Next Article in Journal
Interval Determination Strategy for Bayesian Inversion of Seismic Source Parameters Under Uncertain Interval Conditions
Previous Article in Journal
Cascaded Ambiguity Resolution for Pseudolite System-Augmented GNSS PPP
Previous Article in Special Issue
Quantifying the Rate and Extent of Urbanization Effects on Vegetation Phenology in Mainland China
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 17
Issue 18
10.3390/rs17183150
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Seasonal Effects of Urban Morphology on the Thermal Environment Based on Automated Machine Learning: A Case Study of Beijing

by
Ni Wang
1,2,
Lidu Shen
1,*,
Wenli Fei
1,
Yage Liu
1,
Hujia Zhao
3,
Luyao Liu
1,2,
Anzhi Wang
1 and
Bao-Jie He
4
1
Key Laboratory of Forest Ecology and Management, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110016, China
2
University of Chinese Academy of Sciences, Beijing 100049, China
3
Institute of Atmospheric Environment, China Meteorological Administration, Shenyang 110166, China
4
School of Architecture and Urban Planning, Chongqing University, Chongqing 400045, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2025, 17(18), 3150; https://doi.org/10.3390/rs17183150
Submission received: 8 July 2025 / Revised: 2 September 2025 / Accepted: 5 September 2025 / Published: 11 September 2025
(This article belongs to the Special Issue Remote Sensing Applications in Urban Environment and Climate)
Browse Figures
Versions Notes

Abstract

Highlights

What are the main findings?
  • Explainable analysis identifies NDVI as the most influential indicator of LST variations, followed by BH with a cooling effect and BCR with a positive impact.
  • 3D indicators exert stronger direct and total effects than 2D indicators within the causal pathways, except in winter.
What is the implication of the main finding?
  • Study provides an AutoML-SHAP framework with structural equation modeling for investigating nonlinear relationships.
  • These results offer seasonally adaptive strategies for surface UHI mitigation.

Abstract

Understanding the seasonal nonlinear relationship between urban heat island (UHI) and multidimensional urban morphological patterns is crucial for regulating the urban thermal environment. To address this, this study quantified the contributions and sensitivities of urban morphology to land surface temperature (LST) variations and revealed their influencing pathways across four seasons in Beijing, using automated machine learning, SHapley Additive exPlanations interpretation, partial dependence analysis, and structural equation modeling. The results showed significant seasonal variations at the grid scale of 200 m. It was revealed that Normalized Difference Vegetation Index (NDVI) emerged as the most significant indicator affecting LST, followed by building height (BH) and building coverage ratio (BCR), while sky view factor and frontal area index had the least impact. BH was more influential than NDVI, affecting LST during winter. Additionally, sensitivity analysis revealed that impervious surface area, BCR, and mean building volume had positive relationships with LST. In contrast, NDVI and BH negatively affected LST with a noticeable cooling effect, particularly in summer. Furthermore, the total effects of all indicators on LST were negative, with the greatest in spring and the least in winter. Three-dimensional indicators generally exhibited more pronounced direct and total effects than two-dimensional indicators, except in winter. These findings can offer valuable insights for regulating seasonal surface UHI to maximize thermal environmental benefits.

1. Introduction

Rapid urban expansion has substantially transformed urban surfaces, resulting in higher temperatures in urban areas compared to surrounding rural regions, namely the urban heat island (UHI) effect [1,2,3,4,5]. As global urbanization is projected to rise from 55% in 2020 to 68% by 2050, the UHI effect is expected to intensify in the future [6,7]. UHI can significantly alter heat regimes in cities, exacerbating heat waves in summer and decreasing cold waves in winter [8,9,10,11]. These shifts in temperature regimes have profound impacts on human health, energy consumption, and air quality in urban environments [12,13,14]. Therefore, understanding the mechanisms and influencing factors of the thermal environment is crucial for developing effective urban planning strategies towards sustainable development.
Urban morphology, which is characterized by the horizontal configuration and vertical structure of built-up surfaces, can significantly alter urban energy balance and thus influence LST variability. Previous studies mainly focused on two-dimensional (2D) urban morphology indicators, such as impervious surface area (ISA) and green space, particularly driven by horizontal urban expansion [15], and found these 2D indicators largely determine the spatial distribution of surface temperature. For example, the proportion of ISA was found to be strongly positively correlated with LST, demonstrating significant warming effect [16,17]. On the contrary, urban green space has a notable cooling effect, due to enhanced evapotranspiration [18,19].
When considering urban vertical structure, the impacts on LST are more complicated. Buildings can alter radiation transfer, heat storage, and air circulation within the city, thereby concentrating urban heat [20,21]. Due to difficulties in accessing building data at a large scale, the relationships between land surface temperature (LST) and three-dimensional (3D) urban morphology indicators have been less investigated. Until recently, with the progress of remote sensing observation and open-access building data, the impacts of 3D urban morphology indicators, including building height (BH), sky view factor (SVF), and building volume (BV), on LST have been revealed [22,23]. Studies have indicated that BH and Building Density (BD) exhibit seasonally stable influence on LST, BH is revealed to be negatively correlated with LST, while BD is positively correlated [24,25]. Moreover, comparative studies of 2D and 3D indicators have shown varying impacts on LST with some findings indicating that 2D indicators generally play a more significant role in UHI modulation across different areas [26]. In contrast, other research highlights that 3D morphology indicators have a greater impact on LST than 2D indicators, with compact 3D urban morphology tending to reduce LST [27,28,29]. Three-dimensional building–vegetation integrated morphology feature has also demonstrated strong explanatory power [30]. The comparison of 2D and 3D indicators’ impact on LST may also show seasonal variations [31,32,33]. It has been revealed that 2D indicators shows a greater impact in warmer seasons, while 3D indicators influence LST more in cooler seasons [34]. Therefore, these findings highlight the necessity of investigating the seasonally varying relationship between urban morphology and LST.
Quantifying the impact of urban morphology on LST in the highly heterogeneous city is challenging. Previous studies have predominantly adopted statistical models such as correlation analysis and multiple regression [35,36]. These approaches often fail to fully address the inherent nonlinearity of these interactions [37]. Machine learning (ML) methods, such as random forest and enhanced regression trees, offer advantages by capturing these nonlinear relationships, which have been widely used in LST studies [38,39,40]. However, developing a machine learning model often presents significant challenges, particularly in choosing the appropriate algorithm and optimizing hyperparameters [41]. The state-of-the-art automated machine learning (AutoML) model provides a promising solution to address this challenge by integrating multiple models and automatically optimizing hyperparameters for LST prediction [42,43].
Despite machine learning models demonstrating strong predictive performance, their inherent complexity often obscures the underlying relationships between input variables and model outputs, limiting interpretability [44,45]. The development of SHapley Additive exPlanations (SHAPs) has significantly advanced model interpretability in machine learning [46]. By applying concepts from cooperative game theory, SHAP provides mathematically rigorous feature attributions, enabling both global model interpretation and local prediction explanations [47]. Recently, this approach has been applied in urban thermal studies [48]. For example, SHAP has been combined with Random Forest (RF) to analyze the impact of adjacent street morphological features on LST [49], while other studies have integrated the Extreme gradient boosting (XGBoost) model with SHAP to explore the relationships between multi-dimensional urban morphology and LST [50].
Nevertheless, few studies have employed an integrated AutoML-SHAP framework to systematically investigate the interactive influences of urban morphology on LST. Additionally. This framework not only has an excellent ability to capture a more nuanced understanding of the contributions of morphology indicators to LST [51], but also can further assess scale dependence [52], and visualize the marginal effects of independent variables with partial dependence plots (PDPs) [53,54]. But these methods cannot fully uncover the pathways through which urban morphology indicators affect UHI. Structural equation models (SEMs) can address this limitation by incorporating intermediate variables to quantify both direct and indirect effects of features on LST, providing a comprehensive understanding of how these effects are mediated through interactions among indicators [55,56].
To address the problems above, this study investigated the seasonal variations in the impacts of urban morphology on LST in urban Beijing based on satellite LST. We first chose the optimal model and grid scale using AutoML. Then, we applied SHAP to identify how the indicators contribute to LST, and discussed how these indicators affect LST across different seasons by Partial Dependence Analysis (PDA). We finally examined how interactions among these indicators influence LST using SEM. This study deepens the understanding of UHI formation and provides references for urban planning to improve the urban thermal environment.

2. Materials and Methods

2.1. Study Area

This study focuses on Beijing, a metropolitan city located in the North China Plain (39°28′–41°05′N, 115°25′–117°0′E) (Figure 1). Beijing has a warm temperate monsoon climate with cold, windy winters and hot, humid summers [57]. The annual temperature ranges from 10 to 12 °C. As the capital and the second largest city in China, Beijing has experienced accelerated urbanization in the 21st century, and heat waves have continued to increase in this region [58]. It has an area of about 16,800 square kilometers and a population of about 22 million. The constructed land is also constantly expanding to accommodate the growing population. This expansion has notably increased impervious surface areas, including roads, buildings, and other infrastructures. The continuous urban expansion would also lead to a pronounced UHI effect in Beijing [59].

