A review on the CFD analysis of urban microclimate

Urban microclimate studies are gaining popularity due to rapid urbanization. Many studies documented that urban microclimate can affect building energy performance, human morbidity and mortality and thermal comfort. Historically, urban microclimate studies were conducted with observational methods such as field measurements. In the last decades, with the advances in computational resources, numerical simulation approaches have become increasingly popular. Nowadays, especially simulations with Computational Fluid Dynamics (CFD) is frequently used to assess urban microclimate. CFD can resolve the transfer of heat and mass and their interaction with individual obstacles such as buildings. Considering the rapid increase in CFD studies of urban microclimate, this paper provides a review of research reported in journal publications on this topic till the end of 2015. The studies are categorized based on the following characteristics: morphology of the urban area (generic versus real) and methodology (with or without validation study). In addition, the studies are categorized by specifying the considered urban settings/locations, simulation equations and models, target parameters and keywords. This review documents the increasing popularity of the research area over the years. Based on the data obtained concerning the urban location, target parameters and keywords, the historical development of the studies is discussed and future perspectives are provided. According to the results, early CFD microclimate studies were conducted for model development and later studies considered CFD approach as a predictive methodology. Later, with the established simulation setups, research efforts shifted to case studies. Recently, an increasing amount of studies focus on urban scale adaptation measures. The review hints a possible change in this trend as the results from CFD simulations can be linked up with different aspects (e.g. economy) and with different scales (e.g. buildings), and thus, CFD can play an important role in transferring urban climate knowledge into engineering and design practice.


