UNIVERSIDAD COMPLUTENSE DE MADRID FACULTAD DE CIENCIAS FÍSICAS TESIS DOCTORAL Modelling the wind flow and dispersion of reactive pollutants in cities at microscale MEMORIA PARA OPTAR AL GRADO DE DOCTORA PRESENTADA POR Beatriz Sánchez Sánchez Directores José Luis Santiago del Río Magdalena L. Palacios Gómez Madrid, 2019 © Beatriz Sánchez Sánchez, 2017 FACULTAD DE CIENCIAS FÍSICAS TESIS DOCTORAL MODELLING THE WIND FLOW AND DISPERSION OF REACTIVE POLLUTANTS IN CITIES AT MICROSCALE SIMULACIÓN A MICROESCALA DEL FLUJO ATMOSFÉRICO Y DISPERSIÓN DE CONTAMINANTES REACTIVOS EN CIUDADES Beatriz Sánchez Sánchez Directores: José Luis Santiago del Ŕıo Magdalena L. Palacios Gómez Tutor: Gregorio Maqueda Burgos Mayo 2018 Agradecimientos Me gustaŕıa agradecer brevemente a todas las personas que me han ayudado en la realización de la tesis, y a las que estoy tremendamente agradecida porque, de una forma u otra, siempre me han brindado su apoyo para que consiguiera alcanzar mi objetivo. Al Dr. José Luis Santiago por formarme en el ámbito de la modelización, por guiarme en el desarrollo de la tesis, por su valioso trabajo en la revisión de las múltiples versiones de la misma y, sobre todo, por su paciencia y apoyo constante. A la Dra. Magdalena Palacios por introducirme en el campo de la qúımica atmosférica, por sus sugerencias y consejos a la hora de elaborar la tesis aśı como, por su apoyo y confianza. Al Dr. Alberto Martilli por su continua disponibilidad para resolver mis dudas e inquietudes, por su ayuda en la discusión de los resultados, por sus conocimientos transmitidos y su valiosa contribución a lo largo de este trabajo. Al Dr. Gregorio Maqueda por introducirme en el campo de la micrometeoroloǵıa y por animarme a iniciar mi camino en el mundo de la investigación, por su interés y labor a lo largo de todo este proceso. Al Dr. Fernando Mart́ın por acogerme en el grupo de modelización y confiar en mi trabajo. A la Dra. Marta G. Vivanco por su disposición a ayudarme siempre que lo he necesitado. También me gustaŕıa agradecer a mis compañeros del grupo de Modelización y Ecotoxidad del CIEMAT por sus continuos ánimos, especialmente a Juan Luis Garrido y Esther Rivas, por animarme siempre cuando créıa que no iba a acabar nunca. Y por supuesto, por todos esos cafés que siempre estábais dispuestos a tomaros conmigo para ayudarme desconectar. Agradecer al CIEMAT por la financiación de este trabajo a través de los proyectos LIFE MINOx-STREET (LIFE12 ENV/ES/000280) y TECNAIRE-CM (S2013/MAE-2972). Me gustaŕıa agradecer a mis compañeros del grupo de Caracterización de la Contaminación Atmosférica y COPs por facilitarme todos los datos necesarios para realizar la validación de las simulaciones. i Dar las gracias también al grupo de investigación en Tecnoloǵıas Ambientales y Recursos Industriales (TARINDUSTRIAL) de la Universidad Politécnica de Madrid, en particular a Christina Quaassdorff, David de la Paz y al Dr. Rafael Borge, por su disponibilidad para discutir los resultados y proporcionarme todos los datos que he necesitado para realizar las simulaciones. Me gustaŕıa agredecer al grupo de Micrometeoroloǵıa y Variabilidad Climática de la Universidad Complutese de Madrid por facilitarme los datos meteorológicos usados en la validación de las simulaciones. Agradecer al Centro Extremeño de Tecnoloǵıas Avanzadas (CETA-CIEMAT) por permitirme usar sus recursos computacionales para realizar todas las simulaciones que constituyen este trabajo. También me gustaŕıa agradecer el apoyo y ánimos que siempre he recibido de ellas durante este proceso, ¡gracias amigas! Finalmente, quiero agradecer inmensamente a mis padres y a mi hermana que siempre me hayan apoyado y animado para que consiguiera todo lo que me propusiera. Y a mi pareja, que, gracias a él, todo siempre es más fácil. Resumen La calidad del aire es, hoy en d́ıa, una de las principales preocupaciones ambientales en las ciudades, donde se concentra la mayor parte de la población y se registran niveles altos de contaminantes, especialmente debido al tráfico. Los NOx (óxidos de nitrógeno), es decir, NO (óxido ńıtrico) y NO2 (dióxido de nitrógeno), son unos de los contaminantes más importantes asociados asociados a las emisiones del tráfico rodado. Sin embargo, sólo está legislado el NO2, por ser perjudicial para la salud. A pesar de las poĺıticas medioambientales adoptadas en los últimos años, los valores ĺımite establecidos por la Unión Europea (EU, en inglés) y la Organización Mundial de la Salud (WHO, en inglés) son ampliamente superados a lo largo de Europa, especialmente en las zonas urbanas. Por este motivo se necesita llevar a cabo estudios exhaustivos, que ayuden a gestionar y planificar estrategias de mitigación para reducir la contaminación atmosférica urbana. La complejidad de los procesos atmosféricos en las zonas urbanas dificulta la evaluación de la calidad del aire, tanto a través de técnicas de monitorización como de modelos numéricos. El gran número de obstáculos existentes en la ciudad (edificios, vegetación, veh́ıculos...) generan complejas distribuciones del flujo de viento en las calles. Esto unido a la heterogeneidad de las emisiones a lo largo de la ciudad, producen fuertes gradientes de concentración. Por esta razón es necesario resolver con alta resolución espacial los fenómenos atmosféricos urbanos. En este sentido, los modelos de mecánica de fluidos (CFD, en inglés), son herramientas potentes capaces de reproducir en detalle la dispersión de contaminantes en zonas urbanas. Por contra, estas simulaciones requieren un elevado número de puntos de malla (varios millones), que conlleva un alto coste computacional. Para ese motivo, este tipo de modelos requiere millones de celdas, que permitan resolver adecuadamente los fenómenos atmosféricos urbanos. Por lo tanto, para llevar a cabo una simulación CFD es necesario disponer de altos recursos computacionales. El objetivo de este trabajo es simular a microescala el flujo de viento y la dispersión de contaminantes reactivos en escenarios urbanos mediante un modelo CFD. La incorporación de reacciones qúımicas a la simulación es un gran reto en términos de modelización numérica y de recursos computacionales. En este estudio se han implementado las reacciones iii qúımicas gaseosas más relevantes que se producen en la atmósfera urbana durante el periodo diurno. Para ello se tienen en cuenta dos mecanismos qúımicos: el mecanismo fotoestacionario (reacciones NOx−O3) y un mecanismo más complejo que incluye las reacciones con compuestos orgánicos volátiles (VOC, en inglés) (reacciones NOx−Ox−VOC). En este trabajo se evalúa la importancia que tiene considerar reacciones qúımicas para simular adecuadamente la dispersión del NO2, tanto en geometŕıas simplificadas como en escenarios reales urbanos. También se estudia el impacto que tendŕıa incluir las reacciones qúımicas con VOC para simular la concentración del NO2 en la calle. Las transformaciones qúımicas dependen, a través de las constantes de reacción, de las variables meteorológicas, como la radiación solar y temperatura del aire. No obstante, existen otros aspectos caracteŕısticos de las zonas urbanas, como las emisiones de tráfico o la ventilación de la calle, que también afectan a la diferencia que resulta de considerar o no reacciones qúımicas en el modelo. Por este motivo, se analiza la dispersión de contaminantes reactivos en diferentes condiciones atmosféricas, estudiándose, tanto temporal como espacialmente, el acoplamiento de los procesos dinámicos y qúımicos a lo largo del periodo diurno. Los resultados muestran cómo la turbulencia y el equilibrio qúımico juegan un papel importante en la diferencia de concentración debida a las interacciones qúımicas. Por último, se presenta una aproximación numérica, basada en el uso de modelos a microescala y mesoescala, que permite obtener mapas detallados de concentración para largos periodos de tiempo. De esta manera, se ampĺıa el alcance del uso de los modelos CFD para evaluar la calidad del aire en las ciudades. Este sistema consiste en utilizar los resultados de la simulación a mesoescala para proporcionar la variabilidad temporal de las condiciones atmosféricas a la simulación de microescala. Esto permite obtener la distribución representativa de los contaminantes en cualquier parte de la ciudad. Por lo tanto, este trabajo contribuye principalmente a comprender el comportamiento de los procesos qúımicos dentro de la ciudad, teniendo en cuenta la variabilidad temporal y espacial de las condiciones atmosféricas. Los resultados obtenidos muestran diferencias significativas en la concentración del NO2 al incluir reacciones qúımicas durante el periodo diurno. El alto coste computacional que requiere incluir estos procesos en la simulación acota el uso del modelo CFD a periodos cortos de tiempo (varias horas). No obstante, estas simulaciones son de gran utilidad para analizar estrategias concretas de reducción de la contaminación en zonas urbanas. Concretamente, para estudiar las condiciones atmosféricas que podŕıan propiciar superaciones de los valores ĺımite horarios. Finalmente, para evaluar largos periodos de tiempo, la aproximación multiescalar de modelos desarrollada proporciona mapas fiables de concentración de contaminantes, necesarios para la gestión de calidad del aire en zonas de alta contaminación. Abstract Air quality is nowadays a priority environmental concern in the cities. High pollution levels have been often observed in urban areas, where both emissions and population are concentrated. The NOx, namely NO and NO2, are the primary traffic-related pollutants in gas-phase, but only the NO2 is legislated given that it is the harmful to human health. Despite the environmental policies adopted in the recent years, the limit values established by the European Union (EU) and the World Health Organization (WHO) are widely exceeded across Europe, especially in urban areas. Accordingly, exhaustive studies are required to manage and plan mitigation strategies for the abatement of urban air pollution. The complexity of atmospheric processes in urban environments hinders the assessment of air quality in the streets using both monitoring techniques and numerical models. The vast number of obstacles found in an urban area (buildings, vegetation, vehicles...) lead to complex flow patterns in the streets. Additionally, the heterogeneity of pollutant emissions, mainly from road traffic, across the city, gives rise to strong gradients of concentration. Hence the irregular distribution among buildings can only be captured by maps of high spatial resolution. In this sense, the Computational Fluid Dynamics (CFD) models are powerful tools for reproducing the dispersion of pollutants considering realistic features of urban environments. This work focuses on simulating the wind flow and dispersion of reactive pollutants in urban environments, at microscale, using a CFD model. The implementation of chemical reactions into the CFD simulations over urban areas faces a big challenge in terms of numerical modelling and computational resources. In this study, the relevant chemical reactions in gas-phase that take place in the urban atmosphere at daytime are implemented in the model. To do that two chemical mechanisms are considered: the photostationary state (NOx−O3 scheme) and a condensed mechanism including Volatile Organic Compounds (VOC) (NOx−Ox−VOC scheme). The influence of the chemical reactions by modelling the NO2 dispersion at street level is extensively analysed in simplified and real urban scenarios. Moreover, the impact of the VOC reactions on the NO2 concentration is compared with the use of the NOx−O3 v mechanism. The reaction rates are closely dependent on the meteorological parameters like the solar radiation and air temperature. Nonetheless, other urban factors, such as traffic emissions or wind flow patterns, bring about variations in the concentration of reactive pollutants. For that reason, the dispersion of reactive pollutants is, temporally and spatially, analysed in urban scenarios under numerous atmospheric conditions. As from this substantial study, the coupling of dynamical and chemical processes show how the turbulence and the chemical equilibrium in the streets can play an important role in the concentration difference caused by chemical conversions. Finally, in an attempt to improve the CFD modelling, a multiscale system is undertaken using the outputs derived from a mesoscale model to provide time-dependent boundary conditions to the microscale model. This system allows to compute the distribution of pollutant concentrations in the streets with high spatial resolution in any part of a city. This comprehensive study contributes to a better understanding of the coupled behaviour of dynamical and chemical processes in urban environments, taking into account the temporal and spatial variability of the atmospheric conditions. The implementation of chemical reactions in the CFD model improves the accuracy of the simulated concentration of reactive pollutants in the streets. Due to the increase of computational time by including chemical reactions, the CFD model are used to reproduce the temporal variation of pollutant dispersion in a short period of time (several hours). It is especially useful for simulating the detailed distributions of the NO2 concentration under unfavourable atmospheric conditions, in which the hourly limit values are exceeded. Moreover, the modelling approach used in this study provides reliable concentration maps for long periods of time (e.g. annual average concentration) demanded for the air quality management in urban hot-spots. Contents Agradecimientos i Resumen iii Acknowledgements iii Abstract v List of symbols xi List of acronyms xv 1 Introduction and research motivation 1 1.1 Urban air pollution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Urban Boundary Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.4 Gas-phase chemistry in urban environments . . . . . . . . . . . . . . . . . . 8 1.5 Management of urban air quality . . . . . . . . . . . . . . . . . . . . . . . . 11 1.6 Research motivation and objectives . . . . . . . . . . . . . . . . . . . . . . . 13 2 The CFD-RANS model 17 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.2 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.2.1 The Reynolds-averaged Navier-Stokes model . . . . . . . . . . . . . . 20 2.2.2 The k -ε turbulence model . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2.3 The Boussinesq approximation . . . . . . . . . . . . . . . . . . . . . . 23 2.3 Chemical Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.1 The NOx−O3 scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.2 The NOx−Ox−VOC scheme . . . . . . . . . . . . . . . . . . . . . . . 26 2.4 Modelling reactive pollutants . . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.4.1 Transport equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 vii viii CONTENTS 2.4.2 Box model validation . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.4.3 Chemical and turbulent time scales . . . . . . . . . . . . . . . . . . . 32 2.5 Modelling urban vegetation . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.6 Features on modelling urban areas . . . . . . . . . . . . . . . . . . . . . . . . 34 2.6.1 Geometry, mesh and boundary conditions . . . . . . . . . . . . . . . 35 2.6.2 Modelling atmospheric inflow conditions . . . . . . . . . . . . . . . . 36 2.6.3 Road traffic emissions . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2.7 Statistical model evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3 Modelling dispersion of reactive pollutants in simplified geometries with the chemical schemes: NOx−O3 and NOx−Ox−VOC 43 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.2 CFD model description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.2.1 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.2.2 CFD model evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.3 Impact of the atmospheric conditions on the dispersion of reactive pollutants 49 3.3.1 Description of the flow patterns . . . . . . . . . . . . . . . . . . . . . 51 3.3.2 Ozone influence on reactive pollutants . . . . . . . . . . . . . . . . . 52 3.3.3 Wind speed influence on reactive pollutants . . . . . . . . . . . . . . 57 3.4 Vertical transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3.5 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4 Modelling the dispersion of reactive pollutants in an urban area 65 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.2 Study case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.3 CFD model description and simulation setup . . . . . . . . . . . . . . . . . . 69 4.3.1 Geometry and mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 4.3.2 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 4.4 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.4.1 Time series at the sampling points . . . . . . . . . . . . . . . . . . . 71 4.4.2 Daytime mean concentration at the sampling points . . . . . . . . . . 76 4.5 Difference between chemical approaches at pedestrian level . . . . . . . . . . 78 4.5.1 Daytime behaviour of NO and NO2 at pedestrian level . . . . . . . . 78 4.6 Chemical and dynamical processes in an urban area . . . . . . . . . . . . . . 82 4.7 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 5 Modelling approach for air quality assessment in an urban hot-spot in winter conditions 89 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 CONTENTS ix 5.2 Experimental Campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 5.3 Geometry, mesh and boundary conditions . . . . . . . . . . . . . . . . . . . 92 5.4 Daytime behaviour of NO and NO2 . . . . . . . . . . . . . . . . . . . . . . . 96 5.4.1 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 5.4.2 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 5.4.3 Impact of the chemical approach on concentration at pedestrian level 100 5.5 Modelling approach for urban air quality assessment . . . . . . . . . . . . . . 105 5.5.1 Modelling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 5.5.2 Evaluation of the modelling approach against measurements . . . . . 108 5.5.3 Validation with passive samplers . . . . . . . . . . . . . . . . . . . . 113 5.6 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 6 Concluding remarks 119 6.1 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 6.2 Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Publications 125 Appendix A. Relation between modelled concentration and wind speed 127 Appendix B. Modelling the NOx−O3 and the NOx−Ox−VOC mechanisms in an urban area 129 Appendix C. Modelling approach in summer conditions 133 Appendix D. Multiscale system of WRF-CMAQ and CFD models in an urban hot-spot 137 References 143 x CONTENTS List of Symbols AEm Emission area Astreet Area of streets Aroof Area of buildings roof Cback Background concentration Clocal Local concentration Cp Specific heat at constant pressure Ctot Total concentration Cd,veg Drag coefficient of vegetation Cε Model constant for ε equation Cµ Model constant d Zero-plane displacement D Molecular diffusivity Da Damkhöler number Dh Dissipation function g Acceleration due to gravity Gk Production of turbulent kinetic energy hcanop Canopy height H Building height H0 Sensible heat flux at the surface Hmax Tallest building height iu Turbulence intensity JNO2 NO2 photolysis constant k Turbulent kinetic energy kNO+O3 Rate chemical constant for the NO + O3 reaction Kt Turbulent or eddy diffusivity FT Total flux Fm Mean flux Ft Turbulent flux xi xii LIST OF SYMBOLS l Large-eddy scale L Obukhov length Lref characteristic lenght scale for Re computation NOT NO concentration as non-reactive pollutant NO2T NO2 concentration as non-reactive pollutant NOR NO concentration as reactive pollutant NO2R NO2 concentration as reactive pollutant p Pressure Prt Turbulent Prandtl number QNO Source emission rate for NO QNO2 Source emission rate for NO2 Re Reynolds number Ri Richardson number S Mean strain-tensor Sem Emission flux at surface Sh Source or sink of heat Sk Source or sink of turbulent kinetic energy SM Source or sink of momentum Straff Traffic emission scenario Sveg,ui Sink of momentum for vegetation modelling Sveg,k Sink of k for vegetation modelling Sveg,ε Sink of ε for vegetation modelling Sφ Source or sink of a scalar variable Sct Turbulent Schmidt number t Time ts Time step T Temperature T0 Reference temperature U Mean velocity component u Instantaneous velocity u′ Turbulent velocity component u∗ Friction velocity uτ Reference velocity (cyclic simulations) W Building width Wsect Wind direction sector z Height above a reference plane z0 Roughness length LIST OF SYMBOLS xiii αh Thermal diffusivity β Thermal expansion coefficient βp Constant for vegetation modelling βd Constant for vegetation modelling ε Turbulent dissipation rate η Small-eddy scale δij Kronecker delta δNO Relative differences from non-reactive approach for NO δNO2 Relative differences from non-reactive approach for NO2 θ Potential temperature θ∗ Scale temperature θref Reference temperature κ Von Karman constant λp Packing density δeq Chemical equilibrium deviation µ Dynamic viscosity µt Turbulent or eddy viscosity ν Kinematic viscosity νt Turbulent or eddy kinematic viscosity ρ Air density σk Turbulent Prandtl number for turbulent kinetic energy σε Turbulent Prandtl number for turbulent dissipation energy σθ Turbulent Prandtl number for temperature τ Shearing stress τ0 Surface shear stress or drag τturb Turbulent time scale τreact Chemical time scale ϕ Zenith angle φ Scalar variable Φ Mean scalar component φ Turbulent scalar component ψ M-O similarity function for normalized velocity xiv LIST OF SYMBOLS List of acronyms ABL Atmospheric Boundary Layer AADT Annual Average Daily Traffic AGL Above Ground Level BEP Building Effect Parametrization BEM Building Energy Model CBM-IV Carbon Bond Mechanism IV CCM1/2 NOx−O3−VOC scheme with emission VOC/NOx=1/2 CCM1/5 NOx−O3−VOC scheme with emission VOC/NOx=1/5 chemBM Chemical Box Model CHEMATA CHEmical Mechanism Adaptation to Tropospheric Applications CFD Computational Fluid Dynamics DNS Direct Numerical Simulation EU European Union EEA European Enviromental Agency FAC2 Factor 2 FB Fractional Bias LAD Leaf Area Density LES Large Eddy Simulation LST Local Solar Time MCM Master Chemical Mechanism NMSE Normalized Mean Square Error PDT Plames-type passive diffusion tubes R Correlation coefficient RACM Regional Atmospheric Chemistry Mechanism RAMS Regional Atmospheric Model System RANS Reynolds-averaged Navier-Stokes RSM Reynolds Stress Model RNG Re-Normalisation Group xv xvi LIST OF ACRONYMS PBL Planetary Boundary Layer PSS Photostationary state UBL Urban Boundary Layer UCL Urban Canopy Layer VOC Volatile organic compound WA CFD-RANS Weighted Average CFD-RANS methodology WHO World Health Organization WRF Weather Research and Forcasting WRF-CHEM Weather Research Forecasting-CHEMistry Chapter 1 Introduction and research motivation 1.1 Urban air pollution Air pollution is nowadays an important environmental and social issue that involves a large and complex challenge in terms of management and mitigation of harmful pollutants (EEA, 2017). The air pollutants can be directly emitted from both anthropogenic and natural sources, called primary pollutants, which in turn chemically react in the atmosphere forming secondary compounds. High pollution levels have often been observed in urban areas, where both emissions and population are concentrated. Nevertheless, the pollutant emitted in cities can be transported and travel long distances, affecting large areas as a result. Not only has air pollution significant impact on long- and short-term health-related issues in the population, but also, it has an important economic repercussion, e.g. in the increase of medical costs. The most harmful pollutants to human health are the particulate matter, nitrogen dioxide (NO2) and ground-level ozone (O3). In 2014, in the European Union (EU), the estimates of the effects in health attributable to air pollution exposure were about 399 000 premature deaths, caused by the long-exposure to the fine particulate matter (PM2.5). Moreover, around 75 000 and 13 600 deaths per year were associated to the exposure to the NO2 and O3 respectively (EEA, 2017). Effective air quality policies call for the cooperation at global, national and local scales in order to act against air pollution. The EU and the World Health Organization (WHO) established specific thresholds for several pollutants, and thus improve air quality and prevent the increasing effects on human health. For that matter, the WHO and EU estimate that the hourly limit value of NO2 is 200 µg m−3, while the annual average value is 40 µg m−3. Even so, between 2013 and 2015, around 8% of the urban population are exposed to values of NO2 above the established threshold. Moreover, the annual limit value of NO2 continues to be widely exceeded across Europe, in around 10% of all the monitoring stations recording 1 2 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION concentrations in 2015 (EEA, 2017). Concerning the hourly limit, it was also exceeded at monitoring stations of some countries, being Spain one of them. The procedure for reducing the impact of air pollution demands extensive knowledge of its causes. Therefore, more detailed studies are required to face the high levels of harmful pollutants recorded at monitoring stations in an attempt to establish suitable mitigation measures. Accordingly, a comprehensive study of the dynamical and chemical processes that take place in the urban atmosphere and affect air pollution is essential. 1.2 Urban Boundary Layer The atmosphere can be classified according to the scales of motions, both spatially and temporally, into broad categories; microscale, mesoscale and macroscale, which are also commonly referred to as local, regional and global scales, respectively (Arya, 2001). Therefore, the field of meteorology that deals with the smallest processes in the troposphere is called micrometeorology. Figure 1.1 illustrates the scales of the atmospheric phenomena and the reference time scales for each category. Figure 1.1: Time and spatial scales of chemical and dynamic processes in the urban atmosphere. Source: Salmond and McKendry (2009). Micrometeorology focuses on the phenomena developed in the atmospheric layer closest to the earth’s surface, known as the Planetary Boundary Layer (PBL) or Atmospheric Boundary CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 3 Layer (ABL). Thus, the PBL can be defined as the part of the troposphere that results from the interaction with the underlying surface. Hence the atmospheric processes in the PBL are primarily dominated by the exchanges of heat, momentum and mass between the atmosphere and the earth’s surface. The thickness of the PBL is fairly variable in time and space due to its entire dependence on several factors such as the wind strength, the surface roughness or the heating and cooling of the surface (Stull, 2012). Therefore, the variation of the PBL height is linked to atmospheric phenomena at different scales; from the synoptic weather conditions, like high or low pressure systems to the surface energy balance resulted from the interaction of the lowest layers with the surface features (Arya, 2001). Moreover, its temporal variability is closely related to the diurnal cycle as a result of the heating and cooling of the land surfaces, produced by solar radiation. The height and structure of the PBL can vary depending on the topography or the land use of the underlying surface resulting in the surface drag (τ0/ρ). Over an urban area, the flow features in the PBL are modified by the great number of obstacles comprising a city. The buildings and other structures considerably increase the surface roughness and modify the surface-energy balance as well. Accordingly, the lowest part of the atmosphere over an urban area is also known as the Urban Boundary Layer (UBL). Figure 1.2 displays a cross section of the scales in the UBL, which is divided into the Mixing Layer and the Surface Layer. In turn, the latter involves the Inertial Sublayer and the Roughness Sublayer, in which the Urban Canopy Layer is found. Figure 1.2: Scales of the Urban Boundary Layer. Modified from Oke (1997). 