2.2. Data

2.2.1. LST Data

We utilized LST data from the Landsat 8 Collection 2 Surface Temperature product. The LST data is derived from the Collection 2 Level 1 Thermal Infrared Sensor band 10, based on the single-channel algorithm version 1.3.0, which meets geometric and radiometric quality requirements at a spatial resolution of 30 m [60]. The LST product is accessed and processed through the Google Earth Engine (GEE) platform [61,62]. To better quantify the UHI effect and ensure the reliability of the seasonal LST, we specifically chose the images with less than 5% cloud coverage and controlled for pixel coverage greater than 80% to minimize the impact of clouds and missing pixels during 2020–2022. Within this dataset, we computed the average LST for each season: spring (March to May), summer (June to August), autumn (September to November), and winter (December to February). Then, we selected the range of 1–99% of the images to minimize the influence of outliers and to provide a clearer view of seasonal LST variations.

2.2.2. Urban Morphology Indicators

Urban morphology contains complex information about the horizontal and vertical structure of the built environment. Based on previous studies, we selected seven urban morphology indicators that have been found to have the most critical influence on LST [19], including 2D: ISA, building coverage ratio (BCR), Normalized Difference Vegetation Index (NDVI) and 3D: BH, SVF, frontal area index (FAI) and mean building volume (MBV). The ISA, BCR, BH, SVF, and FAI were from a national urban morphology dataset in 2020 [63,64]. The dataset is developed through a two-step rasterization step from building footprint data with height. With a spatial resolution of 100 m, this dataset is suitable for LST studies. In addition, we calculated the MBV using BH and BCR as follows:
MBV = ∑ i = 1 n BCR   ×   BH   ×   A BN
where A is the area of the grid scale, and BN is the building number within a grid.
We also calculated the NDVI for attribution analysis using the Landsat images from the same period as the LST data by season, as follows:
NDVI = NIR − R NIR + R
where NIR and R are near-infrared band (NIR) and the visible red band (R), namely SB4 and SB5 in the Landsat 8 Collection 2 product, respectively.
Then, we conducted a Variance Inflation Factor (VIF) analysis to assess multicollinearity. All morphology indicators showed VIF values below 10, indicating sufficient non-collinearity and enhancing the robustness of our subsequent analysis [65]. The information on the morphology indicators used in this study is summarized in Table 1.

2.2.3. Other Data

Land cover data were used to mask urban morphology data to exclude the influence of non-urban surfaces, especially water bodies. Yearly land cover data was acquired from the European Space Agency (ESA) World Cover 10 m 2020 product [66]. The product provides a global land cover map for 2020 with a 10 m resolution and 11 land cover classes derived from Sentinel-1 and Sentinel-2 data. We resampled the data to 100 m and masked the morphology indicators to remove the non-urban pixels.

2.3. Method

The framework used in this study includes four steps (Figure 2): (1) constructing an AutoML model to select the optimal scale and fit the model; (2) quantifying the contributions of building indicators to LST based on AutoML-SHAP interpretation; (3) analyzing the marginal effects of urban morphology indicators on LST using PDPs; and (4) understanding the direct and indirect effects of indicators on LST through a pathway analysis using SEM. To reveal the seasonal effect, we did these analyses for each season.

2.3.1. Constructing an AutoML Model to Select the Optimal Scale

We used AutoML to identify the best grid scale for the subsequent analysis. Compared to the traditional ML methods that require manual parameter setting, AutoML can automate the process of building and optimizing ML models, consisting of algorithm selection and hyperparameter optimization to identify the best algorithm combination that performs optimally in cross-validation from raw datasets to ML models [67,68]. AutoML enhances the efficiency and performance of ML models by reducing human intervention, thus saving time and resources [69].
We employed the H2O AutoML model integrates various common ML algorithms. Since the widespread application of tree-based models in combination with SHAP has demonstrated significant advantages in this field of study [28,39,45,52], we selected Gradient Boosting Machines (GBMs), Extreme Gradient Boosting (XGBoost), and Distributed Random Forests (DRFs) as candidate models in AutoML to enable SHAP interpretation of predictions in this study [70]. The dataset was randomly divided into a 70% training set and a 30% testing set using random sampling. Then we used the AutoML functionality of H2O in the R package “h2o” (version 3.42.0.2) with the above seven urban morphology indicators as independent variables and LST as the dependent variable [71,72]. During the training process, the H2O AutoML ran ten kinds of machine learning algorithms simultaneously for over 1200 s based on the 5-fold cross-validation results.
Using the AutoML method, we developed a model for each season. Models were ranked based on model evaluation indicators, including Root Mean Square Error (RMSE) and Mean Squared Error (MSE) [73]. Using the best-performing GBM model, which with optimal Parameters, we tested the scale dependence of the developed model at five grid scales ranging from 100 m to 500 m with an interval of 100 m. Model performance metrics, including RMSE and Coefficient of Determination (R2), were calculated based on predictions from the testing set [43]. Additionally, we compared these performance metrics with those of RF to further validate the high accuracy of AutoML.

2.3.2. Quantifying the Contributions of Urban Morphology Based on SHAP

Based on the AutoML results, we employed the SHAP interpreter to assess the contributions of urban morphology indicators to LST, aiming to identify the key indicators affecting predictions. SHAP is a game theory-based framework that explains the output of any machine learning model [74]. The method calculates the SHAP values to quantify the contribution of each feature to the prediction for individual instances. Furthermore, SHAP provides global interpretability by aggregating the contributions of each feature across all instances in the dataset [28]. SHAP values numerically allocate feature contributions, where positive values indicate warming effects and negative values represent cooling effects on LST [75].
In this study, the SHAP values, calculated using h2o.shap_summary_plot() functions [76], represent the importance of each feature influencing the average LST.

2.3.3. Quantifying Marginal Effects Between LST and Urban Morphology Indicators

We quantified the sensitivity of LST to each morphology indicator using partial dependence plots based on AutoML results to analyze the marginal effect of various indicators on LST. The marginal effect measures the independent impact of each variable, considering the average effects of other variables while keeping other covariates constant [20]. The response results were centralized to compare the different responses of a single indicator to LST in different seasons: the greater the absolute value, the greater the impact. This approach provides an in-depth understanding of the mitigation effect and degree in a quantitative way, and reveals the dynamic nonlinear relationship between LST and urban morphology, so as to optimize its pattern and alleviate the thermal environment to the greatest extent [53,77].

2.3.4. Quantification of the Direct and Indirect Causal Relationships

We finally used the structural equation modeling (SEM) to reveal the direct and indirect causal relationships between LST and urban morphology indicators and visualize the influencing pathway. SEM is a multivariate analysis method based on probabilistic statistics [78]. It explores causal relationships between indicators by analyzing the covariance matrix of these variables, combining measurement models and structural models [56,79]. The modeling pathways allow simultaneous quantification of direct and indirect causal relationships among multiple variables [31,80].
We used the “lavaan” package in R to establish SEMs for each season. Initially, we established the theoretical pathways for the SEM based on the literature, which defines the expected relationships between LST and morphology indicators. Then, the pathways adjustment was modified according to the model fitting indices and expert judgment. In order to evaluate the model fit, we used the following parameters: Goodness of Fit Index (GFI) > 0.90; Comparative Fit Index (CFI) > 0.90; and Root Mean Square Error of Approximation (RMSEA) < 0.10 [55,81]. Through SEM analysis, we identified the complex interactions among these indicators [82].

3. Results

3.1. Spatial Patterns of Urban Morphology Indicators

Urban morphology indicators present obvious spatial variations in the study area (Figure 3). The BCR and ISA were high in the central region and low around, but the ISA had larger values because ISA contains BCR as well as other impervious surfaces, such as roads (Figure 3a,b). The average BCR and ISA were 0.21 ± 0.15 (mean ± standard deviation; similarly hereinafter) and 0.81 ± 0.21, respectively. The spatial distributions of BH, FAI, and MBV were similar, with the urban–rural area clustered (Figure 3c–e), mainly related to the height of buildings. BH varied from 3 m to 220 m, with an average of 18.90 ± 15.90 m, whereas FAI and MBV had an average of 0.16 ± 0.16 and 7.60 ± 13.30 km3, respectively. The spatial distribution of SVF was opposite to that of BH and FAI, with higher values in the outer periphery than in the urban center, due to the high degree of occlusion of buildings and low sky visibility. The mean SVF was 0.65 ± 0.26 (Figure 3f). The mean NDVI values for spring, summer, autumn, and winter were 0.25 ± 0.10, 0.39 ± 0.16, 0.32 ± 0.14, and 0.15 ± 0.07, respectively. In Beijing’s core area, the NDVI was lower, while it was higher around the periphery, with this pattern being particularly evident in the summer season (Figure 3g–j).

3.2. Spatial Patterns of LST Across Four Seasons

Spatially, the LST presented evidence of urban–rural differences and seasonal variations (Figure 4). Generally, LST showed a concentric pattern with high values in the middle and low in the periphery, namely a significant UHI effect. The average LST was 30.88 ± 2.15 °C in spring, 39.24 ± 2.78 °C in summer, 21.08 ± 1.45 °C in autumn, and 2.51 ± 1.25 °C in winter, respectively. There was a similarity between the distribution of LST and BH. The maximum spatial variation in LST occurred in summer (11.80 °C), while the minimum occurred in winter (5.60 °C), indicating stronger UHI in summer and weaker UHI in winter. The spatial patterns of LST in spring and autumn were similar with relatively uniform variation, but the overall LST in spring was higher than that in autumn. Specifically, the LST difference between urban and suburban areas still existed. The highest LST was primarily concentrated in the central urban area, with numerous dense buildings and residential areas. The urban fringe was not fully urbanized, resulting in slightly lower LST.