Introduction
The United Nations (UN) and the World Bank anticipate a rapid increase of the percentage of the world population living in urban areas within the course of the 21st century [1,2] (Fig. 1). This change is expected to occur due to the increase in the number of cities, migration from rural to urban areas and transformation of some rural settlements into urban areas [3]. Recently, making "cities and human settlements climate resilient and sustainable" is marked as one of the sustainable development goals by the UN [4]. As a result, research on sustainable habitats and related topics is gaining importance and will continue to do so in the coming years [5].
Urban settlements are formed by replacing natural surroundings by urban environments and the latter create their own, unique microclimates. 1 In his pioneering publication "the Climate of London", Luke Howard [6] documented that urban microclimates can be substantially different from their rural counterparts as the former tend to produce and retain more heat and are therefore characterized by higher temperatures. This phenomenon is commonly known as the Urban Heat Island (UHI) effect, a term first used by Manley in 1958 [7], although Erell et al. [8] mention it might have been coined earlier.
Interpreting the impact of the UHI effect merely as an "increase of temperature inside urban areas" would be an oversimplification. The UHI effect in particular and urban microclimate in general can yield a wide range of impacts on health and energy use and these impacts are not necessarily negative. For instance, UHIs can reduce building energy demand depending on the city [9], location within the same city [10], type of building [11,12] or meteorological conditions [13]. Knowledge of urban microclimate and its dependency on key physical parameters is important as input for urban designers, architects and engineers to design and plan built environments [14][15][16][17][18][19][20][21]. The inherent complexity and multiscale character evidently requires a multiscale approach for its analysis.
Urban microclimate as a research topic has a long history. The paper by Mills [22] identifies the work by Luke Howard [6] as the starting point. Mills [17] specifies six distinct periods of urban microclimate studies based on the research methodology followed: 1) Since the 1900s: Observation of urban-rural temperature differences with conventional meteorological measurement devices; 2) Since the 1960s: Measurement of urban microclimate process variables such as turbulent heat exchanges and the use of statistical methods to document UHI intensity; 3) Since the 1970s: Use of early energy budget models for the physical explanation of the UHI effect, early use of computer modeling techniques; 4) Since the 1980s: Adoption of experimental approaches, scaledphysical models and flux measurements (e.g. latent heat flux, storage heat flux); 5) Since the 1990s: Understanding the relationships between real urban forms and their effect on urban microclimate, organized field projects; 6) Since the early 2000s: Development of realistic urban microclimate models and employment of new techniques for the analysis of urban microclimate.
As suggested by these six periods, nowadays, a wide range of approaches can be employed for urban microclimate studies.  Figure modified from reference [3]. and Haghighat [15] and Mirzaei [23] distinguish two main categories: (a) observational approaches and (b) simulation 2 approaches. Observational approaches refer to measurement techniques such as field measurements, thermal remote sensing (e.g. satellite imagery) or small-scale physical modeling (e.g. atmospheric boundary layer windtunnel tests). Traditionally, observational approaches dominated urban microclimate analysis [24,25]. More recently, the increasing availability of computational resources has strongly advocated the application of numerical simulation approaches [26,27], where a distinction can be made between Energy Balance Models (EBM) and Computational Fluid Dynamics (CFD). The main advantage of the numerical simulation approaches compared with their observational counterparts is the opportunity to perform comparative analyses based on different scenarios [19,28]. In addition, while measurements are generally only performed at a limited number of points in space, numerical simulations can provide information on any investigated variable in the entire computational domain [16,24,29]. EBMs, which are based on the law of energy conservation for a control volume, have been used extensively in the past [30] and have increased in popularity with the pioneering article by Oke [31] entitled "The Energetic Basis of the Urban Heat Island". Later, several studies utilized EBM for validation and model development purposes [32][33][34][35][36][37][38]. In the early 2000s, new validated models were proposed by Masson [39], Martilli et al. [40] and Kanda et al. [41]. Throughout this period of new EBM developments, the use of observational approaches, such as heat flux measurements, has continued [42][43][44], mostly to support the validation of newly developed models.
From an urban climate research point of view, CFD offers two advantages compared to EBM: (1) CFD is capable of performing simulations with the explicit coupling of velocity and temperature fields and if necessary, with the addition of humidity and pollution fields; (2) With CFD, it is possible to resolve the flow field at finer scales (e.g. building or even human scale) than EBM [45]. On the other hand, CFD simulations require a high-resolution representation of the urban geometry, the knowledge of boundary conditions for all relevant flow variables and adequate computational resources [15,16,19].
With the increased necessity for simulations incorporating higher spatial and modeling details and driven by the advances in computing power [27], CFD has continued to gain popularity as a tool for urban microclimate research, in particular from the 1990s. In the concluding chapters of two urban climate review papers by Souch and Grimmond [26] and Kanda [28], the increasing popularity of the CFD approach is pointed out with the following quotes: "The development and use of CFD is a very active area of inquiry. The models are becoming more sophisticated in terms of numerical methods, mesh structures and turbulence modeling approaches." (Souch and Grimmond [26]); "CFD technologies that explicitly resolve urban buildings are the most complex representation of urban surfaces. Such technologies will play an important role not only in pure application studies but also in guiding the improvement of simpler models" (Kanda [28]).
Computational simulations can be employed to study urban microclimate at different spatial scales, ranging from the meteorological mesoscale over the meteorological microscale to the building scale and the indoor environment [16,19,29] (Fig. 2).
The smallest scale at which CFD is employed for climatic analysis in urban areas is the building indoor environment, where typical horizontal distances are around 10 m and the focus is on indoor climate. Studies at this scale have employed CFD mainly for ventilation studies [127][128][129] and for topics related to HVAC design and building services engineering [130]. Natural ventilation studies with CFD can also be performed at multiple scales, by combining building and indoor scales, which enables researchers to conduct coupled analyses [94,95,[131][132][133][134][135][136][137][138][139][140][141].
Some review papers on the analysis of urban microclimate such as Fig. 2. Schematic representation of the spatial scales in climate modeling, with typical horizontal dimensions. 2 Here, the authors refer to numerical simulations, as opposed to physical simulations in e.g. atmospheric boundary layer wind tunnels. Erell and Williamson [142], Ooka [143], Mochida et al. [51] and Lun et al. [144] have evaluated numerical models in general (including EBMs), without a specific focus on any CFD approach. Mochida and Lun [81] have reviewed CFD microclimate studies, but without focusing on the coupling of velocity and temperature fields. As CFD studies on urban microclimate are gaining popularity, it is important to document the achievements and trends in this field for future research, and this paper serves this purpose. This paper reviews studies on the CFD analysis of urban microclimate. The scope of the review covers studies published in refereed journals, in English, with 3D computational domains and with coupling of velocity and temperature fields. To the best of our knowledge, the first study that fits to this scope is from 1998. Therefore, this review covers studies from what we consider as the first study in this field until the ones from 2015. In Section 2, the investigated studies are listed and classified based on the type of the urban area considered (generic versus real urban) and methodology followed (with or without validation study). Section 3 contains a further analysis of the reviewed studies and Section 4 presents a discussion with future perspectives. Finally, Section 5 contains the conclusions.

Overview of studies on the CFD analysis of urban microclimate
Within the above-mentioned scope of the review, a total of 183 studies are identified and investigated. The earliest study is from 1998 and the latest is from 2015. Fig. 3 shows the yearly distribution of the studies, indicating the increasing popularity of the field. The figure shows that the number of studies considered only in the last three years constitute more than half of all the studies (104 of 183 studies).
The papers are categorized based on the type of urban area (generic versus real) (see Fig. 4) and the methodology, without validation versus with validation. We remark that a study containing validation of at least one parameter from velocity and/or temperature field is classified as a study with validation. Fig. 5 shows that most studies are focused on real urban areas and are conducted without validation.
The studies are summarized in tables with the following entries: -Author(s) and publication year; -Reference number as listed in this paper; -Urban setting/location ￮ For studies with generic urban areas, the urban geometries are classified as follows (Fig. 6  Later, keywords with similar or interchangeable use are grouped and in total 37 keyword categories are identified. 3 For papers with less than five keywords in the list of keywords, first the title and then the abstract is scanned for selecting suitable keyword categories. Referring to the title and the abstract for selecting keywords has its limitations but it was adopted due to the lack of a better alternative. Some very general keywords, such as microclimate, CFD and urban heat island (effect) are omitted from the categories. At the end, five keywords categories per study are identified. In the remainder of this paper, we will use the word "keyword" to refer to these "keyword categories". In alphabetical order, the categories are:

Studies for generic urban areas
CFD studies for generic urban areas typically comprise simple building shapes, such as cubes or rectangular prisms. Early CFD models employed for microclimate analysis considered generic domains for model development and validation purposes. Later studies were generally conducted to investigate generic aspects of fluid flow and/or heat transfer in urban areas that can provide basic insights that subsequently can be translated to understanding these processes in real urban areas.

Studies without validation
Most early studies on generic urban areas did not consider validation. For example, this is the case for the five of the oldest studies in this review: Bruse [172]. The subcategory "generic urban areaswithout validation" contains 40    Table 1 Overview of studies in the sub-category "generic urban areas -without validation".    studies, which are summarized in Table 1.
Most early studies focused on new model developments, demonstrating the suitability of CFD for microclimate analysis. For instance, the study by Bruse and Fleer [168] focused on surface, plant and air interaction at the microscale and is considered as the original documentation of the CFD microclimate software ENVI-Met, which is a tool increasingly employed by researchers in later years.
In the late 1990s, EBMs were the main tool used for the numerical analysis of urban microclimate. In the early years of CFD microclimate analysis, the influence of EBM methodology on CFD simulations is evidenced in the following ways: (1) CFD studies have averaged most of the turbulent heat fluxes in urban canopies similar to the way they were used in EBMs [169,170,172,173]; (2) the terms, which were popular in EBMs, such as "energy budget" and "turbulent heat fluxes," were very often used as the main keywords in these early CFD studies.
The five most commonly used keywords in this sub-category are vegetation (19 of 40

Studies with validation
Validation of CFD studies for generic urban areas is typically performed with data from wind-tunnel measurements [15,19] whereas validation with field measurements is less common [206]. Many studies have been performed on the CFD validation of urban flow patterns in terms of velocity fields [29] but these studies are often conducted for isothermal conditions and as such are not within the scope of this review. As mentioned at the beginning of this section, a study is considered "with validation" as long as there is at least one parameter related to velocity or temperature fields, which is compared with measurement data. The sub-category "generic urban areaswith validation" contains 21 studies, which are summarized in Table 2.
Some of the validation studies are conducted to investigate and demonstrate the suitability and accuracy of newly developed CFD approaches [207][208][209][210][211][212][213]. Others focus on the CHTC of individual buildings in urban areas [210,214,215], which in turn can be used for coupling CFD with BES [210]. Validation studies on generic urban areas can be the first step towards justification of a CFD approach in modeling the cooling effect of adaptation measures, before implementing the same approach on real urban areas. Some of the studies for instance propose validated approaches for the cooling effect from vegetation sources [209] and from water bodies [213] on generic urban domains.
The five most commonly used keywords in this sub-category are wind (flow) (10 of 21 studies), canyon (9 of 21 studies), case comparison (7 of 21 studies), model development (7 of 21 studies) and surface heating (6 of 21 studies). Note however that keywords such as climate sensitive design, energy budget, optimization, seasonal variation, sustainability and urban design do not occur as keywords in any of these 21 studies. 5

Studies for real urban areas
The term "real urban areas" can cover only a few buildings to a portion of a city. CFD simulations on real urban areas are performed either as practical case studies orin case of studies with validationto investigate the possibilities and limitations of CFD for real urban areas that are generally characterized by a complexity that substantially exceeds that of generic urban areas.

Studies without validation
CFD studies on real urban areas without validation are generally comparative studies, where several urban configurations, design parameters, neighborhoods (districts) within the same urban area are compared. Most of these studies aim to reach best-case scenarios based on the optimization of a target parameter (e.g. outdoor thermal comfort). The sub-category "real urban areaswithout validation" contains 65 studies, rendering this sub-category the most popular one. These studies are summarized in Table 3.
The table indicates that studies in this sub-category have two characteristics. First, all studies are conducted using the RANS equations, which may be attributed to the high computational requirements of LES. Second, most studies are characterized by practical rather than fundamental keywords. The five most commonly used keywords in this sub-category are vegetation (39 of 65 studies), case comparison (36 of 65 studies), thermal comfort/heat stress (29 of 65 studies), urban design/planning (27 of 65 studies) and materials/ albedo (19 of 65 studies). Keywords which might indicate that a particular study's focus is on fundamental aspects, such as CHTC, heat transfer or wind (flow), are not used often (1 of 65, 2 of 65 and 9 of 65 studies respectively). Keywords such as energy budget, optimization, statistical analysis, surface heating and thermal stability do not occur as keyword 6 in any of these 65 studies.

