Comprehensive Impact of Different Urban Form Indices on Land Surface Temperature and PM2.5 Pollution in Summer and Winter, Based on Urban Functional Zones: A Case Study of Taiyuan City

Abstract

Urban form plays a crucial role in regulating urban thermal environments and air pollution patterns. However, the indirect mechanisms through which urban form influences PM2.5 concentrations via land surface temperature (LST) remain poorly understood. This study investigates these pathways by analyzing representative two- and three-dimensional urban form indices (UFIs) in the central urban area of Taiyuan, China. Multiple log-linear regression and mediation analysis were applied to evaluate the combined effects of urban form on LST and PM2.5. The results indicate that UFIs significantly influence both LST and PM2.5. The frontal area index (FAI) and sky view factor (SVF) emerged as key variables, with LST playing a significant mediating role. The indirect pathways affecting PM2.5 via LST, along with the direct LST-PM2.5 correlation, exhibit pronounced seasonal differences in direction and intensity. Moreover, different urban functional zones exhibit heterogeneous responses to the same form indices, highlighting the spatial variability of these linkages. These findings underscore the importance of incorporating seasonal and spatial differences into urban design. Accordingly, this study proposes targeted urban form optimization strategies to improve air quality and thermal comfort, offering theoretical insights and practical guidance for sustainable urban planning.

1. Introduction

Since the beginning of the 21st century, accelerated global urbanization has led to a projected 2.5-fold increase in built-up areas by 2050 compared to 2000 [1]. This rapid spatial expansion has transformed natural land cover into artificial surfaces, intensifying urban climate effects such as the urban heat island (UHI) and air pollution [2,3,4]. Currently, over 55% of the global population is exposed to the combined stress of UHI and PM2.5 pollution, with more severe challenges in developing countries due to high-intensity construction [5]. As the world’s fastest urbanizing country, China’s megacities are exhibiting increasingly negative impacts on climate regulation capacity [6,7,8]. This contemporary challenge necessitates a critical re-examination of the foundational urban theories that have shaped our cities. The functionalist visions, for instance, promoted a rational order of segregated zones, a morphology whose unforeseen adverse consequences on local thermal environments have become a primary concern [9]. Similarly, the “image of the city” prompts us to consider how these invisible environmental stressors—elevated temperatures and poor air quality—degrade the experiential quality of urban spaces [10]. Therefore, understanding the coupled dynamics between thermal environment and atmospheric pollution is crucial for achieving the United Nations Sustainable Development Goals [11].

Rapid urbanization exacerbates fine particulate matter (PM2.5) pollution through multiple pathways, primarily driven by anthropogenic sources such as industrial emissions, vehicular exhaust, and construction activities [12,13,14]. Empirical studies show that PM2.5 not only positively correlates with urban sprawl but also exhibits a significant positive relationship with haze pollution [15,16,17].

Land surface temperature (LST), a key indicator of the urban thermal environment, exhibits complex spatial and seasonal interactions with PM2.5 concentrations. Studies in Beijing and the Yangtze River Delta have shown that high temperatures can both promote secondary aerosol formation and alter diurnal UHI patterns, subsequently affecting pollution levels [18,19,20]. LST–PM2.5 correlations are often nonlinear, with UHI potentially facilitating pollutant removal via enhanced turbulence at low pollution levels, but failing under heavy pollution conditions [21,22,23].

Notably, urban expansion influences PM2.5 and LST not only through increased human activity but also indirectly by altering urban form, which affects pollutant dispersion and deposition [24,25,26]. This physical structure’s influence is multi-faceted. Two-dimensional (2D) attributes, such as high building density and extensive impervious surfaces, can impede ventilation and increase surface temperatures, thus trapping both heat and pollutants [16,27,28]. Three-dimensional (3D) attributes introduce further complexity: tall buildings and high frontal areas can enhance mechanical turbulence but create deep street canyons that limit dispersion at the same time [29,30,31,32]. Low sky view factor reduces radiative cooling at night, intensifying the UHI effect, which in turn affects atmospheric stability and pollution patterns [33,34]. These dynamics are particularly acute within street canyons, the very spaces that Jane Jacobs championed as the heart of a vibrant, safe, and diverse city life [35]. From a modern perspective, the viability of complex “street ballet” is fundamentally dependent on the micro-environmental quality—thermal comfort and clean air—shaped by urban morphological details.

Research has significantly advanced the understanding of urban form’s impacts on thermal environments and air pollution, with an increasing variety of urban form parameters identified as influential [36], and methods like structural equation modeling and geographically weighted regression applied to explore these complex relationships [32,37,38]. Recent studies on urban morphological blocks increasingly use dynamic 3D modeling and generative optimization over static 2D metrics [39]. This advancement stems from fusing diverse datasets, including remote sensing imagery, points of interest (POIs), and human mobility data. Such approaches have enabled more holistic classification schemes like urban functional zones (UFZs) and enhanced local climate zone (LCZ) mapping. These provide a more nuanced characterization of urban blocks and entire structures, also laying the foundation for studying the regulatory effects of blocks on the thermal environment [40,41,42]. However, the existing body of research exhibits three key limitations: (1) a lack of focus on the mediating role of thermal environment (i.e., how does LST act as an “intermediary” to affect the urban form indices on pollution?) in the form–pollution relationship [43]. This omission is critical because LST is not merely a consequence of urban morphology, but also regulates chemical reactions and atmospheric stability; (2) insufficient consideration of seasonal reversals in form effects, such as the paradoxical role of height-to-width ratios in both beneficial summer shading and detrimental winter heat retention [44,45]; and (3) the inadequate consideration of the strong dependence of urban form–environment relationships across different urban functional zones, treating cities as homogeneous spaces or using uniform grids such as the local climate zones (LCZs), which obscure the significant spatial heterogeneity of the influence of urban form indices on thermal and air quality outcomes [46,47].

To fill these gaps, this study aims to answer the following key scientific questions: (1) Which urban form indices are the main drivers of LST and PM2.5? Do these relationships differ in kind and magnitude across functional zones and seasons? (2) What is the specific mediating pathway of LST in the relationship between urban form and PM2.5? What is the nature (e.g., partial, full, or masking) and intensity of this indirect effect? (3) Based on these findings, what differentiated and locally adapted urban planning strategies can be proposed to synergistically improve the thermal environment and air quality? By answering the central research questions, this study develops an integrated analytical framework, which contributes to sustainable urban planning by offering a methodology to empirically assess and optimize different urban design paradigms (Figure 1).

2. Materials and Methods

2.1. Study Area

The central city of Taiyuan, the capital of Shanxi Province, was selected as the study area (Figure 2). Geographically, the area is situated within a basin between the Taihang Mountains and Luliang Mountains on the eastern edge of the Loess Plateau, serving as a key industrial center and transportation hub in North China.