3.3. The Optimal Model and Scale for Urban Morphology in Modeling LST

We selected the best grid scale depending on performance metrics, including RMSE and R2 of the stacked ensemble model. The comparative analysis of seasonal prediction models at different scales revealed that the gradient boosting machine approach consistently outperformed other models, demonstrating superior predictive accuracy. The top five models were GBMs with different hyperparameter settings (Table 2). The models were ranked in descending order based on the selected evaluation indicators (RMSE and MSE).
The performances of the optimal AutoML model (GBM) varied broadly across grid scales from 100 to 500 m for all seasons (Figure 5). The general trend of R2 was that the larger the grid scales, the higher the modeling accuracy when the grid size was less than 200 m. As the grid size increased, the modeling accuracy might continue to improve in summer but decline in spring and winter. Specifically, the maximum value of R2 was 0.48 at 200 m in spring, 0.58 at 400 m in summer, 0.40 at 200 m in autumn, and 0.26 at 200 m in winter, respectively (Figure 5a). The seasonal ranking of R2 was summer > spring > autumn > winter. Meanwhile, the fluctuation amplitude of RMSE was largest in summer (Figure 5b). The minimum RMSE was 1.30 °C at 200 m in spring, 2.10 °C at 200 m in summer, 1.08 °C at 200 m in autumn, and 0.93 °C at 400 m in winter. The seasonal ranking of RMSE was summer > spring > autumn > winter, consistent with that of R2. R2 reached its maximum at 200 m in spring and winter, and RMSE also reached its minimum at 200 m in all seasons.
Generally, the AutoML model had a higher R2 and lower RMSE compared to the RF model, indicating that AutoML provides a stronger explanatory power. The RF model also achieved the highest accuracy at 200 m scale in spring, autumn, and winter, but their accuracy was much smaller than that of the AutoML model (Table 3). Additionally, given that the raw resolution of LST and morphological dataset were all less than 100 m, applying a lower grid scale tends to reduce the uncertainties during interpolation and help to keep the spatial information of the data. Considering all these points, we chose the AutoML model developed at the grid scale of 200 m for the following analyses.

3.4. Contributions of Urban Morphology Indicators to LST

Figure 6 displays the seasonal contributions of various urban morphology indicators on LST through the summary plot, revealing substantial variations in their effects across different seasons.
Firstly, it can be concluded that NDVI was the most critical indicator affecting LST in all seasons, followed by BH and BCR, while SVF and FAI typically ranked the lowest. On the other hand, due to the larger contributions of NDVI and BCR, 2D indicators demonstrated greater importance compared to 3D indicators without considering the influencing direction, except during winter. Furthermore, the ranking of the importance of indicators in spring and autumn exhibited similarities, although the least influential indicators were different.
Specifically, in spring, NDVI and BH were the dominant indicators of LST (Figure 6a). High values of NDVI and BH corresponded predominantly to negative SHAP values (Figure 6b), suggesting a cooling contribution, while high BCR and MBV exhibited a weak positive contribution. FAI and SVF exhibited narrower value distributions and had varying directional contributions.
During summer, NDVI emerged as the most influential indicator, followed by BCR and ISA (Figure 6c). High values of BCR and ISA were strongly associated with positive SHAP values (Figure 6d), reflecting the pronounced warming effects of artificial surfaces under lush seasonal conditions. Whereas high BH values had a slight cooling effect.
By autumn, NDVI and BH regained dominance, with BCR as the third most important indicator (Figure 6e). The influence of ISA diminished. The contribution direction of NDVI and BH remained positive yet less intense compared to spring, indicating a consistent but moderated impact (Figure 6f). FAI and SVF had the least influence in summer and autumn.
In winter, BCR and BH became the primary drivers, followed by SVF and MBV, while the influence of NDVI greatly diminished. (Figure 6g). High BCR and SVF values largely contributed to warming. Moreover, FAI exhibited a clear directional effect that higher FAI values were associated with lower LST, while the contribution of SVF was the largest in winter among all four seasons.

3.5. The Seasonal Sensitivity of UHI to Urban Morphology Indicators

The partial dependence graph (Figure 7) further quantified the distinct seasonal sensitivity of LST to various indicators. Generally, ISA, BCR, and MBV were positively associated with LST, while NDVI and BH negatively affected LST. NDVI, BCR, and BH exerted the greatest impact on LST variations (Figure 7a–c), followed by ISA and MBV, while the indicators with the smallest marginal effect were SVF and FAI (Figure 7e,f), which was consistent with the relative contribution results.
Specifically, there was a robust negative relationship between NDVI and LST variation. NDVI exhibited a significant cooling effect when NDVI exceeded 0.4, especially in summer. Lush vegetation drives strong cooling through high transpiration, effectively decreasing LST. The maximum LST decrease in summer was about 4 °C. In spring and autumn, a negative correlation existed, but with a gentler slope than in summer. Weaker vegetation coverage and transpiration capacity during these seasons diminished the regulatory effect on LST. NDVI had minimal impact on LST in winter.
The mean response of ISA was nearly linear with minor fluctuations, showing an upward trend in different seasons (Figure 7b). The LST variation was largest in summer, and became weaker in other seasons. ISA values below 0.5 had a cooling effect, while the values above 0.5 contributed to a warming effect. This indicates that higher ISA amplifies urban heat, especially during warmer months.
Although BCR exhibited a nonlinear marginal effect, there was a strong positive relationship between BCR and LST in all seasons (Figure 7c). The variation curve initially remained flat before rising sharply, and then leveled off. Low BCR values had a cooling effect, especially in summer (below 0.50). However, the warming effect intensified as BCR exceeded this threshold, and then the LST variation essentially plateaued as BCR approached around 0.70. This suggests that lower building density helps reduce LST, while higher density contributes to increasing LST.
BH had a clear cooling effect, with a gradual decrease in LST as BH increased (Figure 7d). The turning point occurred at approximately 50 m, where BH began to reduce LST, stabilizing at heights above 60 m. The seasonal response was consistent, except for lower BH in summer, indicating that taller buildings reduced LST, while shorter buildings increased LST.
The response of MBV increased first and then leveled off (Figure 7e). When the MBV was less than 50 km3 in summer and 46 km3 in spring and autumn, a weak cooling effect was observed. In contrast, MBV had the least impact on LST during winter. FAI and SVF showed minimal influence on LST, with weak seasonal fluctuations. In all seasons, their impact on LST was less than 0.5 °C, indicating that FAI and SVF played a minor role in UHI dynamics compared to other indicators. (Figure 7f,g).

3.6. The Seasonal Influencing Pathways of Urban Morphology Indicators on LST

We used SEM analysis to assess how urban morphology directly and indirectly impacts LST, revealing significant seasonal variations in these influencing pathways (Figure 8). The pathways in SEM produced normalized total effects and explained 0.40 (Figure 8a), 0.57 (Figure 8b), 0.35 (Figure 8c), and 0.19 (Figure 8d) of the variances in LST for spring, summer, autumn, and winter, respectively, indicating the model had the best explanatory power in summer. In terms of direct effects, NDVI exerted the strongest direct effect on LST, followed by BH and BCR, while ISA and MBV had a weak effect. Specifically, NDVI displayed the highest path coefficient in winter (0.10) and the lowest in summer (−0.59), indicating its dominant cooling role in summer and warming effect in winter. BCR exhibited a positive direct effect on LST, indicating a warming influence, with its strongest impact observed in spring. Conversely, BH demonstrated a significant and consistent cooling effect, which was most pronounced in winter (−0.40) (Figure 8d). ISA and MBV exhibited relatively weak direct effects. SVF displayed seasonally divergent roles; it exerted a minor warming effect in spring and winter but a cooling effect in summer and autumn, with its strongest cooling influence observed in summer (−0.23). FAI showed a significant negative effect, particularly in spring (−0.18). Additionally, regarding indirect effects, BCR and NDVI influenced LST indirectly through ISA. BH also indirectly affected LST through its substantial influence on SVF and MBV.
The total effects of all indicators on LST were overall negative (Figure 9), with −0.64 (Figure 9a), −0.53 (Figure 9b), −0.49 (Figure 9c), and −0.02 (Figure 9d), respectively. This indicates an overall mitigating influence of the urban morphology on LST, with the greatest collective cooling effect observed in spring and the weakest in winter. The total effect of 3D indicators generally demonstrated more pronounced direct and total effects than that of 2D indicators, except during winter. Specifically, the total effect of 3D indicators was most pronounced in autumn (−0.53) and less pronounced in winter (−0.25). In contrast, 2D indicators showed a weak negative total effect in spring and summer but transitioned to a positive effect in autumn and winter. The direct effects of all indicators were greater than the indirect effects, except in winter, where 2D indicators showed a strong positive direct effect (0.27) on LST but a weak indirect positive effect (−0.04), leading to the total effect being positive in winter (Figure 9d). Additionally, both 2D and 3D building structures demonstrated stronger direct effects than indirect effects, except in autumn.