Studies with validation
Validation of CFD studies for real urban areas is typically performed with data from on-site (field) measurements. Wind-tunnel measurements focusing on real urban areas can be challenging as the representative models of real, complex urban formations are more difficult (and often more expensive) to build than generic forms and they are often larger, hence might be difficult to fit into a wind tunnel.
CFD urban microclimate studies in the sub-category "real urban areaswith validation" include validation based on one or more of the simulation parameters with measurements. We note that studies on the CFD validation of urban flow patterns in terms of velocity fields are often conducted for isothermal conditions and as such are not within the scope of this review. For real urban areas, the field measurement of  4 Although these words might be used within the text itself, here, they are not present in the keyword categories as extracted from abstract and the list of keywords. 5 Although these words might be used within the text itself, here, they are not present in the keyword categories as extracted from the abstract and the list of keywords. 6 Although these words might be used within the text itself, here, they are not present in the keyword categories as extracted from the abstract and the list of keywords.

Y. Toparlar et al.
Renewable and Sustainable Energy Reviews 80 (2017) 1613-1640 Table 3 Overview of studies in the sub-category "real urban areas -without validation".  air temperature is relatively straightforward and especially in the last years, many campaigns are undertaken for measurement data, although the suitability and availability of these data for scientific use and especially for validation can be a limitation. The reason is that the complexity and inherent variability of the meteorological conditions not only require careful measurement of a large number of parameters (to be used as boundary conditions in the simulations) but also a very complete reporting of urban area, measurement set-up, measurement accuracy, etc., without which a detailed and thorough validation exercise will not be possible [296,301,306,307]. The sub-category "real urban areaswith validation" contains 57 studies, 47 of which use air temperature as one of the validation parameters. The studies belonging to this sub-category are summarized in Table 4. Similar to the studies on real urban areas without validation, all the studies in this sub-category employed the RANS equations. However, different from the studies without validation, studies in this subcategory have radiation as a relatively popular keyword (16 of 57 studies). That is mostly because recent studies on human thermal comfort demonstrated the importance of thermal radiation (e.g. mean radiant temperature) on thermal comfort levels. Therefore, studies try to validate their CFD simulation results based on radiation parameters.
The five most commonly used keywords in this sub-category are thermal comfort/heat stress (29 of 57 studies), vegetation (28 of 57 studies), materials/albedo (19 of 57 studies), case comparison (20 of 57 studies) and radiation (16 of 57 studies). On the other hand, keywords such as energy budget, economy, optimization, surface heating and thermal stability do not occur as keywords 7 in any of these 57 studies.

Urban setting/location investigated
Considering generic urban areas, five studies in this sub-category [177,191,216,218,223]  Studies for generically distributed building blocks are mostly case comparisons without validation and they focus on the effect of different urban geometries (e.g. orientation, density) [174,176,178,189,197,223], vegetation patterns [171,193,199], building materials [188] and building forms [190,191]. Studies on building blocks that include validation are in most of the cases targeted at more fundamental fluid flow or heat transfer aspects [207,210,214,215,225]. Studies for street canyons and urban street canyons typically investigate canyon related aspects, such as the effect of aspect ratio [175,180,194,224] and wind/ventilation [172,182,184,185,195,196,222].
The majority of the studies on real urban areas are conducted for locations in mid-latitude climates and in the developed regions of the world (Fig. 9).    Table 4 Overview of studies in the sub-category "real urban areas -with validation.".