4 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION The Inertial Sublayer is a more homogeneous area, where flows are well developed and pollutants are expected to be well mixed. In air pollution, it is commonly referred to as the mixing height or depth, because the emitted pollutants are blended here, giving rise to a strong contrast with the free atmosphere (Arya, 2001). The Roughness Sublayer is characterised by intermittent motions, where the flow is still adjusting to the effects produced by the perturbations found on the surface (Britter and Hanna, 2003). And lastly, below the mean height of buildings is defined the lowest atmospheric layer, known as the Urban Canopy Layer (UCL). The processes in the lower atmosphere are strongly influenced by the drag force exerted by urban obstacles. Moreover, the energy balance is significantly affected by the heat changes brought about by the activities associated with urban areas. The additional heat fluxes and the heat storage of the urban surface trigger the widely known urban heat island phenomenon (Britter and Hanna, 2003). The daytime behaviour of the UBL is largely based on the physical processes caused by the wind shear and the buoyancy effects in the city. The vertical gradients of temperature tend to decrease in the roughness sublayer and reach almost zero above it, giving rise to unstable atmospheric conditions. In contrast, the UBL at night comes down on account of the strong stability that suppress the turbulent vertical mixing (Arya, 2001). Therefore, the turbulence processes together with the heat exchanges between surface and atmosphere lead to a thermal stratification of the surface layer, which is subsequently transferred to the entire mixing layer. The Monin-Obukhov Similarity Theory established an approach to describe the mean flow profiles in a horizontally homogeneous surface layer, by means of the atmospheric stability based on the scale variables of the length, velocity and temperature. The scale velocity is the so-called friction velocity (u∗), related to the surface drag through τ0/ρ ≡ u∗ 2. The surface kinematic heat flux (H0/ρCp) is expressed as H0/ρCp ≡ u∗/θ∗, where θ∗ is the scale temperature; H0, the sensible heat flux at the surface; ρ, the air density and Cp, the specific heat at constant pressure. Accordingly, the stability parameter (z/L) is calculated on the length scale L known as Monin-Obukhov length, which is defined as follows: L = − u∗ 3 κ(g/T0)(H0/ρCp) (1.1) where κ is the von Karman constant (0.4). g is the gravity and T0, the reference temperature and, thereby, g/T0 the buoyancy variable. In this way, the ratio z/L provides the relative importance of the buoyancy forces versus the shear effects. Therefore, positive values of z/L stand for unstable atmospheric conditions whereas negative values, do it for stable conditions. CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 5 Finally, the wind profiles over a homogeneous surface layer can be described by the following logarithmic vertical profile: u(z) = u∗ κ ln ( z z0 + ψ(z/L) ) (1.2) where κ is the Von Karman constant (0.4) and z0 the roughness parameter (Arya, 2001). The expression ψ(z/L) is the dimensionless function of the Monin-Obukhov theory, that is zero in neutral atmospheric conditions. In the Inertial Sublayer over urban areas, the atmospheric processes might be adapted to the standard expressions described above (Stull, 2012). The average shear stress over the urban surface gives rise to the friction velocity, which can be used to obtain both the wind and turbulence profiles. In neutral atmospheric conditions, taking into account the effect of buildings, the wind profile is properly defined through the displacement height (d) as follows: u(z) = u∗ κ ln ( z−d z0 ) . Figure 1.3 illustrates an approximation of the logarithmic velocity profile in urban environments. Figure 1.3: The spatially averaged mean velocity profile near an urban area. Source: Britter and Hanna (2003). Finally, the non-uniformity of buildings layout in urban areas lead to the motions in the UCL are always in a turbulent flow regime. 1.3 Turbulence The motions in the low atmosphere are defined by the mathematical expressions of the fundamental laws of conservation mass, momentum and energy of an incompressible 6 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION newtonian fluid. The continuity equation (Eq. 1.3) states that the divergence of velocity is zero at all times and at every point in the flow. The momentum equations (Eq. 1.4) establish that the rate of change of momentum for a fluid particle is equal to the sum of all the forces on the particle (Newton’s Second Law). The energy conservation is based on the First Law of thermodynamics that can be expressed by the temperature equation (Eq. 1.5) (Versteeg and Malalasekera, 2007). ∂ui ∂xi = 0 (1.3) ∂ui ∂t + uj ∂ui ∂xj = −1 ρ ∂p ∂xi + ν ∂2ui ∂xj∂xj + SM (1.4) ∂θ ∂t + uj ∂θ ∂xj = αh ∂2θ ∂xj∂xj +Dh + Sh (1.5) where ui is the component of velocity using the Einstein notation, where i and j is 1, 2 and 3 representing the x, y and z directions. θ is the potential temperature; t, the time; ν, the kinetic viscosity (ν = µ/ρ, where µ is the dynamic viscosity) and αh, the thermal diffusivity. SM is the source or sink of momentum that includes contributions due to the body forces, such as Coriolis. In the Equation 1.5, Dh stands for the dissipation function and Sh is the heat source term. The viscosity brings about the frictional resistance between the adjacent fluid layers (Arya, 2001). The resistance force is referred to as shearing stress (τij), which is proportional to the velocity gradient following the Newton relationship: τij = µ ∂ui ∂xj (1.6) In the lower atmosphere, most of the physical processes are involved in a turbulent flow regime with high Reynold number (Re). In the PBL, turbulent flows typically have Re ≈ O(107) (Vilà-Guerau de Arellano et al., 2004). The Re number is expressed as ULref/ν, where U and Lref are the characteristic velocity and the length scale of the mean flow. The turbulence generation might be caused either by the shear production that transforms the mean flow energy into turbulent kinetic energy (k) or by the buoyancy production due to the transformation of potential energy of unstable stratification into turbulent kinetic energy (similarly, stable stratification suppresses turbulence). The contribution of each process is determined by the Richardson number (Ri), expressed by the ratio of the buoyancy and shear terms. Ri = − g T0 ∂θ/∂z (∂u/∂z)2 (1.7) CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 7 The turbulent flow is characterised to be highly irregular and rotational, as well as dissipative, because k is continuously dissipated by viscosity. Therefore, flow variables exhibit intermittent variations both in time and space. These fluctuations represent the turbulence contribution, which are commonly represented applying the Reynolds decomposition (Reynolds, 1895). Thus, the instantaneous variable (φ) of a turbulent flow is defined as the sum of the mean component (Φ ≡ φ) and its time fluctuation (φ′), resulting in φ = Φ + φ′. The mean component is calculated using Φ = 1 ∆T ∫ ∆T 0 φ dt, where ∆T is the sampling time and, the time average of fluctuations is always zero (φ′ = 0). Figure 1.4 represents the turbulent fluctuation of velocity at a local point. Figure 1.4: Representation of the turbulent fluctuation of the velocity at a local point. Source: Versteeg and Malalasekera (2007). Therefore, the mean components provide information about the mean structure of the flow, and its fluctuation represents the turbulence structure. The statistical parameters calculated on the turbulent fluctuations report the exchanges processes in the turbulence structure. The variances of the simplest fluctuations of the velocity components (u′2, v′2, w′2) provide the turbulent kinetic energy through k = 1 2 (u′2 + v′2 +w′2). In addition, the covariances of time fluctuations (u′iu ′ j, u ′ iθ ′...) give rise to the turbulent fluxes of momentum, heat or mass. The momentum turbulent fluxes are the result of the covariances of the velocity fluctuations in all directions, τij = −ρu′iu′j, so-called Reynolds Stress tensor. These turbulent stresses have been reported to be much larger in magnitude than the corresponding mean viscous stresses (Arya, 2001). Turbulent flows are highly rotational and involve three dimensional vortex-like structures called eddies, which are characterised in a wide range of scales depending on the Re number. The characteristic parameter representative of the size of the turbulent eddies is defined by a length scale included in the Re number expression. In this way, two broad categories can be distinguished, the large-eddy scale (l) and the small-eddy scale (η). The former receives the most energy from the mean flow and l is comparable to its characteristic scale without 8 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION depending on the fluid properties. In contrast, the small-eddy scale is entirely dependent on the fluid viscosity and the energy dissipation rate (ε). From the Energy Cascade Hypothesis, the essential characteristic of the turbulence is based on a continuous transfer energy from the largest to the smallest scale of eddies. And, assuming the Kolmogorov Hypothesis, the small-scale length is exclusively dependent on η = ν3/4ε−1/4. 1.4 Gas-phase chemistry in urban environments The urban atmospheric composition is mainly dominated by the incoming air mass to the city, the local emissions associated to the human activities and the chemical processes that take place (Bloss, 2009). All together affect not only the primary emitted pollutants, but also the secondary species formed in short-timescale by chemical reactions. The key pollutants in the urban atmosphere are mostly the nitrogen oxides (NOx), namely NO (nitric oxide) and NO2, carbon monoxide (CO) and volatile organic compounds (VOC) and, also, the particulate matter. The concentration of these pollutants are basically dominated by exhaust emissions and the releases from industrial or domestic activities. Figure 1.5 shows the main sources of NOx, where the road transport sector is the largest contributor, and CO and VOC are mainly emitted from the commercial and industrial sectors, respectively (EEA, 2017). (a) (b) (c) Figure 1.5: Percentage of the source emissions of (a) NOx, (b) VOC (non-methane) and (c) CO. Source: EEA (2017). At local scale, the dynamical processes in urban environments play an important role in air composition due to the different time scales of chemistry and turbulence. The turbulent processes can be faster than the chemical reaction. Thus, only the rapid chemical conversions have influence on the pollutant concentrations in the streets. Hence, the reactivity of the compounds is primarily dependent on the turbulent processes, which dominate the mixing CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 9 of pollutants. It particularly occur close to the sources where the largest concentrations are found (Vilà-Guerau de Arellano et al., 2004). The chemical processes in an urban area are fundamentally related with the NO and NO2. Therefore, the dominant chemical mechanism is the photostationary state that is constituted by the following reactions: O + O2 + M −→ O3 + M (R1) O3 + NO kNO+O3−−−−−→NO2 + O2 (R2) NO2 + hν JNO2−−−→NO + O (R3) where JNO2 is the photolysis constant and kNO+O3 the reaction rate constant. M represents a molecule that absorbs the excess of energy without reacting and, thereby, stabilizes the formed ozone (O3) (Seinfeld and Pandis, 1998). This three-reaction system has timescales of a few minutes or less in daylight conditions in mid-latitude (Bloss, 2009). The reaction R1 is the relevant source of O3 in the atmosphere and, once formed, the O3 reacts with NO and regenerates NO2 (reaction R2). The oxygen atom (O) formed from the NO2 photolysis (reaction R3) disappears as fast as the reaction R1 occurs since the O is very reactive. This NOx −O3 scheme will reach the point where the NO2 destroyed is reformed so fast that the steady-state cycle is maintained (Seinfeld and Pandis, 1998). The Leighton Relationship represents the chemical equilibrium as follows: [O3][NO] [NO2] = JNO2 kNO+O3 (1.8) This relation determines the anticorrelation between the NO2 and O3. It can be clearly observed in urban areas close to the emissions, where the high levels of NOx, simultaneously lead to the minimum values of O3 in air. In addition to the NOx −O3 system, the NO and NO2 are involved in chemical conversions with other compounds found in the atmosphere, such as the hydroxyl radical (OH). The OH formation is caused by the reaction of a water molecule and an excited oxygen atom, which was previously produced by the O3 photolysis in the free atmosphere. The OH is the main oxidant in the atmosphere and, it is the trigger of starting the degradation of the most of organic compounds. In this way, the OH may react with CO or with VOC, turning out the formation of hydroperoxil (HO2). In presence of NOx, the HO2 reacts with NO producing NO2 and again OH: HO2 + NO −→ NO2 + OH (R4) 10 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION The intermediate reactions resulting from the interaction of VOC with OH produce organic peroxy radicals that by reacting with the NO also produce NO2. And thus, this chain drives to the HO2 formation and the subsequent reaction R4. Finally, this rise of NO2, without losing O3 (reaction R2), leads to an increase of O3 production by the photolysis of NO2. Figure 1.6 illustrates a simplified scheme of the relevant chemical reactions that involves the NOx and VOC. Note that the main sink of NOx in an urban area is the nitric acid (HNO3) that disappear by rapidly changing to a condense phase (Bloss, 2009). Figure 1.6: Simplified scheme of the reactions involving the NOx, HOx and VOC. In urban environments, the O3 production can be related either by an airmass rich in O3 that reaches the city from elsewhere or by the local sources of O3 precursors: VOC or NOx. The regime of the O3 production is dependent on the response of the O3 to the changes in the abundance of NOx or VOC. If the O3 production depends on the increase or decrease of the VOC levels, the O3 generation is defined as VOC-limited. That occurs with high NOx levels, the rate of O3 formation increases both with the CO or VOC levels and with the formation rate of HOx. However at low NOx levels, which usually occurs moving further away from the city center, the O3 formation barely varies with the VOC levels. It is due to the NOx is the limiting reactant and, the response of the O3 production is controlled by the NOx behaviour and, thereby, this system is termed NOx-limited. Figure 1.7 shows the diagram of O3 production as function of the dominate regime. It noteworthy that the O3 formation by the contribution of NOx and VOC is highly dependent on the local atmospheric composition, in terms of the NOx and VOC levels, and the meteorological variables (Bloss, 2009). CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 11 Figure 1.7: At local scale the tendency of production of O3 by the NOx and VOC contribution. Source: Council et al. (1992). As mentioned before, the chemical time scale is important on the understanding of the chemical processes in the urban environments. Typically, the time scales of the reactions involved in the NOx−O3 photostationary state are around minutes, whereas the O3 production can range at larger scales (Bloss, 2009). Even so, the influence of the VOC reactions on the O3 formation in urban areas is primarily dependent on the VOC levels and the NOx levels of the city, as well as the dynamical behaviour of the flow in the streets. 1.5 Management of urban air quality The emitted pollutants undergo significant changes caused by the atmospheric processes that take place in the urban atmosphere. In addition, the urban obstacles can improve or worse the dispersion of pollutants giving rise to strong concentration gradients in the streets. All of these factors modulate the air quality in urban areas, which is an important concern for the local and regional governments. Therefore, exhaustive studies are required in an attempt not to exceed the hourly and annual limit values established for pollutants in air. To that end, the understanding of the atmospheric processes that trigger off periods of high concentration levels in the city is needed. Moreover, representative maps of dispersion of pollutants in the streets are very helpful for the appraisal of the annual average concentration. For that reason, the assessment of urban air pollution is tackled from multiple channels in order to establish the most effective strategies for reducing it and mitigating the effects thereof. Figure 1.8 shows a simple scheme of the procedure to be followed for the management of urban air quality. 12 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION Figure 1.8: Scheme of the management to improve urban air quality. At city scale, the research methods used to evaluate urban air pollution are fundamentally based on monitoring stations deployed throughout the city and numerical models developed to simulate the relevant processes of the urban atmosphere. The monitoring techniques are based on several air quality stations that record most pollutants all over the city. The advantage of the experimental data is that these provide the actual values of several species at each location. However, their spatial representativeness is limited. And covering the entire city is not feasible since it would entail a big investment in devices and huge outlay. Therefore, the local networks of air quality monitoring stations cannot capture the strong heterogeneities of the distribution of pollutants in the streets. The lack of spatial information can be complemented with modelling techniques. Several numerical models at regional and local scales are used to better understand the physical and chemical processes in urban environments. The mesoscale models provide both the dynamical phenomena and the chemical conversions at city scale with a spatial resolution around 1 km2. Some meteorological mesoscale models are the Weather Research and Forcasting (WRF) or the Regional Atmospheric Model System (RAMS). They provide the meteorological inputs to the mesoscale chemistry-transport models such as the Community Multiscale Air Quality Model (CMAQ), WRF-CHEM, or CHIMERE models so as to complete the pollutants dispersion given by the transport and chemical transformations. These models are suitable for studying large areas or even several countries, and they are able to identify the hot-spots in the city. However, the buildings, parks, vehicles, tunnels or other components of the city produce irregular patterns of flow and dispersion of pollutants within the area of 1 km2. To address the CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 13 problem, microscale models, such as the Computational Fluid Dynamics (CFD) models, can be used to solve the turbulent flow equations at higher spatial resolution (order of meters) in order to reproduce in detail the atmospheric processes in the urban canopy layer. For that reason, performing a CFD simulation demand a large computational load and, thereby, they are used to model short periods of time (several hours). Currently, the CFD models are widely used to investigate specific aspects that affect urban air quality. Figure 1.9 displays the full system of research methods for the urban air quality assessment. Whereas a mesoscale model provides a value of concentration in 1 km x 1 km, the CFD models resolve in that resolution a detailed distribution of the pollutant concentrations. Subsequently, these results can be evaluated with high precision against the local measurements at the specific point. Figure 1.9: Multiscale system of the research methods for urban air quality assessment. 1.6 Research motivation and objectives Modelling urban meteorology and air quality with high resolution is a growing need. The spatial resolution of CFD simulations provides detailed maps of atmospheric variables in urban areas that are validated at the measurements locations. In a city, this valuable information can be used to assess the dispersion of pollutants at street level and, thus, to determine the pedestrian exposure to urban air pollution. It is really helpful for taking mitigation measures against the high levels of pollutant concentrations in specific areas, e.g. redistribution of the traffic in the streets or planning new strategies in the urban vegetation layout. The first studies using CFD models were focused on examining the urban atmospheric processes in isolated streets in an attempt to save computational time. Hence the importance of the most relevant processes in the dispersion of pollutants has been extensively studied, e.g. thermal effects, chemical reactions, street configurations, traffic emissions and meteorological 14 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION variables (Vardoulakis et al., 2003; Li et al., 2006; Zhong et al., 2016). The influence of the buildings geometry on the ventilation of the street (Santiago and Martin, 2005) as well as the thermal effects on the wind circulations below the canopy (Sini et al., 1996) are widely analysed. Moreover, other components of urban environments can affect the dispersion in the street, e.g. the urban vegetation or the turbulence induced by vehicles. The former brings on both an aerodynamic impact on the flow and a deposition effect reducing the air pollution (Abhijith et al., 2017; Buccolieri et al., 2018). As for the effect of vehicles, their motion increase the turbulence in the street improving the mixing of pollutants (Solazzo et al., 2008). Additionally, the production or loss caused by the chemical reactions on the pollutant concentrations are thoroughly studied in simplified geometries using different chemical mechanisms (Baker et al., 2004; Baik et al., 2007; Bright et al., 2013). Therefore, the composition of the traffic emissions is also an important aspect in the concentration of reactive pollutants (Kwak and Baik, 2012). However, for reliable and meaningful modelling of pollutant dispersion in urban areas, it is sometimes necessary to perform studies at urban neighbourhood rather than at street canyon scale (Gromke and Blocken, 2015). The rapidly development of the computational resources promotes the use of CFD models to understand the behaviour of atmospheric processes in real urban environments (Blocken et al., 2012; Kwak et al., 2015). Nonetheless, CFD simulations need still be optimized in order to find the best compromise between computational time and accuracy of the results. The levels of NO2 in urban environments, mainly caused by the vehicle emissions, are one of objectives to be tackled in the management of urban air quality. Thus far, CFD modelling over urban areas is commonly addressed to study the dispersion of non-reactive pollutants (Amorim et al., 2013; Sanchez et al., 2017; Santiago et al., 2017a). Nevertheless, in case of the NO2, the fastest chemical reactions can be important in the accuracy of modelling its concentration at street level. The implementation of chemical reactions in the CFD simulation leads to an important rise of computational time. Hence the need to in-depth understand the impact of the chemical interactions on the concentration of pollutants in real streets. This has not been fully addressed in previous studies and, therefore, an exhaustive study about the influence of atmospheric conditions on chemical reactions is essential. This research is aimed at simulating with high spatial resolution the dynamical and chemical processes that take place in the urban atmosphere using a CFD model. To that end, the most relevant chemical reactions, in gas-phase, for the NO2 at daytime are implemented in the model. Therefore, the importance of including chemical reactions on modelling the dispersion of pollutants in the streets can be tackled. Additionally, a modelling approach is presented to face the challenge of obtaining detailed maps of concentration over several days. It is worth noting that all simulated results are carefully validated with the available CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION 15 experimental data in every setting. The main purpose of this research is achieved by fulfilling the following objectives: • The implementation of chemical reactions in the CFD model : The three-reaction system (NOx−O3 reactions) and a more complex chemical mechanism of 23 reactions and 25 species (NOx−Ox−VOC reactions) are both included in the CFD model. And the chemical terms are evaluated with a chemical box model. • The estimation of the resulting error by modelling the NO2 dispersion under a non-reactive approach instead of using a chemical mechanism: The deviation in the NO2 concentration is properly assessed in several urban settings under multiple atmospheric conditions. • The analysis of the impact of meteorological variables on the dispersion of reactive pollutants : The variability of the atmospheric conditions is exhaustively analysed in both simplified geometries and real urban scenarios. • The evaluation of the coupled behaviour of turbulent processes and chemical conversions in an urban setting : The influence of turbulence in the street on the reaction rate is thoroughly examined, both in time and space, in a real urban scenario. • The new development of a modelling approach using a meteorological mesoscale and a CFD models in order to obtain detailed concentration maps over long period in real urban areas : This methodology can be applied in any place of a city and its validation is carried out in an urban hot-spot. The present work is broken down into chapters that achieve the above-mentioned objectives. The complex system of mathematical expressions and the turbulence closure of the CFD model is fully described in Chapter 2. The implementation of chemical reactions in the CFD model and the subsequent comparison with a chemical box model are also herein explained. Lastly, the procedure to be followed in order to undertake a CFD simulation over urban areas is described in detail. In Chapter 3, the NO2 dispersion is modelled using the two chemical mechanisms in idealized geometries under different atmospheric conditions. In this way, the impact of specific meteorological scenarios on the chemical activity is analysed through the NO and NO2 concentrations. It provides quantitative information that helps to select the optimal chemical scheme in terms of computational requirements. In Chapter 4, the influence of the atmospheric chemistry on the NO2 concentration is evaluated over time throughout a diurnal cycle in a real urban scenario. In this analysis, the coupled behaviour of the chemical and dynamical processes is thoroughly studied. Finally, the ability of CFD models to assess the dispersion of pollutants over long periods of time using a modelling approach is studied in Chapter 5. 