Taiyuan confronts dual environmental pressures arising from its industrial legacy and recent, rapid urbanization. As a traditional heavy industrial base, its economy has been dominated by resource-intensive sectors such as coal mining, coking, and metallurgy, which constitute significant sources of atmospheric pollutants [48]. These industrial emissions, compounded by energy consumption from urban expansion and winter heating, have resulted in severe PM2.5 pollution. For example, during the winter of 2021, the average PM2.5 concentration reached 98 μg/m³, far exceeding China’s national secondary standard of 35 μg/m³ [49].

Concurrently, the period from 2001 to 2021 was marked by rapid urbanization, during which the city’s population grew from 2.59 to 4.53 million and its built-up area expanded from 366 km² to 625 km² [48]. This dramatic expansion has intensified the Urban Heat Island (UHI) effect, evidenced by a discernible warming trend where the maximum air temperature rose from 32.5 °C to 34.2 °C over the same period [50]. This context of co-existing industrial pollution and urban-induced thermal stress renders Taiyuan a critical case for this study.

The central urban area of Taiyuan is characterized by high urbanization levels, dense population, and a well-defined functional zoning structure that includes residential, industrial, commercial, and public service zones. The city’s unique urban morphology, in conjunction with its distinct basin structure, establishes a distinctive environment for investigating the interactive effects of urban form on both LST and PM2.5 pollution.

2.2. Data Sources

This study utilized multiple datasets to investigate the impact of urban form indices on LST and PM2.5 concentrations in downtown Taiyuan, China (

Table 1. Summary of data sources.

Data	Spatial Resolution	Source
PM2.5	1 km	https://data.tpdc.ac.cn/zh-hans/data/6168e75d-93ab-4e4a-b7ff-33152e49d0bf (accessed on 12 March 2025)
LST	30 m	https://www.usgs.gov/ (accessed on 12 March 2025)
Road	-	https://www.opestreetmap.org (accessed on 20 March 2025)
Point-of-Interest (POI)	-	https://lbsyun.baidu.com/ (accessed on 20 March 2025)
Building	-	https://lbsyun.baidu.com/ (accessed on 23 March 2025)

). To reconcile the differing native resolutions of the source data, all variables were resampled to a common 1 km grid to ensure a consistent spatial framework for analysis. A common resolution of 1 km was adopted for the analysis. This scale aligns with the native resolution of the PM2.5 dataset and also adheres to established research practices. All datasets are publicly available, and no restrictions apply to their accessibility unless otherwise stated. The data sources, spatial resolutions, and access details are summarized below:

Landsat 8 Collection 2 Level-1 OLI/TIRS images for 17 July 2023 and 24 December 2023 were acquired from the U.S. Geological Survey. Scenes with cloud cover less than 10% were selected to minimize atmospheric interference.

Gridded PM2.5 concentration data were obtained from the “ChinaHighPM2.5—A Long-term, High-resolution, High-quality PM2.5 Dataset in China (2000–2023)”. For this study, PM2.5 data corresponding to 17 July 2023 and 24 December 2023 were extracted for the study area.

The two dates were deliberately selected to represent archetypal peak summer thermal stress and winter atmospheric pollution, not merely for their lack of cloud cover, but because they represent archetypal meteorological scenarios. The summer date (July 17, 2023) was chosen for its optimal UHI conditions, characterized by high temperature and calm winds that minimize advective cooling and highlight the influence of urban form. Conversely, the winter date (24 December 2023) was distinguished by strong thermal inversions and stagnant air, which are optimal conditions for severe pollution accumulation. This offered a viable case for analysis, which was not feasible for other high-pollution days due to prohibitive cloud cover.

2.3. Data Processing and Variable Generation

2.3.1. Identification of Urban Functional Zones

Urban functional zones in downtown Taiyuan were delineated using road data from OSM and POI data from Baidu Maps Open Platform. The methodology followed these steps [51]:

• Road Network Construction: The primary road network, consisting of main roads and primary to tertiary roads, was extracted from OSM to define the boundaries of the analysis units. Secondary roads were excluded to refine the zoning resolution.
• POI Data Processing: POI data were collected via a Python 3.7-based web crawler. The raw data underwent a cleaning process consistent with the Urban Land Use Classification Standard (GB 50137-2011) [52], involving de-duplication, coordinate correction, and projection to the WGS84 coordinate system. POIs were categorized into functional types (e.g., residential, commercial).
• Functional Zone Classification: Two indicators were calculated for each grid cell using ArcGIS: “frequency density” (the total count of POIs within a unit) and “category ratio” (the proportion of each POI type). The dominant function type for each unit was assigned based on the majority POI category, resulting in six functional zones: residential, industrial, commercial, public service, green, and water areas.

This classification provides a spatially consistent framework for analyzing the effects of urban form on LST and PM2.5. The accuracy of this classification was validated, achieving high performance across the primary functional zones: residential zones (overall accuracy = 0.88), commercial service zones (0.83), public service zones (0.74), green zones (0.73), industrial zones (0.71), and water zones (1.00).

2.3.2. LST Inversion and Delineation

LST was derived from Landsat 8 OLI/TIRS imagery acquired on 17 July 2023 and 24 December 2023 using the radiative transfer equation (RTE) method. The processing workflow, implemented in ENVI 5.6, consisted of the following steps [53].

• Radiance Calculation: Thermal band data were converted to spectral radiance using calibration parameters.
• Atmospheric Correction: Atmospheric transmittance, upwelling, and downwelling radiance were estimated using the NASA Atmospheric Correction Parameter Calculator.
• LST Calculation: The RTE was applied as follows:

$$L_{\lambda }\left(\theta ,\varphi \right)={ϵ}_{\lambda }{B}_{\lambda }\left(T_s\right){\tau }_{\lambda }\left(\theta \right)+L_{\lambda }^{\uparrow }\left(\theta \right)+\left(1-{ϵ}_{\lambda }\right)L_{\lambda }^{\downarrow }\left(\theta \right){\tau }_{\lambda }\left(\theta \right)$$

(1)

$$B_{\lambda }\left(T_s\right)={\displaystyle \frac{2h{c}^2}{{\lambda }^5}}{\displaystyle \frac{1}{e^{\frac{hc}{\lambda k{T}_s}}-1}}$$

(2)

$$T_s={\displaystyle \frac{K_2}{\mathit{ln}\left(\frac{K_1}{B\left(T_s\right)}+1\right)}}$$

(3)

where $L_{\lambda }\left(\theta ,\varphi \right)$ is the radiance received by the sensor; ${ϵ}_{\lambda }$ is the surface emissivity estimated from NDVI; $B_{\lambda }\left(T_s\right)$ is the blackbody radiance; ${\tau }_{\lambda }\left(\theta \right)$ is the atmospheric transmittance; $L_{\lambda }^{\uparrow }\left(\theta \right)$ and $L_{\lambda }^{\downarrow }\left(\theta \right)$ are the upwelling and downwelling atmospheric radiance, respectively. For Landsat 8 Band 10, $K_1=774.8853 \mathrm{W}/\left({\mathrm{m}}^2\mathsf{\mu }\mathrm{m}\mathrm{s}\mathrm{r}\right)$, $K_2$ = 1321.0789 K.