CFD equations/models
Among the investigated 183 studies, 7 of them did not specify the approximate form of the governing equations used. As for the remaining 176 studies, 169 (96.0%) used only RANS, 5 used only LES (2.8%) and 2 used both LES and RANS (1.1%) as approximate form of the governing Navier-Stokes equations.
A microclimatic CFD simulation that couples the temperature and velocity fields has a higher computational cost and the choice of LES over RANS evidently will increase this cost. It is expected that the increased computational cost and the often sufficient accuracy of RANS [146,304,363,364] are the two main reasons why the vast majority of studies was performed with RANS, even though LES is generally considered to be more accurate than RANS [19,29,73,86 ,89,292,300,304,363-368]. Apart from RANS and LES, the third approach often used in CFD simulations, that is Direct Numerical Simulation (DNS) is not utilized among the microclimate studies investigated here. Due to its dominant use in the investigated studies, the remainder of this section will focus on the RANS approach. Fig. 10 shows the distribution of the use of turbulence models in these studies. Among the 171 studies using RANS, 6 of them [201,210,220,291,324,353] have considered two or more turbulence models.
The most commonly used turbulence model is the Yamada and Mellor E-ε [158] turbulence model (used in 86 studies, or 49% of total). Although this turbulence model is not explicitly recommended or adopted in the CFD best-practice guidelines [19,[75][76][77][78], its popularity results from it being the only available turbulence model option in the microclimate simulation tool ENVI-Met [168]. The second most popular turbulence model is the standard k-ε [157] model (used in 45 studies -25%). The RNG k-ε [155], Realizable k-ε [154] and Modified k-ε [131] turbulence models appear to have similar popularity compared with each other. The standard k-ε model [157] is one of the most popular turbulence models among CFD studies [76] but as argued in several publications [75,95,364], some improved models, such as the realizable k-ε model [154] can show better performance in resolving the mean flow field. Fig. 11 illustrates the use of turbulence models over the years 2010-2015 and the popularity of YMEE (the turbulence model used in ENVI-met) as mentioned before. The figure also shows that except for the year 2014, more recent CFD studies are now using turbulence models other than standard k-ε more often.

Target parameters
As shown in the foregoing tables, the majority of the studies considered more than one target parameter. Fig. 12a shows the distribution of the target parameter categories used and Fig. 12b shows the distribution of the seven most used target parameters. Most studies considered temperature related parameters, especially air temperature for comparison, evaluation or validation purposes. This parameter is followed by wind velocity, surface temperature and mean radiant temperature.
Target parameters related to fundamental fluid flow or heat transfer, such as TKE and CHTC, are mostly used in studies with generic urban areas. Among the eight studies that considered CHTC as a target parameter, seven were conducted in generic urban domains and for the TKE, this ratio is found to be 4/5. This is further evidence that studies on real urban areas do not typically consider parameters related to fundamental flow aspects, which can be explained by their larger complexity or by their difficulty for collecting measurement data. Economic parameters and statistical performance indicators can be used in communicating the results from urban microclimate studies to professionals from other disciplines, such as policy makers, as these aspects may lead to more generalized conclusions. However, although these target parameters are considered in some recent CFD urban microclimate studies, their use is still limited, with only 2 of 183  As CFD has demonstrated its capability for involving multiple scales, there is an increasing interest in linking the results from microclimate analysis with building scale aspects, such as building energy consumption and indoor air temperature. Even though these parameters are not yet considered quite often (BEC 16 of 183 studies and IAT 1 of 183 studies), while the oldest study specifying one of these parameters is only from the year 2011 [182], 11 of these studies are from the last two years, demonstrating the increasing interest.

Keywords
As described in Section 2, keywords are selected either directly from the provided keywords list or from the titles and abstracts of investigated papers. Keywords with similar or interchangeable use were grouped in 37 keyword categories and five keywords per study were identified, as listed in Tables 1-4. According to this procedure, the most used three keywords in CFD urban microclimate studies are: vegetation (90 studies), case comparison (78 studies) and thermal comfort/heat stress (77 studies).
Apart from the number of studies, in this paper, we suggest and apply additional metrics for documenting the annual use of the keywords. One of them is the first year a keyword is introduced in CFD urban microclimate studies (First Year Index = FYI). A second metric called "Weighted Year Index" (WYI) is defined as the weighted average year of a particular keyword's usage: with y the year, y 0 the year of the earliest study, y n the year of the latest study (as investigated in this paper), and k y is the number of times a keyword is used in the y th year. This metric indicates the year associated with a keyword's average use. A lower WYI means that the use of a particular keyword is mainly situated in earlier years, while a higher WYI means that the keyword use is mainly situated in recent years. According to the analysis, of all 37 keywords, the keyword with the highest FYI and WYI is "statistical analysis" (FYI = 2014, WYI = 2014.8), while that with the lowest FYI is "wind / flow" (FYI = 1998) 8 and that with the lowest WYI is "energy budget" (WYI = 2000). The annual use of these keywords along with that of the most common keyword (vegetation) is illustrated in Fig. 13.
A more comprehensive view of the relationship between the historical use of keywords and the associated number of studies, or between the FYI and the WYI and number of studies, is given in Fig. 14. This chart graphically demonstrates the number of times each keyword is used (by size of the circle), whether a keyword is relatively new or old (indicated by FYI on horizontal axis) and whether a keyword is used more often in earlier or more recent years (indicated by the WYI on the vertical axis). The average overall metrics for the ensemble of all keywords are: average number a keyword is used: 24.7, average FYI: 2004.3 and average WYI: 2011.9. The latter two numbers define the origin of the coordinate system in Fig. 14. Fig. 14 shows that the keyword use in CFD studies of urban microclimate has transitioned from keywords related to model development (heat energy budget, turbulent heat fluxes, model development) to keywords related to urban scale adaptation measures (e.g. adaptation, climate sensitive design) and energy (e.g. building energy, sustainable). Highly used keywords with large circles in Fig. 14, such as vegetation, case comparison, thermal comfort, materials/albedo and urban design/planning are keywords which are used since the early years of this research field and are still very commonly employed.