16 CHAPTER 1. INTRODUCTION AND RESEARCH MOTIVATION Chapter 2 The CFD-RANS model 2.1 Introduction The application of CFD models become largely used for environmental flows in a great variety of fields, e.g urban air quality and meteorology (Vardoulakis et al., 2003), wind and thermal comfort (Mochida and Lun, 2008), effects of urban vegetation (Buccolieri et al., 2018), emergency response (Antonioni et al., 2012), indoor/outdoor climate design (Zhai and Chen, 2005) and renewable energy such as wind energy (Li et al., 2012). The main objective of this chapter is to present a brief review of the mathematical equations resolved by the CFD model. Additionally, the procedure to be followed is thoroughly described including each new contribution to simulate the urban atmosphere. Finally, the statistic metrics and the acceptance criteria for an accurate evaluation of modelling results are also presented herein. This research is carried out using a commercial CFD model (STARCCM+ from SIEMENS) which has been modified in order to include additional phenomena such as the vegetation effects and atmospheric chemistry. 17 18 CHAPTER 2. THE CFD-RANS MODEL 2.2 Governing equations CFD models are based on the motion equations for the fluid described by the fundamental laws of conservation of mass, momentum and energy expressed by the following equations (Versteeg and Malalasekera, 2007): ∂u ∂x + ∂v ∂y + ∂w ∂z = 0 (2.1) ∂u ∂t + u ∂u ∂x + v ∂v ∂y + w ∂w ∂z = −1 ρ ∂p ∂x + ν∇2u+ SMx (2.2) ∂v ∂t + u ∂u ∂x + v ∂v ∂y + w ∂w ∂z = −1 ρ ∂p ∂y + ν∇2v + SMy (2.3) ∂w ∂t + u ∂u ∂x + v ∂v ∂y + w ∂w ∂z = −1 ρ ∂p ∂z + ν∇2w + SMz (2.4) ∂θ ∂t + u ∂θ ∂x + v ∂θ ∂y + w ∂θ ∂z = −αh∇2θ + Sh (2.5) where ρ is the air density, ν is the kinematic viscosity related to the dynamic viscosity (µ) through ν = µ/ρ, αh is the thermal diffusivity (or molecular diffusivity of heat). SM represents the momentum source for each component x, y, z and Sh is the source term of heat. The Equation 2.1 represents the mass continuity equation in three dimensions. The Equations 2.2-2.4 are the momentum equations widely known as Navier-Stokes equations. And the energy equation (Eq. 2.5) is expressed by means of the transport of temperature. The atmospheric processes in urban environments are mainly characterized by the turbulent phenomena (Section 1.3). In a turbulent flow, the appearance of eddies with a wide range of length and time scales produce dynamically complex processes (Versteeg and Malalasekera, 2007). Therefore, the mathematical expressions have to be solved applying different numerical methods. There are several models that are able to resolve the turbulent flow equations. Depending on the computational load required, they can be classified from more to less requirements as: the Direct Numerical Simulation (DNS) model, the Large Eddy Simulation (LES) model and the Reynolds-averaged Navier-Stokes equations (RANS) model. • The DNS model: It numerically solves the flow equations without considering any modelling assumption (Coceal et al., 2007). It means that both the mean flow and all turbulent fluctuations of velocity are computed in the simulations. The unsteady Navier-Stokes equations are solved on very fine spatial grids and with time steps sufficiently small to resolve the fastest fluctuations (Versteeg and Malalasekera, 2007). Therefore, DNS simulations require a large computational load that prevents their use in complex geometries, like urban areas, with the current resources. CHAPTER 2. THE CFD-RANS MODEL 19 • The LES model: This model involves a spatial filtering operation of the fluid dynamics equations prior to the computations, which tracks the behaviour of larger eddies. The method consists of solving the mean flow and the large eddies. And the effects caused by the smaller structures are included through the so-called sub-grid scale model. Hence, the flow variables are decomposed in the resolved or filtered component (large scale) and the unresolved or sub-filtered component (small scale). • The RANS model: This numerical method is based on resolving the mean flow and the turbulence fluctuations that appear as additional terms in the time-averaged flow equations. To do that, the turbulent terms are not directly computed by the model, but they are parametrized adding a turbulence closure (Versteeg and Malalasekera, 2007). The main difference between LES and RANS models is that the former need to perform a time-average concentration in order to recover the mean component, which involves simulating a long period of time. In contrast, RANS models directly provide the mean value with no need to do the time average. As for computing requirements, the LES model requires a great volume of calculations demanding a great storage and, thereby, its computational time is fairly large. In contrast, the RANS model does not directly solve the small-scale turbulent fluctuation. It gives a solution for the mean flow variables and, thereby, substantially reduces its computational cost. Tominaga and Stathopoulos (2010) concluded that the RANS method underestimates the turbulent diffusion near the obstacle in comparison with the LES model. Although for mean velocity, the difference between the LES and RANS results are small. Despite the difference in the convergence methods between LES and RANS, they estimated the computational time of LES was around 25 times longer than the RANS model used. This aspect is a very important issue to be considered by modelling flow equations in complex geometries. Santiago et al. (2010) and Dejoan et al. (2010) compared the dispersion of a tracer in an array of obstacles between the LES and RANS results with experimental field data. They concluded that the mean concentration was similar between modelling results and both of them were in a good agreement with the measurements. The main discrepancies between models were close to the edge of the plume due to the effects of small fluctuations. Hence, the LES model provides very accurate results in areas where the instantaneous fluctuations are more relevant. However, the outputs are similar for the mean values evaluated in points further away from the proximity of walls or ground. The fact that CFD-RANS models require moderate computational load makes them an effective tool to solve the turbulent flow equations in urban environments. Therefore, the CFD-RANS approach is selected to carry out this work. 20 CHAPTER 2. THE CFD-RANS MODEL 2.2.1 The Reynolds-averaged Navier-Stokes model A CFD-RANS model is based on the Reynolds-averaged Navier-Stokes equations that takes into account the time average of the fluid dynamics equation and the turbulence effects on the mean flow properties. By applying the Reynolds decomposition (Section 1.3) to the instantaneous variables of turbulent flow, turns out the addition of the mean component (U, V, Φ...) and its time fluctuation (u′, v′,φ′...): u = U + u′ (2.6) v = V + v′ (2.7) w = W + w′ (2.8) φ = Φ + φ′ (2.9) where φ represents a scalar variable. If the instantaneous variables in the equations (Eqs. 2.1-2.4) are replaced by the previous definition, result in the so-called Reynolds-averaged Navier-Stokes equations: ∂U ∂x + ∂V ∂y + ∂W ∂z = 0 (2.10) ∂U ∂t + U ∂U ∂x + V ∂V ∂y +W ∂W ∂z = −1 ρ ∂P ∂x + ν∇2U + 1 ρ [ ∂(u′2) ∂x + ∂(u′v′) ∂y + ∂(u′w′) ∂z ] + SMx (2.11) ∂V ∂t + U ∂U ∂x + V ∂V ∂y +W ∂W ∂z = −1 ρ ∂P ∂y + ν∇2V + 1 ρ [ ∂(u′v′) ∂x + ∂(v′2) ∂y + ∂(v′w′) ∂z ] + SMy (2.12) ∂W ∂t + U ∂U ∂x + V ∂V ∂y +W ∂W ∂z = −1 ρ ∂P ∂z + ν∇2W + 1 ρ [ ∂(u′w′) ∂x + ∂(v′w′) ∂y + ∂(w′2) ∂z ] ︸ ︷︷ ︸ Reynolds Stress tensor +SMz (2.13) Using the Einstein notation, the momentum equations can be rewritten as follows: ∂Ui ∂xi = 0 (2.14) ∂Ui ∂t + Uj ∂Ui ∂xj = −1 ρ ∂P ∂xi + ∂ ∂xj [ ν ∂Ui ∂xj ] − 1 ρ ∂u′iu ′ j ∂xj + SMxi (2.15) The six additional terms, τij = −ρu′iu′j, comprise the Reynolds stress tensor. Based on the Newton’s law of viscosity, the viscous stress are taken to be proportional to the rate of deformation of fluid elements. Thus, the Boussinesq hypothesis about eddy viscosity and eddy diffusivity proposed that the Reynolds stress are proportional to the mean strain-rate tensor (Boussinesq, 1877). τij = −ρu′iu′j = µt ( ∂Ui ∂xj + ∂Uj ∂xi ) − 2 3 ρkδij (2.16) CHAPTER 2. THE CFD-RANS MODEL 21 where the mean strain-rate tensor (Sij) is: Sij = 1 2 ( ∂Ui ∂xj + ∂Uj ∂xi ) (2.17) The turbulent kinetic energy (k) is defined as k = 1 2 u′iu ′ i and the turbulent or eddy viscosity (µt) is related to the turbulent or eddy kinematic viscosity as νt = µt/ρ, which needs to be determined‡. For a scalar variable, the transport equation is expressed as follows: ∂Φ ∂t + Uj ∂Φ ∂xj = D ∂2Φ ∂xj∂xj − ∂u′jφ ′ ∂xj + Sφ (2.18) The terms on the left-hand side represents the rate of change and the convective term. On the right-hand side, the first term stands for the diffusive term, where D is the molecular diffusivity. The second term denotes the turbulent transport and the last one, Sφ, the source of the scalar. The turbulent fluxes of any scalar variable, like heat or mass, are also considered proportional to the gradient of the mean value of the scalar transported. −ρu′iφ′ = Kt ∂Φ ∂xi (2.19) where Kt is the turbulent or eddy diffusivity. This parameter for heat and mass is expected to be analogous to the eddy viscosity µt for momentum since the turbulent transport of these variables are also controlled by same mechanism of eddy mixing. Therefore, the eddy diffusivity for mass and heat are defined as: Kt = µt σt (2.20) where σt stands for the turbulent Prandtl number (Prt) for heat or the turbulent Schmidt number (Sct) for mass. Therefore, the closure issue is reduced to solve the distribution of µt and, for that, it is necessary a turbulence model to resolve the turbulent terms of the RANS equations. The common turbulence closures are classified depending on the number of additional equations needed to solve the RANS flow equations (Versteeg and Malalasekera, 2007). The most simplest model with no extra equation is the mixing length model. The main assumption is that the Reynolds stress and the viscous stress are proportional to the fluid deformation ‡The second term in Eq. 2.16 includes the Kronecker delta (δij=1 if i=j and δij = 0 if i 6=j ) 22 CHAPTER 2. THE CFD-RANS MODEL rate through the turbulent viscosity (µt) as proportionality constant. The Reynolds Stress Model (RSM) is the most complex model with six additional equations, one for each Reynold stress, which increases the computational load. Finally, an intermediate turbulence model in terms of numerical complexity is the k -ε model. It only adds two more equations for the turbulent kinetic energy (k) and the turbulent dissipation rate (ε) that are solved together with the RANS equations. The k -ε turbulence closure is selected in this work for being the most appropriate model that optimise the computing resources without losing accuracy in solving the turbulent processes. It is commonly used for modelling urban air quality (Kim and Baik, 2004; Solazzo et al., 2008; Kwak and Baik, 2012). 2.2.2 The k-ε turbulence model The k -ε turbulence model assumes that the Reynolds stress is proportional to the rate of deformation of fluid elements and, the turbulent viscosity is the proportionality constant for the Reynold stress. This approach is based on modelling two transport equations for the turbulent kinetic energy (k) and for the turbulent dissipation rate (ε). On the basis of the k -ε approach, there are several turbulence closures primarily focused on solving the computation of the effective viscosity and the ε transport equation. The most common turbulence closures are: the Standard k -ε, the Realizable k -ε and the Re-Normalisation Group (RNG) k -ε. The turbulence closure used in this work is the Realizable k -ε. It improves the Standard k -ε using the turbulent viscosity deduced by a new analytical expression and the ε equation derived from an exact equation for the transport of the mean-square vorticity fluctuation. Thus, the turbulent viscosity is defined as follows (Shih et al., 1995): µt = ρCµ k2 ε (2.21) Unlike the Standard k -ε closure, the parameter Cµ is not constant. It is defined by Cµ = 1/(A0 + AsU (∗) k ε ), where A0 = 4, AS = √ 6cos(ξ) and U (∗)= √ S : S−W : W. S is the mean strain-rate tensor, W is the rotation rate tensor and ξ = 1/3 acos( √ 6W ). Finally, the governing equations for the Realizable k-ε model are the following: ∂k ∂t + uj ∂k ∂xj = 1 ρ ∂ ∂xj [( µ+ µt σk ) ∂k ∂xj ] + Gk ρ − ε+ Sk (2.22) ∂ε ∂t + uj ∂ε ∂xj = 1 ρ ∂ ∂xj [( µ+ µt σε ) ∂ε ∂xj ] + Cε1Sε− Cε2 ε2 k + √ νε + Sε (2.23) CHAPTER 2. THE CFD-RANS MODEL 23 where σk = 1 and σε = 1.2 are the turbulent Prandtl numbers for k and ε respectively and Cε2 is a model constant equal to 1.9. These values for the model constants are shown to be the most suitable for a wide range of turbulent flow at microscale (Launder and Spalding, 1974). The term Gk represents the production of turbulent kinetic energy that is defined as Gk = µ2 tS 2, where S is the modulus of the mean strain-rate tensor (S ≡ √ 2SijSij) and the constant Cε1 = max [ 0.43, η η+5 ] , with η = S k/ε. In words, the terms on the left hand side of the Eqs. 2.22-2.23 represent the rate of change and the convective transport of k and ε. On the right hand side, the first term is the transport of k and ε by diffusion and, the following two terms the production and destruction of k and ε, accordingly. Sk and Sε represent the source term of k and ε respectively. The differences with other k -ε models are mainly in the production and destruction terms of the ε equation (Eq. 2.23). The present form of the production term is considered to better represent the energy transfer. Unlike other models, the destruction term of ε never vanishes, even if k becomes small values or zero. From the solution of k and ε, the length scale (l) and the time scale (τt) of the turbulence processes can be defined as (Pope, 2001): l = k3/2/ε (2.24) τturb = k/ε (2.25) 2.2.3 The Boussinesq approximation In non-neutral atmospheric conditions, the thermal effects are important and, thereby, the buoyancy terms are included in the CFD model. By considering the air as an incompressible fluid, the Boussinesq assumption consists of the difference between the actual fluid density ρ and the standard aerostatic density ρn is small: |ρ − ρn|/ρ << 1. The use of ρn as reference density instead of the actual density produce an error in the description of the inertial forces. In this way, this difference ρ − ρn may be important in the description of the gravitational body forces. The Boussinesq approximation holds for the most of the environmental applications in the lower atmosphere (Sini et al., 1996). The density deviation ρ− ρn is related to the potential temperature (θ) through the equation of state as follows: ρ− ρn ρn = −β(θ − θn) (2.26) where β is the thermal expansion coefficient and it is defined inversely proportional to temperature. Hence, the buoyancy force is defined by Fbuoyancy = ρgβ(θref − θ), where gi = (0, 0,−g). 24 CHAPTER 2. THE CFD-RANS MODEL Finally, the flow equations including the buoyancy forces are expressed as: ∂Ui ∂xi = 0 (2.27) ∂Ui ∂t + Uj ∂Ui ∂xj = −1 ρ ∂P ∂xi + µ ρ ∂2Ui ∂xj∂xj − ∂ ∂xj (u′iu ′ j)− ρgiβ(θref − θ) + SMxi (2.28) ∂Θ ∂t + Uj ∂Θ ∂xj = αh ∂2θ ∂xj∂xj − ∂u′jθ ′ ∂xj + Sh (2.29) Concerning the turbulence closure, the equations for k and ε are (Hanjalic and Launder, 1972): ∂k ∂t + Uj ∂k ∂xj = 1 ρ ∂ ∂xi [( µ+ µt σk ) ∂k ∂xj ] + Gk ρ + βgi µt σθ ∂θ ∂xj − ε+ Sk (2.30) ∂ε ∂t + Uj ∂ε ∂xj = 1 ρ ∂ ∂xj [( µ+ µt σε ) ∂ε ∂xj ] + C1εSε+ C1ε ε k β gi µt σθ ∂θ ∂xj − C2ε ε2 k + √ νε + Sε (2.31) where σθ is the turbulent Prandtl number for the temperature. Table 2.1: Summary of equations using the CFD-RANS model with the k -ε turbulence closure. Continuity ∂Ui ∂xi = 0 Momentum equations ∂Ui ∂t +Uj ∂Ui ∂xj = −1 ρ ∂P ∂xi + µ ρ ∂2Ui ∂xj∂xj − ∂ ∂xj (u′iu ′ j)−giβ(θ−θn)+SMxi Energy equation ∂Θ ∂t + Uj ∂Θ ∂xj = k ∂2θ ∂xj∂xj − ∂u′jθ ′ ∂xj + Sh Transport equation ∂Φ ∂t + Uj ∂Φ ∂xj = D ∂2Φ ∂xj∂xj − ∂u′jφ ′ ∂xj + Sφ k-ε turbulence model Turbulent kinetic energy ∂k ∂t + Uj ∂k ∂xj = 1 ρ ∂ ∂xj [( µ+ µt σk ) ∂k ∂xj ] + Gk ρ + βgi µt σθ ∂θ ∂xj − ε+ Sk Turbulent dissipation rate ∂ε ∂t +Uj ∂ε ∂xj = 1 ρ ∂ ∂xi [( µ+ µt σε ) ∂ε ∂xj ] +C1εSε+C1ε ε k β gi µt σθ ∂θ ∂xj − C2ε ε2 k + √ νε + Sε CHAPTER 2. THE CFD-RANS MODEL 25 Therefore, the CFD model used in this work is based on the Reynolds-averaged Navier-Stokes equations with the Realizable k -ε turbulence closure. Table 2.1 summarizes the equations to be solved using the CFD-RANS model. These governing equations are discretized by means of the finite volume method in order to be solved using the segregated method. The pressure-velocity coupling is solved by means of Semi-Implicit Method for Pressure-Linked Equations algorithm (SIMPLE). It is based on an iterative process that includes a solution to a Pressure-Correction equation to ensure mass conservation (Patankar, 1980). 2.3 Chemical Mechanisms The most important reactions of the gas-phase chemistry for the NOx in urban environments are implemented in the CFD model. The NOx chemistry is mainly dominated by the photostationary state involving the NOx−O3 interactions. However, there are many other pollutants in the urban atmospheric composition that may interact with NOx, e.g. the VOC. For that reason, a condensed chemical mechanism involving the relevant interactions of VOC in an urban atmosphere is developed. Accordingly, the two chemical mechanisms implemented in the CFD-RANS model are: the NOx−O3 scheme and the reduced NOx−Ox−VOC scheme. 2.3.1 The NOx−O3 scheme The photostationary state is the most simple chemical mechanism implemented in the CFD-RANS model: NO2 + hν JNO2−−−→NO + O (R5) O + O2 + M −→ O3 + M (R6) O3 + NO kNO+O3−−−−−→NO2 + O2 (R7) The steady-state cycle using the NOx−O3 scheme is maintained (Section 1.4) and, thereby, the O3 concentration is related to the NO and NO2 as follows: [O3] = JNO2[NO2] kNO+O3 [NO] (2.32) The square brackets ([ ]) stands for the concentration of pollutants in ppb (part per billion) or µg m−3 §. The photolysis rate constant, JNO2 , is dependent on the zenith solar angle (ϕ) §In atmospheric conditions of 1 atm and T=298 K: 1 ppb of NO= 1.25 µg m−3, 1 ppb of NO2= 1.88 µg m−3 and 1 ppb of O3 = 2 µg m−3 26 CHAPTER 2. THE CFD-RANS MODEL and is defined by JNO2 = A exp(B/cosϕ) (s−1). And the thermal constant rate (kNO+O3) is computed as kNO+O3 = A exp(−(E/R)/T) (ppb−1 s−1), where T is air temperature and A, B and E/R are chemical constants established for each chemical reaction (Table 2.3). 2.3.2 The NOx−Ox−VOC scheme The chemical processes in the atmosphere involve a vast number of chemical reactions. The most complete chemical scheme is the Master Chemical Mechanism (MCM) with 4531 species and 12691 reactions for representing the degradation of 123 primary VOC species (Saunders et al., 2003; Jenkin et al., 2003). It is highly detailed and it is not feasible to incorporate it in the air quality models because demands a very large computational load. Gery et al. (1989) developed a condensed chemical mechanism focused on urban and regional photochemistry named the Carbon Bond Mechanism IV (CBM-IV) with 33 species and 82 chemical reactions. Stockwell et al. (1997) also presented other chemical mechanism developed to be used in the air quality models known as the Regional Atmospheric Chemistry Mechanism (RACM) which consists of 77 species and 237 reactions. In any case, these chemical mechanisms are computationally expensive to be implemented in the CFD-RANS model and to carry out many simulations. For that reason, a condensed NOx−Ox−VOC mechanism has been developed taking into account the VOC reactions without increasing too much the computational time. This NOx−Ox−VOC scheme has been reduced from the RACM as far as possible without losing much accuracy. It is performed during 30 minutes of simulation for urban conditions in a chemical box model. To that end, the software CHEMATA (CHEmical Mechanism Adaptation to Tropospheric Applications) is used (Junier et al., 2005; Kirchner, 2005). A wide set of different urban pollution conditions was investigated. Special emphasis was given to cases where the results for the RACM mechanism considerably differ from the results of the photostationary steady state. The species list of the NOx−Ox−VOC scheme is shown in Table 2.2. CHAPTER 2. THE CFD-RANS MODEL 27 Table 2.2: List of species Species of the NOx-Ox-VOC scheme O3 ozone NO nitrogen oxide NO2 nitrogen dioxide HNO3 nitric acid CO carbon monoxide SO2 sulfur dioxide N2 nitrogen O2 oxygen H2O water vapour H2 hydrogen O1D excited state oxygen atom HO hydroxyl radical HO2 hydroperoxyl HCHO formaldehyde PAN peroxyacetyl nitrate ONIT organic nitrate MO2 (CH3O2) methyl peroxy radical ACO3 (RC(O)O2) acetyl peroxy and higher saturated acyl peroxy XO2 accounts for additional NO to NO2 conversions XO2N accounts for additional NO to NO2 conversions ALK lumping group of alkanes RACM group OLE lumping group of olefin RACM group ARO lumping group of aromatic RACM group RACM species used in new integrated species Alkanes CH4 methane ETH ethane HC3 alkanes, alcohols, esters, and alkynes † HC5 alkanes, alcohols, esters, and alkynes ∗ HC8 alkanes, alcohols, esters, and alkynes ∗∗ Olefins ETE ethene OLT terminal alkenes OLI internal alkenes Aromatic TOL toluene XYL xylene Radicals RO2 organic peroxy radical †with HO rate constant (298 K, 1 atm) less than 3.4·10−12 cm3 s−1 ∗with HO rate constant (298 K, 1 atm) between than 3 .4·10−12 cm3 s−1 and 6.8·10−12 cm3 s−1 ∗∗with HO rate constant (298 K, 1 atm) greater than 6.8·10−12 cm3 s−1 28 CHAPTER 2. THE CFD-RANS MODEL Three main manners of the mechanism reduction are applied: • Modification of the lumping groups: Several chemical RACM lumping groups are united to a new lumping group. For example, the RACM alkanes (ETH, HC3, HC5 and HC8) are summed up in the new species ALK. The RACM olefins (ETE, OLT and OLI) are lumped to the new specie OLE and the RACM aromatic compounds, TOL and XYL, are integrated in the new species ARO. • Modification of the organic peroxy radical parametrisation: instead of using the 20 RACM RO2 species, the approach described by Kirchner (2005) is applied for the RO2 species CH3O2 (called MO2) and RC(O)O2 (called ACO3), as well as for the two operator species, XO2 and XO2N, which work in the following way shown bellow. The real reactions: R−H + HO −→ R∗ + H2O (R8) R∗ + O2 −→ RO2 (R9) RO2 + NO −→ a RONO2 + b (RO + NO2) (R10) RO −→ Products (R11) are parameterised by: R−H + HO −→ a XO2N + b (XO2 + products) (R12) XO2 + NO −→ NO2 (R13) XO2N + NO −→ RONO2 (R14) • Elimination of reactions and species which are not crucial for simulating under the given conditions (time of about 30 minutes, urban pollution conditions, daytime). Focusing on daytime simulations, the species NO3 (nitrate radical) and all its reactions are eliminated because of the NO3 is rapidly photolyzed in presence of light (Finlayson-Pitts and Pitts Jr, 1999). Also, low reactive species as methane (CH4) and H2 can be eliminated under urban conditions. The resulting NOx−Ox−VOC scheme consists of 23 species and 25 chemical reactions involved in Table 2.3. Compared to the NOx−O3 scheme, it includes the peroxy radical production from VOC, CO and SO2 as well as NOx loss processes by the formation of PAN, HNO3 and organic nitrates. Whereas the peroxy radical production increases the NO2/NO ratio and, thus, the NO2 concentrations, the NOx loss processes decrease NOx and, therefore, NO2 concentrations. As well, O3 loss by the reaction with olefins is considered. This condensed chemical mechanism is particularly developed for this work with aim of optimizing the computational time by undertaking the CFD-RANS simulations. CHAPTER 2. THE CFD-RANS MODEL 29 Table 2.3: The NOx −Ox −VOC scheme: The rate constants of first order reactions are calculated on J = A ∗ exp(B/cosϕ) (s−1). The rate constants for the second order reactions called ”thermal” by k = A ∗ exp[(−E/R)/T], for the ”Troe” reactions by kreac = {k0(T)[M])/(1 + (k0(T)[M]/k∞(T))}0.6{1+[log10(k0(T )[M ]/k∞(T ))]2}−1 , and for the ”troe-equilibrium” by k = Aexp(−B/T)× {k0(T)[M]/(1 + (k0(T)[M]/k∞(T)))} ·0.6{1+log10[k0(T)[M]/k∞(T)]2}−1 , where k0(T) = k0 300(T/300)−n and k∞(T) = k∞ 300(T/300)−m. The units of the second order reactions are cm3 s−1. NO2 NO + O3 photodissociation (A=1.45E-2, B=-0.4) O3 O1D photodissociation (A=1.48E-4, B=-1.65) HCHO CO photodissociation (A=6.65E-5, B=-0.6) HCHO CO + 2 HO2 photodissociation (A=5.40E-5, B=-0.79) ALD CO + HO2 + MO2 photodissociation A=1.35E-5, B=-0.94 O1D + N2 O3 thermal(A=1.80E-11, E/R=-110.0) O1D + O2 O3 thermal(A=3.20E-11, E/R=-70.0) O1D + H2O 2 HO thermal(A=2.20E-10, E/R=0.0) HO + NO2 HNO3 troe(k0 300=2.60E-30,n=3.2, k∞ 300=2.40E-11, m=1.3) HO2 + NO NO2 + HO thermal(A=3.70E-12, E/R=-250.0) O3 + NO NO2 thermal(A=2.00E-12, E/R=1400.0) HO + SO2 HO2 troe(k0 300=3E-31,n=3.3, k∞ 300=1.50E-12, m=0) CO + HO HO2 K = 1.5d-13*(1.+2.439e-20*airc) ALK + HO 0.931 XO2 + 0.842 HO2 + 0.011 CO + 0.011 HO + 0.019 HCHO + 0.051 MO2 + 0.378 ALD + 0.096 XO2N thermal(A=8.05E-12, E/R=237.0) OLE + HO 1.001 XO2 + 0.998 HO2 + 1.011 HCHO + 0.002 ACO3 + 0.747 ALD thermal(A=5.69E-12, E/R=-474.5) ARO + HO 0.950 XO2 + 0.950 HO2 + 1.860 ALD + 0.050 XO2N thermal(A=5.35E-12, E/R=-355.0) ALD + HO ACO3 thermal(A=5.55E-12, E/R=-331.0) HCHO + HO HO2 + CO thermal(A=1.00E-11, E/R=0.0) OLE + O3 0.344 HO2 + 0.383 CO + 0.303 HO + 0.135 XO2 + 0.682 HCHO + 0.092 MO2 + 0.007 ACO3 + 0.630 ALD thermal(A=1.28E-15, E/R=907.1) NO2 + ACO3 PAN troe(k0 300=9.70E-29, n=5.6,k∞ 300=9.30E-12, m=1.5) PAN NO2 + ACO3 troe-equil (k0 300=9.70E-29, n=5.6, k∞ 300=9.30E-12, m=1.5, A=1.16E28, B=13954.) XO2 + NO NO2 thermal(A=4.00E-12, E/R=0.0) MO2 + NO NO2 + HO2 + HCHO thermal(A=4.20E-12, E/R=-180.0) ACO3 + NO NO2 + 0.046 HO2 + 0.046 CO + 0.954 MO2 thermal(A=2.00E-11, E/R=0.0) XO2N + NO 0.939 ONIT thermal(A=4.45E-12, E/R=-39.9) 30 CHAPTER 2. THE CFD-RANS MODEL 2.4 Modelling reactive pollutants 2.4.1 Transport equation Modelling a reactive pollutant requires to include the chemical term in the transport equation. These additional terms represent the formation and depletion, depending on the chemical reactions in which the reactant appears. Hence, the transport equation is added in the CFD-RANS for each compound (φi) as follows: ∂φi ∂t + Ui ∂φi ∂xj = D ∂2φi ∂xj∂xj + ∂ ∂xj ( Kc ∂φi ∂xj ) + ∆φi ) Chem + Sφi (2.33) where φi is the concentration of the ith specie and D and Kc are molecular diffusivity and eddy diffusivity for mass, respectively. The chemical term, ∆φi ) Chem , is modelled as the rate of production or loss of each chemical compound depending on the chemical mechanism. The chemical term of the NO2 using the photostationary state is written as: [∆NO2]Chem = −JNO2[NO2] + kNO+O3 [NO][O3] (2.34) In contrast, in the NOx−Ox−VOC scheme, its chemical term increases as a consequence of the number of chemical reactions (i) in which the NO2 intervenes. Therefore, its chemical term in the transport equation increases as: [∆NO2]Chem = −JNO2[NO2]± ∑ i kiRA RB (2.35) where RA and RB are the reactants of the ith chemical reaction involving the production or loss of NO2. The same numerical solver for resolving the governing equations is used to model the dispersion of reactive pollutants. The terms in each transport equation, including the chemical terms, are solved with a second-order discretization. Therefore, to properly solve the transport considering chemical reactions, a sensitivity test of the time step required is evaluated using the NOx−O3−VOC scheme in a single street-canyon (Sanchez et al., 2016). The concentration of reactive pollutants is analysed for different time steps (0.1 s, 1 s and 2 s). During the first period of the simulation, the turbulent dynamic is simulated at a time step of 1 s, without including chemical interactions. After that, the chemical reactions are implemented in the CFD simulation carrying out every time step case (ts =0.1 s, ts = 1 s and ts = 2 s). Finally, in the steady state, the deviation of concentration was computed from the case ts = 1 s, and the the differences are lower than 0.007 % in comparison with CHAPTER 2. THE CFD-RANS MODEL 31 ts=0.1 s (Sanchez et al., 2016). Therefore, the time step of 1 s was considered appropriate to solve both the turbulence and chemical reactions through the CFD-RANS model. In terms of computational time, modelling reactive pollutants increases the time required not only caused by the additional chemical terms in the transport equation, but also with the number of coupled transport equations. The estimated time for modelling NO2 as a reactive pollutant, with the NOx−O3 scheme, instead of a non-reactive in urban areas might approximately reach a factor 1.5, whereas might be four times as much using the NOx−Ox−VOC scheme. It is noteworthy that it is an approximation because it also depends on the number of grid points, the complexity of the domain and the available computational resources. 