$$P_v={\left({\displaystyle \frac{NDVI-NDV{I}_{soil}}{NDV{I}_{veg}-NDV{I}_{soil}}}\right)}^2$$

(4)

$${ϵ}_{\lambda }=\left\{\begin{array}{c}{ϵ}_{w\lambda }\\ {ϵ}_{w\lambda }{P}_v+{ϵ}_{s\lambda }\left(1-P_v\right)\\ {ϵ}_{v\lambda }\end{array}\right.\begin{array}{c}NDVI<NDV{I}_{soil}\\ NDV{I}_{soil}\le NDVI\le NDV{I}_{veg}\\ NDVI\ge NDV{I}_{veg}\end{array}$$

(5)

where $P_v$ is the proportion of vegetation; NDVI is the normalized difference vegetation index value for a given pixel; $NDV{I}_{soil}$ is the NDVI value for bare soil surfaces, typically set to a threshold of 0.2; $NDV{I}_{veg}$ is the NDVI value for fully vegetated surfaces, typically set to a threshold of 0.5. ${ϵ}_{w\lambda }$, ${ϵ}_{v\lambda }$ and ${ϵ}_{s\lambda }$ are the emissivities of water (0.991), vegetation (0.986), and soil/urban surfaces (0.970), respectively, for Landsat 8 Band 10.

2.3.3. Urban Form Indices (UFIs)

Seven urban form indices were selected to characterize two- and three-dimensional urban form indices. The selection of 2D (BD, NB, SO), 3D (MBH, FAI, SVF) urban form, alongside DEM, was based on their extensive application in prior studies [30,31,32,33,34], strong correlation with LST and PM2.5, and variance inflation factor (VIF < 5) to avoid multicollinearity. All indices were calculated using ArcGIS and ENVI 5.6. The selected UFIs are defined as follows:

• Number of Buildings (NB): The total count of buildings within a grid cell.
  $$\mathrm{NB}=\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{B}}_{\mathrm{i}}$$

  (6)

  where ${\mathrm{B}}_{\mathrm{i}}$ represents the building i in the grid cell.
• Site Orientation (SO): The average building orientation relative to north.
  $$\mathrm{SO}={\displaystyle \frac{1}{n}}\sum _{i=1}^n{\theta }_i$$

  (7)

  where ${\theta }_i$ represents is the orientation angle of the building i.
• Building Density (BD): The ratio of total building floor area to grid cell area.
  $$\mathrm{BD}={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\mathrm{B}{\mathrm{S}}_{\mathrm{i}}}{{\mathrm{S}}_{\mathrm{i}}}}$$

  (8)

  where ${\sum }_{\mathrm{i}=1}^{\mathrm{n}}\mathrm{B}{\mathrm{S}}_{\mathrm{i}}$ is the total area of the buildings in the grid cell i. ${\mathrm{S}}_{\mathrm{i}}$ is the total area of the grid cell i.
• Sky View Factor (SVF): The fraction of visible sky relative to total hemispherical sky.
  $$\mathrm{SVF}=1-{\displaystyle \frac{{\sum }_{i=1}^n\mathit{sin}{\theta }_i}{n}}$$

  (9)

  where ${\sum }_{i=1}^n\mathit{sin}{\theta }_i$ is the influence of the terrain height angle on the azimuth angle i; and n is the number of calculated azimuth angles.
• Mean Building Height (MBH): The average height of buildings within a grid cell.
  $$\mathrm{MBH}={\displaystyle \frac{{\sum }_{i=1}^n{\mathit{BH}}_i}{n}}$$

  (10)

  where ${\sum }_{i=1}^n{\mathit{BH}}_i$ is the total height of buildings in the grid cell i.
• Frontal Area Index (FAI): The ratio of windward building area to grid cell area.
  $$\mathrm{F}\mathrm{A}\mathrm{I}={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}} S_{\left\{p,i\right\}}\left(\theta \right)}{S_j}}$$

  (11)

  where ${\sum }_{\mathrm{i}=1}^{\mathrm{n}}{S}_{\left\{p,i\right\}}\left(\theta \right)$ is the total projected area of the building i’s facade onto a vertical plane perpendicular to the wind direction in the grid cell j; $S_j$ is the total area of the grid cell j.
• Digital Elevation Model (DEM): The grid-based elevation of terrain.

The DEM data originates from the 30 m SRTM data provided by NASA Earthdata and was processed in ArcGIS.

2.4. Statistical Analysis

2.4.1. Correlation Analysis

Spearman rank correlation analysis was conducted to assess the strength and direction of bivariate associations between all variables (UFIs, LST, PM2.5). This non-parametric method was chosen because some variables were discrete, not all variables followed a normal distribution, and it has reduced sensitivity to outliers compared to Pearson correlation.

2.4.2. Multiple Regression Modeling

To quantify the aggregate effect of UFIs on LST and PM2.5 for summer and winter periods, multiple linear regression models using the ordinary least squares (OLS) estimation method were employed. Given the potential for non-linear relationships between urban form and LST/PM2.5 [28], all independent variables (UFIs) were natural log-transformed. This transformation serves to linearize the relationships and improve model fit. The general forms of the models are as follows:

$$\mathrm{LS}{\mathrm{T}}_{\mathrm{i}}={\mathsf{\alpha }}_0 +{\mathsf{\alpha }}_1 \ln \mathrm{N}{\mathrm{B}}_{\mathrm{i}} + {\mathsf{\alpha }}_2 \ln \mathrm{B}{\mathrm{D}}_{\mathrm{i}} + {\mathsf{\alpha }}_3 \ln \mathrm{S}{\mathrm{O}}_{\mathrm{i}} + {\mathsf{\alpha }}_4 \ln \mathrm{MB}{\mathrm{H}}_{\mathrm{i}} + {\mathsf{\alpha }}_5 \ln \mathrm{FA}{\mathrm{I}}_{\mathrm{i}} + {\mathsf{\alpha }}_6 \ln \mathrm{SV}{\mathrm{F}}_{\mathrm{i}} + {\mathsf{\alpha }}_7 \ln \mathrm{DE}{\mathrm{M}}_{\mathrm{i}} + {\mathsf{ϵ}}_{\mathrm{i},1}$$