4. Discussion

4.1. Influencing Mechanism Across Seasons

Urban morphology affects the local thermal environment by altering surface energy balance [83]. Solar radiation is the primary energy source for the formation of LST. Both its intensity and angle exhibit significant seasonal variations, thus affecting the urban morphology’s impact on LST. Our interpretable AutoML-SHAP framework provides robust insights into these seasonal mechanisms.
It was revealed that NDVI and BH consistently emerged as the most influential factors across seasons. NDVI exerted a prominent cooling effect, attributable to enhanced evapotranspiration, while its effect slightly reversed in winter due to reduced solar radiation and physiological activity. BH also played a cooling effect at values greater than 50 m, as higher BH can cast more shadows and shield more solar radiation. Meanwhile, a larger height difference makes it easier to form vents, which is conducive to heat dissipation, resulting in lower LST [53]. Beijing’s low winter solar elevation angle maximizes these shading benefits, explaining why BH was the most pronounced indicator during cold seasons. In contrast, the solar elevation angle is high in summer, leading to a minimal impact on shading.
Furthermore, our results suggested that SVF and FAI were the indicators with the smallest contributions and exhibited varying directional effects. Previous studies have shown that the impact of sky on LST can be positive, negative, or neutral, depending on the dominant processes [84,85]. In this study, SVF showed the strongest negative effect on LST in summer, yet a slight positive impact on LST in winter. This is because areas with high SVF receive more solar radiation but also benefit from better air circulation and heat dissipation [55,86]. Conversely, in areas with low SVF, heat becomes trapped under tree canopies and buildings, exacerbating the UHI effect. FAI affects LST by influencing wind patterns and heat diffusion. We observed that the contribution rate of FAI was higher in spring and winter than in summer and autumn. This is likely because the strong northwesterly winds in spring and winter mean that the long sides of buildings face the wind. High-FAI building clusters may enhance the penetration of cold air, thereby lowering local temperatures [84].
Notably, 2D indicators primarily affect LST through changing the energy allocation. Quantitative assessments revealed that ISA and BCR were positively associated with LST, which is in line with previous studies [20]. These effects were most pronounced in summer, when the shading’s impact becomes smaller and horizontal features more directly affect LST as the increase in solar altitude angles. Impervious surfaces, characterized by materials such as asphalt and concrete, significantly modify the local microclimate. These artificial surfaces exhibit low albedo and minimal water retention capacity, playing a warming effect, which collectively reduces evapotranspiration rates while enhancing heat storage and sensible heat flux. Conversely, green spaces increase evapotranspiration, playing a cooling effect [87,88].
Moreover, findings from SEM indicated that the overall effect of 3D indicators was greater than that of 2D indicators, with direct effects outweighing indirect ones, except in winter. The advantage of 3D morphology indicators over 2D is consistent with previous research results [32]. This may be because 3D indicators directly affect LST by altering the physical structure of the urban microclimate, while 2D indicators may more indirectly influence LST by changing surface cover and land use types. The 2D indicators primarily involve a trade-off between the warming effects of BCR and ISA and the cooling effect of NDVI. Except in winter, NDVI generally has a more substantial contribution. However, NDVI has a positive impact on LST in winter, and when combined with the effects of BCR, their collective influence can surpass the direct impact of 3D indicators.

4.2. Implications for Urban Planning

Our findings on the seasonally divergent roles of 2D and 3D urban morphology provide valuable insights for urban planning to regulate UHI. Effective urban planning must consider the specific thermal characteristics of each season and adopt strategies that maximize the thermal benefits.
In summer, where our results identified BCR and ISA as the dominant drivers of LST increase, planning strategies should prioritize mitigating the warming effect of horizontal surfaces [39]. Given Beijing’s limited available land, it is suggested to rationally distribute urban green space, such as vertical greening and rooftop greening to break the concentration of the impervious surfaces in high-density building clusters. Dispersed building layouts with taller structures can also alleviate the mitigation of heat accumulation [89]. Additionally, optimizing street canyon geometry to maintain adequate sky view can facilitate longwave radiative cooling at night, while ensuring that building spacing does not excessively hinder ventilation.
In contrast, in spring and autumn, the seasons’ NDVI and BH emerged as influential cooling factors. The focus should shift to balancing shading and ventilation. BH can be utilized to provide shading while maintaining reasonable height differences to promote ventilation. Areas with high BCR should be prioritized for summer-focused interventions (greening, cool materials), whereas low-rise areas may benefit from redevelopment introducing taller, well-spaced structures to promote cooling in spring and autumn. Additionally, maintaining reasonable FAI can control airflow and promote heat diffusion [53].
In winter, greater emphasis should be placed on the influence of 3D indicators. Taller buildings can provide solar shading and reduce heat loss, potentially improving thermal comfort [51]. Enhancing building insulation with high-performance materials reduces heat loss and improves comfort. Seasonally adaptive green infrastructure and optimized building orientation to maximize solar absorption are essential for natural heating.
Overall, this study underscores that effective UHI mitigation requires a seasonally adaptive and spatially tailored planning strategy. A comprehensive approach integrates building form, green spaces, material selection, and ventilation. Ultimately, these tailored measures can effectively mitigate UHI and create more livable, sustainable cities.

4.3. Uncertainties

While this study provides valuable insights into the seasonal dynamics of urban heat islands, several limitations should be acknowledged, which also present opportunities for future research. First, we used LST to quantify the thermal environment in this study, as the remote sensing-based LST data is easy to access. However, LST can sometimes be poorly associated with air temperature and moist heat stress, which are more relevant for heat exposure [90]. To better inform policy makers for urban heat regulation, future research should prioritize the integration of high-resolution air temperature data and incorporate the influence of anthropogenic heat emissions on human thermal comfort. Second, the precision of our model is constrained by the accuracy of the urban form data we used and our selection of indicators. Besides morphology, other factors, such as anthropogenic heat emissions, can also affect UHI, potentially introducing disturbances into the results [25]. Third, although we considered the optimal scale, some studies chose block-scale or local climate zones as the research unit [19,91]. Appropriate research area and research unit size are crucial to the study. Fourth, this study was limited to a single city without accounting for different climatic environments and social development conditions. Variations in climate and architectural structures may lead to different outcomes in other cities. It is necessary to further explore multiple cities with various climate backgrounds [10]. Finally, this study combined AutoML and SEM to quantify the contributions and visualize the influencing pathways, but the two approaches remain separate at this point. Future study needs to merge SEM with AutoML to establish an advanced hybrid model framework that leverages the strengths of both models to enhance predictive performance.

5. Conclusions

This study integrated a framework combining AutoML, SHAP, PDA, and SEM to systematically quantify the contributions and sensitivities of urban morphology to LST, and revealed distinct seasonal pathways by which urban morphology indicators influence LST in the city of Beijing, using LST data from the Landsat 8 Collection 2 during 2020–2022. The findings from this study are summarized as follows:
The results indicated that the AutoML model outperformed the RF model in accuracy. The GBM model was identified as optimal, with 200 m confirmed as the best grid scale for capturing seasonal dynamics.
Furthermore, NDVI emerged as the most critical indicator affecting LST in all seasons, followed by BH and BCR, while SVF and FAI had the least impact. Additionally, 2D indicators demonstrated greater importance compared to 3D indicators, regardless of the influencing direction, except during winter.
Sensitivity analysis revealed ISA, BCR, and MBV had positive relationships with LST. In contrast, NDVI and BH negatively affected LST. A noticeable cooling effect was observed for BH when values exceeded 50 m, and for NDVI when values surpassed 0.4. SVF and FAI had minimal impact on LST variations. The sensitivity gradients confirmed that lower building density combined with taller structures most effectively reduces LST.
Overall, path analysis showed that the total effects of all indicators on LST were negative, with the greatest impact in spring (−0.62) and the smallest in winter (−0.02). The direct effects of both 2D and 3D indicators were generally greater than their indirect effects, except in autumn. The total effects and direct effects of 3D indicators were larger than those of 2D indicators, except in winter.
The findings underscore the significant seasonal variation in the role of urban morphology in influencing LST, which provides support and recommendations for urban planning strategies aimed at improving the urban thermal environment regarding its seasonal variation.

Author Contributions

Conceptualization, N.W. and L.S.; methodology, W.F. and N.W.; data curation, N.W.; validation, N.W.; visualization, N.W. and L.L.; writing—original draft preparation, N.W.; writing—review and editing: L.S., Y.L., H.Z. and B.-J.H.; supervision, L.S. and W.F.; project administration, A.W. and L.S.; funding acquisition, W.F., Y.L. and H.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (42305205), the Natural Science Foundation of Liaoning Province (Grant Nos. 2023-BSBA-312 and 2023-BSBA-319), and the National Science and Technology Major Project (SQ2025AAA090080).