2.4.2 Box model validation The implementation of a chemical scheme in the CFD model is evaluated with the chemical box model (chemBM) used to developed the NOx−Ox−VOC scheme. To that end, the CFD model is also run as a box model using the NOx−Ox−VOC scheme using a time step of 1 s. The same conditions both in the CFD model and in the chemical box model are used to perform the simulation for 30 min. The chemical constants are calculated with a constant air temperature T=293 K and a zenith angle fixed in ϕ = 40◦. The imposed initial conditions are based on concentrations similar to atmospheric conditions and traffic emissions of an urban area (Table 2.4). Table 2.4: Initial values of concentration of pollutants in ppb. NO 5 OLE 1 NO2 12 ALK 2 O3 20 H2O 107 CO 300 O2 2.095 108 HCHO 0.1 N2 7.808 108 ARO 1 Figure 2.1 displays the time evolution and the scatter plots of several pollutants resulted from the chemical box model and the CFD model. 32 CHAPTER 2. THE CFD-RANS MODEL (a) (b) (c) (d) (e) (f) Figure 2.1: Time series and scatter plots of NO, NO2, O3, HCHO, ALD and ARO (in ppb) from the chemical box model and from the CFD model. The results obtained from the CFD model are practically the same that the outputs of the chemical box model. Indeed, the maximum relative errors of concentration for NO, NO2 and O3 are respectively 0.28 %, 0.18 % and 0.085 %, associated to simple numerical errors. Therefore, the chemical terms included in the transport equations are appropriately solved by the CFD model. 2.4.3 Chemical and turbulent time scales The pollutants dispersion in urban environments is mainly dominated by the turbulent processes. In addition, the reaction rate is dependent on the ability of the turbulence to mix the reactants in the PBL (Vilà-Guerau de Arellano et al., 2004). The relation between the turbulent mixing and the chemical reactions is given by the Damkhöler number (Da), which is defined by the ratio of their corresponding time scales. Da = τturb τreact (2.36) τturb represents the time scale of the turbulent processes and τreact the time scale of the chemical reactions. Da < 1 means that the chemical processes are relatively slow respect to the turbulent phenomena and, the species become well mixed more rapidly than the chemical CHAPTER 2. THE CFD-RANS MODEL 33 reaction take place. Values of Da≈ O(1) represent that the turbulence is not efficient enough to properly blend the reactants, being able to affect the chemical balance. For Da > 1, the influence of the chemical reactions grows in importance against the turbulent phenomena. As previously defined, the turbulent time scale can be computed by the turbulent kinetic energy, k, and its dissipation rate, ε, as follows (Pope, 2001): τturb ≡ k ε (2.37) Concerning the chemical time scale, for a first-order reaction, it is computed by means of the photolysis rate as τreact = (Ji) −1. For a second-order chemical reaction, the time scale is expressed as τreact = (kiRi) −1, where ki is the reaction rate of the ith chemical reaction and Ri the concentration of the reactant in that reaction. To compute the Da number in a chain of chemical reactions, the characteristic time scale is obtained from the slowest chemical reaction. Additionally, the behaviour between the turbulent processes and the chemical activity is influenced by the chemical equilibrium (Jonker et al., 2004). The reactive species tend to reach a chemical balance between the loss and production of the compounds involved in the chemical mechanism. Nevertheless, in urban environments, the continuous emissions on the surface give rise to a permanent perturbation of the chemical state of the overlying air. 2.5 Modelling urban vegetation The urban vegetation is an important component that affects the dispersion of pollutants in the streets. In fact, the impact of vegetation in urban areas and its use as mitigation measure for urban air pollution is actually matter of debate. The motivation is related to the complex processes that make difficult to draw general conclusions due to the high dependency on the local conditions. Several modelling techniques, particularly through the CFD, are being applied to study the impact of the urban vegetation on the dispersion of pollutants (Vos et al., 2013; Gromke and Blocken, 2015; Jeanjean et al., 2015; Santiago et al., 2017c; Abhijith et al., 2017). The impact of vegetation is caused by aerodynamic effect, modifying the flow, and by the deposition effect, removing pollutants from air through their deposition on the leaves. Several studies have found that the aerodynamic effect is larger than the pollutants removal by deposition (Vos et al., 2013; Santiago et al., 2017b; Buccolieri et al., 2018). In the CFD-RANS model, the vegetation is considered as a porous medium, which is limited to simplified geometries of the real trees located in the streets. Its aerodynamic effect is modelled as a sink of momentum caused by the drag force exerted that lead to a sink and 34 CHAPTER 2. THE CFD-RANS MODEL source of k and ε (Santiago et al., 2017b). Hence, these extra terms are added to the equations of momentum (Sveg,ui), turbulent kinetic energy (Sveg,k) and dissipation rate (Sveg,ε). Sveg,ui = −ρLADCd,veg |U |ui (2.38) Sveg,k = ρLADCd,veg ( βp|U |3 − βd |U |k ) (2.39) Sveg,ε = ρLADCd,veg ( Cε4 βp ε k |U |3 − Cε5βd|U |ε ) (2.40) |U | is the magnitude of wind speed and ui is the wind component in i direction. LAD is the leaf area density of vegetation, which depends on the type of trees (evergreen or caduceus) and, in general terms, on the season. Cd,veg is the drag coefficient of vegetation (0.2). βp is the fraction of the mean kinetic energy converted into k by the drag and takes values between 0 and 1. Based on previous studies (Santiago et al., 2013, 2017b), βp is equal to 1. The constant βd is a dimensionless coefficient for the turbulence cascade short-circuiting and Cε4 and Cε5 (Cε4 = Cε5) are based on the following analytical expressions (Sanz, 2003): βd = C1/2 µ ( 2 α )2/3 βp + 3 σk (2.41) Cε4(= Cε5) = σk ( 2 σε − C 1/2 µ 6 ( 2 α )2/3 (Cε2 − Cε1) ) (2.42) where the constant α is 0.5 (Dalpé and Masson, 2009). For the vegetation expression, the values of Cµ and Cε1 are regarded as constant values, 0.09 and 1.44 respectively. 2.6 Features on modelling urban areas To perform a CFD simulation over urban areas requires to follow a determined procedure related to the urban geometry and the computational domain (Section 2.6.1). The first step in the CFD simulation is the geometry of the computational domain. In real urban areas, the complexity of the geometry increases and there are too many obstacles to be designed (buildings, trees, tunnels and so on). After that, the domain is meshed with the appropriate type of cell and also the proper grid resolution. Both steps are carried out bearing in mind the optimization of the CFD simulation in order not to increase too much the number of grid points. Lastly, the features of boundary conditions are selected in the domain. Modelling the atmospheric processes requires to carry out several assumptions explained in Section 2.6.2. In addition, the implementation of the road traffic emissions are described in Section 2.6.3. CHAPTER 2. THE CFD-RANS MODEL 35 2.6.1 Geometry, mesh and boundary conditions The size of the computational domain is related to the geometry features of the area of interest and the boundary conditions that will be imposed. The COST Action 732 report (Franke, 2007) encompasses the basic guidelines for performing CFD simulations applied to micrometeorology. The height of the domain top should be at least at 4-5Hmax above the roof, where Hmax is the tallest building height. This prevents the influence of the top in the study area. For small blockage, it is acceptable to establish the top of the domain at 4Hmax and for larger blockage around 10Hmax. In urban areas, with a vast number of buildings, the vertical extension is recommended to impose it at 5Hmax. The lateral extension is dependent on the height of the obstacles and, besides, on the boundary conditions of the flow. In urban areas, the lateral boundaries should be placed at 5Hmax to the buildings region in order not to affect the solution in the research area (Fig. 2.2a). For that reason, commonly, the CFD domain is horizontally reduced to size about 1 km x 1 km. (a) (b) Figure 2.2: (a) Computational domain over an urban area. (b) Polyhedral mesh. The computational domain is discretized in grid points and the mesh resolution may affect the results. Accordingly, the mesh has to be fine enough to capture the dynamical processes with high resolution without increasing too much the computational load. To that end, the optimization of the mesh is important. The grid stretching should progressively change with a ratio below 1.3 to keep a small truncation error. Depending on the geometry, a recommendable type of the mesh is the tetrahedral shape since introduce small truncation errors and show better convergence (Hirsch et al., 2002). But currently, there are better types of mesh like the polyhedral grid with dodecahedra shapes, very useful for complex and non-uniform geometries (Figure 2.2b). As for the grid resolution, it is completely determined by the research purpose, geometry and the boundary conditions. Thus, a mesh dependency test is recommended in order to check that the outputs are independent of the grid resolution. Additionally, parallel grid lines to the walls with small resolution provide accurate results close to the buildings (Franke, 2007). 36 CHAPTER 2. THE CFD-RANS MODEL The boundary conditions represent the influence of the surroundings on the study area. For that reason, the inflow conditions are imposed at a certain distance in order not to alter the solution inside and, besides, the inlet conditions are able to fully develop. Thus, the top conditions should be imposed to prevent the change on the inflow conditions and, thereby, the symmetry boundary condition is commonly used. This feature enforces a parallel flow to the boundary by forcing the normal component to vanish and, besides, prescribes zero normal derivatives for all dynamics variables. This condition is also often imposed in the lateral boundaries, when the flow is parallel to them to facilitate the numerical convergence. In relation to the outflow, the open boundary conditions are used with a constant static pressure, where the derivative of the flow variables are forced to vanish. As for the wall boundary conditions, smooth and rough walls are considered in urban areas by means of a wall function approach (Wall All Y+ treatment). The ground is considered as a rough surface with a roughness parameter, which should be lower than the vertical grid resolution (Blocken et al., 2007). In urban environments, the effect of the wall functions on the solution away from the wall is considered small (Franke, 2007). In this work, smooth walls on buildings are assumed since the impact on the atmospheric flow caused by its geometry are larger than the roughness of the buildings walls. 2.6.2 Modelling atmospheric inflow conditions The meteorological variables required at inlet of the CFD simulation are established in base to some assumptions. The lack of the vertical information from experimental data lead to estimate the inflow profiles in a theoretical manner. Assuming near-neutral atmospheric conditions, the inflow is defined by the logarithmic profile law (Eq. 1.2), whereas the inlet conditions of the turbulence parameters, k and ε, are computed as follows (Richards and Hoxey, 1993): uin(z) = u∗ κ ln ( z + z0 z0 ) (2.43) kin = u∗ 2 Cµ 1/2 (2.44) εin(z) = Cµ 3/4kin 3/2 κz (2.45) In this way, in urban areas, the flow vertical profiles can be derived from the wind speed measured at a certain height. It should be noted that the roughness length has to be established in the inlet wind profile according to the roughness imposed at ground. In urban environments, this assumption can be suitable when the thermal effects are small. Consequently, the UBL is thermally in near-neutral conditions and the wind flow is basically CHAPTER 2. THE CFD-RANS MODEL 37 modified by the surface roughness (buildings or any obstacles). However, this atmospheric condition does not strictly occur. Indeed, the wind and temperature conditions irregularly change in height on account of the thermal gradients between surface and atmosphere. In stability conditions further away from neutral atmosphere, the vertical profiles of wind and turbulence considerably differ from the previous expressions (Equations 2.43-2.45). In these cases, the mesoscale models can provide the meteorological variables in order to impose more accurate boundary conditions into the CFD-RANS simulations. Therefore, in these cases, in addition to the wind and turbulent parameters, the vertical profile of temperature is also required at inlet. 2.6.3 Road traffic emissions Road traffic is the main source of air pollution in urban areas, being the primary emitted pollutants in gas-phase: NOx (NO and NO2), CO and VOC. The emission rate per vehicle is dependent on several factors linked to the vehicle or the driving behaviour, e.g. the motor internal combustion, namely, the kind of combustible petrol or diesel, the type or size of vehicle and, as well, the driving speeds. The emission factors are broadly classified taking into account the velocity and the type of vehicle (HBEFA, 2010). In general terms, the standard emission of NOx for a passenger car travelling at 30 km h−1 is around 0.5 g km−1 (Baker et al., 2004), but this NOx rate changes depending on the type of vehicle (motorcycle, light or heavy vehicle, bus and so on). Initially, few years back, the most part of NOx emissions was considered NO, meaning that the emitted NO was markedly superior to the NO2 (NO−to−NO2=10:1) (Buckingham, 1997). However, the trend of this ratio shows a decrease in last years (Carslaw, 2005). Nowadays, the emission ratio of NO2/NOx might be considered around 30% (Borge et al., 2014). Additionally, the VOC are other important component in the vehicle exhaust emissions. Overall, the proportions of the emitted pollutants associated with the road traffic of Madrid are shown in Fig. 2.3a. Regarding to the driving in heavy trafficked areas in the city center, the emission ratio VOC-to-NOx may be around 1/6 or 1/5 according to the emission inventories reports of Madrid (Madrid-Council, 2014). Nevertheless, this ratio may differ not only as a consequence of the driving features (different speed limits, traffic lights and so on), but also of the type of vehicle. 38 CHAPTER 2. THE CFD-RANS MODEL (a) (b) Figure 2.3: (a) Ratio of the emitted pollutants by road transport reported in the Emission Inventories of Madrid. Source: (Madrid-Council, 2014). (b) Representative daily traffic pattern in a European city, e.g. Madrid (Quaassdorff et al., 2016). Figure 2.3b illustrates a representative traffic flow pattern in an urban hot-spot in Madrid (Quaassdorff et al., 2016). The traffic flow (veh/h) usually varies depending on the features (population, commercial activities, etc) and the human activities of the urban area. However, the daily trend of a weekday is usually marked by two maximum values in the morning and in the evening with another slight maximum at the central hours of the day. In the CFD simulations, the traffic emissions are limited to the road areas and the emission rate is added as a source term in the transport equation (Eq. 2.33). On the one hand, the emission rate is commonly computed through the number of vehicles on each street. Overall, this information is usually provided by the annual average daily traffic (AADT) or, otherwise, it can be also derived from specific experimental campaigns. Therefore, the emission rate is computed with the previous emissions factors and is uniformly distributed along the streets. On the other hand, microscale traffic models combined with an emission model provide the traffic emission with high spatial and temporal resolution in the streets (Abou-Senna et al., 2013; Quaassdorff et al., 2016). The traffic model computes detailed trajectories that describe the behaviour of individual vehicles considering drivers’ reaction times, route selection logic and lane change logic. It provides essential information about braking-acceleration patterns that have a huge influence on the amount of pollutants emitted and their distribution in space and time. The emission estimations are based on very detailed fleet composition and vehicle flow data in the research area obtained from an intensive measurement campaign with cameras (Quaassdorff et al., 2016). Therefore, the NOx release is obtained with resolution of meters and, besides, the temporal variation of the spatial distribution over time. Figure 2.4 displays the spatial difference of the NOx emissions for a particular scenario by using both procedures. In the first case, the pollutant emission is uniformly distributed along CHAPTER 2. THE CFD-RANS MODEL 39 the street depending on the number of vehicles at that hour. In contrast, for the same hour, in the outputs from the microscale traffic models, the NOx emissions spatially change due to the specific features in this region, such as traffic lights or crosswalks. (a) (b) Figure 2.4: Examples of (a) uniform traffic emission layout and (b) detailed spatial distribution of the traffic emissions. 2.7 Statistical model evaluation This section is addressed to present the statistical analysis for the CFD models applied to air quality assessment. The management of urban air quality in the decision-making on air pollution mitigation calls for reliable results of the modelling techniques. In urban environments, the dispersion of pollutants in the streets is fundamentally controlled by turbulent processes. Thus, the random nature of the turbulence gives rise to a great spatial and temporal variability, which hinders a fully accurate simulation of the atmospheric processes. In addition to that, the internal numerical errors and the assumptions related to boundary conditions also contribute to the uncertainty of the modelling results. Nevertheless, the outputs of CFD models have been widely validated with experimental data in multiple scenarios, turning out good results of the atmospheric processes. Therefore, the CFD models are able to reproduce the dynamical and chemical processes in distinct scenarios but its evaluation with measurements is an essential requirement. For further quantitative detail, a statistical analysis helps to determine the accuracy of the modelling results. In the framework of the microscale modelling for urban air quality, Chang and Hanna (2004) proposed criteria for the statistical parameters in order to characterise the 40 CHAPTER 2. THE CFD-RANS MODEL reliability of the outputs. To that end, the most characteristic statistical indicators are: the Normalized Mean Square Error (NMSE), the Fractional Bias (FB), the correlation coefficient (R) and the the factor 2 (FAC2), whose mathematical expressions are the following: NMSE = ∑n i=1(Oi − Pi)2∑n i=1(OiPi) (2.46) FB = P −O 0.5(P +O) (2.47) R = ∑n i=1[(Oi −O)(Pi − P )][∑n i=1(Oi −O)2 ]1/2 [∑n i=1(Pi − P )2 ]1/2 (2.48) FAC2 = fraction of data that satisfy 0.5 ≤ Pi/Oi ≤ 2 (2.49) where Oi represents the observed values and Pi the modelling results, and O and P their corresponding mean values. The NMSE estimates the overall deviation between the modelling results and the observations. The FB indicates the underestimations or overestimations of the observed values. The correlation coefficient reflects the linear relationship between the outputs and measurements, and it is not individually a decisive parameter, but it is necessary to conclude good outcomes. Lastly, the FAC2 only indicates a deviation range, and it is not influenced by the high and the low outliers. A perfect outcome would mean a value of R and FAC2 equal to 1, and the FB and NMSE equal to 0. The influence of the random nature of the atmospheric processes together with the modelling uncertainties prevent from obtaining a perfect model for air quality assessment. The previous statistical metrics are classified in ranges that determine the accuracy of the results based on the values associated to a perfect model. Table 2.5 shows the acceptance criteria established in Chang and Hanna (2004) and in Goricsán et al. (2011) for the evaluation of air quality modelling results. In a recent study, Hanna and Chang (2012), proposes new acceptance criteria for urban applications with ranges more flexible for the modelling results. Nevertheless, in this work, the stricter ranges are considered to evaluate the outputs of the CFD-RANS simulations. It is worth noting that the approximation of the model to the real atmospheric processes might be considered appropriate or not depending on the study purpose. For instance,. The management of urban air quality is interested in the distribution of pollutants in the streets during long periods (weeks, months or even a year). Thus, in these cases, the exact values of high concentrations are usually less important than their locations for regulatory applications (Chang and Hanna, 2004). CHAPTER 2. THE CFD-RANS MODEL 41 Table 2.5: Acceptance criteria of the statistical metrics for the modelling evaluation (Chang and Hanna, 2004, 2005). Adapted from Goricsán et al. (2011). Validation metrics Range Membership function Normalise Mean Square Error (NMSE) NMSE < 4 Good 925 Poor Fractional Bias (FB) -0.31.33 Poor Correlation Error (R) | R |>0.8 Good 0.5<| R |<0.8 Fair |R|<0.5 Poor FAC2 0.5< FAC2 <2 Good 0.3< FAC2 <0.4 Fair FAC2 <0.2 Poor 42 CHAPTER 2. THE CFD-RANS MODEL Chapter 3 Modelling dispersion of reactive pollutants in simplified geometries with the chemical schemes: NOx−O3 and NOx−Ox−VOC 3.1 Introduction In urban environments, the proximity between sources and receptors in the street entails that only the fastest chemical reactions have an impact on pollutant concentrations. First studies focus on modelling chemical reactions on the pollutant dispersion at street level in simplified geometries. Baker et al. (2004) and Baik et al. (2007) studied the dispersion of reactive pollutants in a street canyon considering the NOx−O3 photostationary state. Both researches concluded that the O3 concentration was faster depleted within the canopy due to the high NO emission at ground level. Moreover, Baik et al. (2007) found that the magnitude of the chemical terms was comparable to the advection or turbulent diffusion for the O3. In contrast, for the NO or for NO2 concentrations, the dynamical terms were larger than the chemical ones. Nevertheless, these studies underlined the importance of including the O3 photochemical reaction in the study of the dispersion of NO and NO2 within the canopy. Additionally, the VOC related to traffic emissions are also involved in urban air pollution. This makes it necessary to include a more complex chemical scheme by modelling the interactions among NO, NO2, O3 and VOC. In recent studies, complex chemical mechanisms have been implemented in CFD models in order to reproduce the NO and NO2 dispersion in isolated streets. Kwak and Baik (2012) included the CBM-IV mechanism in a CFD model and evaluated the O3 sensitivity with respect to different emissions of VOC and NOx. The 43 44 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC high levels of NO in regards to the NO2 at street level concluded that the O3 loss, by reaction with NO, prevails over its formation by NO2 photolysis. Hence the O3 sensitivity is weakly related to the VOC emission level and negatively correlated with the NOx behaviour. Kwak et al. (2013) obtained the same results about the relation between the O3, NOx and the VOC emission levels. They examined the dispersion and photochemical evolution of reactive pollutants in a street canyon with different canyon aspect ratios. Note that it is the ratio between building-height and the street-width. They found a relation with the wind speed whereby its increase enhances the exchange with overlying air that promote the downward O3 transport into the street. These studies highlight the importance of the O3 oxidation process by assessing the photochemical reactions in the street. Similar major aspects of the relation NOx, VOC and O3 are obtained in three-dimensional street canyons by Park et al. (2015). However, small variations are locally found caused by the flow patterns resulting from the geometry of the domain. The model sensitivity to the chemical mechanism used in the simulation is also thoroughly analyzed. Kim et al. (2012) compared, in identical conditions, the hourly concentration of NO, NO2 and O3 resulted from the photostationary steady state and a full chemical mechanism (110 species and 343 reactions). The outputs provided similar concentrations of NO and NO2 by using any of the mechanisms, whereas for the O3 concentration, with the simple mechanism turns out lower than with the full scheme. Bright et al. (2013) compared the O3 − NOx scheme with a more complex mechanism (51 chemical species and 136 reactions) using a LES model. They found that the values of NO, NO2 and O3 differ from the photostationary state by using a more detailed chemistry, which reflects the additional transformation of NO into NO2, resulting from the intrinsic VOC degradation processes. The purpose of this chapter is to study the importance of chemical reactions on the dispersion of reactive pollutants in two simplified settings in a 2D- and 3D- geometries. Focusing on the chemistry of NO and NO2, two chemical schemes are considered: the NOx−O3 and the NOx−Ox−VOC mechanisms. Hence the dispersion of NO and NO2 are modelled: with no chemical scheme, with the NOx−O3 scheme and with the NOx−Ox−VOC mechanism. Additionally, the influence of atmospheric conditions on the chemical activity is particularly examined through different cases of wind speed, solar position and VOC/NOx emission ratio. The implementation of chemical reactions in the CFD model increases the computational load. Therefore, this study is also addressed to quantify the chemical influence in several atmospheric scenarios and, thus, to find the best compromise between the purpose to be reached and the available computational resources. It is worth mentioning that the NOx−Ox−VOC scheme involves more than twice the computational time required by the photostationary steady state. CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 45 Accordingly, the outcomes of this work provide further details about the coupled behaviour of chemical and dynamic processes in order to be able to respond the following research questions: • How does the dispersion of NO and NO2 in the urban canopy layer change by including the NOx−O3 and NOx−Ox−VOC schemes in the CFD simulation? • Which is the impact of the meteorological conditions on the reactive pollutants dispersion? 3.2 CFD model description The CFD-RANS model with the Realizable k -ε closure, described in Section 2.2.2, is used to analyse the reactive pollutant dispersion in simplified geometries in order to characterise the chemical behaviour by imposing different atmospheric conditions. 3.2.1 Simulation setup Figure 3.1 illustrates the computational domains: a single street-canyon (2D-geometry) and a staggered array of cubes (3D-geometry) and their respective sizes in x, y and z directions are: 24 m x 40 m x 64 m and 64 m x 64 m x 64 m. (a) (b) Figure 3.1: (a) Domain of the street canyon with H = W =16 m (2D-geometry). (b) Domain of the staggered array of cubes with H = W = L =16 m (3D-geometry). Pink region stands for the emission location. 