Global Trends in Air Pollution Modeling over Cities Under the Influence of Climate Variability: A Review

Abstract

The objective of this article is to conduct a review to analyze global trends in the use of air pollution models under the influence of climate variability (CV) over urban areas. Five scientific databases were used (2013–2024): Scopus, ScienceDirect, SpringerLink, Web of Science, and Google Scholar. The frequency of citations of the variables of interest in the selected scientific databases was analyzed by means of an index using quartiles (Q). The results showed a hierarchy in the use of models: regional climate models/RCMs (Q3) > statistical models/SMs (Q3) > chemical transport models/CTMs (Q4) > machine learning models/MLMs (Q4) > atmospheric dispersion models/ADMs (Q4). RCMs, such as WRF, were essential for generating high-resolution projections of air pollution, crucial for local impact assessments. SMs, such as GAM, excelled in modeling nonlinear relationships between air pollutants and climate variables. CTMs, such as WRF-Chem, simulated detailed atmospheric chemical processes vital for understanding pollutant formation and transport. MLMs, such as ANNs, improved the accuracy of predictions and uncovered complex patterns. ADMs, such as HYSPLIT, evaluated air pollutant dispersion, informing regulatory strategies. The most studied pollutants globally were O₃ (Q3) > PM (Q3) > VOCs (Q4) > NOx (Q4) > SO₂ (Q4), with models adapting to their specific characteristics. Temperature emerged as the dominant climate variable, followed by wind, precipitation, humidity, and solar radiation. There was a clear differentiation in the selection of models and variables between high- and low-income countries. CTMs predominated in high-income countries, driven by their ability to simulate complex physicochemical processes, while SMs were preferred in low-income countries, due to their simplicity and lower resource requirements. Temperature was the main climate variable, and precipitation stood out in low-income countries for its impact on PM removal. VOCs were the most studied pollutant in high-income countries, and NOx in low-income countries, reflecting priorities and technical capabilities. The coupling between regional atmospheric models and city-scale air quality models was vital; future efforts should emphasize intra-urban models for finer urban pollution resolution. This study highlights how national resources and priorities influence air pollution research over cities under the influence of CV.

1. Introduction

Atmospheric pollution is defined as the presence of chemical, physical, or biological agents that alter the natural characteristics of the atmosphere, causing adverse effects on human health and the environment [1,2]. Climate change and variability are interrelated phenomena that significantly impact air quality over cities. Climate variability (CV) refers to the natural fluctuations in climate that occur over short time scales, ranging from seasons to decades, while climate change involves a long-term alteration in the Earth’s average climate conditions, beyond natural variability [3,4]. The primary difference between the two lies in the time scale of study. Moreover, CV is a natural and cyclical phenomenon, whereas current climate change is primarily driven by human activities and is characterized by sustained global warming [5]. Understanding this distinction is crucial for determining whether an extreme event, such as a heatwave or drought, is a result of natural variability or is amplified by climate change, and how this extreme event can influence air pollution over cities. Currently, it is evident that climate variability, manifested in the spatiotemporal variability of precipitation and temperature, significantly influences human activity, generating severe socioeconomic and environmental effects [6,7]. The characterization of CV and its behaviors has two main applications: first, climate modeling uses recovered data to improve the understanding of climate manifestations and their future variability in a scenario of anthropogenic climate change; second, climate dynamics and their extreme manifestations are incorporated into the historical interpretation of climate [8,9,10].

Climate processes influence the formation, transformation, and removal of air pollutants over cities. For example, aerosol formation is a complex process involving the condensation of gases such as SO₂ and VOCs on solid or liquid particles suspended in the air [11]. These particles can be primary, emitted directly from sources such as road dust, or secondary, formed through chemical reactions in the atmosphere [12]. Chemical reactions, such as the oxidation of NOx and VOCs under the influence of solar radiation, lead to the creation of secondary pollutants, such as tropospheric O₃ [13]. The removal of these pollutants occurs primarily through dry and wet deposition. Dry deposition refers to the direct accumulation of air pollutants on land surfaces, while wet deposition involves the removal of air pollutants through precipitation, which cleans the air but transfers pollutants to soil and water bodies [14]. Moreover, urban characteristics directly influence air quality and how CV effects manifest themselves. Topography could act as a barrier that traps pollutants, while land use, such as the presence of impermeable surfaces, increases local temperature and exacerbates the urban heat island phenomenon [15]. Population density affects the concentration of emission sources, such as vehicular traffic and industrial activities, increasing the air pollutant load in densely populated areas [16]. Vegetation in urban areas also plays a dual role, helping to capture air pollutants through dry and wet deposition and reducing temperatures through evapotranspiration, which can partially mitigate the negative effects of air pollution over cities [17].

Air pollution models are used to forecast contamination levels by addressing key processes and emission sources. These models represent the evolution of gas and particle concentrations, influenced by climate factors obtained from prediction models [18]. Emissions, both anthropogenic and natural, serve as sources of chemical species, and transport processes distribute these species horizontally and vertically [19]. Concentrations change due to resolved processes (advection and sedimentation) and unresolved processes (convection and turbulent diffusion) by the model, in addition to being subject to photolysis and chemical reactions [20,21]. Air pollution models are essential tools for understanding and predicting how pollutants disperse, transform, and deposit in the urban environment [22]. Atmospheric dispersion models simulate how pollutants move through the atmosphere from their emission sources, considering factors such as wind, topography, and chemical interactions [18]. Statistical models are used to identify and quantify the relationships between CV and air pollution over cities [23]. Regional climate models allow the simulation of atmospheric phenomena on smaller scales, providing detailed predictions on how CV may affect the dispersion and concentration of air pollutants in specific areas [24]. Chemical transport models simulate the movement and transformation of chemical species in the atmosphere, providing detailed insights into the spatial and temporal distribution of air pollutants [19]. Machine learning models leverage large datasets to predict air quality by identifying complex patterns and relationships between various environmental factors [25].

Despite significant advances in the understanding of air pollution over cities under the influence of CV, a knowledge gap persists in the synthesis and critical evaluation of the modeling approaches employed. Previous research (e.g., [26,27,28,29,30,31]), although extensive, possibly lacked a comprehensive and comparative overview of the simulation models used, limiting the identification of best practices and the understanding of their relative strengths and weaknesses. Specifically, the lack of a systematic analysis of the frequency of use and temporal evolution of the models, together with a detailed evaluation of the climate parameters and air pollutants considered, probably hinders the ability to discern global trends and areas of scientific consensus. Moreover, there is a lack of integration of disparate regional studies, preventing the identification of universal patterns and the adaptation of models to specific urban contexts. Consequently, the scientific community possibly lacks robust guidance for model selection and prioritization of key variables, which compromises the effectiveness of mitigation and adaptation strategies in addressing the challenges of air pollution over cities in a changing climate.

The main objective of this paper is to conduct a review to analyze the global trends in the use of air pollution modeling under the influence of CV over urban areas. The most commonly used types of models are identified based on their frequency of citations in scientific databases. The climate parameters and air pollutants considered by each of these types of models are also studied. In addition, the temporal trend of the variables considered is explored. Specifically, this study seeks to address the following fundamental questions: What are the air quality modeling approaches that incorporate CV and have received the most attention in specialized scientific literature? What climate variables and air pollutants are consistently considered in these models within the urban context? And how has the evolution in the selection and treatment of these variables manifested over time in published research? The study’s focus on reputable sources of information ensures a representative and up-to-date view of the state of the art in this field of research.

2. Materials and Methods

2.1. Scientific Databases

This literature review was global and covered a period of twelve years (1 January 2013–31 December 2024). This time period was selected in order to obtain an adequate sample size for each of the variables considered in this study. The following five scientific databases were used for information search and collection: Scopus (https://www.scopus.com/, accessed on 10 January 2025, Science Direct (https://www.sciencedirect.com/, accessed on 10 January 2025), Springer Link (https://link.springer.com/, accessed on 10 January 2025), Web of Science (https://clarivate.com, accessed on 10 January 2025), and Google Scholar (https://scholar.google.com/, accessed on 10 January 2025). These databases were initially chosen for their ability to provide a comprehensive and accurate view of the current state of research according to the objective of this study (17,866 documents detected). The Scopus, Science Direct, Springer Link, Web of Science (WoS), and Google Scholar databases were associated with 16.6%, 3.49%, 3.76%, 2.27%, and 73.9% of the total number of documents detected, respectively (

Table 1. Order of importance of the types of models detected for modeling air pollution over cities under the influence of CV.