Data Availability Statement

The data presented in this study are available on request from the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
2DTwo-dimensional
3DThree-dimensional
AutoMLAutomated machine learning
BCRBuilding coverage ratio
BHBuilding height
BNBuilding number
BVBuilding volume
CFIComparative Fit Index
DLDeep Learning
DRFDistributed Random Forests
ESAEuropean Space Agency
FAIFrontal area index
GBMGradient Boosting Machines
GEEGoogle Earth Engine
GFIGoodness of Fit Index
GLMGeneralized Linear Models
IPCCIntergovernmental Panel on Climate Change
ISAImpervious surface area
LSTLand surface temperature
MBVMean building volume
MLMachine learning
MSEMean Squared Error
NDVINormalized Difference Vegetation Index
NIRNear-infrared band
OLSOrdinary Least Squares
PDAPartial Dependence Analysis
RRed band
R2Coefficient of Determination
RMSERoot Mean Square Error
RMSEARoot Mean Square Error of Approximation
SEMStructural equation modeling
SHAPSHapley Additive exPlanations
SVFSky view factor
UHIUrban heat island
VIFVariance Inflation Factor
XGBoostExtreme Gradient Boosting
XRTExtreme Random Trees

References

  1. Cao, C.; Lee, X.; Liu, S.; Schultz, N.; Xiao, W.; Zhang, M.; Zhao, L. Urban heat islands in China enhanced by haze pollution. Nat. Commun. 2016, 7, 12509. [Google Scholar] [CrossRef]
  2. Chakraborty, T.; Lee, X. A simplified urban-extent algorithm to characterize surface urban heat islands on a global scale and examine vegetation control on their spatiotemporal variability. Int. J. Appl. Earth Obs. Geoinf. 2019, 74, 269–280. [Google Scholar] [CrossRef]
  3. Kotharkar, R.; Dongarsane, P.; Keskar, R. Determining influence of urban morphology on air temperature and heat index with hourly emphasis. Build. Environ. 2023, 233, 110044. [Google Scholar] [CrossRef]
  4. Manoli, G.; Fatichi, S.; Schläpfer, M.; Yu, K.; Crowther, T.W.; Meili, N.; Burlando, P.; Katul, G.G.; Bou-Zeid, E. Magnitude of urban heat islands largely explained by climate and population. Nature 2019, 573, 55–60. [Google Scholar] [CrossRef]
  5. Zhou, D.; Sun, S.; Li, Y.; Zhang, L.; Huang, L. A multi-perspective study of atmospheric urban heat island effect in China based on national meteorological observations: Facts and uncertainties. Sci. Total Environ. 2023, 854, 158638. [Google Scholar] [CrossRef] [PubMed]
  6. Intergovernmental Panel on Climate Change (IPCC). Climate Change 2021–The Physical Science Basis: Working Group I Contribution to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change; Cambridge University Press: Cambridge, UK; New York, NY, USA, 2023. [CrossRef]
  7. Zhao, L.; Oleson, K.; Bou-Zeid, E.; Krayenhoff, E.S.; Bray, A.; Zhu, Q.; Zheng, Z.; Chen, C.; Oppenheimer, M. Global multi-model projections of local urban climates. Nat. Clim. Change 2021, 11, 152–157. [Google Scholar] [CrossRef]
  8. Yang, M.; Ren, C.; Wang, H.; Wang, J.; Feng, Z.; Kumar, P.; Haghighat, F.; Cao, S.-J. Mitigating Urban Heat Island through Neighboring Rural Land Cover. Nat. Cities. 2024, 1, 522–532. [Google Scholar] [CrossRef]
  9. Zhao, L.; Lee, X.; Smith, R.B.; Oleson, K. Strong contributions of local background climate to urban heat islands. Nature 2014, 511, 216–219. [Google Scholar] [CrossRef]
  10. Dev Roy, S.; Kuffer, M.; Wang, J. Exploring the Influence of Building Morphology on Surface Temperatures: A Multi-City Analysis in Europe. Build. Environ. 2025, 282, 113274. [Google Scholar] [CrossRef]
  11. Zheng, Z.; Zhao, L.; Oleson, K.W. Large model structural uncertainty in global projections of urban heat waves. Nat. Commun. 2021, 12, 3736. [Google Scholar] [CrossRef]
  12. Kim, S.W.; Brown, R.D. Urban heat island (UHI) intensity and magnitude estimations: A systematic literature review. Sci. Total Environ. 2021, 779, 146389. [Google Scholar] [CrossRef] [PubMed]
  13. Li, H.; Sodoudi, S.; Liu, J.; Tao, W. Temporal variation of urban aerosol pollution island and its relationship with urban heat island. Atmos. Res. 2020, 241, 104957. [Google Scholar] [CrossRef]
  14. Massaro, E.; Schifanella, R.; Piccardo, M.; Caporaso, L.; Taubenböck, H.; Cescatti, A.; Duveiller, G. Spatially-optimized urban greening for reduction of population exposure to land surface temperature extremes. Nat. Commun. 2023, 14, 2903. [Google Scholar] [CrossRef]
  15. Yang, Y.; Guangrong, S.; Chen, Z.; Hao, S.; Zhouyiling, Z.; Shan, Y. Quantitative analysis and prediction of urban heat island intensity on urban-rural gradient: A case study of Shanghai. Sci. Total Environ. 2022, 829, 154264. [Google Scholar] [CrossRef]
  16. Dai, Z.; Guldmann, J.-M.; Hu, Y. Thermal impacts of greenery, water, and impervious structures in Beijing’s Olympic area: A spatial regression approach. Ecol. Indic. 2019, 97, 77–88. [Google Scholar] [CrossRef]
  17. Wang, Y.; He, Z.; Zhai, W.; Wang, S.; Zhao, C. How Do the 3D Urban Morphological Characteristics Spatiotemporally Affect the Urban Thermal Environment? A Case Study of San Antonio. Build. Environ. 2024, 261, 111738. [Google Scholar] [CrossRef]
  18. Schwaab, J.; Meier, R.; Mussetti, G.; Seneviratne, S.; Bürgi, C.; Davin, E.L. The role of urban trees in reducing land surface temperatures in European cities. Nat. Commun. 2021, 12, 6763. [Google Scholar] [CrossRef]
  19. Wang, C.; Ren, Z.; Dong, Y.; Zhang, P.; Guo, Y.; Wang, W.; Bao, G. Efficient cooling of cities at global scale using urban green space to mitigate urban heat island effects in different climatic regions. Urban For. Urban Green. 2022, 74, 127635. [Google Scholar] [CrossRef]
  20. Han, D.; An, H.; Wang, F.; Xu, X.; Qiao, Z.; Wang, M.; Sui, X.; Liang, S.; Hou, X.; Cai, H.; et al. Understanding seasonal contributions of urban morphology to thermal environment based on boosted regression tree approach. Build. Environ. 2022, 226, 109770. [Google Scholar] [CrossRef]
  21. Wu, W.-B.; Yu, Z.-W.; Ma, J.; Zhao, B. Quantifying the influence of 2D and 3D urban morphology on the thermal environment across climatic zones. Landsc. Urban Plan. 2022, 226, 104499. [Google Scholar] [CrossRef]
  22. Chen, C.; Bagan, H.; Yoshida, T.; Borjigin, H.; Gao, J. Quantitative analysis of the building-level relationship between building form and land surface temperature using airborne LiDAR and thermal infrared data. Urban Clim. 2022, 45, 101248. [Google Scholar] [CrossRef]
  23. Wu, X.; Zhang, X.; Liao, W.; Chen, G.; Luo, M.; Liu, X. Contrasting 2D/3D Urban Morphology Drivers of Surface and Canopy Temperatures under Different Spatial Scales. Build. Environ. 2025, 283, 113389. [Google Scholar] [CrossRef]
  24. Cao, S.; Cai, Y.; Du, M.; Weng, Q.; Lu, L. Seasonal and diurnal surface urban heat islands in China: An investigation of driving factors with three-dimensional urban morphological parameters. GIScience Remote Sens. 2022, 59, 1121–1142. [Google Scholar] [CrossRef]
  25. Ding, L.; Xiao, X.; Wang, H. Temporal and Spatial Variations of Urban Surface Temperature and Correlation Study of Influencing Factors. Sci. Rep. 2025, 15, 914. [Google Scholar] [CrossRef]
  26. Huang, X.; Wang, Y. Investigating the effects of 3D urban morphology on the surface urban heat island effect in urban functional zones by using high-resolution remote sensing data: A case study of Wuhan, Central China. ISPRS J. Photogramm. Remote Sens. 2019, 152, 119–131. [Google Scholar] [CrossRef]
  27. Tang, G.; Du, X.; Wang, S. Impact Mechanisms of 2D and 3D Spatial Morphologies on Urban Thermal Environment in High-Density Urban Blocks: A Case Study of Beijing’s Core Area. Sustain. Cities Soc. 2025, 123, 106285. [Google Scholar] [CrossRef]
  28. Wang, Z.; Zhou, R.; Yu, Y. The Impact of Urban Morphology on Land Surface Temperature under Seasonal and Diurnal Variations: Marginal and Interaction Effects. Build. Environ. 2025, 272, 112673. [Google Scholar] [CrossRef]
  29. Luo, P.; Yu, B.; Li, P.; Liang, P.; Liang, Y.; Yang, L. How 2D and 3D built environments impact urban surface temperature under extreme heat: A study in Chengdu, China. Build. Environ. 2023, 231, 110035. [Google Scholar] [CrossRef]
  30. Chen, Y.; Ma, W.; Shao, Y.; Wang, N.; Yu, Z.; Li, H.; Hu, Q. The Impacts and Thresholds Detection of 2D/3D Urban Morphology on the Heat Island Effects at the Functional Zone in Megacity during Heatwave Event. Sustain. Cities Soc. 2025, 118, 106002. [Google Scholar] [CrossRef]