46 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC The aspect ratio, defined as the ratio of building height (H) to the street width (W), is H/W = 1 in both geometries. The domain top is fixed at 4H considered high enough to not affect the dynamical processes at street level (Coceal et al., 2006). The 3D-geometry is represented by staggered array of cubes with the width of the buildings equal to their height (L=H), giving rise to a building packing density λp = 0.25, which is a typical value of real urban areas (Grimmond and Oke, 1999). Note that λp is expressed by the ratio between the plan area of buildings and the total surface area. The mesh resolution of 1 m in all directions was selected after performing a grid independence test and selecting an optimized mesh for carrying out multiple simulations. The same boundary conditions are imposed to both geometries. Symmetry conditions are established in the spanwise direction (y-direction) and at the top of the domain, which enforce parallel flow and zero normal derivatives for the dynamic variables. Cyclic boundary conditions are imposed in the streamwise direction, which represents the simulation of an infinite number of streets. Assuming isothermal condition (T=298 K), the flow is driven in x-direction by a pressure gradient equal to ρu2 τ/4H, where uτ is a reference velocity (Santiago and Martilli, 2010). In this study, two values of uτ are considered: uτ=0.45 m s−1 and uτ=0.225 m s−1. In this way, the wind speed at 1.5H resulted from using uτ=0.225 m s−1 is 1.9 m s−1 and 1.5 m s−1 in the 2D-geometry and the 3D-geometry, respectively. As the wind speed is proportional to uτ , imposing uτ=0.45 entails that the wind speed is twice as large as the previous case. Regarding the pollutant concentrations, this study focuses on modelling the NO and NO2 using different chemical approaches: (a) no chemistry, (b) the NOx−O3 scheme and (c) the NOx−Ox−VOC scheme. The chemical mechanisms are described in detail in Section 2.3.2. And, as explained in Section 2.4.1, the same time step, ts = 1 s, is used to simulate both chemical and dynamical processes, reaching the quasi-steady state in all simulations. The source of pollutants is located at ground level in both domains (Fig. 3.1). The NOx emission is assumed at 0.5 g km−1 per vehicle (Baker et al., 2004; Baik et al., 2007). The emission ratio used was based on previous studies with ratio NO−to−NO2 equal to 10:1 (Buckingham, 1997). Hence the NO and NO2 rate are fixed at 112 µg m−1 s−1 and 17 µg m−1 s−1 respectively, which is equivalent to 930 vehicles per hour. In addition, the VOC emissions are included by applying the NOx−Ox−VOC scheme. In this study, two emission ratios of VOC/NOx are simulated: 1/5 and 1/2. The VOC/NOx = 1/5 is the last value registered in Madrid (Madrid-Council, 2014) but, even so, a higher value (VOC/NOx = 1/2) is also considered. As explained in Section 2.3.2, some VOC are lumped in terms of specific functional groups. The VOC species related to traffic emissions are: OLE, ARO, ALK, ALD and HCHO; and their volumetric proportions are established, based on the emission factors of HBEFA (2010), at: 28.6 %, 23.1 %, 38.6 %, 4.0 % and 5.6 %, respectively. CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 47 The boundary conditions for solving the transport equations are considered as cyclic and symmetry conditions along x- and y-directions respectively. In these conditions, the pollutant concentrations imposed at the top of the domain plays an important role on the chemical reactions, except for the NOx and VOC, because their concentrations in the street are practically due to the traffic emissions. Therefore, the concentrations fixed at the top for NO, NO2, CO and SO2 are respectively 16, 35, 200 and 2 ppb in all scenarios, similar values can be found in the urban atmosphere. As for the VOC concentration, the same ratio VOC/NOx imposed for the emissions is used at the top, and it is deduced from the NOx value (51 ppb) holding the emitted proportion for each specie of VOC. Hence the concentration of VOC in air is considered fully associated to the traffic emissions. Finally, the O3 concentration at the top of the domain is computed by the Leighton Relationship (Eq. 2.32). To analyse the influence of the available O3 on the NO and NO2 levels in the street, two particular cases of background O3 are studied, at rush hour, when high concentrations of NOx are expected. The photolysis rate, JNO2 , is calculated depending on the solar zenith angle (Table 2.3). Hence, a representative solar zenith angle (in mid-latitudes) in winter (ϕw = 78◦) and another one in summer (ϕs = 46◦) are considered to compute the O3 established at top of the domain. This results in two different cases of background O3 at the top of the domain: 10 ppb and 40 ppb. In this chapter, the solar position has been only used to compute the photolysis rate and, the thermal effects on flow and the shadow effects of building on photochemistry are neglected. Including shading effects would imply to study the shadows and the temperature distribution within the street canyon, depending on solar position and street orientation. In shaded zone, the reaction rates decrease (slower chemical reactions) due to the lower radiation and temperature in comparison with the sunlit space. Kwak and Baik (2014) considered the impact of the shadows, which were represented in a simple way, on photochemistry in a street canyon. This is an interesting issue to be study in future works, however, in order to limit the atmospheric conditions analysed, it has been considered out of scope of this research. 3.2.2 CFD model evaluation Controlled experiments of pollutant dispersion in simplified urban configurations are carried out in wind-tunnel, where only non-reactive pollutants can be studied. Note that the implementation of the chemical mechanisms has been previously evaluated in a box model configuration (Section 2.4.2). Therefore, in this section, the dispersion of the non-reactive pollutants, in the 2D-geometry, is compared against the measurements from a wind tunnel experiment performed in a similar street-canyon (Meroney et al., 1996). 48 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC Figure 3.2: Location of the comparison points with measurements from Meroney et al. (1996) The experimental system consists on a single 2D street canyon with H/W = 1 (Fig. 3.2). The wind blows perpendicular to the canyon in x-direction and the emission source is located at the bottom of the domain. This scenario is similar to CFD simulation in the 2D-geometry without imposing the cyclic boundary conditions in streamwise direction. For the comparison with experimental data, just the concentration of the traffic emissions are considered and the concentration gradients within the canyon are evaluated among the points nearby the walls (Fig. 3.2). Therefore, the concentration at each measurement point is normalized with respect to the concentration at the point 7 within the street (C/C7). The same normalization was used in Santiago and Mart́ın (2008) to evaluate the CFD model results with the same experiment. Figure 3.3 displays the comparison of the normalized concentration from the simulation against the experimental values. (a) (b) Figure 3.3: Comparison between simulation results and experimental data (Meroney et al., 1996). (a) Comparison between modeled concentration and experimental measurements at each location. (b) Linear fit between experimental measurements and modeled concentrations. Overall, a good agreement is obtained in spite of the slight difference at points 4 and 13, where CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 49 the concentration is lower than experimental data. Besides, the linear regression concludes a good correlation coefficient (R=0.92) between the measurements and the simulated results. For further quantitative information, the statistical parameters of NMSE, FB and the FAC2 are computed following the Eqs. 2.46-2.49. The statistical results are: NMSE=0.03 (NMSE<1.5), FB= 0.068 (-0.3) of the normalized wind speed (U/uτ ) and the turbulent kinetic energy (< k > /u2 τ ). Larger turbulence and, thereby, greater turbulent mixing below the canopy top, is turned out in the 3D-geometry (Fig. 3.5). 52 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC (a) (b) Figure 3.5: Vertical profiles of (a) < U > /uτ and (b) < k > /u2 τ in the 2D-geometry and 3D-geometry. 3.3.2 Ozone influence on reactive pollutants As explained in Section 1.4, the O3 concentration is generally dependent on the photochemical regime but, in the street, it is also affected by the chemical interactions with the NOx and VOC levels due to traffic emissions. In addition, the photochemical reactions are dependent on the solar position and, as a result of the equilibrium state, two values of the O3 background are imposed at the top (Case 2 and Case 4 in Table 3.1). Therefore, NOR and NO2R, from each chemical scenario, are compared against the NOT and NO2T in order to determine the atmospheric conditions that give rise to the largest impact on the concentration of the reactants. Figure 3.6 shows a vertical section of NOT and δNO for the Case 2 and Case 4 in the 2D-geometry. In the Case 4, δNO obtained from any of the reactive approaches is less than 3-4%, without a significant variation among them below the canopy top. In contrast, δNO increases to nearly 10% in the Case 2 and, only a small difference of 2-4% exists between the scenario CCM1/2 and the others (PSS and CCM1/5). Likewise, δNO2 is calculated with the PSS, CCM1/5 and CCM1/2 for the Case 2 and Case 4 (Fig. 3.7). In the latter, δNO2 shows analogous values from all chemical scenarios, below 15-16% in the street. Accordingly, reactions involving VOC have hardly an impact on the NO and NO2 levels in comparison with the result of the PSS. In the Case 2, δNO2 rises up to 50% of difference from the non-reactive scenario. Actually, in this case, the use of the CCM1/2 results in larger deviation for the NO2 concentration against the NOx−O3 mechanism (PSS). CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 53 (b) (c) (d) (a) (e) (f) (g) Figure 3.6: (a) Vertical section of NOT in ppb. Vertical section of δNO (%) obtained respectively from the PSS, CCM1/5 and CCM1/2 chemical scenarios for (b-d) the Case 2 and (e-g) the Case 4. (b) (c) (d) (a) (e) (f) (g) Figure 3.7: (a) Vertical section of NO2T in ppb. Vertical section of δNO2 (%) obtained respectively from the PSS, CCM1/5 and CCM1/2 chemical scenarios for (b-d) the Case 2 and (e-g) the Case 4. 54 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC Similar to the 2D-geometry, the chemical deviation through δNO and δNO2 are analysed in a horizontal section, at 3 m above the ground level (AGL), in the 3D-geometry. Given that the largest chemical deviations are previously obtained in the Case 2, δNO and δNO2 are just depicted in these conditions for the 3D-geometry in Fig. 3.8. (a) (b) (c) (d) (e) (f) (g) (h) Figure 3.8: Horizontal section at 3 m AGL for the Case 2 of: (a) NOT (ppb); (b-d) δNO (%) from PSS, CCM1/5 and CCM1/2 respectively; (e) NO2T (ppb) and (f-h) δNO2 (%) from PSS, CCM1/5 and CCM1/2 respectively. At street level (3 m AGL), both δNO and δNO2 hardly show differences by implementing either the PSS or the CCM1/5. By increasing the VOC emissions with the CCM1/2, the concentrations of NO and NO2 faintly rise from the PSS. The overall values in terms of the spatial average, at 3 m, of δNO with the PSS, CCM1/5 and CCM1/2 yields 8.9%, 9.4% and 10.9% respectively, whereas for δNO2 reaches values of 47%, 49% and 55% in the Case 2. Hence, an important growth in the reactive NO2 levels is obtained in this case. In addition, similar differences are achieved in the 2D-geometry close to the ground. Regarding to the VOC influence, the O3 gained by including VOC reactions is analysed in the 3-D geometry, at 3 m, for the case 2 (Fig. 3.9). CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 55 (a) (b) (c) Figure 3.9: Maps at 3 m of the O3 concentration (ppb) for the (a) PSS, (b) CCM1/5 and (c) CCM1/2 in the Case 2. The O3 map of the CCM1/2 show the increase of the O3 concentration with the CCM1/2. The VOC reactions increase the NO2 levels and, thus, the photolysis of NO2 promotes the O3 production. For an overview of all chemical scenarios in the Case 2 and Case 4, the normalized concentrations are computed (Eqs. 3.2). Hence, the profiles of the horizontal spatial average of the normalized concentration, < NON > and < NO2N >, are depicted in Fig. 3.10. Note that the red line represents the normalized concentration of the non-reactive pollutant, which is the same for the NO and NO2. The differences on the < NON > and < NO2N > profiles between the domains are basically caused by the flow pattern. Whereas in the 3D-geometry the pollutants are progressively reduced from the bottom, in the 2D-geometry, the average concentration holds practically constant in height up to the top of the canopy. It corresponds to the vertical profiles of wind speed and k, resulted from the interaction of the flow and the obstacles (Fig. 3.5). Below the canopy top, the values of k are larger in the 3D configuration than in the 2D geometry, which encourages the turbulent mixing. Nevertheless, regarding to the chemical influence, the same general conclusions are obtained in both domains, independently of the type of configuration. Secondly, the vertical distributions bear out the previous outcomes, where the difference from the non-reactive approach is clearly larger in the Case 2 (with larger background O3). On the other hand, there is no difference among the chemical scenarios (PSS, CCM1/5, CCM1/2) in the Case 4 (with lower background O3) whereas in the Case 2, small variations in the NO and NO2 are obtained with the CCM1/2 with respect to PSS and CCM1/5. Even the results of CCM1/5 hardly differ from the PSS. 56 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC (a) (b) (c) (d) Figure 3.10: Vertical profiles of horizontal spatial average of < NOR,N > and < NO2N > for all chemical scenarios in Case 2 (solid line) and Case 4 (dotted line) in (a,b) the 2D-geometry and (c,d) the 3D-geometry. To summarize these outcomes, the atmospheric conditions related to the solar position and the ambient O3 give rise to important variations on the NOR and NO2R concentrations. With lower O3 (Case 4), the relative differences of concentration between the reactive and non-reactive approaches are smaller with values below 15% (Fig. 3.7). Nonetheless, in the Case 2, the concentration of the NO2R significantly increases up to values above 60% from the NO2T in both geometries (Fig. 3.7-3.9). Concerning the influence of the VOC, the NO2 and O3 levels from these atmospheric cases reveal a slight impact using the NOx−Ox−VOC scheme, even with greater ambient O3 (Case 2). However, by increasing the VOC release, slight differences are obtained on the NO and NO2 levels in comparison with using the NOx−O3 scheme. It is due to the NOx levels are substantially elevated against the VOC. Hence, the NOx levels tend to inhibit the O3 formation and, thus, to limit the O3 production owing to the VOC reactions. In turn, the available O3, from the background, promotes the loss of NO. For that reason, the influence of the VOC in the lower O3 case is negligible whereas, with enough available O3, its importance CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 57 is slightly noticeable. 3.3.3 Wind speed influence on reactive pollutants In this section, the impact of wind speed on the chemical influence is analysed with the largest background O3 (Case 1 and Case 2 in Table 3.1). To do that, different wind speeds, uτ=0.45 m/s and uτ=0.225 m/s, are simulated keeping constant the rest of atmospheric variables. Based on previous results, the cases of the greatest background O3 are here studied given that larger chemical influence is obtained within the canopy. The concentration of a tracer (non-reactive compound) is inversely proportional to wind speed by subtracting the background concentration (Appendix 6.2). However, the non-linearity of the chemical reactions brings about changes in this behaviour. The deviation in concentration of reactive pollutants, at street level, is dependent on both the dispersion within the canopy and the exchange with the overlying air, linked to the street ventilation. Concerning the chemical influence, the same conclusions are obtained for the NO and NO2. Hence, just the NO2 results are shown in this section (more detail about the NO results in Sanchez et al. (2016). Figure 3.11 displays the comparison of δNO2 for the Case 1 and Case 2 in the 2D-geometry. (a) (b) (c) (d) (e) (f) (g) (h) Figure 3.11: For the Case 1: (a) NO2T (ppb) and (b-d) δNO2 (%) from PSS, CCM1/5 and CCM1/2. For the Case 2, (e) NO2T (ppb) and (f-h) δNO2 (%) from PSS, CCM1/5 and CCM1/2. 58 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC Despite the NO2T within the canopy is lower in the Case 1 than in the Case 2, around 50 ppb versus 70 ppb, δNO2 turns out close values. Nevertheless, the values of δNO2 are around 55-60% in the Case 1 and slightly lesser in the Case 2. The main particularity between both cases focuses at the bottom, close to the emissions. In the Case 1, the faster wind speed enhances the street ventilation, which boosts the mixing of pollutants and triggers the main interactions between NOx and O3. As a consequence, greater chemical influence from the non-reactive approach are obtained. On the other hand, the larger differences between CCM1/2 and PSS are found in the Case 2, even above the canopy. At street level, δNO2 is examined in the 3D-geometry. Figure 3.12 displays δNO2 resulted from the PSS, CCM1/5 and CCM1/2, both in the Case 1 and the Case 2. (a) (b) (c) (d) (e) (f) (g) (h) Figure 3.12: Horizontal section at 3 m AGL of: (a) NO2T (ppb) and (b-d) δNO2 (%) from PSS, CCM1/5 and CCM1/2 respectively, in the Case 1. For the Case 2: (e) NO2T (ppb) and (f-h) δNO2 (%) respectively from the PSS, CCM1/5 and CCM1/2. In the 3D-geometry, comparing with the non-reactive approach, δNO2 shows larger differences in the Case 1, as in the 2D-geometry. The rapid flow enhances the pollutant withdrawal from the street and encourages the pollutant mixing with the overlying air. Thus, the O3 inflow to the street rises triggering the conversion of NO into NO2. As a result, greater deviations of reactants with respect to the non-reactive approach is obtained. As for the influence of VOC reactions, both cases show similar δNO2 for the PSS and CCM1/5 and it is slightly larger from CCM1/2. CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 59 The average concentration at 3 m of δNO2 yields 57%, 57% and 59% for the PSS, CCM1/5 and CCM1/2 in the Case 1 whereas in the Case 2, it is respectively 47%, 49% and 55%. There is low difference in δNO2 from both wind cases, being greater with larger O3 entrainment into the street (Case 1). However, the influence of VOC on the NO2 concentration is more relevant with lower wind speed (Case 2). Whereas in the Case 1 the chemical activity in the street is basically dominated by the photostationary state, in the Case 2 the VOC interactions become important in the NO conversion to NO2. Lastly, the vertical profiles of < NO2N > in each domain is depicted in Fig. 3.13. (a) (b) Figure 3.13: Vertical profiles of < NO2N >, for all chemical scenarios, in the Case 1 (solid line) and Case 2 (dotted line) for: (a) the 2D-geometry and (b) the 3D-geometry. The behaviour of the chemical influence in response to vary the wind speed is examined comparing the vertical distributions. For a non-reactive pollutant, the same profile of < NO2N > is obtained for the different wind speeds (red line). However, the < NO2N > profiles from any of the chemical schemes exhibit the non-linearity of chemistry by increasing the wind speed twice as much. Therefore, the reduction of the wind speed by half does not result in the double of the chemical influence. The differences between chemical approaches are consistent with the results previously discussed. This reveals that the impact of the chemical reactions on pollutant concentrations is dependent on a set of atmospheric factors, particularly in this study, on the wind speed and the available O3 jointly . 3.4 Vertical transport In this section, all atmospheric cases are analysed by means of the vertical concentration fluxes. In this way, both the removal of pollutants from the street and entrainment into the 60 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC street are analysed in each atmospheric condition. Given that the global conclusions of the chemical behaviour are similar in both domains, this section only depicts the vertical fluxes in the 3D-geometry. The total flux (FT,i) is expressed by the sum of the mean flux (Fm,i) and the turbulent flux (Ft,i). The horizontal spatial average (<>) of the vertical concentration fluxes for the NO, NO2 and O3 are performed every 1 m in height. Hence, FT,i for ith species is computed as follows: FT,i =< wC >= Fm,i + Ft,i (3.3) Fm,i =< wCi > (3.4) Ft,i =< w′C ′i >= − < Kc ∂Ci ∂z > (3.5) Here w and w′ are the mean component of the vertical velocity and its fluctuation respect to the mean. And Ci and C ′i are the corresponding components to the concentration of ith species. For a harmonious study of the NO and NO2, the vertical fluxes are normalized in order to properly compare the results with the non-reactive approach. To do that, the concentration imposed at the top is dismissed in the computation of the fluxes. The concentracion fluxes are normalized following the method developed for an inert pollutant in Martilli et al. (2013). Hence, the FT,i for non-reactive species, considering an homogeneous emission, the steady state and neglecting vertical changes in the air density, is normalized as: below the canopy top, < wC >N = < wC > + < w′C ′ > Sem (3.6) above the canopy top, < wC >N = < wC > + < w′C ′ > Sem Astreet + Aroof Astreet (3.7) where Sem is the constant emission flux (ppb m2s−1) at the surface. Astreet and Aroof are the horizontal areas of the streets and the building roofs respectively. Note that the average total flux of a non-reactive pollutant is equal in all atmospheric conditions, turning out one only profile for the NOT and another one for the NO2T. Hereafter, the horizontal spatial average of the normalized total fluxes of NO and NO2 is referred to as < wNO >N and < wNO2 >N . Figure 3.14 illustrates the vertical profiles of < wNO >N , < wNO2 >N and < wO3 > for the CCM1/2 and PSS in the 3D-geometry. The results from the CCM1/5 are not shown since they are practically equal to that the PSS. CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 61 (a) (b) (c) (d) (e) (f) Figure 3.14: Vertical profiles of< wNO >N , < wNO2 >N and< wO3 > (ppb m s−1) respectively for (a-c) the PSS and (d-f) the CCM1/2. The vertical transport of a non-reactive pollutant above the canopy top is mainly dominated by the turbulent processes because the mean vertical velocity is zero, but instead, in the street, the vertical motions are mainly controlled by the mean flux. The dispersion of the reactive pollutants is also managed by these dynamical processes, being in turn affected by chemical interactions with O3 and VOC. Both in the CCM1/2 and PSS scenarios, the largest deviation of < wNO >N and < wNO2 >N from the non-reactive profile is obtained for the Case 1 and Case 2 (green lines), essentially associated with a highest O3 inflow in the street (Fig. 3.14c-f). In contrast, for the Case 3 and Case 4 (blue lines), the vertical transport of the reactants is fairly similar to the non-reactive approach within the canopy. In these cases, the influence of introducing chemical mechanisms just can be observed above the canopy top, where the O3 concentration is not rapidly reduced by the high NO levels in the street. Therefore, the greatest impact of the chemical activity on pollutants is obtained in atmospheric conditions that promote the vertical mixing of pollutants. Comparing < wNO >N and < wNO2 >N obtained from the PSS and CCM1/2, VOC reactions barely produce changes in the outflows of NO and NO2 for the Case 2. Lastly, the rapid conversion of NO into NO2 is studied by the ratio NO2N/NON for the PSS, CCM1/5 and CCM1/2 in all atmospheric cases in the 3D-geometry. Figure 3.15a shows 62 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC the average concentration within the canopy of the ratio NO2N/NON in comparison with the total flux of O3 at roof level (< wO3 >). Thus, the chemical influence is distinguished as a consequence of the deviation from the non-reactive behaviour (NO2N/NON=1). With larger O3 entrainment, the chemical deviation from the non-reactive behaviour is larger, being the maximum differences (NO2N/NON ≈3.5) associated with faster wind speed (red filled symbols). In contrast, when the O3 levels are not enough with respect to the strong NOx emissions, the reactants behaviour barely differ from the non-reactive approach. (a) (b) Figure 3.15: (a) Average concentration within the canopy of NO2N/NON against the total flux of O3 at roof level and (b) variation of the average concentration within the canopy of NO2N with CCM1/5 and CCM1/2 from that the PSS over the average concentration of VOC. Secondly, the deviation of NO2N, after including the VOC reactions, is analysed depending on the VOC levels within the canopy (Fig. 3.15b). These results show that the influence of the VOC on the NO2 concentration depends on both the NOx levels and the available O3 in the street. With higher O3 within the canopy, the NO2 levels from the CCM1/2 scenario increasingly differ from the PSS as the VOC emissions increase. In addition, with lower wind speeds, that deviation rises caused by the emitted VOC hold retained within the canopy and, thereby, the residence time of pollutants in the street increases. However, the results from the CCM1/5 show similar outcomes to PSS since the VOC levels are not enough to provide a significant deviation in the NO2 resulted from the PSS. 3.5 Summary and conclusions Modelling the dispersion of NO and NO2 require to include the chemical reactions that take place in the urban atmosphere. The NO2 concentration is mainly dominated by the interactions with NO and O3, according to the photostationary state. Nonetheless, there is a more complex chain of chemical reactions, in which the VOC are involved, that should also be considered. CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC 63 The purpose of this study is to evaluate the importance of the chemical reactions on the NO and NO2 dispersion considering three chemical approaches in the CFD simulation: (a) non-reactive pollutants, (b) the NOx −O3 scheme and (c) the NOx−O3−VOC scheme. For the latter scenario, two options of traffic emission ratios are deemed: VOC/NOx=1/2 and VOC/NOx=1/5. In this way, the effect of including the VOC interactions on analysing the reactive NO2 is compared with the results from the photostationary state. In addition, the influence of the background atmospheric conditions on the chemical reactions is tackled through the variation of the wind speed and the solar position, which is linked to the ambient O3. This analysis is carried out both in a single street canyon (2D-geometry) and in the staggered array of cubes (3D-geometry) in order to draw general conclusions with no dependence on the geometry. Therefore, the impact of several atmospheric variables on the dispersion of reactive pollutants can be examined within the canopy in simplified urban configurations. The results show that the available O3 in the street determines the difference in the dispersion of the NO and NO2 as reactive species against the non-reactive approach. In turn, the wind speed controls the vertical transport and therefore, the pollutants exchange between the street and the overlying air, which is related with the inflow of O3 to the street. Based on this, the greatest differences of the reactive NO and NO2 from the non-reactive scenario are obtained with larger background O3 (Case 1 and Case 2). Moreover, the faster wind speed promotes that the influence of the chemical activity increases because improving the mixing of pollutants in the canopy. This entails a large O3 entrainment into the street and outflow of the emitted NOx. The vertical fluxes proves that the available O3, regardless of the wind speed in the lowest background O3 cases (Case 3 and Case 4), is not enough to produce a significant influence of including chemical reactions on the NO and NO2 concentration. And, just above the canopy top, the chemical deviation from the non-reactive behaviour is perceptible on the vertical profiles. Concerning the chemical scheme to be used for modelling the NO and NO2 dispersion, the influence of the VOC interactions is analysed against the results of just using the photostationary state. The greatest deviation of the reactants between the chemical schemes is observed in the Case 2, with low wind speed and large background O3. Larger VOC release (VOC/NOx=1/2) helps to rise the difference of NO2 concentration obtained from the photostationary state. The weaker wind flow retains the VOC concentrations within the canopy, which compete with the O3 for the NO oxidation and, thereby, it produces NO2. Hence, the O3 loss in the street is reduced and, besides, the NO2 production slightly increases by using the complex chemical scheme. In contrast, the NO and NO2 concentrations by implementing the NOx−Ox−VOC scheme with the emission ratio VOC/NOx=1/5, turn out similar values to those derived from the NOx−O3 scheme. 