(12)

$$\mathrm{P}{\mathrm{M}}_{2.5,\mathrm{i}}={\mathsf{\beta }}_0 +{\mathsf{\beta }}_1 \ln \mathrm{N}{\mathrm{B}}_{\mathrm{i}} + {\mathsf{\beta }}_2 \ln \mathrm{B}{\mathrm{D}}_{\mathrm{i}} + {\mathsf{\beta }}_3 \ln \mathrm{S}{\mathrm{O}}_{\mathrm{i}} + {\mathsf{\beta }}_4 \ln \mathrm{MB}{\mathrm{H}}_{\mathrm{i}} + {\mathsf{\beta }}_5 \ln \mathrm{FA}{\mathrm{I}}_{\mathrm{i}} + {\mathsf{\beta }}_6 \ln \mathrm{SV}{\mathrm{F}}_{\mathrm{i}} + {\mathsf{\beta }}_7 \ln \mathrm{DE}{\mathrm{M}}_{\mathrm{i}} + {\mathsf{ϵ}}_{\mathrm{i},2}$$

(13)

where i represents each analytical unit, ${\mathsf{\beta }}_0$ and ${\alpha }_0$ are the intercepts, ${\mathsf{\beta }}_1$ … ${\mathsf{\beta }}_7$ and ${\alpha }_1$ … ${\alpha }_7$ are the regression coefficients for the respective log-transformed UFIs. Specifically, ${\mathsf{\beta }}_{1~7}$ indicates the percentage change in Y for a 1% change in ${\mathrm{X}}_{\mathrm{i}}$, holding all other variables constant. The absolute value of ${\mathsf{\beta }}_{1~7}$(|${\mathsf{\beta }}_{1~7}$|) reflects the strength of the relationship between ${\mathrm{X}}_{\mathrm{i}}$ and Y in terms of percentage changes. A larger |${\mathsf{\beta }}_{1~7}$| suggests a stronger proportional effect of ${\mathrm{X}}_{\mathrm{i}}$ on Y. ${\mathsf{ϵ}}_{\mathrm{i},1}$ and ${\mathsf{ϵ}}_{\mathrm{i},2}$ is the random error term.

2.4.3. Mediation Effect Analysis

To disentangle the complex interaction mechanisms, mediation analysis was performed. This analysis tested the hypothesis that LST mediates the relationship between UFIs and PM2.5. The analyses were performed separately for the summer and winter periods. The bootstrapping method [54] was employed for this analysis due to its robustness and suitability for non-normally distributed data. A total of 5000 bootstrap resamples were generated to construct empirical distributions of the mediating effects, and 95% bias-corrected confidence intervals (BCIs) were calculated. A mediating effect was considered statistically significant if its 95% BCI excluded zero. In Figure 3, a is the effect of IV on M; b is the effect of M on DV; c’ is the direct effect of IV on DV with M in the model; c is the total effect of IV on DV without M in the model.

The mediating effect can be divided into four types: (1) Full mediation: a and b significant, c’ non-significant (even if 95% BCI includes 0), or a or b non-significant but 95% BCI excludes 0 and c’ non-significant. (2) Partial mediation: a and b significant, c’ significant, ab and c’ same sign, or a or b non-significant, 95% BCI excludes 0, c’ significant, ab and c’ same sign. (3) Masking effect: a and b significant, c’ significant, ab and c’ opposite signs, or a or b non-significant, 95% BCI excludes 0, c’ significant, ab and c’ opposite signs. (4) Non-significant mediation: a or b non-significant, 95% BCI includes 0 (Figure 4).

3. Results

This section systematically presents the empirical results of our study. First, we illustrate the foundational spatial heterogeneity of urban functional zones and morphological indices in Taiyuan (Section 3.1). Subsequently, we reveal the distinct seasonal and spatial patterns of LST and PM2.5, highlighting a critical divergence between thermal and pollution hotspots (Section 3.2). Finally, through statistical analysis (Section 3.3–3.5), we quantify the complex, seasonally-dependent relationships between urban form, LST, and PM2.5, with a particular focus on the mediating role of the thermal environment.

3.1. Spatial Distribution of Urban Functional Areas and Urban Form Indices in Taiyuan

Using OSM road network and Baidu Maps POI data, the central urban area of Taiyuan was divided into 3849 functional zone units, categorized into six land-use types: residential zones (RZs), industrial zones (IZs), commercial service zones (CSZs), public administration and service zones (PSZs), green spaces and open areas (GZs), and water zones (WZs). The spatial distribution of these zones exhibits distinct differentiation (Figure 5). RZs are concentrated in the south-central part of the city, transitioning outward from the center. IZs are primarily located in the northern and southern regions, while CSZs are centered in the city core, intermingled with RZs. PSZs are scattered throughout, with higher density in the central area. The Fenhe River WZ runs north-south through the city center.

The spatial distribution of the UFIs reveals significant heterogeneity across functional zones in both two-dimensional (2D) and three-dimensional (3D) structures (Figure 6). The central region, predominantly a mix of RZs and CSZs, is characterized by a compact urban form. Here, 2D indices such as high BD and a large NB reflect intensive land use. Concurrently, 3D indices show substantial vertical differentiation, with buildings in CSZs being generally taller than those in RZs. This area exhibits high FAI and low SVF values, indicative of concentrated high-rise structures, which create a high degree of spatial enclosure and elevate the risks of heat accumulation and pollutant retention.

In contrast, the IZs display a distinct morphology: the NB is relatively high, the MBH is moderate, typically ranging from 5 to 27 m, accompanied by moderate BD. The IZs located in the western, northern, and southern peripheries feature high SVF and low FAI, a configuration that suggests greater potential for natural ventilation and pollutant diffusion. As expected, GZs exhibit very low BD, minimal building heights, and SVF values near 1.0, reflecting an open and permeable structure that supports urban ventilation and thermal regulation.

3.2. Surface Thermal Environment and Air Pollution Patterns Across Taiyuan’s Functional Zones

The spatial distribution of LST and PM2.5 concentrations in Taiyuan exhibits significant seasonal and functional zone heterogeneity, analyzed here across thermal environment and aerosol pollution dimensions (Figure 7).

In summer, the UHI effect is pronounced in the built-up area, with LST exhibiting strong clustering. Areas with high temperatures (LST ≥ 35 °C) are concentrated in the central city and southwestern industrial areas. Conversely, cooler zones (LST ≤ 28 °C) are predominantly located in the northern and northwestern areas of GZs and WZs. In winter, spatial disparities in LST become more pronounced: the thermal boundaries of northern GZs appear more sharply defined, while RZs and IZs converge to form a composite urban heat island (UHI) structure characterized by increased thermal intensity and broader spatial extent compared to summer.

PM2.5 concentrations also exhibit seasonal variation. In summer, pollution levels are low to moderate. The areas with high concentrations (PM2.5 ≥ 20 μg/m³) are localized in the northern and southern industrial zones, indicating pollution agglomeration near sources. In winter, pollution intensifies dramatically. In winter, pollution intensifies, with PM2.5 peaking above 55 μg/m³ in the northern IZs, and average concentrations exceeding 40 μg/m³ across many functional units. GZs exhibit a pollution reduction effect in both seasons, though this effect is weaker in winter.