  31. Chen, J.; Zhan, W.; Du, P.; Li, L.; Li, J.; Liu, Z.; Huang, F.; Lai, J.; Xia, J. Seasonally disparate responses of surface thermal environment to 2D/3D urban morphology. Build. Environ. 2022, 214, 108928. [Google Scholar] [CrossRef]
  32. Yang, C.; Zhu, W.; Sun, J.; Xu, X.; Wang, R.; Lu, Y.; Zhang, S.; Zhou, W. Assessing the effects of 2D/3D urban morphology on the 3D urban thermal environment by using multi-source remote sensing data and UAV measurements: A case study of the snow-climate city of Changchun, China. J. Clean. Prod. 2021, 321, 128956. [Google Scholar] [CrossRef]
  33. Martinelli, L.; Matzarakis, A. Influence of height/width proportions on the thermal comfort of courtyard typology for Italian climate zones. Sustain. Cities Soc. 2017, 29, 97–106. [Google Scholar] [CrossRef]
  34. Han, S.; Hou, H.; Estoque, R.C.; Zheng, Y.; Shen, C.; Murayama, Y.; Pan, J.; Wang, B.; Hu, T. Seasonal effects of urban morphology on land surface temperature in a three-dimensional perspective: A case study in Hangzhou, China. Build. Environ. 2023, 228, 109913. [Google Scholar] [CrossRef]
  35. Azhdari, A.; Soltani, A.; Alidadi, M. Urban morphology and landscape structure effect on land surface temperature: Evidence from Shiraz, a semi-arid city. Sustain. Cities Soc. 2018, 41, 853–864. [Google Scholar] [CrossRef]
  36. Hou, L.; Yue, W.; Liu, X. Spatiotemporal patterns and drivers of summer heat island in Beijing-Tianjin-Hebei Urban Agglomeration, China. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 7516–7527. [Google Scholar] [CrossRef]
  37. Gu, X.; Wu, Z.; Liu, X.; Qiao, R.; Jiang, Q. Exploring the nonlinear interplay between urban morphology and nighttime thermal environment. Sustain. Cities Soc. 2024, 101, 105176. [Google Scholar] [CrossRef]
  38. Tanoori, G.; Soltani, A.; Modiri, A. Machine learning for urban heat island (UHI) analysis: Predicting land surface temperature (LST) in urban environments. Urban Clim. 2024, 55, 101962. [Google Scholar] [CrossRef]
  39. Liu, Y.; Zhang, W.; Liu, W.; Tan, Z.; Hu, S.; Ao, Z.; Li, J.; Xing, H. Exploring the Seasonal Effects of Urban Morphology on Land Surface Temperature in Urban Functional Zones. Sustain. Cities Soc. 2024, 103, 105268. [Google Scholar] [CrossRef]
  40. Shen, Y.; Kong, W.; Fei, F.; Chen, X.; Xu, Y.; Huang, C.; Yao, J. Stereoscopic Urban Morphology Metrics Enhance the Nonlinear Scale Heterogeneity Modeling of UHI with Explainable AI. Urban Clim. 2024, 56, 102006. [Google Scholar] [CrossRef]
  41. Wang, Q.; Wang, X.; Zhou, Y.; Liu, D.; Wang, H. The dominant factors and influence of urban characteristics on land surface temperature using random forest algorithm. Sustain. Cities Soc. 2022, 79, 103722. [Google Scholar] [CrossRef]
  42. Zheng, Z.; Fiore, A.M.; Westervelt, D.M.; Milly, G.P.; Goldsmith, J.; Karambelas, A.; Curci, G.; Randles, C.A.; Paiva, A.R.; Wang, C.; et al. Automated machine learning to evaluate the information content of tropospheric trace gas columns for fine particle estimates over India: A modeling testbed. J. Adv. Model. Earth Syst. 2023, 15, e2022MS003099. [Google Scholar] [CrossRef]
  43. Zhang, Y.; Jing, Z.; Feng, Y.; Chen, S.; Li, Y.; Han, Y.; Feng, L.; Pan, J.; Mazarji, M.; Zhou, H.; et al. Using Automated Machine Learning Techniques to Explore Key Factors in Anaerobic Digestion: At the Environmental Factor, Microorganisms and System Levels. Chem. Eng. J. 2023, 475, 146069. [Google Scholar] [CrossRef]
  44. Cheng, J.; Yang, D.; Qie, K.; Wang, J. Analysis of Land Surface Temperature Drivers in Beijing’s Central Urban Area across Multiple Spatial Scales: An Explainable Ensemble Learning Approach. Energy Build. 2025, 338, 115704. [Google Scholar] [CrossRef]
  45. Foroutan, E.; Hu, T.; Li, Z. Revealing Key Factors of Heat-Related Illnesses Using Geospatial Explainable AI Model: A Case Study in Texas, USA. Sustain. Cities Soc. 2025, 122, 106243. [Google Scholar] [CrossRef]
  46. Wang, Z.; Zhou, R.; Rui, J.; Yu, Y. Revealing the Impact of Urban Spatial Morphology on Land Surface Temperature in Plain and Plateau Cities Using Explainable Machine Learning. Sustain. Cities Soc. 2025, 118, 106046. [Google Scholar] [CrossRef]
  47. Yin, S.; Shi, R.; Wu, N.; Yang, J. Measuring the Impact of Technological Innovation on Urban Resilience through Explainable Machine Learning: A Case Study of the Yangtze River Delta Region, China. Sustain. Cities Soc. 2025, 127, 106457. [Google Scholar] [CrossRef]
  48. Zeng, J.; Cai, Y.; Yang, C.; Liu, Y.; Liu, W.; Xing, H. Revealing the Impact Patterns of Urban Morphology on Land Surface Temperature in the Guangdong-Hong Kong-Macao Greater Bay Area, China. Cities 2025, 165, 106147. [Google Scholar] [CrossRef]
  49. Zhu, S.; Yan, Y.; Zhao, B.; Wang, H. Assessing the Impact of Adjacent Urban Morphology on Street Temperature: A Multisource Analysis Using Random Forest and SHAP. Build. Environ. 2025, 267, 112326. [Google Scholar] [CrossRef]
  50. Yuan, B.; Zhou, L.; Hu, F.; Wei, C. Effects of 2D/3D Urban Morphology on Land Surface Temperature: Contribution, Response, and Interaction. Urban Clim. 2024, 53, 101791. [Google Scholar] [CrossRef]
  51. Wang, Q.; Wang, X.; Meng, Y.; Zhou, Y.; Wang, H. Exploring the impact of urban features on the spatial variation of land surface temperature within the diurnal cycle. Sustain. Cities Soc. 2023, 91, 104432. [Google Scholar] [CrossRef]
  52. Li, Z.; Hu, D. Exploring the relationship between the 2D/3D architectural morphology and urban land surface temperature based on a boosted regression tree: A case study of Beijing, China. Sustain. Cities Soc. 2022, 78, 103392. [Google Scholar] [CrossRef]
  53. Guo, A.; Yue, W.; Yang, J.; He, T.; Zhang, M.; Li, M. Divergent impact of urban 2D/3D morphology on thermal environment along urban gradients. Urban Clim. 2022, 45, 101278. [Google Scholar] [CrossRef]
  54. Niu, J.; Chen, J.; Zang, Y.; Wang, K.; Ni, D.; Xu, Z.; Wang, P. Deciphering the Roles of 2-D and 3-D Urban Landscape Metrics in Diurnal Surface Thermal Environment Along Urban Gradients. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 2024, 17, 16449–16466. [Google Scholar] [CrossRef]
  55. Ren, Z.; Li, C.; Fu, B.; Wang, S.; Stringer, L.C. Effects of Aridification on Soil Total Carbon Pools in China’s Drylands. Glob. Change Biol. 2024, 30, e17091. [Google Scholar] [CrossRef] [PubMed]
  56. Wang, S.; Li, R.; Wu, Y.; Wang, W. Estimation of Surface Soil Moisture by Combining a Structural Equation Model and an Artificial Neural Network (SEM-ANN). Sci. Total Environ. 2023, 876, 162558. [Google Scholar] [CrossRef]
  57. Wang, Y.; Xie, X.; Zhao, X.; Liang, S.; Zhu, B.; Tursun, A.; Jiang, F.; Liu, Y.; Zhang, X. Four-decade response of land surface temperature to urban expansion in Beijing. Agric. For. Meteorol. 2023, 341, 109653. [Google Scholar] [CrossRef]
  58. Liu, X.; Zhou, Y.; Yue, W.; Li, X.; Liu, Y.; Lu, D. Spatiotemporal patterns of summer urban heat island in Beijing, China using an improved land surface temperature. J. Clean. Prod. 2020, 257, 120529. [Google Scholar] [CrossRef]
  59. Lemoine-Rodríguez, R.; Inostroza, L.; Zepp, H. Does urban climate follow urban form? Analysing intraurban LST trajectories versus urban form trends in 3 cities with different background climates. Sci. Total Environ. 2022, 830, 154570. [Google Scholar] [CrossRef]
  60. Cook, M.; Schott, J.; Mandel, J.; Raqueno, N. Development of an operational calibration methodology for the landsat thermal data archive and initial testing of the atmospheric compensation component of a land surface temperature (LST) product from the archive. Remote Sens. 2014, 6, 11244–11266. [Google Scholar] [CrossRef]
  61. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [Google Scholar] [CrossRef]
  62. Ermida, S.L.; Soares, P.; Mantas, V.; Göttsche, F.-M.; Trigo, I.F. Google Earth Engine Open-Source Code for Land Surface Temperature Estimation from the Landsat Series. Remote Sens. 2020, 12, 1471. [Google Scholar] [CrossRef]
  63. Li, H.; Liu, Y.; Zhang, H.; Xue, B.; Li, W. Urban morphology in China: Dataset development and spatial pattern characterization. Sustain. Cities Soc. 2021, 71, 102981. [Google Scholar] [CrossRef]
  64. Li, H.; Yuan, F.; Shen, L.; Liu, Y.; Zheng, Z.; Zhou, X. Improving the WRF/urban modeling system in China by developing a national urban dataset. Geosci. Front. 2022, 13, 101385. [Google Scholar] [CrossRef]