64 CHAPTER 3. MODELLING REACTIVE POLLUTANTS: NOx−O3 and NOx−Ox−VOC These conclusions of the behaviour of the reactive pollutants can be generalized to atmospheric conditions in urban environments. In atmospheric conditions of weak solar radiation, the ambient O3 formed is usually low, and the chemical reactions are relatively slower than the rapid dynamical processes characteristic of urban areas. In addition to that, the high concentrations of NOx, associated to the traffic emissions, give rise to the quick reduction of the O3 concentration at street level. Accordingly, the implementation of any chemical scheme to simulate NO and NO2 dispersion may provide small differences in comparison to the non-reactive approach. In contrast, in meteorological conditions of high solar radiation, the background O3 increases promoting the chemical interactions with the NO and NO2. It may induce an important impact on the reactive pollutant concentration in the streets with respect to not use a chemical scheme in the CFD simulation. As for considering of NOx−Ox−VOC scheme instead of the NOx−O3 mechanism, the difference in the NO and NO2 concentrations is dependent on both the presence of the VOC levels in air, in comparison with those of NOx, and the background O3. Based on these results, the difference between chemical schemes is intensified by increasing the amount of VOC within the canopy. The computational resources required to implement a chemical mechanism in the CFD model should be considered. Hence the importance of chemical reactions on the pollutant dispersion has to be analysed in several conditions, since it is affected by many key factors, i.e. the meteorological conditions, the traffic emissions, the urban air composition or the flow pattern representative of the urban area. Therefore, this study provides a better understanding of the chemical influence to seek the best compromise between the complexity of the chemical scheme needed and the computational load required in order to simulate the NO2 dispersion in urban scenarios. Acknowledgements This study has been supported by European Project LIFE MINOx-STREET (LIFE12 ENV/ES/000280) funded by European Union. Authors thank to Frank Kirchner (GAIASENS Technologies) for providing the condensed chemical mechanism (NOx−Ox−VOC scheme) to be implemented in the CFD model for this study. I would like to grateful with Extremadura Research Centre for Advanced Technologies (CETA-CIEMAT) by helping in using its computing facilities for the simulations. CETA-CIEMAT belongs to CIEMAT and the Government of Spain and is funded by the European Regional Development Fund (ERDF). Chapter 4 Modelling the dispersion of reactive pollutants in an urban area 4.1 Introduction The urban air quality is controlled both by the background characteristics of the atmosphere at larger scale and by the dynamical and chemical processes at local scale, basically driven by the turbulence. In urban environments, turbulence primarily dominates the dispersion and mixing of pollutants in the streets. In turn, the emitted pollutants undergo rapid chemical transformations modifying the air composition. Hence, the pollutants advected into the urban area interact with the emitted species from local releases, triggering multiple chemical conversions that can lead to relevant changes in pollutant concentrations. Accordingly, the dispersion of reactive pollutants in urban areas is jointly driven by the turbulence and the chemical processes occurred in the streets. Many studies focus on analysing the impact of turbulent processes on the chemical activity by means of the Damkhöler (Da) number in the atmospheric boundary layer (Galmarini et al., 1995; Krol et al., 2000; Vilà-Guerau de Arellano et al., 2004). These studies reveal the role that the turbulent mixing plays on the chemical reactions, together with the importance of the chemical balance among reactive species. In particular conditions of turbulence and chemical activity, the chemical terms included in the governing equations can be comparable to the dynamical terms (Vilà-Guerau de Arellano et al., 2004). It means that the time scales of chemical reactions and turbulent mixing are of the same order. Krol et al. (2000) showed that the chemical equilibrium among pollutants also determine the relation between turbulence and chemical activity. In chemical equilibrium among pollutants, the effects of the turbulence on the chemical reaction are reduced. Jonker et al. (2004) concluded that the reactive species are dependent on both the Da number and the chemical balance among 65 66 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA reactants. Vilà-Guerau de Arellano et al. (2004) thoroughly examined the influence of the turbulence on the chemical activity using dimensionless numbers, such as the Da number. They concluded that the turbulent processes dominate the dispersion and mixing of the reactive pollutants in the ABL and, thereby, control the chemical activity of the reactants. The inhomogeneity of turbulence in urban settings leads to also analyse the influence of the adveccion-diffusion processes on the chemical conversion. In a street canyon, Baker et al. (2004) proved that the chemical terms may be comparable to the advection and diffusion terms in the transport equation for a reactive pollutant. They also proposed the deviation factor from the chemical equilibrium in order to study the efficiency of the reacting mixing at local scale. Zhong et al. (2015) examined the chemical behaviour with a NOx−O3 mechanism in an idealized deep street canyon (aspect ratio equal to 2). They concluded that the largest deviation from the chemical equilibrium is obtained close to the NOx emissions, where the imperfect pollutant mixing is greater. In real urban environments, the impact of dynamical processes on the chemical activity is difficult to study since there are many factors that can affect the chemical transformation, e.g. the meteorological conditions, the ambient concentration, the irregularity of the buildings layout or the traffic emissions. For that reason, the influence of the chemical reactions are analysed by comparing with a non-reactive approach, under the same conditions. Currently, some studies are addressed to model the dispersion of reactive pollutants in urban scenarios through the CFD simulations. Santiago et al. (2017a) examined the effect of including chemical reactions in the CFD model in a peak and off-peak traffic conditions in an urban area at wintertime. They concluded that the modelled concentration of NO2 shows similar results comparing the reactive and non-reactive assumptions, with differences below 20% at street level. Kim et al. (2012) studied the sensitivity of a CFD model to include chemical mechanisms for simulating the dispersion of the NO, NO2 and O3 throughout the day, in an irregular street. Similar concentrations of NO and NO2 were obtained after considering both a full chemical mechanism and the photostationary state. In contrast, a significant difference in the O3 concentration under lower pollution conditions was captured. It is worth noting that the VOC/NOx emission ratio of vehicles was established at 2.7. In Chapter 3, the overall outcomes revealed small differences on the NO2 concentration at street level between the NOx−O3 and NOx−Ox−VOC schemes. In this study, the emission factors of NOx and VOC are obtained from the emission inventories for the road transport in Spain (Madrid-Council, 2014), which report larger emissions of NOx against the VOC release. To ascertain the deviation on the NO2 concentration in real atmospheric conditions, the use of the NOx−Ox−VOC scheme was compared with the photostationary state, at 1200 LST (Local Solar Time), in a real urban scenario (see Appendix 6.2). The NO and NO2 results were evaluated with measurements in situ. The conclusions of the modelled CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 67 concentrations revealed small differences between the use of both chemical mechanisms. It may be related to the small VOC levels in air in regards to the NOx since the VOC/NOx emission ratio is 1/5. Kwak and Baik (2012) analysed the impact of the VOC/NOx emission ratio concluding that, as the VOC emission level increased, the NO2 and O3 became larger in the street canyon. Accordingly, the need to include the VOC interactions for modelling the NO2 dispersion is dependent on the atmospheric background composition and the emissions traffic, representative of the urban area to be simulated. This chapter is addressed to thoroughly analyse the atmospheric factors that affect the influence of the atmospheric chemistry on the NO2 concentration in a real urban area. To that end, the dispersion of the reactive NO2 is modelled over time, from 0600 LST to 1600 LST (from 08h to 18h local time) at early-autumn in a real urban scenario. Based on the results obtained in Appendix 6.2, just the NOx−O3 scheme is herein considered. Thus, the NO and NO2 are simulated under the two chemical approaches: no chemistry and NOx−O3 scheme. Once the modelling results are validated with measurements, the chemical influence on pollutant concentrations are examined, both in time and space, taking into account the atmospheric variability. Lastly, the impact of dynamical processes on the chemical effect is studied in detail, at local scale, in a real urban environment. In this way, this work attempts to answer the following research questions: • Which is the impact of the atmospheric conditions at daytime on modelling the dispersion of NO2 as a reactive specie in a real street? • How the turbulence mixing and the chemical processes control the NO2 dispersion in a real street? 4.2 Study case This study focuses on an urban area located in Alcobendas, close to the North of Madrid, with geographic coordinates 40◦ 32’ N and 3◦ 37’ W (Fig. 4.1a). The research area (blue square in Fig. 4.1b) is located in a residential area, where the mean buildings height is around 18 m and the tallest one is almost 20 m. The main street within the research area includes a dual carriageway separated by a median strip. The representative traffic intensity, given by the AADT for the studied street is around 2000-4000 veh/day (Alcobendas-Council, 2005). 68 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA (a) (b) Figure 4.1: (a) Location of the research area around Madrid. (b) Locations of the measurement points (red and yellow points) deployed for the experimental campaign. The blue point represent the location of the camera used for recording the traffic flow in specific days. In this area, an intensive experimental campaign was carried out from 11th September to 25th October in 2015 (Pujadas et al., 2016) within the framework of the LIFE MINOx-STREET European Project (LIFE12 ENV/ES/000280). For this study, the 29th September 2015 is selected because all the variables needed to undertake the CFD-RANS simulation were collected §. The data used for this study are derived from measurements taken at several points deployed along the street and, also, at one sampling point placed on the building’s roof, at 20 m AGL. At the reference point (red point in Fig.4.1b), the ambient atmospheric conditions around the research area are intensively recorded every 5 min. In this way, the meteorological variables (wind speed and direction, temperature and solar radiation) together with the concentration of NO, NO2 and O3 provides full information of the background conditions. Additionally, six measurement points (yellow points in Fig.4.1b) are deployed on the median strip along the studied street, at 0.4 m AGL, in which both the NO and NO2 concentrations are taken every 2 min. Besides, the actual traffic flow in the study street was recorded over time for several days throughout the experimental campaign (Pujadas et al., 2016). During this day, the camera was located at east of the street close to the roundabout (blue point in Fig. 4.1b). §For this day, a photocatalytic coating covered a small area (1000 m2) in the studied street as part of the scope of the project. Both the experimental and the modelling approaches proved the efficiency of the photocatalytic material have a negligible effect on the NOx concentration in air (Pujadas et al., 2016). CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 69 4.3 CFD model description and simulation setup 4.3.1 Geometry and mesh Figure 4.2a displays the computational domain, whose size is 1600 m x 1000 m, with the top of the domain established at 8Hmax, where Hmax is the height of the tallest building (20 m) (Franke, 2007). The research area (black square in Fig. 4.2a) is vertically quite homogeneous with a packing density of buildings λp = 0.3 and the mean height of buildings around 18 m. A polyhedral mesh is applied with the grid resolution of 2 m in the research area and, externally, it is progressively increased up to 6 meters. The grid size is further reduced to 1 m close to the ground, emissions and buildings resulting in a total of 2.3 106 grid points. (a) (b) Figure 4.2: (a) Geometry of the computational domain and the research area limited by the black square. (b) Zooming in on the meshed research area and location of the measurements points. 4.3.2 Simulation setup The CFD-RANS simulation is performed in unsteady conditions for the day 29th September 2015, from 0600 LST to 1600 LST (from 08h to 18h local time). Inlet and outlet boundaries are defined depending on wind direction and symmetry conditions are imposed at the top of the domain. The inlet conditions are derived from the measurements taken at the reference point on the building’s roof. From the experimental data, the wind speed and direction is averaged every 30 min and, thus, the vertical profiles of wind speed, turbulent kinetic energy and turbulent dissipation rate are computed from these data using the Eqs. 2.43-2.45. In these atmospheric conditions, the thermal effects become increasingly important and, thereby, affect the flow distribution. Nonetheless, this aspect has not been considered in this simulation because it 70 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA would require include the time-dependent conditions on the solar position, i.e. the variability of the surface heat fluxes. Assuming that this factor modify the dispersion of pollutants in the streets, the isothermal condition has to be assumed in order to perform the simulation in a reasonable computational time. Concerning the background atmospheric composition, the concentrations of NO, NO2 and O3 are obtained from the reference point. The concentration averaged every 30 min is uniformly established at the inlet of the domain (Fig. 4.3). In this study, NO and NO2 are simulated both as reactive and non-reactive pollutants. For the reactants, the NOx−O3 scheme is implemented in this simulation changing the chemical constant rates (JNO2 and kNO+O3 , Table 2.3) every hour depending on the solar position and air temperature. Lastly, the same dynamic and chemical time step is 1 s, varying the inlet conditions every 30 min. Figure 4.3 illustrates the time series of wind speed and direction (at 20 m), the air temperature, the solar position and the concentration of NO, NO2 and O3, established as inlet conditions. (a) (b) (c) Figure 4.3: Time series of the inlet conditions of: (a) wind speed and direction at 20 m, (b) air temperature and (c) background concentrations of NO, NO2 and O3, in ppb. CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 71 Lastly, the traffic-related emissions imposed in the simulation are based on the actual traffic flow, recorded in the studied street during the experimental campaign (Pujadas et al., 2016). Hence, this number of vehicles is clustered every hour giving rise to the daily traffic pattern (Fig. 4.4a). Based on that evolution, the amount of vehicles travelling on the rest of the streets is computed keeping the relation of the traffic intensity established by the AADT, reported by the City Council (Alcobendas-Council, 2005). The emitted NOx per vehicle is established at 0.5 g km−1 and the latest emission ratio NO2/NOx is considered 0.3 (Section 2.6.3). According to the number of vehicles on each street, the emission rate is uniformly distributed along each street as a result of the computed releases through the traffic intensity (Fig. 4.4b). (a) (b) Figure 4.4: (a) Time series of the traffic intensity in the reference street. (b) Uniform distribution of the NOx emissions (ppb s−1) at 0730 LST depending on the traffic flow representative of each street. 4.4 Validation 4.4.1 Time series at the sampling points In this section, the modelled NO and NO2 concentrations are validated with the measurements from 0600 LST to 1600 LST at the sampling points, located along the study street (Fig. 4.5). 72 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA Figure 4.5: Location of the measurement points The proximity of the sampling points to the emission source may introduce perturbations in the measurements that are not reproduced by the model. For that reason, the experimental data of NO and NO2 are, firstly, filtered by means of the hourly average concentration and its standard deviation. Those values that fall out of the deviation range are dismissed in order to recalculate the average concentration at every hour. Subsequently, the simulated concentrations of NO and NO2 are also averaged every hour and compared with the measurements at the six points placed along the street. First of all, the NOx, resulted from the sum of NO and NO2 is evaluated over time in comparison with the experimental data (Fig. 4.6). Note that the NOx can be considered as a non-reactive pollutant. (a) (b) (c) (d) (e) (f) Figure 4.6: (a-f) Time series of the NOx concentration measured and simulated at the measurement points, in order from P1 to P6. Vertical bars indicate the standard deviation of the measurement, at each hour. CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 73 At all the points, the recorded NOx presents the maximum concentrations at the first and last hours, following the evolution of the traffic intensity and PBL behaviour. The modelled NOx underestimates those values, specifically at P6. This point is located very close to the roundabout, where accurate data of the traffic flow are not available to estimate the emission and, thereby, it may be the cause of that underestimation. Nevertheless, during the central hours of the day, the modelled NOx reproduces the temporal evolution, approximately, within the error range of the measurements. The deviation in the simulated NOx concentration may be caused either by an underestimation of the emissions at that time or by an error in the inlet meteorological variables (e.g. wind direction) or by the atmospheric stability, particularly at first hours of the day. At sunrise, the nocturnal stability is destroyed as the solar radiation increases. Hence, the vertical structure of the lowest atmosphere changes and the temperature and wind profiles differ from the assumed profiles in neutral atmospheric conditions. For further quantitative details, the statistics NMSE, FB and R (Eq. 2.5) are computed at each point. The NMSE provides information about the overall deviation from the experimental data and, for a good simulation, the threshold is limited below 1.5 (Chang and Hanna, 2004). The FB<0 indicates an underestimation of the simulated concentration against the experimental value. In this case, the resulting NMSE is below 1.5, with a general underestimation and correlation coefficients above 0.6 at all points (Table 4.1). It concludes that the simulation for the NOx dispersion with respect to the measured NOx is appropriate. Table 4.1: Statistical parameters of NMSE, FB and R of the NOx validation at every points. NOx (P1) NOx (P2) NOx (P3) NOx (P4) NOx (P5) NOx (P6) NMSE 0.15 0.09 0.25 0.23 0.11 0.48 FB -0.18 -0.22 -0.19 -0.11 -0.53 -0.18 R 0.78 0.79 0.66 0.80 0.79 0.65 The NO and NO2 obtained from both chemical approaches are also evaluated at the sampling points. Hereinafter the NO modelled as reactive and non-reactive pollutants are referred to as NOR and NOT respectively. The same notation is used for the NO2 (NO2R and NO2T). Figure 4.7 displays the validation of the hourly average of the measured NO, NOR and NOT at every point. Likewise, the hourly average concentration of NO2R and NO2T against the measured NO2 is depicted in Fig. 4.8. 74 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA (a) (b) (c) (d) (e) (f) Figure 4.7: (a-f) Hourly average values of the experimental NO, NOR and NOT at the sampling points (P1-P6). Vertical bars indicate the standard deviation of the measurements at each hour. (a) (b) (c) (d) (e) (f) Figure 4.8: (a-f) Hourly average values of experimental NO2, the NO2T and NO2R at sampling points (P1-P6). Vertical bars indicate the standard deviation of the measurements at each hour. CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 75 As the NOx validation revealed, there is an overall underestimation of the modelled concentrations of the NO and NO2 at first hours, caused by some uncertainties assumed in the boundaries conditions. Nonetheless, throughout the day, the modelled NO resulted from both the reactive and non-reactive approaches are in a good agreement with the measured NO. Concerning the NO2 concentration, the NO2R fits slightly better to the experimental value than the NO2T, generally, within the range of the measurement variation. For a complete ascertainment, the statistical parameters NMSE, FB and R are computed, at each point, for the results of NO (Table 4.2) and NO2 (Table 4.3). Table 4.2: Statistical parameters: NMSE, FB and R of NOT and NOR at every points. P1 P2 P3 NOT NOR NOT NOR NOT NOR NMSE 0.25 0.50 0.29 0.36 0.57 0.88 FB 0.0030 -0.35 0.17 -0.14 -0.010 -0.32 R 0.62 0.66 0.57 0.65 0.46 0.52 P4 P5 P6 NOT NOR NOT NOR NOT NOR NMSE 0.42 0.70 0.23 0.28 0.61 1.24 FB 0.028 -0.28 0.15 -0.21 -0.47 -0.76 R 0.68 0.72 0.74 0.83 0.32 0.21 Table 4.3: Statistical parameters NMSE, FB and R of NO2T and NO2R at every points. P1 P2 P3 NO2T NO2R NO2T NO2R NO2T NO2R NMSE 0.19 0.047 0.16 0.047 0.27 0.089 FB 0.39 0.045 0.32 -0.018 0.46 0.14 R 0.86 0.78 0.72 0.76 0.73 0.60 P4 P5 P6 NO2T NO2R NO2T NO2R NO2T NO2R NMSE 0.29 0.11 0.23 0.11 0.50 0.13 FB 0.42 0.11 0.40 0.082 0.61 0.27 R 0.73 0.69 0.69 0.67 0.64 0.71 From the simulated concentrations, the NMSE results in all cases below 1.5. For the NOR, the negative values of FB denote an underestimation of the experimental data at all locations, whereas the NOT slightly overestimates the measurement. The statistics of the modelled NO2 concentrations from both chemical approaches yield values of the NMSE below 1.5 and 76 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA correlation coefficients above 0.6 in all points. However, unlike the NO2T, the FB for the NO2R holds within the acceptance criteria (-0.3) at every hour, within the selected area (Fig. 4.12). Figure 4.13 displays the < NOT >, < NOR >, < NO2T >, < NO2R > and, also, the relative differences < δNO > and < δNO2 > (Eq. 3.1). (a) (b) (c) Figure 4.13: At 3 m AGL, time series of: (a) < NOT > and < NOR > in ppb, (b) < NO2T > and < NO2R > in ppb and (c) < δNO > and < δNO2 > in %. In the early hours (from 0600 LST to 0800 LST), the time series of < δNO > and < δNO2 > show a decrease in the chemical influence from 50% to 30%. It is directly caused by the temporal evolution both of the traffic emissions and the solar radiation. At first hour, the solar radiation is lower and, thereby, the NO2 loss by photolysis is quite small. In turn, the traffic emissions are higher and the emitted NO reacts with O3, rising the NO2 production. It hinders the chemical equilibrium among pollutants, which gives rise to an increase of the difference between the reactive from non-reactive approaches. This behaviour is reduced as the solar radiation increases and, thus, the NO2 formation tends to balance the loss of NO2. At certain hours, < δNO2 > reaches values above 50%, which is mainly related to the flow pattern in the street. At 1330 LST, east wind blows parallel to the study street (Fig. 4.14a) and drag the pollutants from the surroundings areas. Consequently, < δNO2 > is 80 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA more affected by the chemical deviation obtained from the pollutants mixing between the emissions and the overlying air in the surroundings. On the other hand, at 1500 LST (Fig. 4.14b), the decrease of the wind speed and the vortexes developed in the street increasing the residence time of pollutants. These conditions promote the chemical interactions among pollutants increasing the value of < δNO2 >. (a) (b) Figure 4.14: Wind flow at pedestrian level at (a) 1330 LST and (b) 1500 LST. Despite the NO and NO2 concentrations are not overly high, because this street is not regarded as a heavily trafficked area, the temporal evolution of the chemical deviation turns out values above 20% at all hours. This value is established as the threshold from which the influence of the chemistry is considered significant. This is based on the legal error allowed for pollutant measurements by European Air Quality Directive 2008/50/EC, which ranges from 15% to 25% depending on the pollutant. The same threshold (20%) is used as the concentration similarity criteria for other purposes, e.g. by computing the spatial representativeness area of air quality monitoring stations (Santiago et al., 2013; Martin et al., 2014; Piersanti et al., 2015). Accordingly, in these atmospheric conditions, the influence of including chemical reactions in order to model the dispersion of NO and NO2 seems to be significant in the daylight hours. At daytime, the traffic emissions are larger and the chemical reactions of the photostationary state mainly reflect the NO and NO2 interactions. However, at night, the photolysis of NO2 ceases and the traffic emissions are considerably reduced. Therefore, the overall deviation obtained at daytime in these atmospheric conditions is quantify through the daytime mean map, from 0600 LST to 1800 LST. Figure 4.15 illustrates the map at pedestrian level of NOT, NOR, NO2T, NO2R and, their corresponding chemical deviations ∆C (CR − CT), δNO and δNO2. Note that the difference (∆C) takes the same value for the NO and NO2 (NOR − NOT=NO2R − NO2T). CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 81 (a) (b) (c) (d) (e) (f) (g) Figure 4.15: Maps of (a) NOT (ppb), (b) NO2T (ppb), (c) NOR (ppb), (d) NO2R (ppb), (e) ∆C (ppb), (f) δNO (%) and (g) δNO2 (%), at pedestrian level. Concerning the dispersion in the street, the mean maps reveal that both the non-reactive assumption and the reactive approach reproduce similar concentration distributions at pedestrian level. However, the concentrations of NOR and NO2R turn out lower and larger values respectively from the non-reactive approach, highlighting the influence of the chemical conversions. In addition, ∆C shows the maximum values in the street over the traffic emission area (around 5-7 ppb). It concludes that the importance of the chemical activity 82 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA varies according to the distance to the source. Overall, both δNO and δNO2 take values above 20% in the entire street, even the 50% is reached in some areas depending on the characteristic flow in the street. Therefore, modelling NO2 as a non-reactive pollutant lead to a relevant difference in the NO2 concentration at daytime. Based on these results, the evaluation of a particular abatement strategy for the NO2 levels would be necessary to consider the chemical reactions in modelling accurate NO2 concentrations. As previously showed, the variability of the atmospheric conditions, e.g. wind flow or the solar position modified the chemical influence on the pollutants concentration in the streets. In the following section, the dynamical processes and its importance in the chemical activity are in-depth studied in the street. 