Validation studies
According to the American Institute of Aeronautics and Astronautics (AIAA), validation is "the process of determining the degree to which a model is an accurate representation of the real world from the perspective of the intended uses of the model" [370]. According to several CFD best practice guidelines [19,[75][76][77][78]292], CFD simulations should be evaluated critically and the results should only be considered reliable after comparing with some measurement data. If measurement data are hard to obtain for a specific case, subconfiguration validation can be performed for the intended use of the model [19]. One example of a sub-configuration validation is the study by Gromke et al. [275] where the authors used a vegetation model which is validated with measurement data in separate CFD simulations and then used the same model in a real urban area. The advantage of sub-configuration validation can be relevant for complex studies with different physical models, as the classical validation of the whole simulation setup can be difficult to conduct due to the uncertainty in parameters.
According to this review, the majority of the CFD urban microclimate studies (105 of 183 studies) are conducted without validation. The percentage share of the studies without validation seems to have remained rather stable in the last years, as demonstrated in Fig. 15. However, it is imperative that CFD urban microclimate studies include validation much more often to ensure the desired reliability and predictive capability.
The most common reason for the absence of validation in CFD microclimate studies might be the lack of relevant and well-documented measurement data. Measurement campaigns on urban areas can have some challenges such as logistic difficulties, data quality issues (e.g. ventilated vs non-ventilated temperature measurements) and problems with spatial representativeness.
Although these challenges remain, difficulties in obtaining relevant data can be overcome in a much better-connected World. Internet resources can be a good alternative for measurement data as large datasets are now within reach. For instance, surface temperature data for various locations around the world can now be obtained from the National Oceanic and Atmospheric Administration's (NOAA) satellite imagery dataset. In one of the plenary session presentations during the 9th International Conference on Urban Climate (ICUC9), Chapman [371] mentioned the possibility of using low-cost air temperature sensors with WiFi network, distributed vastly in urban areas. Such a network could have the potential of becoming a part of Internet of Things [372,373] and consequently could provide a large amount of measurement data for microclimate researchers. In the future, researchers can investigate such resources carefully to find relevant measurement datasets to validate their CFD simulation results. Fig. 9 illustrated the limited variety of studied urban locations. In the article entitled Urban Climatology: History, status and prospects, Mills [17] identifies one of the major challenges in the urban climatology field as: "acquiring information on the climates of cities in less prosperous regions, which are growing rapidly and many of which have tropical climates in which there have been few studies". Although this statement was made referring to urban climatology  studies in general, CFD urban microclimate studies are following a similar route. According to the UN World Urbanization Prospects [2], most of the urbanization expected to occur in the 21st Century will take place in the developing regions of the World. Therefore, in the future, more studies are needed focusing on urban areas in these regions, especially on the ones with high/increasing populations (e.g. African continent, India, Latin America). Another observation in the location analysis is the relation between the study location and the physical location of the respective research groups. Mostly, research groups with expertise in CFD urban microclimate studies investigate the city in which they are located or to which they are closely situated. In the future, researchers should aim at expanding the CFD knowledge to various parts of the World and should not limit themselves to their own vicinity.

Urban locations
Most of the reviewed papers focus on the negative consequences of the UHI effect (e.g. on heat stress, energy demand). Therefore, adaptation/mitigation measures aiming mostly at temperature reductions are proposed and scientifically tested for implementation in the various parts of the World. However, the UHI effect may not necessarily always cause negative consequences. The numerical study by Hirano and Fujita [11], focusing on the climate and building stock of Tokyo has shown that for residential buildings, the UHI effect can have a net positive effect on the building energy demand on a yearly basis. It is safe to assume that if a city is located in colder climate zones, the UHI might actually be beneficial. Future CFD studies can investigate in more detail the consequences of the UHI effect in the colder parts of the World.
Furthermore, the reviewed studies are mostly conducted for cities in the mid-latitudes and cities near the arctic circles are not considered often. Among the reviewed studies, the urban area closest to the Arctic or Antarctic Circle, was Glasgow [352]. Some higher latitude cities, such as Oslo, Stockholm or Moscow can be considered in future CFD urban microclimate studies and new information can be gained. Note that the lack of urban planning and urban microclimate studies in the arctic regions is mentioned in a recent study by Ebrahimabadi et al. [374]. Similarly, (sub)tropical regions are not often considered in CFD microclimate studies. According to Roth [375], among all the urban climate studies, studies on sub(tropical) regions constitute less than 20% and according to the present review, this ratio is 8% (15 of 183 studies).