3.3. Descriptive Statistics and Correlation Analysis

Table 2. Descriptive statistics of the variables for each block.

UFIs	Unit	17 July 2023				24 December 2023
		Mean	Max	Min	Std.	Mean	Max	Min	Std.
PM2.5	μg/m3	20.64	23.33	0.00	3.830	55.39	63.04	0.00	1.343
LST	Degrees Celsius (°C)	45.69	63.04	0.00	4.273	−6.90	0.49	−11.43	1.247
NB	-	23.20	762.00	0.00	31.374	23.20	762.00	0.00	31.374
SO	Degree (°)	73.41	179.59	0.00	37.060	73.41	179.59	0.00	37.060
BD	Percentage (%)	14.13	93.66	0.00	11.222	14.13	93.66	0.00	11.222
MBH	Meters (m)	14.39	98.18	0.00	13.304	14.39	98.18	0.00	13.304
FAI	-	0.15	0.64	0.00	0.120	0.15	0.64	0.00	0.120
SVF	-	0.87	1.00	0.57	0.069	0.87	1.00	0.57	0.069
DEM	Meters (m)	781.18	1117.91	0.00	76.256	781.18	1117.91	0.00	76.256

summarizes the descriptive statistics for PM2.5, LST, and UFIs across 3849 spatial units on 17 July 2023 and 24 December 2023.

During the winter season, the mean PM2.5 concentration (55.39 μg/m³) surpasses the World Health Organization’s recommended 24-h limit of 15 μg/m³ by a substantial margin, reflecting a critical level of air pollution. In summer, the mean PM2.5 (20.64 μg/m³) still surpasses this standard, reflecting persistent air quality challenges. In summer, the average LST reached 45.69 °C, with a maximum of 63.04 °C, showing significant spatial variation. In winter, the thermal landscape exhibited even greater contrast, with an average LST of −6.90 °C and a minimum of −11.43 °C.

The Spearman correlation analysis reveals distinct relationships for LST and PM2.5 (Figure 8). For LST, nearly all urban form indices (UFIs), except for site orientation (SO), showed significant correlations in both seasons. These correlations were consistently stronger in summer, suggesting a more pronounced influence of urban form on the thermal environment during periods of high heat. In contrast, the relationship with PM2.5 was highly seasonal. This suggests that urban form acts as a persistent thermal regulator but its role as a pollution modulator is seasonally activated. In summer, only the sky view factor (SVF) and digital elevation model (DEM) had significant correlations, indicating limited morphological influence. In winter, however, over half of the UFIs became significantly correlated with PM2.5, with key factors like frontal area index (FAI) and SVF showing relationships opposite to those in summer. This suggests a much stronger role for urban form in pollution accumulation under colder conditions. Finally, the low to moderate bivariate correlations among the independent variables confirmed that multicollinearity was not a significant concern for the subsequent regression analyses.

3.4. Total Effects of Urban Form Indices on PM2.5 and LST

Multiple log-linear regression results (

Table 3. Multiple log-linear regression results.

Variables	17 July 2023		24 December 2023
	LST	PM2.5	LST	PM2.5
NB	−0.357 1	0.011	−0.015	−0.140
SO	−0.004	−0.021	−0.008	−0.123
BD	1.545	0.025	0.088	0.361
MBH	−0.481	0.058	−0.042	−0.121
FAI	10.977	−0.488	3.438	4.998
SVF	15.074	2.491	1.114	−3.160
DEM	5.405	1.561	−0.895	4.095
Intercept	−2.579	8.647	−2.165	29.772
N	3849	3849	3849	3849
R2	0.616	0.367	0.215	0.342
RMSE	2.647	1.069	1.105	3.108
p-value	*** 2	***	***	***

¹ The values represent the regression coefficients corresponding to this independent variable in the multiple logarithmic regression equation. ² *** represents p-value <= 0.001.

) assess the total effects of UFIs on LST and PM2.5 for 17 July 2023 and 24 December 2023 (N = 3849). The models fit well, with R² values of 0.616 (summer) and 0.215 (winter) for LST, and 0.367 (summer) and 0.342 (winter) for PM2.5. The root mean square errors (RMSE) are reasonable, indicating robust explanatory power.

The effects of UFIs on LST exhibit seasonal heterogeneity. In summer, FAI, SVF, and DEM positively affect LST, while NB and MBH have negative effects, likely due to shading and thermal radiation blocking by dense, tall buildings. In winter, FAI and SVF maintain positive effects, stabilizing the thermal environment, but DEM shifts to a negative effect, possibly due to cold air and wind impacts in higher-elevation areas.

Seasonal differences in the factors affecting PM2.5 are evident. In summer, sky view factor (SVF) and digital elevation model (DEM) exhibit positive correlations with PM2.5 concentrations, whereas fractional anthropogenic imperviousness (FAI) shows a negative relationship, facilitating pollutant dispersion. In winter, DEM retains a positive effect, but FAI shifts to a significant positive effect (β = 4.998), likely due to pollutant accumulation in high windward areas from heating emissions, while SVF shifts to a negative effect (β = −3.160), indicating better dispersion in open spatial structures during colder months.

From the absolute values of the regression coefficients of the UFIs, FAI, SVF, and DEM are identified as the main UFIs affecting both LST and PM2.5. They have different directions and intensities observed across the seasons. This highlights the seasonal and spatial heterogeneity in urban thermal and pollution dynamics.

3.5. Mediating Effects of Urban Form Indices on PM2.5 via LST

The bootstrap mediation analysis evaluates the indirect effects of UFIs on PM2.5 through LST for summer and winter (

Table A1. Results of the bootstrapping mediation analysis.

Date	IV	M	DV	a	b	Indirect Effect		Direct Effect		Total Effect		Proportion Mediated (ab/c) 1	Effect
						ab	95%BCI	c’	95%BCI	c	95%BCI
17 July 2023	NB	LST	PM2.5	0.017	0.173	0.003	0.045~0.093	−0.002	−0.003~−0.001	0.001	−0.000~0.002	147.168%	Masking effect
	SO			0.013	0.174	0.002	0.043~0.086	−0.002	−0.003~−0.001	0.000	−0.001~0.001	100.962%	Masking effect
	BD			0.149	0.202	0.030	0.189~0.297	−0.030	−0.033~−0.027	0.000	−0.004~0.004	100.854%	Masking effect
	MBH			0.015	0.172	0.003	0.003~0.044	−0.007	−0.010~−0.004	−0.005	−0.008~−0.001	35.742%	Masking effect
	FAI			10.653	0.193	2.060	0.136~0.220	−2.602	−2.901~−2.302	−0.542	−0.895~−0.189	79.173%	Masking effect
	SVF			−5.593	0.175	−0.977	−0.071~−0.028	2.247	1.730~2.763	1.269	0.651~1.887	43.504%	Masking effect
	DEM			0.025	0.137	0.003	0.098~0.257	0.004	0.004~0.005	0.008	0.007~0.008	44.255%	Partial Mediation
24 December 2023	NB	LST	PM2.5	0.004	−0.888	−0.004	−0.042~−0.018	0.004	−0.000~0.007	0.000	−0.004~0.004	100%	Full Mediation
	SO			0.002	−0.877	−0.002	−0.030~−0.008	−0.001	−0.004~0.002	−0.003	−0.006~0.001	100%	Full Mediation
	BD			0.021	−0.967	−0.020	−0.076~−0.042	0.052	0.041~0.062	0.031	0.021~0.042	39.514%	Masking effect
	MBH			0.008	−0.901	−0.007	−0.038~−0.015	0.026	0.017~0.035	0.019	0.009~0.028	28.084%	Masking effect
	FAI			2.804	−1.090	−3.057	−0.115~−0.073	8.162	7.192~9.132	5.104	4.109~6.100	37.460%	Masking effect
	SVF			−2.968	−0.966	2.867	0.036~0.067	−9.742	−11.430~−8.054	−6.875	−8.628~−5.122	29.426%	Masking effect
	DEM			−0.006	−0.591	0.003	0.048~0.085	0.014	0.012~0.015	0.017	0.015~0.018	20.581%	Partial Mediation

¹ ab/c represents the proportion of the total effect (c) accounted for by the indirect effect (ab), indicating mediation magnitude. Interpret cautiously: if c is small, or if ab and c have opposite signs, this ratio can exceed 100% or be negative; in such cases, prioritize the significance and effect size of the indirect effect (ab) itself. Effect size (proportion mediated): complete mediation: 100%; partial mediation: (ab/c); masking effect: (|ab/c|) and non-significant mediation: 0%.

for details) to disentangle this indirect mechanism and quantify its nature and intensity. Most UFIs exhibit significant mediating effects, with distinct seasonal variations.

In summer, FAI exhibits the strongest mediating effect, with a unit increase indirectly raising PM2.5 by 2.060 units via LST (ab = 2.060, 95% BCI: 0.136 to 0.220). However, its direct effect (c’ = −2.602) offsets this, resulting in a total effect of −0.542, indicating a masking effect. SVF reduces PM2.5 indirectly through LST (ab = −0.977), but its direct effect is positive (c’ = 2.247), highlighting LST’s buffering role. DEM’s indirect effect is small (ab = 0.003) but accounts for 44.26% of its total effect, suggesting a clear but limited topographic influence. LST exhibits inconsistent mediation in the effects of NB, SO, BD, and MBH on PM2.5: their LST-mediated indirect effect increasing PM2.5 is offset by their direct PM2.5-reducing effect, leading to non-significant total effects. For NB, SO, and MBH, the absolute magnitudes of these opposing indirect and direct effects are both relatively small. In contrast, the absolute magnitudes of BD’s LST-mediated indirect effect (ab ≈ 0.030) and its direct effect (c’ ≈ −0.030) are significantly larger. This indicates that BD influences PM2.5 through two stronger, opposing mechanisms: LST-mediated pollution exacerbation and its own direct pollution reduction. Therefore, despite a non-significant total effect, BD plays a more potent, counterbalanced dual role in modulating local PM2.5 than NB or SO, and the differing strengths of its influence pathways should not be overlooked.

Conversely, the mediation structure in winter shifts significantly. FAI’s indirect effect on PM2.5 via LST is negative (ab = −3.057), but its direct effect is strongly positive (c’ = 8.162), yielding a total effect of 5.104, suggesting LST mitigates FAI’s pollution-enhancing effect in colder conditions. SVF’s indirect effect becomes positive (ab = 2.867), but its direct effect turns negative (c’ = −9.742), indicating that physical ventilation dominates over thermal regulation in winter. DEM shows a small positive indirect effect on PM2.5 via LST (ab = 0.003, 95% BCI: 0.048 to 0.085), with 20.58% mediated, indicating a primarily direct influence (β = 4.095, Table 3). In contrast, BD has a larger mediation effect (ab = −0.020, 95% BCI: −0.076 to −0.042), with 39.51% mediated, due to its stronger effect on LST (a = 0.021) and LST’s impact on PM2.5 (b = −0.967), despite a smaller total effect (β = 0.361). This suggests that the influence of building density (BD) on PM2.5 is primarily mediated through thermal pathways, whereas the effect of digital elevation model (DEM) appears to be more direct, likely driven by topographic conditions associated with Taiyuan’s basin setting. NB and SO show full mediation (direct effect non-significant), while MBH exhibits a masking effect with a positive total effect, suggesting height-related pollution retention due to winter inversions.

LST mediates the UFI-PM2.5 pathway with seasonal variation. In summer, higher LST correlates with increased PM2.5 (b ≈ 0.17–0.20), indicating inhibited pollutant dispersion under high temperatures. In winter, LST negatively correlates with PM2.5 (b ≈ −0.88 to −1.10), suggesting a dilution effect in colder conditions. These findings clarify indirect pathways of urban form on pollution, highlight LST’s regulatory role, and provide a basis for multi-path pollution control models and zone-specific governance strategies.

4. Discussion

4.1. Differential Responses of LST and PM2.5 Across Urban Functional Zones

The preceding analysis (Section 3.1 and Section 3.2) demonstrated that Taiyuan’s urban functional zones possess significant structural heterogeneity, from the dense, medium-rise morphology of industrial zones (IZs) to the compact forms of residential zones (RZs). This spatial heterogeneity, coupled with varying intensities of human activity intensity and emission profiles, logically leads to distinct zonal responses of LST and PM2.5 to UFIs. To systematically investigate these differential responses, we constructed separate multiple log-linear regression models for the four most active functional zones—RZ, IZ, CSZ, and PSZ. This enables a nuanced analysis of how UFIs moderate the thermal-pollution relationship within each specific context (Figure 9).

RZ’s high-density, high-enclosure morphology (high FAI, low SVF) exacerbates heat and pollution retention. In summer, FAI significantly increases LST (β = 14.397) and PM2.5 (β = 0.135), likely due to reduced ventilation and heat dissipation [24]. In winter, FAI’s effect on PM2.5 intensifies (β = 5.868), reflecting pollution retention from heating emissions in enclosed spaces. MBH positively affects PM2.5 in winter (β = 0.259), possibly due to deep street canyons weakening wind disturbance [55,56]. The RZ regression models show the highest R² values (e.g., 0.502 for summer LST), indicating strong explanatory power of urban form in these zones, where vertical morphology amplifies heat-pollution interactions [57,58].