Stage	Keywords	Scopus		Science Direct		Springer Link		WoS		Google Scholar		Average QI	Average Quartile	Quartile Variation
		DD	QI	DD	QI	DD	QI	DD	QI	DD	QI
Stage 1: General search	Model, urban, air pollution, climate variability	2965	1	624	1	672	1	405	1	13,200
Stage 2: Types of models	Regional climate model (RCM)	310	0.401	51	0.295	30	0.357	134	0.435	970	0.389	0.375	Q3	Q3
	Statistical model (SM)	163	0.211	58	0.335	29	0.345	60	0.195	756	0.303	0.278	Q3	Q3–Q4
	Chemical transport model (CTM)	212	0.274	35	0.202	14	0.167	55	0.179	572	0.229	0.210	Q4	Q3–Q4
	Machine learning model (MLM)	72	0.093	25	0.145	8	0.095	26	0.084	161	0.065	0.096	Q4	Q4
	Atmospheric dispersion model (ADM)	17	0.022	4	0.023	3	0.036	33	0.107	34	0.014	0.040	Q4	Q4
	Total	774		173		84		308		2493

Note: DD = detected documents; QI = Q index.

). These databases were also selected because of their frequent use in literature review articles related to the objective of this study [18,32]. The Scopus database was also used to detect the most frequently used types of models, climate parameters, and air pollutants over cities. Moreover, this database was used to study the temporal trend of the variables considered by means of the “keyword” information analysis tool [33].

2.2. Review Methodology

The literature review methodology followed the guidelines established by Rodríguez-Santamaria et al. [34] and Zafra et al. [35]. The review methodology was structured into five stages, in which key variables related to the study topic were detected. In the first stage, a general search for air pollution models over cities in the context of CV was conducted. This initial stage aimed to obtain the universe of information in each selected database. In the second stage, the types of models were detected and analyzed. In the third phase, the main models associated with the categories of model types were detected and studied. In the fourth phase, the main climate parameters and air pollutants associated with the detected models were analyzed. Lastly, in the fifth stage, the temporal trends of model types, climate parameters, and air pollutants were detected and studied. In this phase, the behavior of the variables was also analyzed in relation to per capita income according to the World Bank’s classification of countries [36]. This analysis was performed on the basis of the citation frequency of the variables considered in the Scopus database (Figure S1).

In this study, a citation frequency index (Q) was used [34,35], which considered the total content of the documents included in each database. To obtain a comprehensive view of each study variable, the citation frequency indices obtained in each database were averaged. Moreover, the range of variation of this Q index was determined to validate the results obtained for each of the databases considered. For the Q index, the following order of importance was established using quartiles: Q4 = 0.00–0.25, Q3 = 0.25–0.50, Q2 = 0.50–0.75, and Q1 = 0.75–1.00. The documents classified in the first quartile (Q1) represented the variables with the highest citation frequency. We assumed that these variables were the most studied and probably the most important in the context of this study. Although this assumption may not be entirely true, it was used as a research hypothesis. Thus, the variables included in the fourth quartile (Q4) corresponded to those with the lowest frequency of citations and, therefore, were assumed to be the least relevant in the context of this study. For example, during the second stage of the review, 51 documents related to regional climate models were detected in the ScienceDirect database, compared to the 173 documents detected in Stage 2 for all model types (Table 1). This resulted in a citation frequency index of 51/173 = 0.295, which classified this variable in the third quartile (Q3) of importance. This approach using the Q index probably allowed for a more structured classification of the variables detected in the context of this study.

The adoption of this literature review methodology across the selected scientific databases constituted a fundamental support of this review article. By employing a systematic five-stage approach—grounded in the identification and analysis of key variables through a citation frequency index (Q)—a comprehensive and representative exploration of global trends in air pollution modeling under the influence of CV over urban environments was ensured. This method enabled the precise identification of the most relevant and frequently studied models, climate parameters, and air pollutants, thereby providing an objective and up-to-date overview of the state of the art. The quantification of variable importance through Q index quartiles facilitated a structured classification and the delineation of research areas with the greatest relevance and impact. This, in turn, provided a robust foundation for the conclusions drawn and for outlining future research directions in air pollution modeling over cities under the influence of CV.

In Stage 1 (general research), the following keyword combinations were used for the literature review: “model”, “urban”, “air pollution”, and “climate variability”. This search detected 17,866 documents across all considered scientific databases. The Google Scholar database had the highest number of detected documents, followed by the Scopus database (Table 1). In Stage 2, the possible types of models used in the context of this study were detected and analyzed. This detection of model types was based on the results of the scientific databases considered and on the classifications reported by Balogun et al. [18], Augusto et al. [32], Seinfeld and Pandis [37], Tahir Bahadur et al. [38], McNider and Pour-Biazar [39], and Menut et al. [40]. In Stage 3, we proceeded to detect and study the possible models included in each of the categories considered for the types of models (Stage 2). This identification of possible models was based on the results of the scientific databases considered and on what was reported by Balogun et al. [18], Augusto et al. [32], Seinfeld and Pandis [37], Tahir Bahadur et al. [38], McNider and Pour-Biazar [39], and Baklanov and Zhang [41]. In Stage 4, the main climate parameters and air pollutants were detected through their citation frequency in the databases considered. For the detection of these variables, the “keywords” tool [33] available in the Scopus database was used.

Additionally, inclusion and exclusion criteria were used to validate the relevance and quality of the documents selected for this study. The criteria for the inclusion of information were as follows: (1) documents related to air pollution over cities and CV modeling, using regional climate models (RCMs), statistical models (SMs), chemical transport models (CTMs), machine learning models (MLMs), and atmospheric dispersion models (ADMs); (2) documents associated with climate parameters such as temperature, wind, precipitation, humidity, and solar radiation, and documents also related to air pollution simulation over cities under the CV context; and (3) documents related to urban air pollutants such as PM, O₃, NOx, VOCs, and SO₂, and documents that were also associated with simulation of air pollution over cities under the CV context. Instead, the information exclusion criteria made it possible to discard documents that did not meet the objectives of this study. The information exclusion criteria considered were as follows: (1) documents devoted exclusively to public health issues, i.e., focused on the effects of air pollution on human health; (2) documents not related to air pollution modeling over cities; (3) documents of low scientific relevance, i.e., not published by indexed scientific journals or that did not meet rigorous evaluation standards prior to publication. Based on these criteria for inclusion and exclusion of information, 155 documents were selected for this study.

In Stage 5, an analysis of temporal trends was carried out in relation to the frequency of citation of model types, climate parameters, and air pollutants. This temporal analysis considered exclusively the information reported by the Scopus database. This scientific database was used because of the large number of filters available to organize the information detected and because of the greater number of documents detected in relation to the Web of Science, Springer Link, and ScienceDirect databases. The temporal analysis considered the time interval between 1 January 2013 and 31 December 2024. This temporal analysis was performed with respect to the total number of documents detected annually by the Scopus database according to the variables considered in this study. Moreover, a comparative analysis of the types of models used, the climate parameters considered, and the air pollutants studied was carried out, stratifying the data according to the World Bank’s per capita income classification [36], in order to examine methodological differences and research priorities between countries at different levels of economic development. In this methodological phase, 2965 documents detected by the Scopus database were considered.