CFD equations/models
Almost half of the investigated studies used the ENVI-Met [168] software. This software combines several physical phenomena (e.g. fluid flow, heat transfer, mass transfer, vegetation interactions) for urban microclimate analysis. Limited modeling options in the software, such as the availability of only one turbulence model (Yamada and Mellor E-ε [158]), the limited options for grid generation and the lack of information about wall functions, can be considered as drawbacks. Such limitations and their possible implications on the results are mentioned in several studies which were included in this review paper [175,312,319,333,339,349].
After the Yamada and Mellor E-ε [158], the second most commonly used turbulence model is the standard k-ε turbulence model [157]. Over the years, the standard k-ε turbulence model has been used in many CFD studies but it is well-known that with this turbulence model, the production of kinetic energy near the frontal corners of buildings is overestimated and that the turbulent kinetic energy in wake regions is underestimated (e.g. [76,376]). Various CFD best practice guidelines [19,76,77] referred to this problem and advised the use of other turbulence models, such as realizable k-ε [154]. Among the reviewed studies, the standard k-ε turbulence model [157] is used many times (45 of 183 studies) but turbulence models such as realizable k-ε [154] (used in 11 of 183 studies) or RNG k-ε [155] (used in 11 of 183 studies) are gaining popularity. Among the 22 studies using one of these two turbulence models (realizable and RNG k-ε), 17 are published in the last three years.
In many CFD review papers and best practice guidelines on urban flow fields, LES is reported as superior to RANS simulations in terms of results accuracy compared with measurements [19,29,73,86,89,292,300,304,[363][364][365][366][367][368]. According to this review, the LES approach is not yet very popular for urban microclimate studies, as only less than 5% of studies followed this approach. Computational requirements and the lack of best practice guidelines for the use of LES can be the two main reasons for this. However, in the near future, rapid developments in the computational resources can and should result in more studies with LES approach.

Target parameters
Most of the CFD urban microclimate studies focused on parameters related to temperature, wind flow, thermal comfort and heat transfer. CFD has repeatedly demonstrated its predictive capability in validation studies focusing on different parameters. The three most commonly used parameters are air temperature ( Measurement data for air temperatures is relatively easier to obtain compared with surface temperatures and thus are used more often. The accurate prediction of surface temperatures can be harder as surface heat transfer modeling requires appropriate near-wall modeling. Therefore, validation of surface temperatures can be a more vigorous test for the predictive capability of CFD simulations.
Some past studies used the percentage difference in predicted and measured temperatures for evaluating the simulation performance. Such comparisons can be improved in the future, as the outcome of percentage difference in temperatures would be different for every temperature unit. For validation studies, dimensionless parameters and/or statistical performance measures can be used more frequently to evaluate the performance of simulation approaches.
CFD and complementary numerical tools can be used to link climate studies from different scales. For instance, the results from urban microclimate simulations can be employed as boundary conditions for simulations at building or indoor scale. This is necessary to accurately assess the effect of urban microclimate on building energy demand and/or indoor ventilation. This review showed that target parameters such as building energy consumption and ventilation rate are not very common; though parameters related to building energy consumption have an increasing popularity. Future studies can focus on these parameters, which can create a deterministic link between urban design and building design [377].
The relevance of urban climatology for urban designers and policy makers is mentioned in several review studies [17,[378][379][380]. Policy makers and/or professionals responsible for public resources (e.g. local municipalities) are interested in the economic consequences of planned adaptations or modifications in urban areas [381]. In the past, microclimate studies (not with CFD approach) have showed the positive economic consequences of adaptation measures on key issues such as energy demand, thermal comfort and human productivity. CFD simulations can be coupled with financial models to obtain deterministic results on the economic aspects of adaptation measures. So far, the reviewed studies showed that the economic aspects are almost never considered with a relevant target parameter. The missing link between fundamental microclimate studies and the economic consequences can be an important aspect in the near future.

Keywords
The keywords can be classified in four categories based on the quadrants in Fig. 14.
Keywords with low FYI and low WYI, such as heat transfer, model development, energy budget and turbulent heat fluxes were mainly used in the earlier studies and are not used very often in the later studies. With more validated CFD approaches, research efforts have shifted from the development of new models to case studies.
Keywords with low FYI and high WYI, such as materials/albedo and vegetation are used in CFD urban microclimate studies since the early 2000s and they are still used very often. Vegetation is the most common keyword as many studies investigated the effect of street trees, urban parks and green roofs/facades since the early years of this research field and still, similar studies are conducted for different cities.
Keywords with high FYI and high WYI are not used very often yet, because they are very recent. Among the new keywords with high WYI values, adaptation/mitigation and climate sensitive design are gaining popularity not only among the CFD studies but also among studies with different methodologies [20,382]. New keywords such as "sustainable" and "climate" demonstrate the effect of the popularized sustainable development challenge on this research field.
Keywords with high FYI but with low WYI values, such as model coupling, thermal stability and optimization refer to studies, which are recent but are not very common. Typically, studies with these keywords are very specialized. For instance, the "optimization" keyword belongs to this group and the parametric optimization of CFD results would require many simulations, with a dedicated campaign, which may have affected its FYI and WYI values.
In the future, many of the new keywords are expected to continue to increase their popularity. Among these new keywords, statistical analysis should play an important role in testing new models and simulation cases more effectively. The economic aspects of adaptation/ mitigation strategies should be evaluated with new methods, possibly by linking multiple scales. Even though the effects of vegetation and materials on urban microclimate seem to be well understood, it might continue being investigated with new case studies and with studies performed on generic urban areas to provide general conclusions (or guidelines) for professionals from other disciplines. CFD is a useful tool for deterministic judgement and researchers in these other disciplines should benefit from this.

Limitations of the review
Given the large scope and large number of publications in this topic, some studies had to be omitted from this review. As denoted in Section 2, this review identified CFD microscale studies, which couple velocity field with temperature field on 3D computational domains. CFD studies on urban microscale investigating pedestrian wind comfort [82,383,384] and thermal comfort [385], are not investigated in this review if their focus was only on the modeling of velocity field, without the coupling with temperature. In addition, this review paper focused only on journal papers which are prepared in English language but surely, valuable studies on CFD urban microclimate analysis have been published in the past as conference papers (e.g. [386][387][388]) or in other languages (e.g. [389,390]).

Conclusions
Considering the trend towards urbanization and the challenge of sustainable habitats, studies on urban microclimate will continue to gain popularity in the 21st century. Numerical methods to analyze urban microclimate are essential tools for engineers, architects, urban planners and policy makers to compare urban design alternatives and to manifest guidelines. CFD is one of these numerical tools, which is frequently used in the urban climate at various spatial scales. CFD studies on the meteorological microscale, where typical spatial distances are less than 2 km, are gaining popularity due to their advantages such as the explicit modeling of urban and building geometry and resolving the flow field with high spatial resolution. Though gaining popularity, to the best of our knowledge, there has been no review paper yet that is dedicated to CFD studies on urban microclimate. This paper presented a systematic review and analysis of the CFD urban microclimate studies that were published in peer-reviewed international journals from 1998 until the end of 2015. A total of 183 studies were identified which include 3D computational domains and couple the velocity and temperature fields. The studies were categorized based on the types of urban areas investigated, real or generic, and on the methodology followed, studies with and without validation.
For every sub-category, the studies were listed in tabular form based on their publication year, location, CFD equations/models, validation parameter (if any), keywords and target parameters. A comparative analysis was provided based on the findings.
From this review paper, the following conclusions can be made: -CFD results should be subjected to detailed validation in the future. This review documented that 58% of the existing CFD microclimate studies have not considered any validation. In order to improve the reliability and the predictive capability of CFD simulations, future studies should collect and employ relevant measurement data to support simulation results. -Even though CFD urban microclimate studies are gaining popularity, the urban locations investigated do not have a large variety. Among the 122 studies focusing on real urban areas, only 74 cities from 30 countries have been the subject of CFD simulations. Especially for the cities located in the developing regions and in the tropics, very few studies can be identified. CFD urban microclimate knowledge needs to be expanded to the developing regions of the World. -The review documented that 96% of the studies considered RANS simulations only, even though LES has the potential to be more accurate in predicting flow field. -Among the RANS simulations, 74% used either Yamada-Mellor E-ε or standard k-ε turbulence models. The choice of turbulence models can be limited with the availability of the software packages. However, detailed validation studies and the resulting increased accuracy and reliability of CFD studies will benefit from the availability and/or use of more turbulence models. -The results from validation studies can be communicated using statistical performance indicators, which were not used in the past studies often. Moreover, the target parameters can be selected from other spatial scales (e.g. building energy demand) or from generic terms (e.g. economy, emissions) for a thorough analysis on the effect of microclimate and adaptation measures on humans, buildings and urban infrastructure. -The themes of CFD urban microclimate studies were documented by investigating the keywords they used. According to this investigation, the early CFD urban microclimate studies had been conducted for model developments and case studies. In the past few years, more studies about urban scale adaptation measures and thermal comfort are conducted. Future studies might focus more on systematic studies with multiple scales (e.g. mesoscale, building scale) and aspects (e.g. economical) to transfer the gained knowledge from urban climatology to routine building and urban design guidelines.
These recommendations are in fact against the current trends observed among CFD urban microclimate studies. Future studies on this topic can consider these recommendations to push the boundaries of this field for acquiring new knowledge on urban climatology.