IZ’s open morphology (high SVF, low FAI) and high emission intensity result in nonlinear thermal-pollution interactions. In summer, SVF strongly increases PM2.5 (β = 7.743), as enhanced radiative absorption may lead to pollutant sinking, contradicting the typical diffusion benefits of openness [59]. In winter, the SVF-LST-PM2.5 pathway reverses, possibly due to cold air stagnation [60]. FAI’s effect on LST is weaker than in RZs, reflecting better ventilation. BD positively affects PM2.5 in both seasons, more so in winter, indicating amplified pollution aggregation under stable conditions [16]. The low R² values (e.g., 0.093 for summer PM2.5) suggest missing variables like emission source strength, limiting explanatory power.

CSZ’s complex morphology, high-rise buildings, and intensive human activity amplify thermal-pollution effects [61]. FAI and SVF strongly influence LST in both seasons, while SVF negatively affects PM2.5 in winter (β = −3.276), indicating open spaces aid pollutant dilution [62]. NB and SO also play significant roles, reflecting the impact of building orientation and building density on thermal-pollution dynamics [61,63]. High BD increases summer LST, but its effect weakens in winter, possibly due to nighttime thermal emissions dominating pollution formation [64].

The typically open morphology of PSZ, often characterized by significant variations in building height, creates unique conditions for wind flow and pollution transport. FAI’s positive effect on PM2.5 in winter is the most pronounced among all zones (β = 8.472), likely attributable to the obstruction of ventilation by large buildings [65,66,67,68]. Conversely, DEM’s negative effect on PM2.5 is evident (β = −17.583 in winter). This is indicative of alterations in wind field patterns, driven by variations in elevation, particularly in relatively flat terrain [69,70].

Overall, key UFIs driving the urban form–thermal–pollution response chain vary significantly across zones, with strong seasonal modulation. These findings underscore the need for zone-specific, seasonally tailored urban form optimization strategies to achieve synergistic improvements in thermal environments and pollution control.

4.2. Indirect Effects of Urban Form Indices on LST and PM2.5

The mediation analysis (Section 3.5) quantitatively demonstrates that in a basin industrial city like Taiyuan, urban form influences PM2.5 concentrations through both direct physical mechanisms and indirect, thermally driven pathways. This highlights the critical mediating role of LST within the coupled “urban thermal environment–air pollution” system. In summer, for instance, FAI indirectly increases PM2.5 by elevating LST (ab = 2.060); however, this is counteracted by a stronger, direct negative effect (c’ = −2.602), resulting in a net masking effect (total effect = −0.542). This aligns with Yangtze River Delta studies, where summer heat enhances photochemical reactions, exacerbating PM2.5 levels [71]. Conversely, it contrasts with findings from Beijing that report a negative UHI–PM2.5 correlation, a discrepancy likely attributable to the dominance of industrial emissions in Taiyuan compared to transportation-related sources in Beijing [72].

SVF exhibits a seasonal reversal in its mediation effect. In summer, low SVF reduces LST by blocking solar radiation, limiting PM2.5 diffusion (ab = −0.977); in winter, low SVF increases LST by suppressing nocturnal longwave radiation loss, enhancing thermal turbulence and PM2.5 dispersion (ab = 2.867). This reflects Taiyuan’s central high-density clusters, which sustain higher LST in winter, reducing near-surface PM2.5 through enhanced mixing by increasing the mixing layer height [58,73].

DEM’s mediation effect exhibits strong geographical dependence, contributing only 20.58% to PM2.5 in winter (ab = 0.003, Table A1), contrasting with studies reporting stronger topographic moderation [74,75]. In Taiyuan’s low-elevation basin, the limited mediation effect likely stems from winter atmospheric stability, which masks DEM’s direct influence on pollutant transport. Taiyuan’s basin topography, characterized by gentle slopes and a relatively flat central area (Section 2.1), fosters cold pool formation and low wind speeds in winter (Section 4.3), reducing LST variability and its role as a mediator [76,77]. Consequently, DEM primarily affects PM2.5 directly through elevation-driven wind field changes, such as channeling pollutants along topographic gradients, rather than indirectly via LST [69,78]. This aligns with findings in other basin cities, where topographic effects on pollutant dispersion are often decoupled from thermal pathways under stable conditions [60]. MBH’s mediation effect (35.74% in summer, 28.08% in winter) is also weak, diverging from studies where upstream building heights significantly alter pollutant paths [56], but consistent with theories of wind speed saturation beyond critical heights in stable boundary layers [55]. These findings suggest that in basin cities like Taiyuan, urban form strategies targeting topographic factors (e.g., DEM) should prioritize direct pollutant transport mechanisms, such as optimizing wind corridors, over thermal mediation pathways, to enhance air quality and thermal comfort [79].

4.3. Seasonal Reversal of the LST-PM2.5 Relationship and Its Driving Mechanisms in Taiyuan

The LST-PM2.5 relationship in Taiyuan exhibits significant seasonal heterogeneity and complex bidirectional feedback, diverging from conventional views that often report a negative correlation [58,59] driven by UHI-enhanced dispersion. Our findings reveal a seasonal reversal: a negative LST-PM2.5 correlation during winter pollution periods (r = −0.313, p < 0.01) and a weak positive correlation in summer (r = 0.144, p < 0.05), indicating fundamentally different dominant atmospheric and physicochemical processes across seasons.

In summer, the weak positive LST-PM2.5 correlation arises from a combination of spatio-temporal coupling of heat and pollution sources and meteorological factors. Firstly, high LST and intense solar radiation accelerate photochemical reactions of gaseous precursors (NO_x, SO₂, VOC_s) into secondary aerosols (sulfates, nitrates, SOA) [80]. This process, often peaking with maximum LST, can lead to concurrent PM2.5 increases, fostering a positive linkage. Secondly, higher summer LST is also associated with stronger convection and an increased boundary layer height [81], which generally promotes pollutant dilution. Thus, the net effect of summer LST on PM2.5 reflects a nuanced interplay between two competing processes: enhanced secondary aerosol formation (a pollution-accumulating effect) and stronger convective dispersion (a pollution-diluting effect). The observed weak positive correlation suggests that, under the conditions studied, the impact of photochemical production slightly outweighs or parallels the benefits of dilution.

Conversely, winter LST exerts a more pronounced negative impact on PM2.5 in Taiyuan. This is driven by the synergistic effects of the urban topography—DEM, meteorology, and altered dispersion dynamics. As a basin city, Taiyuan experiences significant surface cooling in winter due to weak solar radiation and enhanced nocturnal longwave radiation [82,83]. The enclosed topography exacerbates this cooling, which in turn facilitates the formation of cold pools and stable atmospheric conditions that strongly inhibit the vertical dispersion of pollutants [84]. Furthermore, the basin terrain diminishes wind speeds and forms frequent calm wind fields, which is confirmed by analysis of local meteorological data for the study period [85,86]. This substantially reduces horizontal dispersion and prolongs the pollutant residence time [87]. Consequently, lower winter LST and lower DEM indicate highly stable conditions where pollutants are trapped near the surface, causing elevated PM2.5 levels. This provides a clear physical mechanism for the significant negative correlation between LST and PM2.5, and explains the negative correlation between DEM and PM2.5 found in regression models. These findings are consistent with pollution dynamics observed in other urban basins [88].