2.3. Statistical Analysis

Different statistical techniques were employed to analyze the information on the variables considered in this literature review. Initially, descriptive statistical tests (mean, median, and standard deviation) were used to study the patterns of the data series. Percentages and ratios [37] were used to study the distribution of categories and the relationships between them (methodology Stages 2–5). Moreover, a temporal trend analysis was conducted using percentages [42], which allowed for the visualization of the variation of the variables over time. A hierarchical cluster analysis [43] was performed using Ward’s linkage method and the Euclidean distance, without scaling. After executing the analysis, tables and a dendrogram illustrating the hierarchical structure of the clusters were generated (methodology Stage 5). The interpretation of the results focused on identifying the optimal number of clusters by examining the dendrogram and agglomeration tables [43]. Cluster analysis and its dendrogram were used to investigate temporal associations (2013–2024) between model types and key climate variables, as well as between model types and the most commonly used air pollutants in air quality modeling studies over cities. All statistical analyses were performed using SPSS V.21 software with a 95% confidence level.

3. Results and Discussion

3.1. Types of Models Detected

3.1.1. Temporal Aspects

The results showed different types of models related to modeling air pollution over cities under the influence of CV (Table 1). In order of importance, the types of models identified were as follows: RCM (Q3 = 0.375) > SM (Q3 = 0.278) > CTM (Q4 = 0.210) > MLM (Q4 = 0.096) > ADM (Q4 = 0.040). The results showed that RCMs provided fine spatial resolution (>1 km), which allowed for a more detailed representation of the geographic and topographic features of an urban region. This higher spatial resolution was essential for capturing local-scale climate phenomena, such as sea breezes, thermal convection, and extreme events [44], which significantly influenced the dispersion of air pollutants in cities. The temporal resolution (>1 h) of these models made it possible to analyze variations in climate conditions, which is essential for understanding air pollution patterns in an urban environment. RCMs could incorporate high-resolution data from surface observations, climate radars, and mesoscale numerical modeling systems [45]. This nesting capability improved the representation of local physical processes and increased the accuracy of simulations. The findings also suggested that RCMs were considered to represent a wide range of physical processes relevant to air pollution over cities, such as atmospheric chemistry, pollutant deposition, and the interaction between the atmosphere and the land surface [46]. This flexibility allowed for the exploration of the influence of different factors on air pollution, such as emissions from mobile and stationary sources, topography, vegetation cover, and atmospheric transport processes. RCMs were valuable tools for assessing the impacts of CV on air pollution over cities [40,47]. By projecting changes in climate variables, such as temperature, precipitation, and wind speed, it was possible to estimate how air pollution levels would evolve in the future and identify the most vulnerable urban regions.

Additionally, the results showed that to model air pollution over cities, regional or global models were coupled with other models. This coupling was necessary because, while these models provided detailed climatological information (wind, temperature, humidity, solar radiation, precipitation, or atmospheric boundary layer), they did not directly simulate the chemical processes of air pollutants or their emission sources. There were primarily two approaches for this integration. Offline (unidirectional) coupling involved first running the regional or global model to generate climatological fields, which then served as input for an air quality model such as a CTM, an ADM or a SM. This approach did not allow for feedback between air quality and climatology [48,49]. In contrast, online (bidirectional or two-way) coupling simultaneously ran the regional or global model and the air quality model, enabling the continuous exchange of information and the consideration of interactions, such as the effect of aerosols on solar radiation and atmospheric stability, which in turn influence the dispersion of other air pollutants [50].

Regarding SMs (Q3 = 0.278, Table 1), the results displayed that the proliferation of air pollution and climate monitoring stations generated vast databases. SMs, especially those designed for time series analysis and large datasets, proved ideal for processing and analyzing this information [51]. SMs also allowed for the identification of patterns and trends in the data [23,52], such as the seasonality of certain urban air pollutants, the influence of extreme weather events on air quality (heatwaves or heavy rainfall), and the relationship between climate variables (temperature, humidity, and wind) and air pollutant concentrations. By establishing quantitative relationships between variables, SMs enabled the quantification of the impact of CV on air pollution over cities. This was relevant for assessing the magnitude of the effects and for developing predictive models. Moreover, the findings revealed that SMs were relatively easy to implement, even using common statistical software. This likely facilitated their adoption by a broad community of researchers. SMs also offered a wide range of complexity, from simple linear models to more sophisticated nonlinear models [53,54,55]. This allowed the analysis to be tailored to the nature of the data and specific research questions. Despite their advantages, SMs also presented some limitations [56,57,58], such as the difficulty in capturing complex nonlinear relationships or incorporating underlying physical processes. This led to a growing interest in the development of hybrid models that combined statistical and physical approaches.

The results showed that the CTMs (Q4 = 0.210, Table 1) were capable of simulating the processes of advection, diffusion, deposition, and chemical reactions experienced by air pollutants. This detailed representation allowed for the evaluation of how climate conditions, such as wind speed and direction, temperature, and humidity, influenced the dispersion and transformation of air pollutants over cities [59,60]. These models also included detailed chemical schemes that described the reactions between different atmospheric compounds, such as NOx, VOCs, and O₃ [22,61]. This was fundamental for understanding the formation of secondary pollutants and assessing the impact of precursor emissions on air quality. The findings also showed that CTMs could be used to simulate future emission scenarios and climate conditions, allowing for the evaluation of the potential impacts of mitigation and adaptation measures on air quality [62,63]. This was fundamental to assessing the vulnerability of cities to extreme climate events and to designing mitigation and adaptation strategies. Thus, these models were useful for analyzing extreme pollution events, such as episodes of high O₃ or PM concentrations, and assessing the factors contributing to their occurrence over urban areas. This type of modeling also allowed detailed analysis of source-receptor relationships [64,65], identifying pollution sources and quantifying their contribution to air quality in different urban regions. This was relevant to designing effective emission control policies.

The relationship between air pollution over cities and CV was highly complex, influenced by a multitude of climate factors (temperature, humidity, wind, and solar radiation), geographical variables, and anthropogenic variables (emissions from mobile and stationary sources). MLMs (Q4 = 0.096, Table 1), particularly Artificial Neural Networks, proved efficient in capturing these nonlinear and high-dimensional interactions [18]. The results indicated that the development of air quality monitoring networks and the increasing availability of high-resolution climate data enabled the training of MLMs with increasingly large and detailed datasets [25,66]. This abundance of information was fundamental in improving the accuracy and robustness of the models. Another relevant factor of these models was their continuous learning capability [67]. MLMs could be continuously updated and refined as new data became available. This feature was particularly valuable in the context of an ever-evolving climate system and constantly changing pollution control policies. Although most MLMs were able to benefit from continuous updating, some presented significant limitations due to their structural rigidity, computational costs, or the static nature of their rules [25,67]. Moreover, MLMs were capable of identifying hidden patterns and relationships in the data that might have been overlooked by traditional statistical analyses [25,68]. This allowed the discovery of new mechanisms explaining air pollution variability and improved our understanding of atmospheric processes. Indeed, MLMs also proved to be powerful tools for predicting short- and medium-term pollutant concentrations [69,70], which was crucial for issuing early warnings and making public health decisions. Lastly, MLMs, together with artificial intelligence (AI), played a crucial role in the study of air pollution over cities under the influence of CV [54,67]. These models integrated climate and air quality data, making it possible to predict pollution episodes with high accuracy. The synergy between AI and MLMs probably transformed pollution modeling, opening new frontiers in research and decision making.

As for the ADMs (Q4 = 0.040, Table 1), the results showed that these models accurately simulated the physical processes governing the movement of air pollutants, such as advection (transport by wind) and diffusion (molecular and turbulent dispersion) [71]. This capability was essential for evaluating how variations in climate conditions, induced by CV, affected the spatial and temporal distribution of air pollutants over cities. ADMs also required detailed climatological input data, such as wind speed and direction, temperature, humidity, and atmospheric stability [22,72]. This allowed for realistic simulations of the influence of different climate conditions on pollutant dispersion, from sea breezes to extreme events such as heatwaves or storms. The findings also indicated that coupling ADMs with RCMs made it possible to assess the potential impacts of CV on air pollution over cities [73,74]. By simulating different emission scenarios and future climate conditions, it was possible to identify the most vulnerable regions and the most effective mitigation measures. ADMs were valuable tools for analyzing extreme pollution events, such as episodes of high O₃ or PM concentrations [65,75]. These events were typically associated with particular climate conditions, and the models allowed for the identification of contributing factors and the development of strategies to mitigate their effects. Lastly, there were reports that ADMs were widely used to evaluate the effectiveness of various pollution control measures [22,76], such as reducing emissions from stationary and mobile sources, changing fuels used, and implementing pollution control technologies.

The findings displayed the following order of importance in the total document growth for the types of models detected (Figure 1): RCM (39.7%) > CTM (27.6%) > SM (21.2%) > MLM (9.44%) > ADM (2.17%). The results showed that RCMs were the most used compared to other types of models in air pollution modeling studies under the influence of CV. RCMs offered much higher spatial resolution (>1 km) [40,44]. This allowed for a detailed representation of local geographic and topographic features, such as mountains, coasts, and land properties, which significantly influenced the dispersion patterns of air pollutants. These models were also capable of capturing small-scale climatological processes [44,77], such as urban heat islands, sea breezes, and extreme events, which had a direct impact on air pollution. The ability to simulate these local phenomena was fundamental to understanding how CV affected the dispersion and concentration of air pollutants over cities. RCMs could also incorporate high-resolution data from surface observations, climate radars, and mesoscale numerical modeling systems [78,79]. This integration improved the accuracy of simulations and allowed for a more realistic representation of local physical processes.

When conducting a comparative analysis of the spatiotemporal resolution among the three most frequently cited types of models (Figure 1), the results suggested that RCMs offered high spatial (>1 km) and temporal resolution (>1 h), suitable for detailed urban studies of CV and its regional impact. CTMs provided variable resolution (1 to several degrees), appropriate for studies of air pollutant transport and chemistry. SMs, although less detailed spatially, were effective in identifying temporal patterns and trends. Regarding the complexity and accuracy of these three types of models, the findings indicated that CTMs were the most complex and precise in simulating chemical and transport processes of air pollutants over cities. RCMs were accurate in simulating CV, while SMs offered greater simplicity and speed in analyzing historical data and predicting trends. When comparing their applicability, the results suggested that RCMs were more suitable for long-term urban studies and future climate change scenarios. CTMs were ideal for studies of air pollution episodes and evaluation of control policies. SMs were useful for sensitivity analyses and short-term predictions. It is important to highlight that in this study it was observed that MLMs started to be used in modeling studies of air pollution over cities under the influence of CV as of 2018 (Figure 1).

3.1.2. Monetary Aspects

The results showed that the most used model type in high-income countries was CTM and in low-income countries was SM (Figure 2). In the former, CTMs probably predominated, thanks to their ability to simulate detailed physicochemical processes, crucial for evaluating emission control policies and predicting air quality [61]. The availability of computational resources and trained personnel probably facilitated their implementation. In contrast, low-income countries may have opted for statistical models due to data and resource limitations [80]. The simplicity and applicability of models such as linear regression probably made them viable, allowing estimation of air quality and its relationship to climate. The choice of models, therefore, reflected the technical and financial capabilities of each country. The results suggested that CTMs, with their complexity and accuracy, were the predominant choice of developed countries, while statistical models, with their simplicity and low cost, were adapted to the constraints of developing countries. The findings also showed the following sequence in the use of model types in high-income countries: CTM (63.7%) > RCM (59.7%) > SM (53.8%) > MLM (33.8%) − ADM (33.8%). In low-income countries the sequence was as follows: SM (5.69%) > RCM (4.52%) > MLM (4.23%) − ADM (4.23%) > CTM (3.28%). The use and application of any type of model to simulate air pollution over cities under the influence of CV is a privilege of upper-middle- to high-income countries. Conversely, low income acts as an impediment to atmospheric modeling over cities.

3.2. Detected Models

The results showed that the most commonly used RCMs were as follows: WRF—Weather Research and Forecasting Model (Q1 = 0.797) > RegCM—Regional Climate Model (Q4 = 0.137) > COSMO-CLM—Consortium for Small-Scale Modeling-Climate Limited-Area Modeling (Q4 = 0.036) > HIRLAM—High-Resolution Limited Area Model (Q4 = 0.030,

Table 2. Order of importance of the models detected for modeling air pollution over cities under the influence of CV.

Stage	Keywords	Scopus		Science Direct		Springer Link		WoS		Google Scholar		Average QI	Average Quartile	Quartile Variation
		DD	QI	DD	QI	DD	Index	DD	DD	QI	DD
Stage 3: Models	Regional climate models (RCMs)
	WRF	390	0.874	52	0.486	45	0.882	29	0.906	1150	0.837	0.797	Q1	Q1–Q3
	RegCM	18	0.040	51	0.477	4	0.078	1	0.031	82	0.060	0.137	Q4	Q3–Q4
	COSMO-CLM	17	0.038	3	0.028	1	0.020	1	0.031	85	0.062	0.036	Q4	Q4
	HIRLAM	21	0.047	1	0.009	1	0.020	1	0.031	57	0.041	0.030	Q4	Q4
	Total	446		107		51		32		1374
	Statistical models (SMs)
	GAM	47	0.296	19	0.317	5	0.313	13	0.565	215	0.319	0.362	Q3	Q3–Q4
	ARIMA	59	0.371	18	0.300	7	0.438	1	0.043	269	0.399	0.310	Q3	Q3–Q4
	GLM	46	0.289	19	0.317	4	0.250	6	0.261	165	0.245	0.272	Q3	Q4
	BHM	7	0.044	4	0.067	0	0.000	3	0.130	25	0.037	0.056	Q4	Q4
	Total	159		60		16		23		674
	Chemical transport models (CTMs)
	WRF-Chem	184	0.307	23	0.299	13	0.302	19	0.373	425	0.277	0.311	Q3	Q3–Q4
	CMAQ	150	0.250	17	0.221	13	0.302	8	0.157	372	0.243	0.235	Q4	Q3–Q4
	GEOS-Chem	97	0.162	21	0.273	9	0.209	12	0.235	362	0.236	0.223	Q4	Q3–Q4
	MOZART	89	0.148	4	0.052	2	0.047	5	0.098	166	0.108	0.091	Q4	Q4
	CAMx	32	0.053	7	0.091	2	0.047	2	0.039	94	0.061	0.058	Q4	Q4
	CHIMERE	30	0.050	3	0.039	3	0.070	1	0.020	77	0.050	0.046	Q4	Q4
	LOTOS-EUROS	18	0.030	2	0.026	1	0.023	4	0.078	37	0.024	0.036	Q4	Q4
	Total	600		77		43		51		1533
	Machine learning models (MLMs)
	ANNs	101	0.616	38	1.652	9	0.125	4	0.108	402	0.754	0.651	Q2	Q3–Q4
	RFs	80	0.488	33	1.435	5	0.069	14	0.378	272	0.510	0.576	Q2	Q2–Q4
	SVMs	35	0.213	17	0.739	8	0.111	2	0.054	168	0.315	0.287	Q3	Q3–Q4
	KNNs	14	0.085	6	0.261	1	0.014	0	0.000	56	0.105	0.093	Q4	Q4
	GBMs	4	0.024	4	0.174	0	0.000	5	0.135	13	0.024	0.072	Q4	Q4
	Total	234		98		23		25		911
	Atmospheric dispersion models (ADMs)
	HYSPLIT	146	0.890	19	0.826	18	0.250	35	0.946	423	0.794	0.741	Q2	Q1
	AERMOD	8	0.049	2	0.087	18	0.250	0	0.000	45	0.084	0.094	Q4	Q4
	CALPUFF	8	0.049	1	0.043	18	0.250	2	0.054	29	0.054	0.090	Q4	Q4
	ADMS	2	0.012	1	0.043	18	0.250	0	0.000	36	0.068	0.075	Q4	Q4
	Total	164		23		72		37		533

Note: DD = detected documents; QI = Q index.

). There were specific reasons for the use of each model. WRF was frequently used due to its ability to simulate a wide variety of climate phenomena and its highly active user community [81]. For example, this model was used in various configurations to study air quality in different cities around the world, adapting to the specific characteristics of each urban area. RegCM was characterized by its ability to simulate small-scale convective processes and its flexibility in model physics configuration [44]. This model stood out for its ability to simulate urban-scale climate phenomena, such as the heat island effect. COSMO-CLM was notable for its high spatial resolution and its ability to simulate complex physical processes, such as the interaction between the atmosphere and the land surface [82]. For instance, COSMO-CLM incorporated advanced urban parameterization schemes to improve the representation of urban processes in climate simulations. HIRLAM was characterized by its ability to simulate mesoscale processes and its application in short-term numerical weather prediction studies [80,83]. This model, for example, was validated in multiple case studies, demonstrating its accuracy in simulating climate and air quality conditions over urban areas. All of the above models were useful because they allowed the identification of patterns in the spatiotemporal variability of air pollution in relation to climate conditions, facilitated the assessment of potential CV impacts on air quality, and contributed to designing mitigation and adaptation strategies to reduce the effects of air pollution on human health and the environment. Despite their advantages, these models also had limitations, such as the difficulty in representing some very small-scale physical processes and the uncertainty associated with the parameterizations used. In recent years, there was a growing interest in the development of coupled terrestrial system models [17,84], which integrated RCMs with air quality models and other components of the terrestrial system, allowing for a more realistic representation of the interactions between the different processes involved.

The findings indicated that the most commonly used SMs were as follows: GAM—Generalized Additive Model (Q3 = 0.362) > ARIMA—Autoregressive Integrated Moving Average Model (Q3 = 0.310) > GLM—Generalized Linear Model (Q3 = 0.272) > BHM—Bayesian Hierarchical Model (Q4 = 0.056, Table 2). GAMs successfully modeled nonlinear relationships between variables [51,85], which was essential for capturing the complexity of atmospheric processes. These models also provided clear interpretations of the influence of each climate variable on air pollutant concentrations over cities. ARIMA models were ideal for analyzing time series of air pollution data [86,87,88], capturing temporal dependence and seasonal patterns. These models allowed for short-term air quality forecasts. GLMs could model a wide variety of error distributions [18,85], making them suitable for different types of response variables (e.g., counts and proportions). GLMs allowed for the inclusion of categorical variables in the model, such as emission source type or season. BHMs were useful for simulating complex hierarchical structures [89,90], such as the spatial and temporal variability of air pollution over cities. These models incorporated uncertainty in model parameters, providing more realistic confidence intervals.

The results showed that the most commonly used CTMs were as follows: WRF-Chem—Weather Research and Forecasting Model Coupled with Chemistry (Q3 = 0.311) > CMAQ—Community Multiscale Air Quality (Q4 = 0.235) > GEOS-Chem—Goddard Earth Observing System Global 3-D Model of Atmospheric Chemistry (Q4 = 0.223) > MOZART—Model for Ozone and Related Chemical Tracers (Q4 = 0.091) > CAMx—Comprehensive Air Quality Model with Extensions (Q4 = 0.058) > CHIMERE—Multi-Scale Chemistry-Transport Model (Q4 = 0.046) > LOTOS-EUROS—Open-Source Chemical Transport Model (Q4 = 0.036, Table 2). These models were fundamental tools for understanding the complex interaction between CV and air quality over cities. WRF-Chem combined the ability to simulate high-resolution climate processes with a detailed atmospheric chemistry scheme [91,92]. This model allowed for the simulation of the interaction between climatology and atmospheric chemistry, which was essential for understanding the processes of pollutant formation and transport under variable atmospheric conditions. CMAQ was one of the most widely used models globally [93,94,95]. It was notable for its ability to simulate a wide range of atmospheric chemical processes at high spatiotemporal resolution. This model allowed for the simulation of the formation and transport of O₃, PM, aerosols, and other pollutants. Moreover, it could be coupled with high-resolution climate models for a more realistic representation of atmospheric conditions. GEOS-Chem was a global chemical transport model used to study atmospheric chemistry and air quality at various spatial scales [62,96,97]. This model integrated climatological and emissions data, allowing for a detailed simulation of CV and its impact on air quality. Its ability to perform historical and future simulations made it a valuable tool for air quality studies.

Additionally, MOZART was a global atmospheric chemistry model that focused on simulating the global distribution of O₃ [98,99]. This model allowed for the simulation of O₃ formation and destruction processes in the troposphere and the assessment of the impact of O₃ precursor emissions. CAMx was a flexible and modular model that allowed for the configuration of different emission scenarios and the evaluation of various pollution control policies [65,100,101]. This model could simulate a wide range of chemical and physical processes, including dry and wet deposition of air pollutants. It also allowed for the assimilation of observational data to improve simulation accuracy. CHIMERE was a regional air quality model notable for its ability to simulate complex chemical processes and its application in pollution studies over cities [102,103]. This model allowed for the simulation of O₃, PM, and other pollutants in complex urban environments. LOTOS-EUROS was a regional air quality model designed to simulate atmospheric pollution at regional and subregional scales [22,104,105]. This model covered a wide range of inert and chemically active constituents, including O₃, NO₂, and fine particles. Its ability to assimilate satellite and ground-based measurements improved the accuracy of simulations.

The results indicated that the most commonly used MLMs were as follows: ANNs—Artificial Neural Networks (Q2 = 0.651) > RFs—Random Forests (Q2 = 0.576) > SVMs—Support Vector Machines (Q3 = 0.287) > KNNs—K-Nearest Neighbors (Q4 = 0.093) > GBMs—Gradient Boosting Machines (Q4 = 0.072, Table 2). ANNs stood out for their ability to learn and model complex and nonlinear relationships between climate and air pollution variables [18,106,107]. ANNs were also useful in identifying hidden patterns in large datasets and accurately predicting pollution levels under different climate conditions. RFs were used due to their robustness and ability to handle large volumes of data with multiple variables [69,108]. This ensemble learning model, based on the construction of multiple decision trees, also captured variability and complex interactions between climate and air pollution variables, providing accurate and reliable predictions. SVMs were preferred for their effectiveness in high-dimensional data classification and regression [54,109]. SVMs used kernel functions to transform nonlinear data into a higher-dimensional space where linear relationships could be identified, resulting in accurate models for predicting air pollution based on climate variables. GBMs were valued for their ability to iteratively improve model performance by combining multiple weak models [25,110]. GBMs, including variants such as XGBoost and LightGBM, proved highly effective in capturing nonlinear relationships and improving the accuracy of air pollution predictions under different climate scenarios. KNNs were used for their simplicity and effectiveness in classification and regression based on data proximity [111,112]. KNN was useful in identifying local patterns and predicting air pollution levels based on observed climate conditions in similar neighborhoods.

Additionally, the use of AI in MLMs was reported. Though, due to its relatively recent application, this type of model possibly did not have a high citation frequency (Q) that would allow its selection in the scientific databases considered in this study. For example, deep learning, with its Convolutional Neural Networks (CNNs) [25], revolutionized spatial data analysis by extracting information from satellites and pollutant maps, which improved the understanding of pollution dispersion. Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) [54] modeled temporal dependency, enhancing short- and long-term predictions. Generative Adversarial Networks (GANs) [113] generated high-resolution synthetic data, useful in areas with limited data and for simulating future scenarios. AI facilitated the integration of heterogeneous data, offering a comprehensive view of the problem. These advances improved prediction accuracy, identified influential climate factors, developed robust models, and facilitated informed decision making.

The findings showed that the most commonly used ADMs were as follows: HYSPLIT—Hybrid Single-Particle Lagrangian Integrated Trajectory Model (Q1 = 0.741) > AERMOD—Steady-State Gaussian Plume Model (Q4 = 0.094) > CALPUFF—Non-Steady-State Meteorological and Air Quality Modeling System (Q4 = 0.090) > ADMS—Atmospheric Dispersion Modeling System (Q4 = 0.075, Table 2). HYSPLIT was a Lagrangian model that tracked the trajectory of individual particles in the atmosphere [114,115,116]. This model was highly flexible and could simulate a wide variety of scenarios, from short-term pollutant dispersion to long-distance aerosol transport. HYSPLIT also allowed for backward air trajectory tracking, which was useful for identifying pollution sources during poor air quality events. This model was also used to assess risks associated with the accidental release of hazardous substances. AERMOD was frequently used for its ability to model pollutant dispersion over urban areas, considering different types of terrain and emission sources [73,117,118]. This model incorporated climate data and terrain characteristics to provide accurate estimates of air pollutant concentrations. Its use in air quality studies over cities was due to its accuracy in simulating pollutant dispersion under variable atmospheric conditions. CALPUFF was valued for its ability to model pollutant dispersion in complex terrains and under changing climate conditions [114,117,119]. This model could simulate a wide variety of physical and chemical processes, including dry and wet deposition, aerosol chemistry, and O₃ formation. CALPUFF allowed for the simulation of chemical transformation and deposition of pollutants, making it suitable for air quality studies over urban environments with complex topography. ADMS was commonly used due to its advanced approach to modeling pollutant dispersion in the atmospheric boundary layer [71,120,121,122]. This model incorporated algorithms that considered the influence of buildings, complex terrain, and chemical reactions in the atmosphere. ADMS was effective in simulating pollutant dispersion in densely populated urban areas, providing detailed estimates of pollutant concentrations at different distances from emission sources.

The results demonstrated that several regional or global models (e.g., WRF: 1–4 km, RegCM: 10–50 km, COSMO-CLM: 1–50 km, or HIRLAM: 1–50 km) had been coupled with air quality models. In offline coupling, the regional model was executed first to generate climate fields, which were subsequently used as input for air quality models. These included CTMs (e.g., CMAQ or CHIMERE) or ADMs (e.g., AERMOD or CALPUFF), which simulated the emission, transport, chemical transformation, and deposition of air pollutants at the urban scale [123,124,125]. Conversely, online coupling integrated both models into a unified simulation framework, enabling continuous information exchange and accounting for dynamic interactions between climate and air quality. Representative examples included WRF-Chem and Enviro-HIRLAM, which incorporated atmospheric physics and chemistry within a single computational architecture [125,126,127,128]. The findings indicated that both coupling strategies were effective. Offline coupling appeared more prevalent due to its lower computational demands and the maturity of existing models, whereas online coupling offered a more coherent representation of climate–air quality interactions, particularly relevant for assessing how CV influenced pollutant formation and transport in urban environments.

3.3. Climate Parameters: Temporal Aspects

The results showed that the most commonly used climate parameters globally in modeling air pollution over cities under the influence of the CV were as follows: temperature (Q3 = 0.335) > wind (Q4 = 0.223) > precipitation (Q4 = 0.208) > humidity (Q4 = 0.146) > solar radiation (Q4 = 0.087,

Table 3. Order of importance for the climate parameters and pollutants detected in the context of air pollution modeling over cities and CV.

Stage	Keywords	Scopus		Science Direct		Springer Link		WoS		Google Scholar		Average QI	Average Quartile	Quartile Variation
		DD	QI	DD	QI	DD	QI	DD	DD	QI	DD
Stage 4: Climate parameters	Temperature	1595	0.380	563	0.294	586	0.296	126	0.406	11,500	0.300	0.335	Q3	Q3
	Wind	852	0.203	416	0.218	423	0.214	83	0.268	8110	0.212	0.223	Q4	Q3–Q4
	Precipitation	983	0.234	409	0.214	477	0.241	35	0.113	9170	0.239	0.208	Q4	Q4
	Humidity	528	0.126	309	0.162	295	0.149	44	0.142	5860	0.153	0.146	Q4	Q4
	Solar radiation	242	0.058	215	0.112	197	0.100	22	0.071	3650	0.095	0.087	Q4	Q4
	Total	4200		1912		1978		310		38,290
Stage 4: Air pollutants	Ozone	1029	0.390	236	0.316	268	0.391	124	0.369	5340	0.383	0.370	Q3	Q3
	Particulate matter	833	0.316	216	0.290	175	0.255	147	0.438	3950	0.283	0.316	Q3	Q3
	Volatile organic compounds	274	0.104	104	0.139	85	0.124	24	0.071	1630	0.117	0.111	Q4	Q4
	Nitrogen oxides	260	0.099	94	0.126	86	0.125	29	0.086	1600	0.115	0.110	Q4	Q4
	Sulfur dioxide	240	0.091	96	0.129	72	0.105	12	0.036	1430	0.103	0.093	Q4	Q4
	Total	2636		746		686		336		13,950

Note: DD = detected documents; QI = Q index.

). These climate parameters were likely used due to their critical roles in atmospheric processes and their direct impact on the dispersion and concentration of air pollutants. Temperature affected the rate of chemical reactions in the atmosphere [129], including the formation of O₃ and other secondary pollutants. Temperature also influenced atmospheric stability and the height of the mixing layer [130], determining the vertical dispersion of air pollutants over cities. Simulation models used wind data to predict how pollutants were transported and diluted in urban air [131,132]. Wind also affected the ventilation of urban areas, influencing air pollutant concentrations. Simulation models also considered precipitation to evaluate how rain events reduced air pollutant concentrations (washout or wet deposition). Moreover, high humidity levels favored the formation of secondary particles and the absorption of gaseous pollutants into water particles [16,19]. Humidity also influenced atmospheric stability and pollutant dispersion. Simulation models used solar radiation data to predict photochemical reaction rates and the production of secondary pollutants [133]. Variability in solar radiation also affected temperature and atmospheric stability.

As mentioned, the results showed that temperature was the dominant climate variable in air pollution modeling studies over cities under the influence of CV, surpassing solar radiation in use (Table 3). This predominance may also have been due to several technical and practical reasons [134]. First, temperature exhibited superior data availability, with abundant historical records and a wider geographical distribution of measurement stations. Its direct relevance to key air pollution processes, such as ozone-forming chemical reactions, atmospheric stability, and VOC emissions, consolidated it as a fundamental variable. Moreover, temperature proved to be more easily modeled compared to solar radiation because it was less susceptible to local variability. Lastly, the high sensitivity of many air pollutants to temperature changes, particularly O₃, reinforced their usefulness in these studies [23]. Thus, these technical and practical factors probably contributed to the preponderance of temperature in the investigation of the influence of CV on air pollution.

Therefore, all the aforementioned climate parameters (temperature, wind, precipitation, humidity, and solar radiation) were essential for the types of models due to their direct influence on the processes of dispersion, transport, transformation, and removal of pollutants in urban air. The results of the temporal analysis in the Scopus database indicated the following order of importance in the total growth of documents for each climate variable under study (Figure 3): temperature (38.2%) > precipitation (23.7%) > wind (19.8%) > humidity (12.5%) > solar radiation (5.83%). The findings suggested that temperature was the climate variable of greatest interest in the types of models, likely due to its direct and multifaceted influence on atmospheric and chemical processes that determined air quality over cities. Indeed, its accurate representation was important for improving the precision of air pollution simulations and predictions.

The results of the cluster analysis showed temporal associations between the types of models and the climate parameters identified. This was based on the temporal evolution in the number of documents detected by the Scopus database. Based on the dendrogram and the proximity matrix, the following three clusters were identified (Figure 4): (1) CTM, SM, ADM, and MLM; (2) RCM and solar radiation; and (3) wind and precipitation. The findings indicated that solar radiation was the main climate variable for RCMs. Solar radiation was a critical component in RCMs due to its direct influence on the energy balance of the atmosphere and the land surface [24,135]. RCMs also used detailed solar radiation data to simulate processes such as evapotranspiration, cloud formation, and aerosol dynamics. The ability of RCMs to adequately represent solar radiation allowed for a better understanding of how CV affected air pollution over cities. Although solar radiation was important for CTMs, especially in photolysis and other photochemical reactions [10,136], these models did not rely as heavily on solar radiation as RCMs. CTMs primarily used climate and emissions data to model the dispersion and transformation of air pollutants. Solar radiation was a variable that could be included in SMs [24,137], but its relevance depended on data availability and its statistical relationship with air pollution. These models were less capable of capturing the complexity of atmospheric processes compared to RCMs.

Additionally, ADMs focused on air flow dynamics and pollutant dispersion, using climate data such as wind speed and direction, temperature, and atmospheric stability [74,138]. Solar radiation played an indirect role in these models, affecting atmospheric stability and the formation of mixing layers, but it was not a central component as it was for RCMs. Solar radiation could be one of many variables used in MLMs, but its relevance depended on the model structure and available data [66,137]. These models were useful for making predictions based on historical data but did not necessarily provide a deep understanding of underlying physical processes. Therefore, solar radiation was more relevant for RCMs, probably due to its fundamental role in climate processes and its ability to influence air pollution modeling and CV. The other types of models considered used solar radiation to a lesser extent or indirectly, depending on their specific objectives and methodologies. According to the cluster analysis, in order of importance, solar radiation had the following sequence for each of the types of models identified: RCM > CTM > SM > MLM > ADM (Figure 4). Indeed, the results suggested a relationship between solar radiation (Q4 = 0.087) and temperature (Q3 = 0.335), and this relationship varied according to the type of model. RCMs heavily depended on this relationship to simulate climate processes and their impact on air pollution. CTMs and ADMs also considered this relationship, but more indirectly. SMs and MLMs used the relationship between solar radiation and temperature to improve their predictions, although they did not capture all the physical processes involved.

The results from the cluster analysis displayed that the most associated climate variables were wind and precipitation (Figure 4). The relationship between wind and precipitation in RCMs was critical for the accuracy of precipitation simulations [102,139,140]. Models with higher horizontal resolution tended to generate improved simulations in both vector wind fields and precipitation, especially in monsoon regions and areas with warm and humid advection. The ability of RCMs to simulate precipitation was closely related to their ability to simulate vector winds. In CTMs, the relationship between wind and precipitation was fundamental for the accurate prediction of air pollutant concentrations [22,102,141]. Winds influenced the horizontal transport and vertical dilution of air pollutants, while precipitation affected the wet deposition of particles and gases. SMs used empirical relationships between climate variables and air pollutant concentrations [52,142]. The relationship between wind and precipitation was modeled through transfer functions and penalized regressions to capture spatiotemporal variability. ADMs focused on pollutant dispersion in the atmosphere, where wind and precipitation played important roles [114,138,143]. Wind determined the direction and speed of pollutant transport, while precipitation influenced particle removal through wet deposition. The relationship between wind and precipitation in these models was essential for predicting the spatiotemporal distribution of air pollutants. MLMs applied advanced machine learning techniques to predict air pollution based on climate variables such as wind and precipitation [18,54,56]. The relationship between wind and precipitation was integrated into the models through derived features and preprocessing techniques, improving prediction accuracy.

3.4. Air Pollutants: Temporal Aspects

The results showed that the most commonly considered air pollutants globally by models in the context of this study were as follows: O₃ (Q3 = 0.370) > PM (Q3 = 0.316) > VOCs (Q4 = 0.111) > NOx (Q4 = 0.110) > SO₂ (Q4 = 0.093, Table 3). These air pollutants were the most studied, likely due to their impact on human health and the environment, and the relatively advanced understanding of the atmospheric processes that generated them. Ground-level O₃ was a potent oxidant and a key component of photochemical smog [18,104,144]. Its formation was closely linked to solar radiation, NOx and VOC emissions, and climate conditions. PM, especially fine particles (PM_2.5 and PM₁₀), had adverse effects on human health and could affect visibility [60,145]. Its formation was related to combustion processes, dust resuspension, and chemical reactions in the atmosphere. VOCs were precursors to O₃ and secondary aerosols [18,146]. Their emission came from a wide variety of sources, including vehicles, industries, and vegetation.

Additionally, NOx were primarily emitted by fossil fuel combustion and were precursors to O₃, nitric acid, and nitrate aerosols [145]. SO₂ was mainly emitted by the combustion of sulfur-containing fossil fuels and was a precursor to sulfuric acid and sulfate aerosols [23]. This study observed that the field of air quality over cities continues to evolve, and it is expected that in the future, more attention will be paid to other emerging pollutants (semi-volatile organic compounds or microplastics) and the synergistic effects of air pollutant mixtures. The results of the temporal analysis in the Scopus database indicated the following order of importance in the total growth of documents for each urban air pollutant under study (Figure 5): O₃ (38.5%) > PM (32.4%) > VOCs (10.6%) > NOx (9.68%) > SO₂ (8.84%). The results suggested comparatively that O₃ and PM were the air pollutants of greater concern. This preponderance was probably based on a convergence of technical factors and social relevance [16]. First, the direct impact of O₃ and PM on public health positioned them as high priority pollutants, supported by epidemiological studies that showed their relationship with respiratory and cardiovascular diseases. The availability of standardized O₃ and PM data from extensive monitoring networks also facilitated long-term research, which is crucial for understanding pollution trends and patterns. Furthermore, their secondary formation in the atmosphere, influenced by climate conditions, and the long-range transport of PM, made them key indicators of the effects of CV [60]. Air quality regulations, which established concentration limits for O₃ and PM, also encouraged their study and monitoring. The need to assess the impact of climate change on compliance with these regulations further reinforced their importance [10]. Therefore, these technical and public policy factors possibly consolidated O₃ and PM as the most relevant pollutants in the models to investigate air pollution over cities under the context of CV influence. Lastly, O₃ was formed from photochemical reactions involving NOx and VOCs under conditions of high solar radiation and temperature [61]. O₃ also acted as a strong oxidant, affecting the chemistry of other pollutants and contributing to the formation of secondary PM [132]. PM, especially PM_2.5, had adverse effects on respiratory and cardiovascular health, and its formation and dispersion were influenced by CV, including temperature and humidity [18].

The results of the cluster analysis showed temporal associations between the types of models and the air pollutants identified. This was based on the temporal evolution in the number of documents detected by the Scopus database. Based on the dendrogram and the proximity matrix, the following three clusters were identified (Figure 6): (1) NOx, SO₂, VOCs, and RCM; (2) CTM, SM, ADM, and MLM; and (3) PM and O₃. The findings showed that in RCMs, NOx, SO₂, and VOCs were relevant due to their fundamental role in the formation of O₃ and secondary PM [147], both key components of air pollution. RCMs were used to simulate the interaction between these precursor emissions and large-scale climate conditions, allowing for the assessment of how CV affected air pollutant formation and dispersion [102]. The ability of RCMs to capture complex atmospheric processes, such as photochemistry and aerosol dynamics, was essential for understanding the influence of NOx, SO₂, and VOCs on air quality. In CTMs, although NOx, SO₂, and VOCs were also important, the main focus was on the detailed simulation of atmospheric chemistry and pollutant transport [61,148]. The accuracy of air quality simulations in CTMs largely depended on the detailed representation of chemical reactions involving NOx, SO₂, and VOCs, as well as local-scale climatology.

Additionally, although NOx, SO₂, and VOCs were considered by SMs, the relevance of these pollutants depended on the availability of historical information and their ability to detect patterns and trends [54,149]. SMs were effective for short-term air quality predictions, but their ability to capture CV and its impact on atmospheric chemistry was limited compared to RCMs and CTMs. NOx, SO₂, and VOCs were relevant for ADMs due to their impact on O₃ and PM formation, but ADMs’ ability to simulate complex atmospheric chemistry was limited [72,73]. ADMs were mainly used to evaluate the spatial and temporal dispersion of air pollutants rather than their chemical transformation. Although NOx, SO₂, and VOCs were included as input variables in MLMs [54,150], the relevance of these pollutants depended on the algorithms’ ability to detect nonlinear relationships and complex patterns in the data. MLMs were effective for making accurate predictions, but their ability to simulate detailed atmospheric and chemical processes was limited compared to RCMs and CTMs.

The findings from the cluster analysis showed that the most associated air pollutants were PM and O₃ (Figure 6). RCMs simulated how climate conditions such as temperature and solar radiation influenced the formation of O₃ from precursors like NOx and VOCs, and how these processes affected secondary PM formation [17,151]. CV, including extreme events like heatwaves, had a significant impact on O₃ and PM concentrations, affecting air quality over cities. CTMs modeled the chemical reactions leading to O₃ and secondary PM formation, considering the influence of climatological factors such as temperature and humidity [141]. The interaction between O₃ and PM was observed in the formation of secondary aerosols, where O₃ acted as an oxidant, transforming gaseous precursors into fine particles. SMs used empirical relationships to capture the interaction between PM and O₃ [23,152]. The relationship between PM and O₃ was modeled through regressions and correlation analyses, allowing for the prediction of pollutant concentrations based on specific climate conditions. ADMs simulated how climate variables such as wind and precipitation affected the dispersion of O₃ and PM, and how these pollutants interacted in the atmosphere [74]. The formation of O₃ and secondary PM was considered based on the availability of precursors and local atmospheric conditions. The relationship between PM and O₃ was integrated into MLMs through derived features and preprocessing techniques, improving the accuracy of air pollution predictions.

3.5. Climate Parameters and Air Pollutants: Monetary Aspects

The results showed that the most used climate variables in high-income countries were temperature and wind, and in low-income countries, it was precipitation (Figure 7a). In high-income countries, temperature and wind were selected as the predominant variables [19]. The availability of high-resolution data, thanks to dense monitoring networks and advanced technology, facilitated their use. The relevance of these variables to air pollution over cities was possibly crucial. Temperature influenced atmospheric chemical reactions and O₃ formation, while wind determined air pollutant dispersion in densely populated areas [133]. In contrast, in low-income countries, precipitation became more important. Data limitations for temperature and wind contrasted with the greater availability of precipitation data [153]. Moreover, its impact on the removal of particulate pollutants, especially in areas with high dust concentrations, made it relevant. Precipitation also facilitated the deposition of air pollutants on the surface [10]. Thus, the choice of climate variables reflected data availability, technological capacity, and the local relevance of each variable in the context of air pollution, showing how resources and priorities influenced the studies. The findings also showed the following sequence in the use of climate variables in high-income countries: temperature (59.4%)–wind (59.4%) > precipitation (57%) > humidity (54.6%) > solar radiation (53%). In low-income countries the sequence was as follows: precipitation (2.43%) > humidity (1.70%) > temperature (1.61%) > wind (1.06%) > solar radiation (1.04%).

Additionally, the findings showed that the most studied air pollutant in high-income countries was VOCs, and in low-income countries it was NOx (Figure 7b). In high-income countries, VOCs received predominant attention [154]. The technical capacity to investigate the complex photochemistry of VOCs, their role in tropospheric ozone formation, and their impact on health drove their study. The stringency of environmental regulations, focused on controlling VOC emissions, also spurred research to develop mitigation technologies. In contrast, low-income countries focused their attention on NOx. NOx emissions from transportation and industry represented a significant urban pollution problem [155]. Measuring NOx concentrations, identifying sources, and assessing their impact on public health were priorities. Resource constraints, which made measurement and analysis of VOCs difficult, also influenced the choice of NOx as the primary air pollutant. Therefore, the choice of the most studied air pollutant reflected study priorities, technical capabilities, and specific pollution conditions in each country, evidencing how resources and priorities influenced the research. Lastly, the results showed the following sequence in the study of air pollutants in high-income countries: VOCs (67.1%) > O₃ (63.2%) > SO₂ (62.7%) > NOx (61.1%) > PM (58.3%). In low-income countries the sequence was as follows: NOx (2.95%) > SO₂ (1.47%) > PM (0.75%) > O₃ (0.69%) > VOCs (0.12%). All of the above findings were framed in the context of air pollution modeling over cities under the influence of CV.

4. Conclusions

In the context of air pollution modeling over cities influenced by CV, this study establishes a hierarchy of model utility: RCM (Q3), SM (Q3), CTM (Q4), MLM (Q4), and ADM (Q4). RCMs are critical for downscaling climate data to generate high-resolution projections, supporting localized air quality assessments and adaptation planning. SMs demonstrate strong capabilities in integrating heterogeneous datasets and identifying complex temporal patterns, making them indispensable for retrospective analyses and trend forecasting. CTMs provide mechanistic insights into pollutant transport and chemical transformation, thereby informing regulatory strategies and emission control policies. MLMs leverage large-scale datasets to enhance predictive accuracy and reveal latent structures in air quality and CV interactions, offering significant advantages in real-time forecasting and scenario evaluation. ADMs, while more specialized, are essential for quantifying pollutant dispersion from point sources, contributing to emergency response planning and regulatory compliance. Collectively, these modeling frameworks offer complementary strengths, forming a robust toolkit for understanding, predicting, and mitigating the multifaceted impacts of air pollution over urban environments under the influence of CV.

The findings suggest that the main RCMs, SMs, CTMs, MLMs, and ADMs are WRF-coupling with other models (Q1), GAM (Q3), WRF-Chem (Q3), ANNs (Q2), and HYSPLIT (Q2), respectively. WRF coupled with other city-scale air quality models is important for high-resolution climate and air pollution projections and localized forecasts. GAM excels in modeling nonlinear relationships between air pollutant concentrations and climate variables. WRF-Chem is fundamental for simulating air pollutant formation and transport. ANNs improve the accuracy of predictions and uncover hidden patterns. HYSPLIT assesses air pollutant dispersion, informing regulatory frameworks and emergency responses.

This study identifies O₃ (Q3), PM (Q₃), VOCs (Q4), NOx (Q4), and SO₂ (Q4) as the most frequently modeled air pollutants globally. RCMs are instrumental in capturing the formation of O₃ and secondary PM from precursors such as NO_x and VOCs. CTMs provide detailed simulations of atmospheric chemistry and pollutant transport, particularly for O₃ and PM. SMs effectively forecast short-term trends in NO_x, SO₂, and VOCs, while ADMs assess the spatial and temporal dispersion of O₃ and PM. MLMs enhance predictive accuracy across all pollutants by identifying complex, nonlinear patterns. Cluster analysis reveals strong temporal associations between model types and specific pollutants, underscoring the critical role of NO_x, SO₂, and VOCs in O₃ and PM formation. The complex interaction between O₃ and PM is captured through diverse modeling strategies, ranging from mechanistic simulations to empirical and dispersion-based approaches.

The results suggest that the climate parameters most commonly used worldwide by the types of models in the context of this study are the following: Temperature (Q3), wind (Q4), precipitation (Q4), humidity (Q4), and solar radiation (Q4). Cluster analysis reveals significant temporal associations between the types of models and the climate variables identified. RCMs stand out for their dependence on solar radiation, essential to simulate atmospheric processes and their impact on air pollution over cities. Solar radiation directly influences the energy balance and aerosol formation, being crucial for RCMs. In contrast, CTMs and ADMs use solar radiation more indirectly. The relationship between wind and precipitation is fundamental in RCMs for the accuracy of precipitation simulations, while in CTMs and ADMs, this relationship is vital for the prediction of air pollutant concentrations. SMs and MLMs integrate these relationships to improve their predictions, although they do not capture all underlying physical processes.

This study reveals that the selection of models, climate variables, and air pollutants directly reflects the technical and financial capabilities of each country. CTMs, with their ability to simulate detailed physicochemical processes, predominate in high-income countries, facilitating the evaluation of emission control policies and the prediction of air quality. In contrast, SMs, because of their simplicity and applicability, are preferred in low-income countries, where data and resource constraints are more pronounced. In high-income countries, temperature emerges as the dominant climate variable due to its data availability and relevance to key pollution processes, while precipitation stands out in low-income countries for its impact on PM removal. VOCs are the most studied pollutant in high-income countries, driven by their complex photochemistry and role in O₃, while NOx dominate in low-income countries, due to their significant public health impact and data availability. These decisions highlight how resources and national priorities condition air pollution research under the influence of CV.

This study provides practical insights into the use of models to analyze air pollution over cities under the influence of CV. By identifying the most commonly used models, it guides researchers in selecting effective approaches, enhancing the quality of future studies. The analysis of climate parameters and air pollutants highlights key research areas, aiding in the development of targeted environmental management and adaptation strategies. The coupling between regional atmospheric models and air quality models proved essential for simulating city-scale pollution. While offline coupling remains prevalent, online approaches offer more physically consistent representations. However, future research should prioritize intra-urban models to better resolve fine-scale processes critical for accurate air pollution assessments. Lastly, this research is also important for understanding and mitigating the effects of air pollution over cities under the influence of CV, ultimately improving public health and the urban environment.