4.6 Chemical and dynamical processes in an urban area In urban environments, the dynamical processes control the transport of pollutants in the streets. The advective term determines the flow pattern and, thus, the residence time of pollutants in the street. In turn, the turbulence is a key factor that dominate the mixing of pollutants. As for the chemical processes, the reactive pollutants tend to seek the chemical equilibrium up to reach the point, where the removal is equivalent to the formation of the reactants. Nevertheless, in urban areas, the fresh emission in the streets is continuously disturbing the reach of a chemical balance. Therefore, the dynamical processes play an important role in the mixing of the emissions and the ambient pollutants that promote to achieve the chemical equilibrium. In this study, the importance of the turbulence in the chemical activity is examined through the concentration differences between the reactive and the non-reactive approaches. That deviation represents the influence of the chemical conversions, involving both the advective and turbulent-diffusive terms (assuming the molecular diffusion is irrelevant with respect to the turbulent). In this way, the contribution of the turbulence to the chemical influence is analysed by the Da number (Section 2.4.3). This dimensionless number provides the relation between the turbulent processes and the reaction rate, as a function of the turbulent time scale (τturb) and the chemical time scale (τreact). Using the NOx−O3 mechanism, the characteristic time scale of the steady state is computed for the slowest chemical reaction; in this case, using the NO concentration and the reaction rate constant kNO+O3 : τreact = (kNO+O3NO)−1 ; τturb ≡ k ε (4.1) Da = τturb τreact = k/ε 1/(kNO+O3 · NO) (4.2) CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 83 Another important aspect is the deviation from the chemical equilibrium (δeq). Based on the the Leighton relation (Eq. 1.8), δeq can be computed as follows: δeq = 1− ( JNO2 · NO2 kNO+O3 · NO ·O3 ) (4.3) If δeq ≈1, the chemical state of pollutants is completely out of the chemical equilibrium, whereas if δeq ≈0, there is a balance between the NO2 photolysis and its formation by the NO and O3 interaction. Nonetheless, if other chemical processes become more important, triggering additional loss processes, the chemical equilibrium does not hold. At sunset and sunrise, the deviation of the chemical equilibrium are expected fairly high. In these conditions, the NO2 photolysis is considerably small and the steady-state relation of chemical equilibrium is not fulfilled (Calvert and Stockwell, 1983; Finlayson-Pitts and Pitts Jr, 1999). Due to the chemical influence varies as a result of the proximity to the source, two locations at pedestrian level are selected for this study (Fig. 4.16). The first point (PS) is placed at centre of the street near the road. The second one (PB) is located further away from the NOx release, which may be representative of the ambient composition. Figure 4.16: Location of the study points in the streets: PS, near the road and PB, further from the traffic emissions. At these points, the relation between turbulent mixing and equilibrium state and its influence on the chemical activity is analysed over time. This allows to understand the reasons that cause the variability of the chemical influence on the dispersion of pollutants in the street. To that end, the difference in NO2 from the reactive and non-reactive approaches (∆NO2=NO2R-NO2T), the Da number and δeq are depicted, at PS and PB, in Fig. 4.17. 84 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA (a) (b) Figure 4.17: Temporal evolution of the ∆NO2 (=NO2R−NO2T), Da number and δeq in (a) PS and (b) PB. Firstly, ∆NO2 exhibits a clear difference between both locations and, therefore, an evidence of the spatial variation owing to the chemical influence in urban environments. In addition, the chemical state of the ambient pollutants (δeq) at PB shows that the species are close to the chemical balance along the day except at first hours. However, at PS, the reactants are usually further from the chemical equilibrium. Likewise, the results of the Da number reveal different behaviours at PS from the PB. In the latter, the Da number takes low values (Da≈0.1) that corresponds to the small variations of ∆NO2. This represents that the turbulent processes are efficiently able to well mix the pollutants up to reach the chemical equilibrium (δeq ≈ 0). At PS, the time series of the Da number correlates with the evolution of the largest variations of ∆NO2, except at the first hour. Therefore, an increase in the chemical influence reveals the turbulent processes are not able to properly merge the emitted pollutants with the overlying air. Thus, at PS, the Da number exhibits that the relation between turbulence and chemical activity affects the dispersion of reactant in the street. At first hours, both at PS and PB, the reactants are completely out of the chemical equilibrium (δeq ≈ 1) since the NO2 photolysis does not make up for the NO2 production by the reaction of NO with O3. In these conditions, the concentration difference with the reactive approach is basically caused by the reaction of the continuous NO emissions and the background O3. Hence there is no influence of the turbulent mixing on the chemical rate as the Da number shows. And ∆NO2 close to the release (PS) hardly differs from the value at PB. The spatial variation in the streets of ∆NO2, δeq and Da number are depicted at 0630 LST, 1200 LST and 1330 LST, at pedestrian level, in Fig. 4.18. CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 85 (a) (b) (c) (d) (e) (f) (g) (h) (i) Figure 4.18: (From left to right) Map at pedestrian level of ∆NO2 (ppb), the Da Number and δeq at (a-c) at 0630 LST, (d-f) 1200 LST and (g-i) at 1330 LST. At 0630 LST (4.18a-c), δeq displays that pollutants are not in a chemical equilibrium throughout the entire domain. At this time, the solar position is fairly low since it corresponds to the sunrise. As a consequence, the NO2 photolysis is slow or almost non-existent, which prevents to reach the chemical balance (Eq. 4.3). Using the photostationary state, ∆NO2 cannot be related to the influence of the turbulence in the street. The loss of O3 is not being chemically replaced by the NO2 photolysis and, therefore, the pollutants mixing cannot improve the chemical conversions. Therefore, the chemical influence is not dynamically affected and ∆NO2 is only dependent on the reaction of NO and O3, depending on the background O3 imposed in the CFD simulation and the emitted NO. 86 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA In these atmospheric conditions, other chemical reactions regarding the NO3 interactions become more important, giving rise to a more complex chemical system (Stockwell and Calvert, 1983). At 1200 LST (4.18e-f), the largest δeq is just located close to the traffic releases area in the streets. In most part of the domain, the chemical balance is maintained being disturbed in the proximity of NOx sources. The maximum importance of the chemical difference, namely maximum ∆NO2, is locally found in areas where the Da number is higher. However, Da number is generally low in the street, which reveals that the turbulence is able to rapidly mix the emissions with the overlying pollutants. Therefore, at this time, the efficient mixing drives to small deviations between chemical approaches, given that NO2 removal is quickly replaced by the reaction of NO and O3. The highest values of ∆NO2 are obtained in the study street at 1330 LST (Fig. 4.18g). As at 1200 LST, the largest deviation from the chemical equilibrium is located over the traffic emissions. However, in this case, the Da number represents the inefficiency of the turbulence to quickly blend the reactants in the emissions areas. The chemical and turbulent time scale are comparable. It entails that the reaction of NO and O3 cannot make up for the loss of NO2. In this case, the flow pattern (Fig. 4.14a) drags the pollutants from the surroundings streets, which are not efficiently mixed with the ambient pollutants, and significantly increases the difference between NO2R and NO2T. Accordingly, in these atmospheric conditions, the importance of the chemical reactions increases with respect to the turbulent phenomena. Therefore, the influence of the chemistry on the reactive pollutants is closely related to the advective-turbulent processes in the streets. In turn, the perturbation of the emissions in the chemical balance from the overlying air determines the concentration difference, given by the chemical reactions, in which the turbulence in the street is a key factor. At sunrise and sunset, the turbulence effect is limited because ∆NO2 is mainly controlled by the reaction rates and the chemical balance of the background pollutants imposed at inlet in the simulation. However, as the solar radiation and temperature increase together with the background O3, the chemical equilibrium is quickly reached. In the streets, that chemical balance is disturbed by the local emissions. With an effective turbulent mixing able to efficiently mix the NOx emission and the ambient pollutants, the influence of the chemical interactions is reduced. In contrast, a poor mixing gives rise to the chemical reactions concentration difference in the streets between the reactive and non-reactive approaches. CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA 87 4.7 Summary and conclusions This chapter is addressed to study the coupled behaviour of dynamic and chemical processes in a real street and, thereby, how the turbulence can modify the influence of chemical interactions on pollutant concentrations. To that end, the dispersion of NO and NO2 is simulated considering two chemical approaches: no chemistry and NOx−O3 mechanism. Accordingly, the effect of the chemistry is examined depending on the atmospheric variability at daytime. In this way, the differences between the chemical approaches are quantified for these conditions at pedestrian level. Firstly, the time series of the pollutant concentrations are compared against the experimental data, at sampling points, along the study street. Despite the assumption of neutral atmospheric conditions, without including thermal effects, the NOx validation shows that the transport is properly reproduced by the simulation. However, at first hour, there are some errors derived from the uncertainties of traffic emissions or the atmospheric stability. Overall, the validation reveals a good agreement of the modelled concentrations for the NO, NO2 and NOx. As for the difference between chemical approaches, small differences are obtained. Nevertheless, the NO2 concentration is better captured using chemical reactions. The improvement of the reactive approach is obtained by comparing the ratio NO-to-NO2 in air with the measurements. Unlike the non-reactive approach, the ratio obtained from the reactive pollutants follows the same tendency than the experimental one. Once the simulation is evaluated, the daytime behaviour of the chemical influence is also analysed, both in time and space, in the study street. The diurnal variability of the atmospheric parameters are a key component in the reaction rates such the solar position and the air temperature. First of all, the relative differences in concentration between the two chemical approaches are analysed over time at pedestrian level. Overall, the temporal evolution of < δNO > and < δNO2 > in the study street yield above 20% throughout the day and, even, < δNO2 > reaches up to 70%, at some hours. On the other hand, the mean maps of NO and NO2 show broadly similar horizontal gradients of the dispersion of pollutants in the street. However, in terms of concentration, the NOR and NO2R take respectively lower and larger values above 20% than those obtained using the non-reactive approach. Therefore, these results conclude that the spatial distribution in the streets is approximately captured using both the reactive and non-reactive approaches. Nonetheless, in these atmospheric conditions, the values of concentration are significantly modified by using chemical reactions in the simulation. Lastly, the relation of the turbulence with the chemical influence is thoroughly examined. To do that, the variability in the difference between the reactive and non-reactive approaches is explained through the changes of the Da number and the deviation from the chemical 88 CHAPTER 4. MODELLING REACTIVE POLLUTANTS IN AN URBAN AREA equilibrium. At sunrise and sunset, the NO2 photolysis decreases and the steady-state system (represented by NOx−O3) is no longer fulfilled. Consequently, other chemical reactions can prevail in the NO2 transformations. In the simulation, the lack of chemical balance among the ambient pollutants reveals that the turbulence has not effect on the deviation caused by chemical reactions. As the solar radiation increases, the chemical equilibrium in the overlying air is rapidly reached, and just disturbed by the local emissions in the streets. Therefore, the difference between the reactive and non-reactive approaches is related to the efficiency of the turbulence to properly blend the emitted and ambient pollutants in the streets. For a turbulent time scale lower enough than the characteristic chemical time, the turbulent processes are able to suitably mix the pollutants in the streets. This involves that the NO and O3 reaction replaces the loss of NO2 as fast as the photolysis occurs. In these conditions, the chemical influence on the dispersion of a reactive pollutant is small in comparison with the non-reactive approach. On the contrary, for an inefficient pollutant mixing, the importance of the chemical reactions increases. In this way, the NO2 photolysis is not balanced by the reaction of NO and O3, giving rise to larger differences in concentration between the approaches. Moreover, that difference rises in case of the perturbation of the traffic emissions increases. Accordingly, the assessment of particular mitigation measure for the abatement of NO2 concentration requires to introduce the chemical reactions in CFD modelling. It improves the value of NO2 concentration obtained from the simulation. Nonetheless, the representative distributions of the pollutants in the streets is similarly reproduced by using any of the chemical approaches in the CFD modelling. Acknowledgments This study has been supported by European Project LIFE MINOx-STREET (LIFE12 ENV/ES/000280). Thanks to the colleagues of the Atmospheric Pollution Characterization and POC Group from CIEMAT for providing the experimental data used in the validation of the CFD-RANS simulation. The authors are also grateful to Extremadura Research Centre for Advanced Technologies (CETA-CIEMAT) by helping in using its computing facilities for the simulations. CETA-CIEMAT belongs to CIEMAT and the Government of Spain and is funded by the European Regional Development Fund (ERDF). Chapter 5 Modelling approach for air quality assessment in an urban hot-spot in winter conditions 5.1 Introduction Broadly half of the global population currently lives in urban areas and this share is projected to increase to two thirds by 2050. In Europe, this percentage is even higher, with a 73% of urban population and a projected share of 82% by 2050 (EEA, 2015). EEA (2015) established that 9% of European urban population lives in areas in which the annual NO2 EU threshold was exceeded in 2013. Current European air quality legislation (Directive 2008/50/EC) establishes the need to assess air quality and apply plans to improve air quality in non-compliant areas. However, high population exposure is often associated to hot-spots with high emission intensity from road traffic and reduced natural ventilation (Vardoulakis et al., 2003). Therefore, more detailed studies about urban air pollution are required by policymakers to effectively mitigate the population exposure. Urban monitoring station networks are not dense enough to capture these heterogeneities, making it difficult to understand and manage urban air quality issues. Indeed, the spatial representativeness of urban air quality monitoring stations (i.e. the area around the station where the pollutant concentration does not differ significantly from the concentration measured at the station) is currently an important concern (Mart́ın et al., 2015). Hence, high resolution maps of average concentration over long periods of time, ideally annual or several weeks at least, are necessary for air quality management and the assessment of local abatement measures taking into account the variability of emissions and meteorological conditions. 89 90 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT Experimental campaigns with passive samplers have been carried out to evaluate air pollution levels in several urban zones (Krochmal and Kalina, 1997; Vardoulakis et al., 2002, 2005; Ott et al., 2008; Parra et al., 2010). This experimental deployment with a large amount of passive samplers allows obtaining time-average concentration maps during several weeks for a given pollutant, usually the NO2. These samplers are cheaper and easier to install than monitoring stations, although they lack temporal resolution. Concentration maps can be obtained applying interpolation methods from the measurements points. However, these results may differ from the real maps owing to the complex geometry with a large number of obstacles in urban areas and the irregular distributions of pollutants sources (Deligiorgi and Philippopoulos, 2011). CFD models are an effective tool to simulate airflow and pollutant dispersion with 1-m resolution in urban environments. However, the temporal evolution of pollutant concentration during several days, weeks or months cannot be simulated using a CFD model due to the huge computational requirements. Nevertheless, recent studies (Parra et al., 2010; Solazzo et al., 2011; Santiago et al., 2013, 2017a) have developed a numerical methodology based on a combination of a set of steady CFD-RANS simulations to generate the time-average concentration map over a long period. In this way, the temporal evolution of the pollutant is composed of several simulated scenarios involving the actual wind data and the traffic emission conditions derived from experimental data at every hour. The main assumption of this methodology is that the pollutants are regarded as non-reactive, which is not fulfilled in case of the NO2. This work focuses on obtaining the spatial distribution of NO2 concentration averaged over several weeks in a heavily trafficked urban area in Madrid (Spain) using a CFD-RANS model. In this scenario, the modelling results are compared against measurements recorded in an intensive experimental campaign in wintertime. Firstly, the influence of chemical reactions on the NO2 concentration is evaluated at daytime for a particular day. To do that, the NO2 is modelled as a inert pollutant and a reactive compound, using the NOx−O3 mechanism. Secondly, the NO2 mean map is obtained using the numerical methodology based on the sequence of steady-state simulations. In this chapter, some novelties are included in the numerical approach for improving the microscale results. On one hand, the use of mesoscale variables for the required meteorological data. On the other hand, the traffic emissions are derived from a microscale simulation system that combines a traffic and emission models (Quaassdorff et al., 2016). In that way, emissions with high spatial resolution are implemented in the CFD simulation. Lastly, the time average map of NO2 concentration is compared against the experimental data obtained from 72 passive samplers deployed throughout this urban hot-spot. CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 91 Therefore, the following research questions are raised to be solved in this chapter: • Which is the error in modelling NO2 as a non-reactive pollutant in winter conditions? • Can we use meteorological mesoscale models in order to improve the CFD modelling of urban air quality? • Are the concentration fields obtained with this methodology reliable? 5.2 Experimental Campaign The research area is located in a heavily trafficked roundabout with a freeway crossing it through a tunnel in Fernández Ladreda Square at south of Madrid (Fig. 5.1a). The selected location is a highly contaminated microenvironment with a very complex geometry and intense presence of pedestrians (more than 12 000 pedestrians/hour at peak hours) that makes this location a representative traffic hot-spot where urban air quality issues commonly occur. The study of the air quality in this location is encompassed within the framework of TECNAIRE-CM Project (S2013/MAE-2972). (a) (b) Figure 5.1: (a) Location of the research area at south-west of Madrid (Spain). (b) Locations of measurements sites in the research area: red dots stand for the passive samplers; yellow dot for the air quality monitoring station; blue dot for the meteorological station and green dot represents the sonic anemometers location. An intensive measurement campaign was carried out from 9th to 27th February, 2015 in this urban area to study the spatio-temporal variations of the main pollutants and its implication 92 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT for population exposure (Borge et al., 2016). This experimental campaign included the characterization of NO2 spatial gradients through a dense network of Palmes-type passive diffusion tubes (PDT). Altogether, 206 PDT were homogeneously distributed throughout the roundabout (in lampposts, traffic signs, etc.) at red points in Fig. 5.1b and approximately deployed at 3 m AGL. The time series of pollutant concentrations are measured at the air quality monitoring stations that belongs to the measurement network of the Madrid Council. This station (yellow point in Fig. 5.1b) is classified as traffic urban station and is called Fernández Ladreda station (hereafter referred to as FL station). In addition, the two closest monitoring stations to the research area are used to obtain the background concentration, named Farolillo and Retiro stations, which are categorised as urban background stations. The former is located at 1.5 km at North-West from the research area, hereinafter referred to as NW-UB station. The second one is placed at 4.5 km of distance from the FL station in the North-East direction, hereinafter called NE-UB station. All monitoring stations measure the concentration of NO, NO2 and O3 every hour, at 3 m AGL. Additionally, the meteorological parameters were recorded from 17th February to 2nd March, 2015 (Borge et al., 2016). The meteorological station was located on the roof of a building, at 18 m AGL (blue point in Fig. 5.1b) and the variables instantaneously recorded were wind speed and direction, air temperature, relative humidity, pressure, precipitation and global solar radiation. The meteorological data were originally integrated into 10-min averages and, subsequently, averaged through a vectorial mean of 1-h periods for modelling evaluation purposes. In addition, the velocity components and temperature were measured by two sonic anemometers deployed at 6 m and 8 m at the second location (green point in Fig. 5.1b). Thus, micrometeorological parameters such as the turbulent kinetic energy and the turbulent heat flux were calculated from the measurements recorded with a high sampling frequency (20 Hz). Accordingly, two periods are considered for this study: for the concentration measurements from 9th to 27th February, 2015 and, as regards the meteorological data, from 17th to 27th February, 2015. 