Figures 12 and 13 present the differences of wind speed (Figures 12a  and 13a) and of PM10 concentrations (Figures 12b and 13b), between the green and the baseline scenarios. These figures present the absolute differences at 1.5 m high. In addition, both Figures 12 and 13 correspond to distinct inflow conditions in terms of emission rates; that is, Figure 12 shows the differences at 4 am, during night and with low emission rates. Figure 13 shows the differences at 8 am, in the morning and with high emission rates (the emission rates for each road segment are presented in Figure 4 for these 2 h, as example). In addition, the obtained maximum of PM10 concentration is equal to 13.6 and 29.6 μg m −3 , in the baseline and the green scenarios, respectively. The differences plotted in Figure 12a correspond to 38% of zero differences, 46% of positive differences, pointing out an overall increase of wind speed in the green scenario, and 16% of negative differences. In addition, we found 3% increases of wind speed of more than 0.1 m s −1 and 1% decreases of more than 0.1 m s −1 . Figure 12b denotes 69% of zero differences in PM10 concentrations, 15% of positive differences, and 16% of negative differences, indicating a similar number of occurrences of increases or decreases of PM10 concentrations. Furthermore, PM10 concentrations show 1% of increase above 0.1 μg m −3 and 0.5% of decrease below 0.1 μg m −3 .   Theoretically, increases of wind speed (ΔU > 0) in a green scenario will led to enhanced dispersion of particulate matter, with lower concentrations in the green scenario (ΔC < 0), compared to the baseline. On the contrary, a decrease of wind speed (ΔU < 0) in a green scenario will promote the retention of pollutants, with an increase of PM10 concentrations (ΔC > 0). However, for positive differences of wind speed, at 4 am, we have found 13% of positive and 10% of negative differences of PM10 concentrations (plus 19% of zero differences of concentrations), while for negative differences of wind speed, we have found 6% of positive and 5% of negative differences of PM10 concentrations (plus 8% of zero differences). Zero differences of wind speed mostly correspond to zero differences of concentrations (36%), with negligible positive and negative differences (~3%). At 8 am, positive differences of wind speed led to 7% of positive and 7% of negative differences of PM10 concentrations (plus 32% of zero differences of concentrations), while for negative differences of wind speed, we have found 2% of positive and 2% of negative differences of PM10 concentrations (plus 12% of zero differences). Zero differences of wind speed mostly correspond to zero differences of concentrations (25%), with a total of positive and negative differences equal to 14%. These results point out a nonlinear relationship between the wind speed and consequent pollutant dispersion patterns.
In conclusion, the overall effects of green infrastructures on air quality show that local air quality is strongly dependent on the linkages between meteorological conditions, the urban morphology and the emission rates.
A new set of CFD simulations were performed to assess the impact of green infrastructures on flow dynamics and, consequently, on PM10 dispersion, under recent past and medium-term future climate, corresponding to step 3 of this study. The climate results presented in section 3.4 pointed out a decrease in the number of days with moderate to strong wind speed and an increase of low wind speed conditions, unfavorable conditions to air pollutant dispersion. Thus, CFD simulations under climate change were performed for the four cardinal directions, north, east, south and west, and the four intercardinal directions, northeast, southeast, southwest and northwest, for inflow wind speed representative of low wind speed conditions, i.e., 1 m s −1 , and for strong wind conditions, 6 m s −1 . The set of emission rates for each road segment obtained at 4 am, as example of low emission rates, and at 8 am, as example of high emission rates, was used as inflow conditions.
The effects of green infrastructures depend on inflow meteorological conditions and emission rates. Considering only the changes on the flow dynamics in medium-term climate, i.e., changes almost negligible in wind speed, the assessment of the effects of green infrastructures procedure is similar to the one performed previously to the baseline climate (in 2010). The obtained conclusions for the baseline climate are also valid for medium-term future climate. The advantages of GI found in Figures 12 and 13, as well as the disadvantages, can be linearly predicted to medium-term climate.

Conclusions
The increased occurrence of extreme weather events and air pollution episodes, because of climate change, can cause a wide range of impacts on society, economy and environment.
This chapter presents a CFD model as a valid tool to assess climate change effects at local scale, and the role of green infrastructures as adaptation measures to increase resilience of urban areas. The CFD model presents a good performance for the simulation of flow dynamics. However, the model presents some weakness to simulate particulate matter dispersion. Therefore, the understanding of the influence of turbulent flow dynamics and the distinct emission sources contributing to fine particulate matter pollution still present a set of uncertainties. Furthermore, the most relevant conclusion of this analysis is the lack of understanding of the overall effects of GI on flow dynamics and pollutant dispersion. Future efforts should focus on the improvement of our understanding of the overall perturbations exerted by urban vegetation on the urban boundary layer.
Climate-driven changes in meteorology may also modify the natural emission patterns. In addition, future climate will also affect human activities, leading to an impact over the anthropogenic emissions. These key parameters should be taken into account, and as ongoing work on future projections of emissions, scenarios are being considered at local-scale simulations.
The results of this study present an important contribution to develop a set of adaptation measures, supported by scientific-based knowledge, from urban to district and street level, considering the geographical, demographical, economic and environmental characteristics of each urban area.