  65. Akinwande, M.O.; Dikko, H.G.; Samson, A. Variance inflation factor: As a condition for the inclusion of suppressor variable(s) in regression analysis. Open J. Stat. 2015, 5, 754–767. [Google Scholar] [CrossRef]
  66. Zanaga, D.; Van De Kerchove, R.; De Keersmaecker, W.; Souverijns, N.; Brockmann, C.; Quast, R.; Wevers, J.; Grosu, A.; Paccini, A.; Vergnaud, S.; et al. ESA WorldCover 10 m 2020 v100 [Dataset]. Zenodo; The European Space Agency: Paris, France, 2021. [Google Scholar] [CrossRef]
  67. Chen, F.; Zhou, B.; Yang, L.; Chen, X.; Zhuang, J. Predicting bacterial transport through saturated porous media using an automated machine learning model. Front. Microbiol. 2023, 14, 1152059. [Google Scholar] [CrossRef] [PubMed]
  68. Song, Y.; Wu, S.; Chen, B.; Bell, M.L. Unraveling near real-time spatial dynamics of population using geographical ensemble learning. Int. J. Appl. Earth Obs. Geoinf. 2024, 130, 103882. [Google Scholar] [CrossRef]
  69. Ren, H.; Zhang, L.; Yan, M.; Chen, B.; Yang, Z.; Ruan, L. Spatiotemporal assessment of forest fire vulnerability in China using automated machine learning. Remote Sens. 2022, 14, 5965. [Google Scholar] [CrossRef]
  70. Li, K.-Y.; Burnside, N.G.; de Lima, R.S.; Peciña, M.V.; Sepp, K.; Cabral Pinheiro, V.H.; de Lima, B.R.C.A.; Yang, M.-D.; Vain, A.; Sepp, K. An automated machine learning framework in unmanned aircraft systems: New insights into agricultural management practices recognition approaches. Remote Sens. 2021, 13, 3190. [Google Scholar] [CrossRef]
  71. Adams, M.D.; Massey, F.; Chastko, K.; Cupini, C. Spatial modelling of particulate matter air pollution sensor measurements collected by community scientists while cycling, land use regression with spatial cross-validation, and applications of machine learning for data correction. Atmos. Environ. 2020, 230, 117479. [Google Scholar] [CrossRef]
  72. Cao, Y.; Wang, Z.; Li, P.; Zhou, Z.; Li, W.; Zheng, T.; Liu, J.; Wu, W.; Shi, Z.; Liu, J. Prediction of rural domestic water and sewage production based on automated machine learning in northern China. J. Clean. Prod. 2024, 434, 140016. [Google Scholar] [CrossRef]
  73. Pande, C.B.; Egbueri, J.C.; Costache, R.; Sidek, L.M.; Wang, Q.; Alshehri, F.; Din, N.M.; Gautam, V.K.; Chandra Pal, S. Predictive modeling of land surface temperature (LST) based on Landsat-8 satellite data and machine learning models for sustainable development. J. Clean. Prod. 2024, 444, 141035. [Google Scholar] [CrossRef]
  74. Lundberg, S.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions. arXiv. 2017. [Google Scholar] [CrossRef]
  75. Fu, X.; Wang, M.; Zhang, D.; Chen, F.; Peng, X.; Wang, L.; Tan, S.K. An XGBoost-SHAP Framework for Identifying Key Drivers of Urban Flooding and Developing Targeted Mitigation Strategies. Ecol. Indic. 2025, 175, 113579. [Google Scholar] [CrossRef]
  76. LeDell, E.; Poirier, S. H2O AutoML: Scalable automatic machine learning. In Proceedings of the 7th ICML Workshop on Automated Machine Learning (AutoML 2020), Vienna, Austria, 17–18 July 2020. [Google Scholar]
  77. Bansal, P.; Quan, S.J. Examining Temporally Varying Nonlinear Effects of Urban Form on Urban Heat Island Using Explainable Machine Learning: A Case of Seoul. Build. Environ. 2024, 247, 110957. [Google Scholar] [CrossRef]
  78. Wang, J.; Zhou, W.; Zheng, Z.; Jiao, M.; Qian, Y. Interactions among spatial configuration aspects of urban tree canopy significantly affect its cooling effects. Sci. Total Environ. 2023, 864, 160929. [Google Scholar] [CrossRef] [PubMed]
  79. Fan, Y.; Chen, J.; Shirkey, G.; John, R.; Wu, S.R.; Park, H.; Shao, C. Applications of structural equation modeling (SEM) in ecological studies: An updated review. Ecol. Process. 2016, 5, 19. [Google Scholar] [CrossRef]
  80. Xie, M.; Chen, J.; Zhang, Q.; Li, H.; Fu, M.; Breuste, J. Dominant landscape indicators and their dominant areas influencing urban thermal environment based on structural equation model. Ecol. Indic. 2020, 111, 105992. [Google Scholar] [CrossRef]
  81. Li, C.; Fu, B.; Wang, S.; Stringer, L.C.; Wang, Y.; Li, Z.; Liu, Y.; Zhou, W. Drivers and impacts of changes in China’s drylands. Nat. Rev. Earth Environ. 2021, 2, 858–873. [Google Scholar] [CrossRef]
  82. Wu, W.; Li, L.; Li, C. Seasonal variation in the effects of urban environmental factors on land surface temperature in a winter city. J. Clean. Prod. 2021, 299, 126897. [Google Scholar] [CrossRef]
  83. Oke, T.R. The energetic basis of the urban heat island. Q. J. R. Meteorol. Soc. 1982, 108, 1–24. [Google Scholar] [CrossRef]
  84. Cao, Q.; Luan, Q.; Liu, Y.; Wang, R. The effects of 2D and 3D building morphology on urban environments: A multi-scale analysis in the Beijing metropolitan region. Build. Environ. 2021, 192, 107635. [Google Scholar] [CrossRef]
  85. Scarano, M.; Mancini, F. Assessing the relationship between sky view factor and land surface temperature to the spatial resolution. Int. J. Remote Sens. 2017, 38, 6910–6929. [Google Scholar] [CrossRef]
  86. Li, H.; Zhou, Y.; Jia, G.; Zhao, K.; Dong, J. Quantifying the response of surface urban heat island to urbanization using the annual temperature cycle model. Geosci. Front. 2022, 13, 101141. [Google Scholar] [CrossRef]
  87. Li, H.; Wolter, M.; Wang, X.; Sodoudi, S. Impact of land cover data on the simulation of urban heat island for Berlin using WRF coupled with bulk approach of Noah-LSM. Theor. Appl. Climatol. 2018, 134, 67–81. [Google Scholar] [CrossRef]
  88. Liu, S.; Wang, Y.; Zhang, G.J.; Gong, P.; Wei, L.; He, Y.; Li, L.; Wang, B. Effects of urbanization in China on the east Asian summer monsoon as revealed by two global climate models. J. Geophys. Res. Atmos. 2024, 129, e2023JD039737. [Google Scholar] [CrossRef]
  89. Lu, H.; Li, F.; Yang, G.; Sun, W. Multi-scale impacts of 2D/3D urban building pattern in intra-annual thermal environment of Hangzhou, China. Int. J. Appl. Earth Obs. Geoinf. 2021, 104, 102558. [Google Scholar] [CrossRef]
  90. Chakraborty, T.; Venter, Z.S.; Qian, Y.; Lee, X. Lower urban humidity moderates outdoor heat stress. AGU Adv. 2022, 3, e2022AV000729. [Google Scholar] [CrossRef]
  91. He, B.-J.; Fu, X.; Zhao, Z.; Chen, P.; Sharifi, A.; Li, H. Capability of LCZ Scheme to Differentiate Urban Thermal Environments in Five Megacities of China: Implications for Integrating LCZ System into Heat-Resilient Planning and Design. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2024, 17, 18800–18817. [Google Scholar] [CrossRef]
Remotesensing 17 03150 g001
Figure 1. Landcover of the city area of Beijing from the European Space Agency (ESA) World Cover 10 m 2020 product.
Figure 1. Landcover of the city area of Beijing from the European Space Agency (ESA) World Cover 10 m 2020 product.
Remotesensing 17 03150 g001
Remotesensing 17 03150 g002
Figure 2. Work flowchart. LST, land surface temperature; NDVI, Normalized Difference Vegetation Index; AutoML, automated machine learning.
Figure 2. Work flowchart. LST, land surface temperature; NDVI, Normalized Difference Vegetation Index; AutoML, automated machine learning.
Remotesensing 17 03150 g002
Remotesensing 17 03150 g003
Figure 3. Spatial patterns of urban morphology indicators in urban Beijing. (a) impervious surface area, (b) building coverage ratio, (c) building height, (d) mean building volume, (e) frontal area index, and (f) sky view factor, (g) spring NDVI, (h) summer NDVI, (i) autumn NDVI, and (j) winter NDVI.
Figure 3. Spatial patterns of urban morphology indicators in urban Beijing. (a) impervious surface area, (b) building coverage ratio, (c) building height, (d) mean building volume, (e) frontal area index, and (f) sky view factor, (g) spring NDVI, (h) summer NDVI, (i) autumn NDVI, and (j) winter NDVI.
Remotesensing 17 03150 g003
Remotesensing 17 03150 g004
Figure 4. Spatial patterns of mean LST in urban Beijing in four seasons during 2020–2022: (a) LST in spring, (b) LST in summer, (c) LST in autumn, (d) LST in winter.
Figure 4. Spatial patterns of mean LST in urban Beijing in four seasons during 2020–2022: (a) LST in spring, (b) LST in summer, (c) LST in autumn, (d) LST in winter.
Remotesensing 17 03150 g004
Remotesensing 17 03150 g005
Figure 5. Variations in (a) R2 and (b) RMSE of the AutoML model with grid scale. R2, Coefficient of Determination.
Figure 5. Variations in (a) R2 and (b) RMSE of the AutoML model with grid scale. R2, Coefficient of Determination.
Remotesensing 17 03150 g005
Remotesensing 17 03150 g006