5.3 Geometry, mesh and boundary conditions The simulations are carried out using the CFD-RANS model with the Realizable k-ε turbulence closure. The size of the computational domain over the research area is, approximately, 1300 m x 1300 m (Fig. 5.2). The geometry consists of several characteristic regions: a tunnel located at the center of the square, vegetation zones (green part), buildings CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 93 (dark grey part) and a traffic emission region (red part). The base resolution of the polyhedral irregular mesh used is around 5 meters, with a smaller resolution of 2 meters within the region of 400 m x 400 m located at the center of the square (black square). Within this zone, the cell size is even smaller than 1 m close to the ground and buildings and in the emission region. As a consequence, the mesh holds 8.3 106 grid points in total. A mesh independence test was performed using fine (14 106 grid points) and coarse (1.8 106 grid points) meshes and the results turn out that the medium grid resolution is enough to properly simulate the flow and dispersion in this scenario. Therefore, the medium mesh was selected for this work in order to obtain the best compromise between accuracy and computational load. (a) (b) Figure 5.2: (a) Domain of the research area (black square). (b) Mesh of the central square. Green regions represent the vegetation zones and the red part, the traffic emission region in the research area. In wintertime, the evergreen urban vegetation may affect pollutants dispersion in the streets. Hence, it is included by means of a simplified geometry (green regions in Fig. 5.2). The vegetation effect is modelled assuming trees as a porous medium, just considering its dynamical effects with a value of leaf area density of 0.5 m2 m−3 (Section 2.5). On the other hand, the red region (Fig. 5.2) marks out the traffic emission zone of 300 m x 300 m with 1m-height AGL, where spatially non-uniform source have been implemented. These detailed emissions of NOx are derived from a microscale traffic emission model (Quaassdorff et al., 2016). The traffic model computes detailed trajectories that describe the behaviour of individual vehicles considering drivers’ reaction times, route selection logic and lane change logic. It provides essential information about braking-acceleration patterns that have a huge influence on the pollutants emitted and their distribution in space and time. For this study, emissions are computed through the TNO ENVIVER high-resolution post-processor that links the emissions model VERSIT+micro, based on VERSIT+ model (Smit et al., 2007), to speed-time profiles generated by the PTV VISSIM microscale traffic model. The emission estimations are based on very detailed and specific fleet composition and vehicle flow data in 94 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT the square, obtained from an intensive measurement campaign with cameras (Quaassdorff et al., 2016). Therefore, 13 emission scenarios of 1-h length, representative of different traffic composition and flow conditions, are computed to complete the weekly pattern (Fig. 5.3). For instance, similar traffic flow and driver’s behaviour is considered for Monday, Tuesday and Wednesday at 08:00 a.m. and, thereby, the same scenario is assigned at that hour (S2). Figure 5.3: Daily traffic flow in local time for each emission scenarios (Quaassdorff et al., 2016). Table 5.1 gathers the allocation of the emission scenario to every hour. Table 5.1: Hourly distribution of the traffic emission scenarios along a week in local time. PPPPPPPPPDay Hour 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Monday S5 S5 S5 S5 S5 S5 S5 S13 S2 S13 S13 S13 S13 S3 S3 S14 S14 S4 S4 S4 S4 S4 S4 S5 Tuesday S5 S5 S5 S5 S5 S5 S5 S13 S2 S13 S13 S13 S13 S3 S3 S14 S14 S4 S4 S4 S4 S4 S4 S5 Wednesday S5 S5 S5 S5 S5 S5 S5 S13 S2 S13 S13 S13 S13 S3 S3 S14 S14 S4 S4 S4 S4 S4 S4 S5 Thursday S5 S5 S5 S5 S5 S5 S5 S13 S6 S13 S13 S13 S13 S7 S7 S14 S14 S14 S8 S8 S8 S8 S8 S5 Friday S5 S5 S5 S5 S5 S5 S5 S13 S6 S13 S13 S13 S13 S7 S7 S14 S14 S14 S8 S8 S8 S8 S8 S5 Saturday S9 S9 S9 S9 S9 S9 S9 S9 S10 S10 S11 S11 S11 S11 S11 S11 S11 S12 S12 S12 S12 S12 S9 S9 Sunday S9 S9 S9 S9 S9 S9 S9 S9 S10 S10 S11 S11 S11 S11 S11 S11 S11 S12 S12 S12 S12 S12 S9 S9 In spatial terms, the NOx emission are calculated in a resolution of 5 m x 5 m considering the specific release into the tunnel and on the square. In the CFD simulation, the traffic emissions are implemented in the domain according to 1-m resolution for each emission scenario (Fig. 5.4). CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 95 Figure 5.4: From left to right, the spatial distribution of NOx emitted in µg s−1 from the traffic scenarios S2, S13 and S10 (see Table 5.1) implemented into the CFD simulations. The daytime variation of air temperature in winter atmospheric conditions is smooth without strong thermal gradients in the streets. For that reason, the assumption of isothermal condition may be appropriate and, thereby, the mathematical expressions for neutral atmosphere stability are applied in the CFD simulations. Thus, the vertical profiles of wind speed (u), turbulent kinetic energy (k) and turbulent dissipation rate (ε) are imposed at inlet, following the expressions: uin(z) = u∗ κ ln ( z + z0 z0 ) kin = u∗ 2 Cµ 1/2 εin(z) = Cµ 3/4kin 3/2 κz (5.1) where κ is the von Karman constant (0.4); Cµ, the constant (0.09); z0, the roughness length (0.03 m) and u∗, the friction velocity (m s−1). The turbulence induced by traffic and the thermal effects give rise to a greater turbulent diffusion. However, they are not considered in these CFD simulations. Accordingly, the diffusion of pollutants is modelled considering a low value of turbulent Schmidt number, Sct = 0.3, in order to minimize the underestimation in the dispersion of pollutants. The most common values used to simulate the pollutant dispersion are between 0.7 and 0.9. Nevertheless, Tominaga and Stathopoulos (2007) concluded that the optimum values of Sct are in a range from 0.2 to 1.3, according to various flow properties and geometries. For instance, Vranckx et al. (2015) observed, by comparing CFD simulations with wind tunnel experimental data, that the optimum turbulent Schmidt number depends on the case and ranges from 0.3 to 1.0. In addition, the research area within the city is also affected by the traffic emission at the surroundings neighbourhoods. Therefore, the background concentration used is derived from the closest air quality stations to the study area, depending on the wind direction (NW-UB and NE-UB). 96 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 5.4 Daytime behaviour of NO and NO2 In this section, the influence of the chemical reactions in the NO2 levels is examined in wintertime under two approaches: with no chemical reactions and with the NOx−O3 mechanism. In winter, maximum NO2 concentrations are usually measured at night, when the PBL height shrinks down due to the nocturnal stability. Nevertheless, the study of photohemical reactions leads to focus at daytime, in presence of solar light. For that reason, the CFD simulation is carried out from 0700 LST to 1700 LST (from 08:00 a.m. to 18:00) on Wednesday, 18th February 2015, within the period of the experimental campaign. This day is selected because the largest NO2 concentrations are recorded at FL station throughout daylight hours, within the period of the experimental campaign. 5.4.1 Simulation setup The CFD-RANS simulation is performed in unsteady conditions with 1 s time step and the boundary conditions are changed at every hour. The inlet meteorological conditions are obtained from the hourly-averaged data recorded at the meteorological station on a building roof at 18 m AGL (blue point in Fig. 5.1b). The wind speed and direction at 18-m-height are used to compute the inlet vertical profiles for wind, turbulent kinetic energy and turbulent dissipation rate assuming neutral atmospheric conditions (Eq. 5.1). For the reactive approach, the chemical constants, JNO2 and kNO+O3 , are computed over time through the hourly variation of air temperature and solar position (Section 2.3.1). As for the background pollutant concentrations, the values of NO, NO2 and O3 are obtained from the air quality station NE-UB station since, for the selected day, north-east wind is the prevailing direction. Lastly, the emission scenarios corresponding to daily pattern on Wednesday are simulated over time. The NO2 release is calculated taking into account the emission ratio NO2-to-NOx equal to 0.3, derived from the Madrid emission inventories (Borge et al., 2014). Figure 5.5 shows the inlet conditions over time for the wind speed and direction (at 18 m AGL), the air temperature and the background concentration of NO, NO2 and O3. CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 97 (a) (b) (c) Figure 5.5: Inlet values of: (a) wind speed and direction at 18 m, (b) air temperature and solar zenith angle and (c) background concentrations of NO, NO2 and O3 derived from NE-UB station. 5.4.2 Validation The meteorological and turbulent parameters obtained from the CFD simulation are evaluated against the experimental data taken in the sonics anemometers, at 8 m and 6 m AGL (green dot in Fig. 5.1b). For the validation, the measurements are hourly averaged. Figure 5.6 illustrates the time series of wind speed and direction and turbulent kinetic energy resulted from the simulation against the experimental value. 98 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT (a) (b) (c) (d) (e) (f) Figure 5.6: Time series of wind speed, wind direction and turbulent kinetic energy (k) respectively, (a-c) at 8 m AGL and (d-f) at 6 m AGL. The errors bars means the standard deviation of the measurement at every hour. At both points, the wind direction is well reproduced by the simulation, being slightly better at 8 m. At that height, the modelled wind speed remains within the variation range of measurement every hour whereas, at 6 m, the outputs show overestimations at certain hours. In contrast, the modelled turbulent kinetic energy clearly underestimate the experimental values. For further quantitative information, the statistics NMSE, FB and R are calculated for wind speed and turbulent kinetic energy at 8 and 6 m (Table 5.2). Table 5.2: Statistical parameters for wind speed (U) and turbulent kinetic energy (k). U (z=8 m) k(z=8 m) U (z=6 m) k(z=6 m) Acceptance Criteria NMSE 0.09 0.57 0.69 0.52 NMSE<1.5 FB 0.19 -0.32 -0.74 -0.62 -0.36 Better results for wind speed and turbulent kinetic energy are obtained at 8 m, where all statistics are within the acceptance criteria established in Chang and Hanna (2004). The NMSE reveals an overall small deviation of the modelled value from the observation. At 8 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 99 m, the FB shows a slight overestimation of the wind speed and a larger underestimation of turbulent kinetic energy close to acceptance threshold. Likewise, for both variables, R above 0.6 exhibits a fair correlation between the outputs and the observations. In contrast, at 6 m, whereas the experimental wind speed and turbulent kinetic energy are slightly lower than at 8 m, the modelling results provide similar values. This little difference between both points is mainly caused by the mesh resolution of the CFD simulation. Improving this error would require to reduce the grid size in order to better reproduce the vertical gradient in that short distance. In addition, there is vegetation close to the experimental deployment, which may introduce an error in the simulation since its features geometry are modelled in a simplified way. The validation of the concentrations of NOx, NO, NO2 and O3 is performed with the measurements recorded at FL station (blue dot in Fig. 5.1b). Figure 5.7 displays time series of the experimental and simulated NOx concentrations. Except to first hour, the simulated NOx follows the same trend of the measured NOx, obtaining values of NMSE and FB equal to 0.17 and -0.064. The error in the NOx concentration at 0700 LST may be related to an overestimation of the traffic emission for that particular day, or to the uncertainties associated to atmospheric stability at that time. Figure 5.7: Time series of measured and modelled concentration of NOx (ppb) at FL station. The error bars represent the 20% of experimental data. Figure 5.8 shows the temporal evolution of NO, NO2 and O3 obtained from the simulation and the measurements at FL station. The NO and NO2 modelled as reactive pollutants are hereafter referred to as NOR and NO2R and, as non-reactive, NOT and NO2T. It is noteworthy that the error bar is the 20% of the measurements related to the legal error allowed for the measurements by the UNION et al. (2008), without considering the variability of concentration during the hour. 100 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT (a) (b) (c) Figure 5.8: Time series of the measured and modelled concentrations (ppb) at FL station for: (a) NO, (b) NO2 and (c) O3. The error bars represent the uncertainty of the measurement (20%). Unlike the first hour, the modelled concentrations reproduce the same behaviour that the measurements at FL station. Overall, the statistical analysis for the NOT and NOR yield similar values of the NMSE, 0.36 and 0.25 respectively and the FB equal to 0.13 and -0.08, which concludes a good approximation of both approaches. Likewise, for the NO2T and NO2R, the NMSE is 0.13 and 0.12 respectively whereas the FB shows a slight underestimation with values of -0.31 and -0.05 respectively. Even so, despite the error made at 0700 LST, both chemical approaches considered in the simulation seem to provide a good result of the daytime behaviour for both the NO and the NO2. 5.4.3 Impact of the chemical approach on concentration at pedestrian level This section is addressed to examine the deviation of modelling NO and NO2 as reactants instead of non-reactive pollutants at pedestrian level over the research area (Fig. 5.9). (a) Figure 5.9: Research area of 300 m x 300 m. CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 101 Firstly, the spatial average (<>) concentrations over the research area (300 m x 300 m) is computed at every hour for: NOT, NOR, NO2T and NO2R and the relative differences, δNO and δNO2 (Eq. 3.1). Figure 5.10 depicts the temporal evolution of: < NOT >, < NOR >, < δNO > and < NO2T > and < NO2R > and < δNO2 >. (a) (b) Figure 5.10: Time series over the research area, at 3 m AGL, of (a) < NOT >, < NOR > and < δNO > and (b) < NO2T >, < NO2R > and < δNO2 >. Overall, the results of NO and NO2 show the same temporal trend with barely differences, below 20%, between the reactive and the non-reactive assumptions, particularly from 0800 LST to 1600 LST. At midday, the ambient O3 and the NO2 photolysis are larger and, thereby, the deviation from the chemical equilibrium is lower in the overlying airmass. However, at sunrise (0700 LST), the NO2 photolysis is smaller and, thereby, there is no possibility to reach a chemical balance using the NOx−O3 mechanism (as described in Chapter 4). In these cases, < δNO > and < δNO2 > increases up to 30% mainly caused by the NO and O3 reaction. Similarly, at last hour, the photolysis of NO2 decays, regarding the reaction between NO and O3 since the temperature and the background O3 is still larger (Fig. 5.5). In this case, < δNO2 > slightly exceeds 20% of difference. Overall, < δNO > and < δNO2 > reveal an small impact of the chemical reactions. Comparing the spatial layout, NO2R and δNO2 are depicted at 0700 LST and 1200 LST at pedestrian level in Fig. 5.11. 102 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT (a) (b) (c) (d) Figure 5.11: Maps at pedestrian level of (a-b) NO2R (ppb) and (c-d) δNO2 (%) respectively at 0700 LST and 1200 LST. Grey arrows indicate the flow fields corresponding to the NE direction. The maps of δNO2 show that the smallest differences are located around the emission region in both cases. However, the largest δNO2 are obtained further away from emissions, where the NO2 concentration is lower. At 0700 LST, δNO2 displays an overall increase in the concentration of NO2R, that provides local deviations significantly higher. At 1200 LST, δNO2 rises in the wind flow direction (NE), as the emissions interact with the ambient pollutants, disturbing the chemical equilibrium of the overlying air. As a consequence of the flow pattern, the concentration of reactive pollutants differ from the non-reactive approach, especially in areas with larger residence time, e.g. in the vortexes produced in the street or in the vegetation region. The average error throughout the day in the NO and NO2 is examined through the daytime mean maps. Figure 5.12 shows the time-average concentration from 0700 LST to 1700 LST CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 103 for the NOT, NOR, NO2T, NO2R in the research area, at pedestrian level. (a) (b) (c) (d) Figure 5.12: Maps at pedestrian level of (a) NOT (ppb), (b) NOR (ppb), (c) NO2T (ppb) and (d) NO2R (ppb). The concentration mean maps show similar distributions of pollutants dispersion in the streets using any of the two chemical approaches, though an increase of NO2R concentration is appreciable. ∆C (=CR − CT) displays the largest values at south-west of the square, due to the flow pattern produced by the prevailing wind direction (North-East) along the day (Fig. 5.13a). It is noteworthy that the maximum difference is around 14 ppb with concentration values of 80 ppb for the NO and 40 ppb for the NO2. The significant values of δNO and δNO2 above 20% are only limited to certain areas, also caused by the flow (Fig. 5.13b-c). In general, low differences between the two modelling chemical approaches are 104 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT obtained in most part of the domain. The spatial average over the research area results in < δNO > around 6.3 %, whereas the < δNO2 > reaches a value of 11.5 %. (a) (b) (c) Figure 5.13: Daytime map at pedestrian level of: (a) ∆C (ppb), (b) δNO (%) and (c) δNO2 (%). Grey color represents relative values below 20% of difference. The conversion of NO to NO2 over the total NOx is analysed by means of the ratio NO2-to-NOx in air. Figure 5.14 displays the NO2T-to-NOx and NO2R-to-NOx at pedestrian level. (a) (b) Figure 5.14: Maps at pedestrian level of: (a) NO2T-to-NOx and (b) NO2R-to-NOx. The ratio NO2T-to-NOx slightly differs from the ratio imposed at the source (NO2/NOx=0.3). The horizontal gradients show that the ratio grows as the distance from the emission increases, which is mainly caused by dynamical processes (Fig. 5.14a). The inclusion of CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 105 chemical reactions displays smoother horizontal gradients but the NO2R-to-NOx takes closer values to the non-reactive approach (Fig. 5.14b). Accordingly, in these atmospheric conditions, the difference obtained between the two chemical approaches is reduced throughout day and, also, in the daytime mean maps (below 20%). However, there are local differences above 20% in certain areas, mainly as a consequence of the flow pattern. Therefore, as the error obtained for the NO2 concentration is small, NO2 can be modelled as a non-reactive pollutant in order to reproduce the NO2 dispersion in the streets for long periods. 5.5 Modelling approach for urban air quality assessment Air quality assessment in urban environments demands representative distributions of wind flow and dispersion of pollutants in the streets during long periods (weeks, months and even years). This section is addressed to develop a modelling approach through the CFD-RANS simulations in an attempt to obtain the NO2 mean map over several weeks in the research area. The resulting NO2 map is validated with the NO2 concentration obtained from the 72 samplers placed around the square, from 9th to 27th in February, 2015 (Section 5.2). 5.5.1 Modelling approach The numerical approach used is based on the Weighted Average methodology using a CFD-RANS model (WA CFD-RANS methodology) proposed by Santiago et al. (2013, 2017a). The main assumptions of this methodology are the following: 1. The pollutants have to be strictly considered as non-reactive specie. In this way, in neutral atmospheric conditions, the modelled concentration is directly and inversely proportional to the emissions and wind speed respectively (see Appendix 6.2). Although some differences are found between the reactive and non-reactive approaches by modelling NO2 concentration, this error can be assumed small in winter conditions (δNO2 <20%). Moreover, the non-linearity of chemical reactions prevents the application of this modelling approach. 2. The pollutant concentration just depends on emissions, wind speed and background concentration at every hour. Therefore, there is no influence from one hour to the next, which is implicitly considered in the background concentration applied at every hour. 106 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 3. Neutral atmospheric conditions have to be assumed and, thereby, thermal effects are not included in the CFD-RANS simulations. The strong mechanical turbulence, generated by the presence of buildings, makes that the thermal effects can be negligible, particularly in wintertime, when the solar forcing is not very intense. The numerical approach consists in modelling neutral atmospheric conditions for each wind direction sector (such as N, NNE, and so on), in which is included all emission scenarios. Subsequently, the temporal evolution is performed from a sequence of those CFD simulations, depending on the wind sector and emission scenario at every hour. Finally, the adjustment of the selected simulations to the actual conditions of every hour is carried out using a reference velocity. Figure 5.15 illustrates a scheme of the procedure to be followed for the application of the numerical approach. Figure 5.15: Scheme of the procedure of the numerical methodology. The CFD simulations are carried out for each wind rose sector, in neutral stability conditions (Eqs. 5.1). To do that, the friction velocity is fixed (u∗=0.22 m s−1) and the steady state is reached. For each wind direction sector, all emission scenarios involved in the weekly traffic pattern are simulated (Table 5.1). For applying this methodology, the temporal evolution of the meteorological conditions at every hour is needed. In this work, the use of variables from a meteorological mesoscale model is proposed to improve the final result. The mesoscale model (WRF) used is particularly adapted to simulate the urban atmosphere (Chen et al., 2011) by including the urban roughness sublayer parametrization (BEP-BEM) (Martilli et al., 2002; Salamanca et al., 2010). In particular, the grid point (1 km x 1 km) corresponding to the research area is classified as Compact Mid rise. This type of urban zone is characterized by a building packing density of 0.58 and building height distribution CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 107 of 25 % of buildings of 10 m, 45 % of 15 m, 20% of 20 m and 10% of 25 m (Stewart and Oke, 2012). Note that these are not the real morpholological parameters of the area. With the aim of covering the entire experimental campaign (Section 5.2), the mesoscale simulation starts at 1800 LST of the 8th February 2015 and ends at midnight of the 28th of February 2015 (Sanchez et al., 2017). Therefore, for the composition of the microscale maps, the wind direction is selected at every hour from the mesoscale results, at the grid point corresponding to the microscale domain. Likewise, for that hour, the emission scenario is chosen based on the traffic flow pattern (Table 5.1). Consequently, the wind direction sector and the emission scenario corresponding to every hour are suitably associated throughout the entire experimental campaign. Csim(t) = Csim(Wsect(t), Straff (t)) (5.2) where Csim is the simulated concentration of a non-reactive pollutant, and Wsect(t) and Straff (t) are the wind direction sector and the traffic emission scenario corresponding to the time t. As a second step, the simulated concentration (Csim(t)) has to be transformed into the local concentration (Clocal(t)) over time through the ratio of reference velocities (Fig. 5.15). This approximation is based on the concentration is inversely proportional to wind speed. Therefore, the concentration is transformed according to the actual wind speed at every hour. In previous studies (Santiago et al., 2013, 2017a), the wind speed is used as reference velocity, which is obtained from a nearby meteorological station that is not perturbed by buildings. However, when thermal effects become significant in comparison with dynamical effects, the shape of the wind profile significantly differ from the neutral profile. In these cases, the use of the wind speed introduce some errors in the computation. To face this issue, the mesoscale model provides additional information about meteorological conditions. For that reason, the friction velocity (u∗) at the top of the canopy is proposed to be used as reference velocity. The height of the canopy top is considered at mean height of buildings. The friction velocity is defined by the turbulent momentum fluxes so that it provides information concerning vertical transport at that level. In the microscale simulations, the u∗ is computed by the spatial average (< >) of turbulent momentum fluxes given by: u∗ = [ (< u′w′ >)2 + (< v′w′ >)2 ](1/4) (5.3) 108 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT where the turbulent momentum fluxes in the CFD-RANS model are computed as: u′iu ′ j = −µt ρ ( ∂ui ∂xj + ∂uj ∂xi ) (5.4) Likewise, the u∗ from the mesoscale grid point is computed through the turbulent momentum fluxes in the horizontal surface at the selected level (around 30 m). Note that the mesoscale model takes into account the momentum and heat exchanges as a consequence of the presence of buildings. However, the u∗ from the CFD simulation just considers the drag forces produced by the obstacles. In this way, several effects neglected in the CFD simulations are implicitly involved in the friction velocity derived from the mesoscale simulation. Finally, the ratio of the velocities obtained from the mesoscale and microscale outputs is computed at every hour. This provides the adjustment of the vertical transport at the height of the canopy top (hcanopy). The ratio u∗,CFD(t)/u∗,WRF (t) is used to transform the Csim(t) into Clocal(t), adjusted to the atmospheric conditions given by the mesoscale simulation at that time t. Moreover, the air composition from the surroundings of the research area is considered in the background concentration, Cback. It is obtained from the NW-UB station since the North-West is the prevailing wind direction during the experimental campaign. Hence the total concentration, Ctot(t), is computed at every hour from the sum of the local contribution and the background concentration, expressed as follows: Ctot(t) = u∗,CFD(t) u∗,WRF (t) ) z=hcanopy · Csim(Wsect(t), Straff (t)) + Cback(t) (5.5) In atmospheric conditions of neutral stability, this approach is equivalent to use the wind speed as reference value since the wind speed is proportional to the u∗ (logarithmic profile). However, when thermal effects become important, the use of u∗ minimizes the error induced by the differences in the vertical wind profile. This discussion about the suitable reference velocity to use in this modelling approach is analysed in the following section. 5.5.2 Evaluation of the modelling approach against measurements The application of the previous modelling approach is firstly evaluated with the meteorological data measured at 18 m height on the building roof (blue point in Fig.5.1b). Secondly, the estimation of the NOx concentration, using both the wind speed at a certain CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT 109 height and the friction velocity, is compared with measurements recorded at FL station. For this analysis, NOx is chosen since it is commonly considered as non-reactive pollutant. Time series of meteorological variables The meteorological variables are not recorded during the entire experimental campaign. For that reason, the evaluation of the modelling results is focused on full days, from 17th to 27th February, 2015. The outputs are compared against the wind data, averaged over 1-h periods, at 18 m height. Once the inlet wind sector is selected, the temporal evolution of the wind direction from the CFD simulation is analysed against the experimental data at the sampling point (Fig. 5.16a). Overall, the outputs approximately capture the experimental wind direction over time, except in some hours. These differences are related to the mesoscale modelling errors as well as to the discretization of the inlet conditions of wind direction in the CFD simulations. Note that the continuous values derived from the mesoscale simulation is classified into the 16 wind sectors simulated with the CFD model. (a) (b) Figure 5.16: (a) Time series of wind direction from CFD results (blue line) and measurements (black line) at meteorological station site. (b) Time series of the computed wind speed from CFD results (blue line) and measurements (black line). Grey area represents the interval between the maximum and minimum values recorded at every hour. In addition, the wind speed is computed following the modelling approach (ucomp), but taking into account that the wind speed is directly proportional to the friction velocity: ucomp(t) = u∗,WRF (t) u∗,CFD(t) ) z=hcanopy · usim(Wsect(t)) (5.6) Figure 5.16b illustrates the temporal evolution of ucomp and the wind speed obtained from 110 CHAPTER 5. MODELLING APPROACH FOR URBAN AIR QUALITY ASSESSMENT experimental data at the same point. The ucomp reproduces the time evolution of the experimental data within the maximum and minimum values (grey area in Fig. 5.16b) recorded at every hour. For further details, the statistical parameters: NMSE, FB, R are computed together with the fraction of predictions within the factor 2 of observations (FAC2). The values obtained for the NMSE ans FB are respectively 0.31 and 0.27, which show a small deviation from the experimental data with a slight overestimation within the range of the acceptance criteria (NMSE<1.5 and −0.3