4.4. Implications for Urban Planning and Design

Synergistically enhancing urban air quality and thermal comfort necessitates planning interventions that address the complex, seasonally variant interactions of UFIs, moving beyond simplistic, single-season strategies. This study demonstrates that the effects of urban form on PM2.5 are not merely direct but also significantly mediated by LST, with these pathways exhibiting profound spatial and seasonal heterogeneity. A “functional zoning-morphology-diffusion” framework can guide differentiated optimization strategies for sustainable urban planning.

At the functional zone level, targeted interventions are needed. RZs should limit FAI and MBH growth, optimizing building spacing for better ventilation. IZs can enhance SVF and choose places with higher DEM to improve pollutant diffusion through wind corridors. CSZs should adjust SO and BD to prevent heat-pollution buildup in deep street canyons. PSZs can leverage spatial flexibility to enhance terrain-driven ventilation (Figure 10). Zone-specific LST-PM2.5 response mechanisms necessitate tailored morphological guidelines, avoiding one-size-fits-all approaches.

From a mediation perspective, LST’s dual role in pollution dispersion requires season-specific measures. FAI’s winter enhancement of PM2.5 via LST suggests minimizing large, enclosed layouts. SVF’s seasonal reversal calls for increased summer shading to reduce heat and maintain winter openness for turbulence diffusion. DEM’s limited effect in basin cities requires integrated wind path and emission source planning. Urban form strategies should address both direct and indirect (via LST) effects on PM2.5.

Given Taiyuan’s basin topography and industrial structure, thermal-pollution issues often spatially coincide. Future urban renewal should prioritize areas with significant mediation effects (e.g., high FAI or critical SVF zones), transitioning from integrated morphology–thermal–pollution management to cross-sectoral systemic control, achieving synergistic thermal and air quality improvements.

4.5. Limitations and Future Research

This study has several limitations. First, its temporal scope is restricted to two days (17 July and 24 December 2023) and spatially limited to Taiyuan. Future research should expand to multi-seasonal, multi-city analyses to capture broader spatial-temporal dynamics. Second, by focusing mostly on urban form indices of building, this study omits the significant counteracting effects of green infrastructure. The ability of vegetation to cool LST via shading and evapotranspiration and improve air quality by capturing pollutants and altering dispersion was not accounted for [89,90,91,92]. Thus, our results represent the gross effect of urban form, with the understanding that the true net effect is modulated by landscape features. Integrating both factors is a key direction for more holistic future research.

Third, the single-functional zoning model simplifies the urban fabric, particularly in mixed-use areas. These heterogeneous zones likely possess unique environmental dynamics. For instance, their blend of commercial and residential activities can create distinct diurnal thermal profiles and more complex pollution emission and dispersion patterns. By assigning a single function, our model may not fully capture these nuanced interactions in such areas. Fourth, focusing on PM2.5 overlooks other pollutants (e.g., O₃, NOₓ), increasingly relevant in Taiyuan, necessitating a multi-pollutant model to explore varied responses [93,94]. In addition, the main limitation of this study is the integration of multiple source data with different native spatial resolutions. To address the large resolution differences and mitigate potential microscale effects, we aggregated spatial variables into a common analysis unit, the UFZ. This approach smoothed fine-grained noise but sacrificed the microscopic details of the finer LST datasets. Future studies should use datasets with higher resolution and smaller resolution differences. Finally, micro-scale morphological parameters (e.g., facade roughness, roof design) and their impact on near-surface dynamics remain unquantified; street-level measurements and CFD simulations can provide deeper insights. Future studies should integrate multi-dimensional elements—“pollutant types, climate factors, 3D morphology, landscape patterns”—into a dynamic coupled model, offering a robust basis for refined, sustainable urban design.

5. Conclusions

Improving urban air quality and thermal comfort represents a critical nexus for improving public health and achieving urban sustainability. While the discrete effects of urban form on air quality and thermal environments are well-documented, a comprehensive understanding of their complex, seasonally dependent interactions is relatively limited. In this study, focusing on the downtown area of Taiyuan City, we selected typical polluted winter days and hot summer days, constructed an integrated dataset of PM2.5 concentrations, LSTs and UFIs based on multi-source data, and systematically explored the integrated mechanism of the urban form on PM2.5 concentrations and LSTs by using multivariate logarithmic regression with mediated effects model.

This study concludes that LST acts as a critical and seasonally reversing mediator in the relationship between urban form and air quality. FAI and SVF contributed 79.17% and 43.50% of the PM2.5 concentration reduction effect through the mediating effect of LST, respectively, while DEM partially mediated its enhancement effect on PM2.5 through the upward effect of LST by 44.26% on the high temperature days in summer. On the winter pollution day, FAI and BD reduced PM2.5 concentration by 37.46% and 39.51%, respectively, through the cooling effect of LST, while SVF partially mediated its negative effect on PM2.5 through the LST with 29.43%, and the mediating effect of DEM accounted for 20.58% of the total effect, indicating that LST partially mediated the increasing effect of DEM on PM2.5.

In addition, the sub-functional area analysis further reveals the key variability in the heat-pollution response chain in different urban functional areas. While FAI, SVF, and DEM were consistently identified as the most influential morphological drivers overall, their specific impacts and relative importance vary significantly by zone. This underscores the conclusion that there is no “one-size-fits-all” morphological solution; effective strategies must be tailored to the specific context of each functional area.

In summary, this study reveals that the urban form–thermal environment–air pollution response chain is highly heterogeneous, contingent upon the specific spatial structure and anthropogenic activities of different functional zones. The influence of dominant morphological factors varies not only in magnitude across these zones but, more importantly, also exhibits seasonal reversals in the direction and nature of their regulatory mechanisms on LST and PM2.5. These findings provide scientific support for the paradigm shift towards seasonally-adaptive zoning regulations and fine-grained urban management, emphasizing that differentiated morphological optimization strategies are essential for enhancing the synergistic thermal and air quality performance of cities (

Table 4. Planning strategy for functional areas.

Urban Functional Zones	Key UFIs	Control Strategy and Layout Recommendation
RZs (Residential Zones)	FAI, MBH, BD	Limit BD and MBH, optimize building spacing; avoid large enclosures
IZs (Industrial Zones)	SVF, DEM	Increase SVF, choose higher DEM, use topography to create wind tunnels
CSZs (Commercial Service Zones)	SO, BD	Adjust SO and BD, avoid deep street canyons
PSZs (Public Service Zones)	FAI, DEM	Enhance and utilize topography-driven ventilation

).