Figure 6. Comparison of the SHAP summary plot of urban morphology indicators to LST in four seasons at the grid scale of 200 m. (a,c,e,g) Global feature importance. (b,d,f,h) Local explanation. Each row in the Local explanation plot represents an indicator, and the x-axis corresponds to the SHAP value. A single dot represents an individual sample, with the color indicating the magnitude of the feature’s value: redder colors signify higher values for the feature, while bluer tones indicate lower values. The Global feature importance was established by averaging the absolute values for interpretation.
Figure 6. Comparison of the SHAP summary plot of urban morphology indicators to LST in four seasons at the grid scale of 200 m. (a,c,e,g) Global feature importance. (b,d,f,h) Local explanation. Each row in the Local explanation plot represents an indicator, and the x-axis corresponds to the SHAP value. A single dot represents an individual sample, with the color indicating the magnitude of the feature’s value: redder colors signify higher values for the feature, while bluer tones indicate lower values. The Global feature importance was established by averaging the absolute values for interpretation.
Remotesensing 17 03150 g006
Remotesensing 17 03150 g007
Figure 7. Variations in LST with the changes in urban morphology indicators in four seasons at 200 m grid scale: (a) NDVI, (b) ISA, (c) BCR, (d) BH, (e) MBV, (f) FAI, (g) SVF. Note: The dashed horizontal line indicates LST variation = 0, where LST equals the average with no change. LST variation > 0 indicates LST above the average (warming effect), while LST variation < 0 indicates LST below the average (cooling effect).
Figure 7. Variations in LST with the changes in urban morphology indicators in four seasons at 200 m grid scale: (a) NDVI, (b) ISA, (c) BCR, (d) BH, (e) MBV, (f) FAI, (g) SVF. Note: The dashed horizontal line indicates LST variation = 0, where LST equals the average with no change. LST variation > 0 indicates LST above the average (warming effect), while LST variation < 0 indicates LST below the average (cooling effect).
Remotesensing 17 03150 g007
Remotesensing 17 03150 g008
Figure 8. The structural equation modeling of the relationship between LST and urban morphology indicators in four seasons at 200 m scale: (a) spring, (b) summer, (c) autumn, (d) winter. Note: Green and blue rectangles represent 2D and 3D variables, respectively. Arrows indicate causal relationships, where red arrows represent negative relationships and blue arrows represent positive relationships. The standardized path coefficients are displayed next to the arrows, reflecting the magnitude of the effects. The thickness and darkness of the arrows represent the absolute values of the path coefficients. Solid lines indicate significance at the level of p < 0.001, and dashed lines indicate significance at p < 0.05. RMSEA, Root Mean Square Error of Approximation.
Figure 8. The structural equation modeling of the relationship between LST and urban morphology indicators in four seasons at 200 m scale: (a) spring, (b) summer, (c) autumn, (d) winter. Note: Green and blue rectangles represent 2D and 3D variables, respectively. Arrows indicate causal relationships, where red arrows represent negative relationships and blue arrows represent positive relationships. The standardized path coefficients are displayed next to the arrows, reflecting the magnitude of the effects. The thickness and darkness of the arrows represent the absolute values of the path coefficients. Solid lines indicate significance at the level of p < 0.001, and dashed lines indicate significance at p < 0.05. RMSEA, Root Mean Square Error of Approximation.
Remotesensing 17 03150 g008
Remotesensing 17 03150 g009
Figure 9. The direct and indirect effects of urban morphology indicators on LST in different seasons at 200 m scale :(a) spring, (b) summer, (c) autumn, (d) winter. Note: All indicates all direct and indirect paths of all factors combined, 2D represents all paths from BCR through NDVI and ISA to LST, and 3D represents all paths from BH through MBV, FAI, and SVF to LST.
Figure 9. The direct and indirect effects of urban morphology indicators on LST in different seasons at 200 m scale :(a) spring, (b) summer, (c) autumn, (d) winter. Note: All indicates all direct and indirect paths of all factors combined, 2D represents all paths from BCR through NDVI and ISA to LST, and 3D represents all paths from BH through MBV, FAI, and SVF to LST.
Remotesensing 17 03150 g009
Table 1. Description of urban morphology indicators used in this study.
Table 1. Description of urban morphology indicators used in this study.
CategoryIndicatorsDescription
2DImpervious Surface Area (ISA)Percentage of impervious surface
Building coverage ratio (BCR)Horizontal space of the building
Normalized Difference Vegetation Index (NDVI)Quantifying the vegetation coverage and health status
3DBuilding height (BH)Mean height of the buildings
Mean building volume (MBV)Total building volume divided by its number
Frontal area index (FAI)Frontal area per unit horizontal area per unit height increment
Sky view factor (SVF)Fraction of the hemisphere occupied by sky
2D, two-dimensional; 3D, three-dimensional; ISA, impervious surface area; BCR, building coverage ratio; NDVI, Normalized Difference Vegetation Index; BH, building height; MBV, mean building volume; FAI, frontal area index; SVF, sky view factor.
Table 2. Seasonal performance of AutoML models at 200 m scale.
Table 2. Seasonal performance of AutoML models at 200 m scale.
Model_idRMSEMSEModel_idRMSEMSE
SpringSummer
GBM_1_AutoML1.3231.752GBM_5_AutoML2.0754.307
GBM_grid_1_AutoML_model_11.3231.752GBM_2_AutoML2.0794.325
GBM_2_AutoML1.3241.753GBM_1_AutoML2.0824.334
GBM_3_AutoML1.3251.757GBM_grid_1_AutoML_model_12.0854.349
GBM_grid_1_AutoML_model_31.3261.758GBM_3_AutoML2.0864.350
GBM_5_AutoML1.3261.758GBM_grid_1_AutoML_model_32.0914.376
GBM_4_AutoML1.3351.783GBM_4_AutoML2.1004.414
DRF_1_AutoML1.3371.788GBM_grid_1_AutoML_model_22.1024.419
XRT_1_AutoML1.3381.792XRT_1_AutoML2.1064.438
GBM_grid_1_AutoML_model_21.3401.796DRF_1_AutoML2.1124.463
AutumnWinter
GBM_2_AutoML0.8320.692GBM_5_AutoML0.9790.959
GBM_5_AutoML0.8330.694GBM_grid_1_AutoML_model_10.9800.962
GBM_1_AutoML0.8360.699GBM_3_AutoML0.9810.963
GBM_grid_1_AutoML_model_10.8360.699GBM_1_AutoML0.9810.963
GBM_grid_1_AutoML_model_30.8360.699GBM_2_AutoML0.9820.964
GBM_3_AutoML0.8370.700GBM_grid_1_AutoML_model_30.9840.969
GBM_4_AutoML0.8380.702GBM_4_AutoML0.9860.974
XRT_1_AutoML0.8390.704XRT_1_AutoML0.9890.979
DRF_1_AutoML0.8420.710DRF_1_AutoML0.9920.985
GBM_grid_1_AutoML_model_20.8510.724GBM_grid_1_AutoML_model_20.9930.985
RMSE, Root Mean Square Error; MSE, Mean Squared Error; GBM, Gradient Boosting Machines; XRT, Extreme Random Trees; DRF, Distributed Random Forest.
Table 3. Comparisons of R2 and RMSE of the optimal AutoML model (GBM) and RF at 200 m.
Table 3. Comparisons of R2 and RMSE of the optimal AutoML model (GBM) and RF at 200 m.
SeasonsAutoML (Best Model)RF
R2RMSER2RMSE
Spring0.461.300.431.43
Summer0.522.100.472.28
Autumn0.401.080.381.10
Winter0.260.930.240.94
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wang, N.; Shen, L.; Fei, W.; Liu, Y.; Zhao, H.; Liu, L.; Wang, A.; He, B.-J. Seasonal Effects of Urban Morphology on the Thermal Environment Based on Automated Machine Learning: A Case Study of Beijing. Remote Sens. 2025, 17, 3150. https://doi.org/10.3390/rs17183150

AMA Style

Wang N, Shen L, Fei W, Liu Y, Zhao H, Liu L, Wang A, He B-J. Seasonal Effects of Urban Morphology on the Thermal Environment Based on Automated Machine Learning: A Case Study of Beijing. Remote Sensing. 2025; 17(18):3150. https://doi.org/10.3390/rs17183150

Chicago/Turabian Style

Wang, Ni, Lidu Shen, Wenli Fei, Yage Liu, Hujia Zhao, Luyao Liu, Anzhi Wang, and Bao-Jie He. 2025. "Seasonal Effects of Urban Morphology on the Thermal Environment Based on Automated Machine Learning: A Case Study of Beijing" Remote Sensing 17, no. 18: 3150. https://doi.org/10.3390/rs17183150

APA Style

Wang, N., Shen, L., Fei, W., Liu, Y., Zhao, H., Liu, L., Wang, A., & He, B.-J. (2025). Seasonal Effects of Urban Morphology on the Thermal Environment Based on Automated Machine Learning: A Case Study of Beijing. Remote Sensing, 17(18), 3150. https://doi.org/10.3390/rs17183150

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop