Study of the Correlation between the Urban Wind–Heat Environment and Urban Development Elements in High-Density Urban Areas: A Case Study of Central Shanghai

Abstract

To prevent damage to human life and health caused by high temperatures and extreme weather and to promote sustainable urban development, it is necessary to optimize the layout of urban development elements to improve the urban wind–heat environment. Taking the high-density urban area of Shanghai as an example, this study used wavelet analysis to investigate the cyclic changes of the average annual temperature of Shanghai from 1950 to 2022 and the average annual wind speed of Shanghai from 2000 to 2020. The correlation between the urban heat environment and the urban development elements was analyzed using geographically weighted regression. The correlation was then examined using BP neural network, and finally, the impacts of different urban spatial patterns on the wind environment were analyzed using CFD numerical simulation. The results show that the average annual temperature of Shanghai city has an overall upward trend, with significant cycles of 44 and 32 years that are shortening over time. The average annual wind speed has a downward trend, with a significant main cycle of 22 years. Greening and water coverage, as well as the floor area ratio, have a significant reducing effect on surface temperature, whereas building density is positively correlated with surface temperature. Building density has a significant reducing effect on wind speed, whereas the effect of floor area ratio is not significant. The effect of building density on wind speed is significantly weakened, whereas the effect of the floor area ratio is not significant. This study provides valuable references for Shanghai and other high-density cities to optimize urban spatial patterns in order to improve the safety and comfort of the urban wind–heat environments. This study is of significant importance and value in promoting sustainable urban development, protecting the health of urban residents, and advancing spatial justice and equal well-being.

1. Introduction

The urban wind–heat environment directly affects the sustainable development of cities, and persistent high temperatures and increasing climate extremes continue to pose a threat to human life and health. As urbanization continues in China, the natural environments in cities are continuously replaced by artificial environments. This increases the solar radiation absorption capacity of urban surfaces, reduces plants’ transpiration, and increases wind stagnation effects, thereby exacerbating the urban heat island effect. Exploring the correlation between greening and water coverage, building density, the floor area ratio, and other urban development elements and aspects of the wind–heat environment can help to optimize the urban spatial layout. This can play a major role in alleviating the urban heat island effect, promoting sustainable development and protecting human life and health.

Scholars have studied the spatial and temporal characteristics, influencing factors, and mechanisms of the urban wind–heat environment from multiple perspectives. Studies have focused on vegetation cover [1,2] surface water bodies [3,4], impervious surfaces [5,6], and other elements of the urban heat environment. They have explored the effects of land use/cover change (LUCC) [7,8,9] and urban surface characteristics [10,11,12]. They have considered the relationship between the urban heat environment and urban morphological characteristics [13] and the spatial distribution characteristics of urban land surface temperature [14], as well as spatial and temporal variation patterns [15,16].

The urban heat island effect is a local-scale microclimate phenomenon characterized by higher temperatures in urban areas compared with the surrounding suburbs or rural regions and is one of the most significant features of the urban heat environment. The relationship between the urban form and the urban heat island effect has been confirmed in several studies, which are mostly at the urban level [17,18]. Most of the studies attempt to use the geometric features of the city to express the urban form or environmental characteristics, such as density, compactness, connectivity, etc. [19,20,21,22,23,24]. To estimate the impact of urban development on the local climate, modeling-based research has been increasingly conducted in major metropolitan areas around the world [25,26]. Studies have shown that urban development significantly alters the surface energy balance and hydrological cycle. Replacing natural/semi-natural surfaces with impermeable surfaces greatly changes surface albedo and evapotranspiration, affecting the heating/cooling potential of the ground. The standing building infrastructure also alters airflow, energy absorption, and atmospheric heat transfer in the city, resulting in profound changes in meteorological environments, such as changes in the temperature, humidity, and wind [27,28,29]. In recent years, more and more studies have explored the impact of three-dimensional development environments on urban microclimates, establishing statistical relationships between observed meteorological elements or land surface temperatures and urban development environments using machine learning methods [26,30,31].

Some scholars have simulated the urban wind–heat environment to explore the effects of different scales of the urban natural and built environment elements on the wind–heat environment. For example, research has evaluated wind comfort and wind safety in urban areas based on computational fluid dynamics (CFD) [32], used the microclimate environment simulation system ENVI-met to evaluate the thermal comfort of sites [33,34], and performed the quantitative simulation of air velocity, temperature, and air pressure of the outdoor environment using Phoenics 2019 software [35]. Meanwhile, many studies have found that optimizing the urban wind environment can improve the problem of the urban heat environment by directly reducing temperature, improving thermal comfort, or mitigating heat pollution [36,37].

The latest research and practice in the field of the urban wind–heat environment has focused on evaluating the wind environment and thermal comfort within the city and proposing optimization strategies based on the evaluation results. However, the correlation and spatial relationships between urban development elements (such as greening and water coverage, building density, and floor area ratio) and the wind–heat environment have not yet been explored. This study suggested that there is a correlation between urban development factors and urban thermal and wind environments that required further investigation. Shanghai is the largest mega-city in China and has undergone rapid urbanization in the past few decades, which has led to an increase in the intensity of the urban heat island effect. Existing research has confirmed that extreme temperatures have had impacts on the population in Shanghai, including increasing the risks of diseases and death [38,39]. Consequently, this study took Shanghai’s high-density urban area, a typical mega-city high-density area, as a case study. Through wavelet analysis, geographically weighted regression, BP neural network analysis, and CFD numerical simulation, we analyzed the effects of urban development factors such as greening and water coverage, building density, and the floor area ratio on the urban wind–heat environment. In today’s world of intensifying global warming, the research findings provide important references for optimizing the layout of urban development elements, improving the urban wind–heat environment and promoting equal well-being, particularly in mega-cities.

2. Methodology

2.1. Study Area

People living in high-density cities are more likely to be at risk from extreme heat. Shanghai, as one of China’s largest and most urbanized cities, is a mega-city that has undergone rapid urbanization in the past few decades, with an urbanization rate of 89.3% in 2022 and a permanent population of 24.759 million. This urbanization process has also led to concerning urban heat island effects in the city [40,41]. The city serves as a typical representative of mega-cities in developing countries. Additionally, Shanghai is located in the southern region of China and experiences a subtropical monsoon climate, with hot and humid summers. Optimizing the urban wind and heat environment is particularly important. Therefore, it is urgent to study the factors affecting the urban wind and heat environment in Shanghai and optimize the urban spatial environment based on research results in order to mitigate the adverse effects of the wind and heat environment on the health of urban residents.

The research scope includes the central districts of Shanghai, including Huangpu, Xuhui, Changning, Jing’an, Putuo, Hongkou, and Yangpu, which are high-density administrative districts. The total area is 289.44 km². The study area accounts for 4.56% of Shanghai’s urban area but has 26.36% of the population, with a population density 5.77 times higher than the average in Shanghai. The concentration of high-density populations and the intense urban development undoubtedly pose greater challenges to the urban wind and heat environment (

Table 1. Area and population overview in 2021.

City (District)		Area (km2)	Population (1000 Million)	Population Density (Thousand Person/km2)
Shanghai		6340.50	24.89	3.93
The study area	Huangpu district	20.46	0.58	28.45
	Xuhui District	54.76	1.12	20.36
	Changning District	38.30	0.70	18.16
	Jing’an District	36.88	0.97	26.19
	Putuo district	54.83	1.24	22.68
	Hongkou district	23.48	0.72	30.44
	Yangpu district	60.73	1.23	20.26
	Total	289.44	6.56	22.66

Data are from the Shanghai 2022 Statistical Yearbook (https://tjj.sh.gov.cn/tjnj/nj22.htm?d1=2022tjnj/C0202.htm, 6 Janurary 2024).

, Figure 1).

2.2. Study Indicators and Data

Land surface temperature (LST) is considered one of the most effective factors in urban ecology and plays an important role in connecting the urban surface energy. Therefore, in this study, LST is used as an indicator of the urban heat environment [42]. In related research on the urban wind–heat environment, urban land cover types are often studied as important 2D influencing factors. In addition, the 3D built environment is also an important research factor. This study focuses on urban development factors, specifically the impact of urban development and construction on the wind–heat environment. Therefore, three indicators, namely building density, the floor area ratio, and green and water coverage, were selected as measures of the degree of urban development. Building density and green and water coverage can cover different types of urban land surfaces, representing the horizontal development of the city, whereas the comparison between building density and the floor area ratio can illustrate the vertical development of the city.

The specific heat capacities of green spaces and water bodies are significantly higher than those of steel and concrete, meaning that more heat energy is required to raise their temperatures by 1 Kelvin. By categorizing green spaces and water bodies under the same land cover type, the aim is to highlight the difference between these two indicators and the building density and plot ratio. Additionally, as not every zone within the study area has water bodies, merging water bodies and green coverage into a single indicator helps to avoid the influence of significant vacant areas on the research model and regression results.

Vector data on the building outline and building height in Shanghai were downloaded from Gaode Map (https://www.amap.com/, 9 July 2023). Data on the annual average temperatures from 1950 to 2020 and annual average wind speeds from 2000 to 2020 were obtained from the website of the National Centers for Environmental Information under the National Oceanic and Atmospheric Administration (https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/, 9 July 2023). The remote sensing image data used for the land surface temperature inversion and greening and water coverage extraction were obtained from the Landsat 8–9 OLI/TIRS C2 L2 satellite digital products of the Geospatial Data Cloud (http://www.gscloud.cn/, 9 July 2023). To ensure the accuracy of the study results, summer Landsat 8–9 OLI/TIRS C2 L2 remote sensing images with good image quality and low cloud cover (cloud cover = 2.91%) were selected. The imaging date was 14 August 2022.

The average wind speed and direction data for Shanghai on 14 August 2022 were obtained from Weather Spark’s website (https://zh.weatherspark.com/, 9 July 2023) for the summer of 2022 in Shanghai. The daily average wind speed in summer 2022 was 18.3 km/h, which is equivalent to 5.08 meters per second, and the daily average wind direction was southeasterly: east (39%), south (36%), north (16%), and west (9%) in order of prevalence.

2.3. Study Method

This study selected the central urban area of Shanghai as a typical high-density urban area for research. Firstly, wavelet analysis was used to explore the temperature change trend in Shanghai over the past 70 years and the periodic change trend of wind speed over the past 20 years. Then, the GWR model was established to explore the correlation between the building density, plot ratio, green space and water coverage, and urban thermal environment. Compared with ordinary regression analysis, GWR can display spatial heterogeneity in the region. On this basis, the BP neural network was introduced for nonlinear modeling to obtain the importance of various factors and verify the results of GWR, improving the accuracy of the results. Secondly, the plots were divided into four urban spatial forms according to the high and low values of the building density and plot ratio. CFD was used to simulate the wind environment and explore the impact of the building density and plot ratio on the wind environment. The innovation and focus of our study lie in exploring the meteorological change trend in the research area, analyzing the relationship between the wind–thermal environment and urban development elements in-depth and crosswise, and verifying the results of spatial regression through machine learning methods to obtain reliable conclusions and find a path to optimize the urban wind–thermal environment.

2.3.1. Land Surface Temperature Inversion and Greening and Water Coverage Extraction

The study used Landsat 8–9 OLI/TIRS C2 L2 remote sensing images for land surface temperature (LST) inversion according to the land surface temperature conversion formula given on the official USGS website (www.usgs.gov/landsat-missions/landsat-collection-2-level-2-science-products, 9 July 2023). As the National Aeronautics and Space Administration (NASA) does not recommend using Band 11 for LST inversion, Band 10 was chosen [43]. The following equation was used:

$$\mathrm{L}\mathrm{S}\mathrm{T}=\mathrm{“}\mathrm{b}\mathrm{a}\mathrm{n}\mathrm{d}10\mathrm{”}\times 0.00341802+149.0$$

(1)

The unit of LST in the above equation is thermodynamic temperature (K), which is converted to Celsius by subtracting 273.15 from the original value.

Landsat 8–9 OLI/TIRS C2 L2 remote sensing images dated 14 August 2022 were selected for greening and water coverage extraction using the supervised classification method. The supervised classification method focuses on classifying images by building discriminant functions based on selected training area samples.

2.3.2. Wavelet Analysis

This study utilizes wavelet analysis to explore the temporal variations and periodic trends of surface temperature and wind speed. The wavelet function originates from multiresolution analysis, and its basic idea is to represent the function f(t) in the expansion as a series of successive approximation expressions, each of which takes the form of the f(t) motion after smoothing. These expressions correspond to different resolutions. The wavelet mother function is $\mathrm{\Psi }\left(\mathrm{t}\right)$ and satisfies the following equation:

$$\int \mathrm{\Psi }\left(\mathrm{t}\right)\mathrm{d}\mathrm{t}=0$$

(2)

where $\mathrm{\Psi }\left(\mathrm{t}\right)$ is the basis wavelet function and the wavelet transform is obtained as follows:

$${\mathrm{W}}_{\mathrm{f}}\left(\mathrm{a},\mathrm{b}\right)={\left|\mathrm{a}\right|}^{-\frac{1}{2}}{\int }_{\mathrm{R}}\mathrm{f}(\mathrm{t})\overline{\mathrm{\Psi }}(\frac{\mathrm{t}-\mathrm{b}}{\mathrm{a}})\mathrm{d}\mathrm{t}$$

(3)

where ${\mathrm{W}}_{\mathrm{f}}\left(\mathrm{a},\mathrm{b}\right)$ represents the wavelet transform coefficients; $\mathrm{f}\left(\mathrm{t}\right)$ is a signal or square productive function; a is the scaling scale; b is the number of translations; and $\overline{\mathrm{\Psi }}\left(\frac{\mathrm{x}-\mathrm{b}}{\mathrm{a}}\right)$ is the $\mathrm{\Psi }\left(\frac{\mathrm{t}-\mathrm{b}}{\mathrm{a}}\right)$ of the complex conjugate function. The discrete wavelet of Equation (3) is transformed as follows:

$${\mathrm{W}}_{\mathrm{f}}\left(\mathrm{a},\mathrm{b}\right)={\left|\mathrm{a}\right|}^{-\frac{1}{2}}∆\mathrm{t}\sum _{\mathrm{k}=1}^{\mathrm{N}}\mathrm{f}\left(\mathrm{k}\mathsf{\Delta }\mathrm{t}\right)\overline{\mathrm{\Psi }}\left(\frac{\mathrm{k}\mathsf{\Delta }\mathrm{t}-\mathrm{b}}{\mathrm{a}}\right)$$

(4)

The wavelet variance is obtained by integrating all wavelet coefficients obtained from the wavelet transform equation for different time scales over the b domain e:

$$\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{a})={\int }_{-\mathrm{\infty }}^{\mathrm{\infty }}{\left|{\mathrm{W}}_{\mathrm{f}}\right(\mathrm{a},\mathrm{b}\left)\right|}^2\mathrm{d}\mathrm{b}$$

(5)

The wavelet variance plot shows the variation of wavelet variance with scale a. From Equation (5), it can reflect the distribution of the energy of signal fluctuations with scale a. Therefore, a wavelet variogram can reflect the main time scale of the signal, i.e., the principal period.

2.3.3. Geographically Weighted Regression

As geographically weighted regression (GWR) can better reveal spatial heterogeneity and local relationships, it is conducive to analyzing the spatial characteristics of the impact of urban development elements on surface temperature. This study utilizes GWR to analyze the correlation between urban development elements and the thermal environment.

Geographically weighted regression was proposed by Brunsdon et al. as a means of modeling the spatial non-stationarity of variables based on nonparametric locally weighted regression [44,45]. The GWR model equation is as follows:

$${\mathrm{y}}_{\mathrm{i}}={\mathsf{\beta }}_0\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)+{\sum }_{\mathrm{k}=1}^{\mathrm{m}}{\mathsf{\beta }}_{\mathrm{k}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right){\mathrm{X}}_{\mathrm{i}\mathrm{k}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(6)

where ${\mathrm{y}}_{\mathrm{i}}$ denotes the response variable at the spatial location $\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$, and ${\mathrm{X}}_{\mathrm{i}\mathrm{k}}$ denotes the observed value of the independent variable at the spatial location $\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$. ${\mathsf{\beta }}_0\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ is the intercept term of the regression relationship, and ${\mathsf{\beta }}_{\mathrm{k}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ is the regression coefficient of the kth independent variable at the spatial position $\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$. The coefficient of the kth independent variable at the spatial position $\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ is the continuous function of ${\mathsf{\epsilon }}_{\mathrm{i}}$, which is the mutually independent random error term.

2.3.4. BP Neural Network

The BP neural network is commonly applied as a predictive model and can also be used to determine the weights of indicators. In this study, the BP neural network is used to validate the analysis results of GWR and explore the correlation between the land surface temperature and urban development elements, as well as the importance of each indicator.

BP neural network analysis is currently a widely used machine learning method. Its learning process consists of two parts. The first is forward propagation, which uses input vectors to generate the corresponding target vectors and compares the desired and actual outputs. The second part of the learning process is backward propagation, which is performed when there is an error between the desired output and the actual output [46]. The completeness theorem of the mapping capability of BP neural networks suggests that a three-layer network can approximate any continuous function with arbitrary accuracy [47]. The structure of a BP neural network contains an input layer, a hidden layer, and an output layer [48]. The number of nodes in the input layer depends on the number of input features, and the number of nodes in the output layer is determined by the type of classification.

2.3.5. Computational Fluid Dynamics

This paper utilizes Sketchup 2020 software to construct architectural block models and uses DEEPUD to simulate the outdoor air velocity in high-density urban areas of Shanghai in order to explore the impact of different urban spatial forms on wind environment.

Computational fluid dynamics (CFD) is a simulation technology whose basic principle is to analyze the discrete distribution of the flow field by solving the differential equations controlling the fluid flow, thereby simulating the fluid flow state. The Reynolds number is a dimensionless ratio used to describe the relative importance of inertial and viscous forces in fluid flow. It is determined by the density, velocity, length, and dynamic viscosity of the fluid [49,50].

DEEPUD is an intelligent deduction and calculation platform for urban space that utilizes cloud computing to obtain real-time statistical data and quantitative analysis results (https://www.shifang.city/, 20 July 2023). DEEPUD uses the lattice Boltzmann method (LBM) to implement fast wind environment analysis. The LBM solver can handle multiple plan configurations and realize the needs of large-scale scene calculation and real-time visualization of results. Automatic Reynolds scaling is used to ensure accurate results, making it possible to apply similarity concepts for large Reynolds numbers.

In the high-density urban areas of Shanghai, representative block units were selected based on three indicators, namely building density, plot ratio, and green coverage, to characterize the overall characteristics of the wind environment analysis. Hence, the boundaries of the typical block units selected in this study were used as the boundaries for CFD simulation analysis.

3. Results

3.1. Analysis of the Meteorological Characteristics of the Study Area

3.1.1. Interannual Variability of Climate

Based on the graph showing the interannual variation of the annual average temperature in Shanghai from 1950 to 2020 (Figure 2a), Shanghai’s annual average temperature in the focal period ranged from 15 °C to 17.5 °C, with an overall increasing trend. From the temperature changes in different stages, the annual average temperature in Shanghai increased insignificantly and with less volatility from 1950 to 1990 and showed a significant fluctuating upward trend from 1990 to 2020.

From the graph of the interannual variation of the annual average wind speed in Shanghai from 2000 to 2020 (Figure 2b), the annual average wind speed in Shanghai in this period ranged from 5.5 km/h to 9.0 km/h. From the changes in wind speed in different stages, the annual average wind speed fluctuated between 2000 and 2010 and the annual average temperature showed an overall increasing trend from 2010 to 2020.

3.1.2. Periodic Variation of Temperature

Wavelet analysis of the annual average temperature in Shanghai from 1950 to 2020 was performed using Matlab. A real-part contour plot was created to plot the real part–time–frequency variation of wavelet transform coefficients, with positive areas representing high-temperature bias and negative areas representing low-temperature bias. The plot revealed changes in the centers of high and low values of wavelet coefficients in some years. Among them, the annual mean temperature wavelet coefficient contours on the interannual variation scale (less than 10 years) were relatively dense, with a time scale of about 5–6 years, and the center of the scale was around 6 years. On the interannual variation scale (greater than 10 years), the central scale of the first main cycle in 1990 changed from a 44-year cycle to a 32-year cycle and the center of the abrupt change point was located around 1990. In addition, there was a significant sub-cycle, with a central scale of about 30 years between 1950 and 1990; after 1990, the central scale decreased to about 15 years. Overall, the primary and secondary cycles both showed a trend of becoming shorter and shorter, and more frequent temperature changes can be expected in the future (Figure 3).

3.1.3. Periodic Variation of Wind Speed

Wavelet analysis of the annual average wind speed in Shanghai from 2000 to 2020 using Matlab is represented in a wavelet transform diagram, which shows that on the interannual variation scale (less than 10 years), the wavelet coefficients of the real part of the annual average wind speed from 2000 to 2010 were relatively dense, with a time scale of about 5–6 years, with the center of the scale around 6 years. From 2010 to 2020, the central scale moved from 6 to 7 years. The interdecadal variability scale (>10 years) showed a significant main cycle of 22 years with a sub-cycle of 11–12 years, and the cycle scale remained unchanged from 2000 to 2020 (Figure 4).

3.2. Analysis of the Spatial Patterns of Urban Development Elements and the Urban Heat Environment

3.2.1. Spatial Pattern of Urban Development Elements

The urban development elements under study included the greening and water coverage, building density, and floor area ratio. A total of 3441 300 m × 300 m grids covering the study area were established using ArcGIS, and the greening and water coverage within the grids were calculated based on Landsat 8–9 OLI/TIRS C2 L2 remote sensing imagery from August 2022, using supervised classification for surface coverage extraction. The building density and floor area ratio within the grid were calculated using the building outline and building height data from Gaode Map.

The greening and water coverage in the study area showed a spatial pattern of high in the west and low in the east, and the coverage along the Huangpu River in the east was also relatively high. Yangpu District, Hongkou District, Jing’an District, southern Xuhui District, and the whole of Huangpu District had low greening and water coverage, and the areas with the highest coverage were the park and university distribution areas in the northern part of Yangpu District and the zoo and park distribution areas in the western part of Changning District. Overall, the study area was found to have a high urbanization rate, but the overall greening and water coverage were high due to the humid and hot climate (Figure 5a,b).

The building density and plot ratio shows a spatial pattern of high in the middle along the road and low around. Since the administrative district boundaries are mostly divided according to the city’s main roads, the areas with the highest building density are mainly concentrated in the boundaries of administrative districts such as the Putuo District, Jing’an District, and Changning District, which are distributed along the city’s main roads such as North Cross Road, Huaihai Road, Yan’an Road, and Zhongshan Road; in addition, the eastern riverside areas of Huangpu District are mostly old urban areas with high building densities (Figure 5c) Areas with a relatively high floor area ratio were located in the south of Putuo District and Jing’an District and distributed in horizontal strips along Beihengdao Road and Yan’an Road (Figure 5d).

3.2.2. Spatial Patterns of the Urban Heat Environment

The land surface temperature was used to characterize the urban heat environment. Based on Landsat 8–9 OLI/TIRS C2 L2 remote sensing imagery from August 2022, the land surface temperature was obtained by inversion.

The results show that the land surface temperature in the study area in August 2022 ranged from 35.62 to 65.10 °C. The temperature inversion and grid zoning statistics were relatively consistent in terms of their spatial distribution patterns, showing an overall distribution trend of high in the west and low in the east. Among them, the highest land surface temperatures were located mainly around Hongqiao International Airport in the southwestern part of Changning District, in industrial and large residential areas in the northwestern part of Putuo District, and in large residential areas in the northern part of Jing’an District. Lower temperatures were mainly found in the eastern part of the study area along the Huangpu River (Figure 6).

3.3. Correlation Analysis between Urban Development Elements and the Urban Heat Environment

3.3.1. Geographically Weighted Regression Modeling

We conducted global spatial autocorrelation analysis on the LST and urban development elements within the study area. Global spatial autocorrelation is an indicator used to reflect whether spatial data exhibits clustering or dispersion trends, as well as the strength and significance of these trends. Moran’s I is the ratio of covariance to variance and a measure of the spatial autocorrelation coefficient. A small p-value (usually p < 0.05) indicates that we can reject the null hypothesis of complete spatial randomness and accept that spatial autocorrelation exists, and z-score values are indicative and can be differentiated based on data [51]. The results show that the Moran’s I and z values of each indicator are greater than 0 and the p-values are less than 0.001, indicating that the global spatial autocorrelation of the outbreak frequency is highly significant and exhibits positive correlation. The Moran’s I values for the LST and greening and water coverage are 0.51 and 0.54, respectively, with z-values of 41.80 and 43.95, respectively, which are higher than the building density and floor area ratio, indicating a higher positive correlation in spatial terms.

The multicollinearity of the greening and water coverage, building density, and floor area ratio within the study area is examined. The variance inflation factor (VIF) is a measure of the severity of multicollinearity in a regression model, which represents the ratio of the variance of the regression coefficient estimate to the variance, assuming no non-linear correlation between the independent variables. The VIF values of all indicators are below the threshold of 5, indicating no or low multicollinearity issues (

Table 2. Global spatial autocorrelation significance and multiple collinearity test.

Projects	Moran’s I	z	p	VIF
LST	0.51	41.80	0.000	-
Greening and Water Coverage	0.54	43.95	0.000	1.27
Building Density	0.29	32.81	0.000	4.16
Floor Area Ratio	029	23.55	0.000	3.91

).

Geographically weighted regression analysis was conducted using ArcGIS to explore the correlation between the urban heat environment and urban development elements. A geographically weighted regression model was constructed by taking the land surface temperature as the dependent variable and the greening and water coverage, building density, and floor area ratio as the independent variables in 3441 grids of 300 m × 300 m within the study area. The determination coefficient R² is a measure of goodness of fit, indicating the explained variation of the dependent variable. Adjusted R² is a measure of goodness of fit that is more reliable than R². The coefficients of R² and adjusted R² were 0.82 and 0.78, respectively, indicating that the geographically weighted regression model explained the correlation between the urban heat environment and urban development factors (

Table 3. Geographically weighted regression model determination coefficients.

R2	Adjusted R2	AICc	Sum of Residual Squares	Bandwidth (km)
0.82	0.78	13,616.07	7997.74	0.94

). The confidence level p-value is specific to individual regression models. Geographically weighted regression (GWR), on the other hand, is essentially composed of regression analyses for multiple spatial units and there is no average p-value to represent the overall confidence level.

3.3.2. Analysis of Geographically Weighted Regression Results

The local R² of the model represents the goodness of fit for each grid, ranging from 0.04 to 0.96, with a median value of 0.77. This indicates that the geographically weighted regression model has strong explanatory power and significant spatial heterogeneity (Table 2). Regions with local R² values greater than 0.5 were considered to have a strong explanatory effect on the model, and the distribution map showed that more than half of the regions in the study area had local R² values greater than 0.5. These were mostly located along the Huangpu River in the eastern part of the study area, in most of Changning District, and in the southern part of Putuo District. The R² values in some regions were less than 0.5, indicating that these areas may be affected by other factors influencing the surface temperature (Figure 7a).

Regression coefficients represent the degree and direction of the correlation between the predictor variables and the outcome variable. Overall, there is a significant negative correlation between the greening and water coverage and surface temperature. The regression coefficients of the greening and water coverage ranged from −18.75 to 1.32, with a mean value of −6.72. The regression coefficients for most of the areas were negative, indicating that the greening and water coverage had a negative correlation with the land surface temperature in these areas because the transpiration effect of greening and water bodies and the strong sunlight reflection effect of water bodies can effectively reduce the land surface temperature (

Table 4. Geographically weighted regression results.

Projects	Average Value	Median	Maximum Value	Minimum Value	Standard Error
Local R2	0.52	0.77	0.96	0.04	0.19
Greening and Water Coverage	−6.72	−3.82	1.32	−18.75	3.17
Building Density	9.38	3.99	34.74	−21.20	5.81
Floor Area Ratio	−1.25	−0.14	5.35	−6.24	1.28

, Figure 7b).

Building density had a significant enhancement effect on the land surface temperature. The maximum value of the regression coefficients was 34.74, the average value was 9.38, and the regression coefficients of building density were positive in most of the areas studied. In areas with higher building densities, buildings and hard surfaces have a stronger capacity to absorb solar radiation and there is a greater discharge of artificial heat sources, resulting in higher surface temperatures in the vicinity. The higher regression coefficients observed around large water bodies may be due to the higher specific heat capacity of water, leading to the formation of local cool islands. When the building density increases around these water bodies, it intensifies the obstruction of water–land circulation by buildings, thereby enhancing the warming effect of building density (Table 4, Figure 7c).

The floor area ratio had a mitigating effect on the land surface temperature. The regression coefficients had values between −6.24 and 5.35, with a mean value of −1.25, and their values were negative in most of the study areas. Due to the shading effect caused by the high floor area ratio in built-up areas, most regions show a characteristic where a higher floor area ratio results in lower surface temperatures. However, there are only a few cases around large parks, green spaces, and water bodies where an increase in floor area ratio is associated with an increase in the surface temperature. This is because the buildings surrounding these areas obstruct the water–land circulation, leading to a warming effect from the increased floor area ratio (Table 4, Figure 7b).

To better understand the regression effects of GWR, we plotted the spatial distribution map of standardized residuals and used the local Moran’s I to estimate the spatial autocorrelation to track potential clustering in the residuals. The results show that the residuals are not significant and are randomly dispersed in most areas. This indicates that GWR has addressed the spatial heterogeneity issues in most locations. Only a few regions exhibit clustering and spatial outliers, indicating the presence of spatial heterogeneity in some areas or the need for other variables to be included in the model [52,53] (Figure 8).

3.4. Validation of the Correlation between Urban Development Elements and the Heat Environment

3.4.1. BP Neural Network Establishment

Machine learning can be used to identify potential relationships between different indicators, particularly non-linear relationships [54,55,56]. Building a BP neural network to analyze the correlation and importance of various indicators between the urban thermal environment and urban development elements can be compared with geographic weighted regression results to assess the performance advantages and disadvantages of the two regression methods and better identify the complex relationships between the urban thermal environment and urban development elements.

A BP neural network was constructed using Matlab to examine the correlation between the urban heat environment and urban development elements. A three-layer back-propagation neural network model was used, with the vegetation coverage, building density, and floor area ratio of a 300 m × 300 m grid within the study area as the input layer; the land surface temperature as the output layer; and two hidden layers. The samples were set as the training set, the validation set, and the test set in the ratio of 70%, 15%, and 15% to build a BP neural network framework with a 3-2-1 topology (Figure 9).

3.4.2. Results of BP Neural Network Analysis

The neural network training effects and regressions were obtained using the Levenberg–Marquardt algorithm with an intermediate process of hidden layers. The results show that the MSE metrics of the BP training process, validation process, and testing process tended to perform consistently in each generation and were most effective at the 202^nd iteration of training (Figure 10).

The training set, validation set, test set, and overall fit probability parameter R were all greater than 0.70, which indicated a good fit. The overall fit probability parameter of this training set was greater than 0.75, and the fit probability parameters of the training set, validation set, and test set were 0.73, 0.72, and 0.81, respectively, indicating that the correlation between the urban heat environment and urban development elements was strong. This validated the results of the geographically weighted regression model (Figure 11).

The importance of urban development elements was estimated using SPSS for neural network radial basis function analysis, inputting urban heat environment data and urban development elements data and setting a 70% training layer, a 15% validation layer, and a 15% test layer. The results show that the importance of the greening and water coverage, building density, and floor area ratio were 0.54, 0.27, and 0.20, respectively, indicating that the greening and water coverage had the greatest influence on the urban land surface temperature, building density had the second greatest influence, and floor area ratio had the least influence (Figure 12). This was generally consistent with the results of the geographically weighted regression (Table 2).

3.5. Correlation Analysis between Urban Development Elements and the Urban Wind Environment

Shanghai is located in a subtropical monsoon climate zone with four distinct seasons, sufficient sunshine, and abundant rainfall. According to the Weather Spark website, the average wind speed in Shanghai on 14 August 2022 was 18.3 km/h, which is equivalent to 5.08 m per second. The daily average wind directions in the order of prevalence are east (39%), south (36%), north (16%), and west (9%), with a southeasterly wind.

The study initially selected the Huangpu District as the CFD test area, simulating a southeast wind with a height of 1.5 m above ground and a wind speed of 5.08 m/s. The results showed that, at the urban scale, only the main and secondary roads in the city form ventilation corridors with higher wind speeds. It was not possible to identify the detailed influence of the building density and floor area ratio on the wind speed. Therefore, it is necessary to reduce the scale of the site simulations (Figure 13).

To further explore the impact of urban development elements on the wind environment, this study selected four urban spatial patterns: low floor area ratio–high building density, low floor area ratio–low building density, high floor area ratio–high building density, high floor area ratio–low building density. A total of eight sites were chosen for computational fluid dynamics simulations. The selection of urban forms was based on a preliminary analysis of the existing urban structure in downtown Shanghai. Our aim was to include various common urban forms and fully consider factors such as building height, density, and orientation. The simulations aimed to illustrate the wind speed conditions under different urban spatial patterns. As DEEPUD cannot adjust the size of the field, it inevitably produces the canyon effect around the site, thus the study only focuses on the wind environment between buildings. Wind speeds around the site are not considered relevant.

3.5.1. Low Floor Area Ratio Scenario

Sites with low plot ratios were selected and two types of urban forms, high building density and low building density, were simulated. In the wind environment simulation for areas with a low floor area ratio and high building density, it was found that in such sites, the overall wind speed was low and it was more difficult to form a good ventilation environment. Wind environment simulation for areas with a low floor area ratio and low building density revealed that the overall wind speed of such sites was relatively high and the airflow formed within sites with better ventilation. The lower the building density, the higher the wind speed of the site, Indicating a negative correlation between building density and wind speed (Figure 14).

3.5.2. High Floor Area Ratio Scenario

Plots with high floor area ratios were selected for wind environment simulation, and they were also divided into two types of urban forms: low building density and high building density. First, wind environment simulation was conducted for areas with a high floor area ratio and low building density, and when the distance between buildings was large, a higher wind speed could form in the gap and the site had a better ventilation environment. When wind environment simulations were conducted for areas with high floor area ratios and high building densities, it was found that the wind speed at such sites was lower and it was more difficult to develop a good ventilation environment. When the building density is too high, it is difficult for airflow to enter the interior of a building complex, reducing the wind speed inside the site. The simulation of the above two scenarios further verified the negative correlation between building density and site wind speed (Figure 15).

3.5.3. Low Building Density Scenario

The sites with low building densities were selected and divided into two types of urban forms, those with a low floor area ratio and those with a high floor area ratio, for simulation. Simulation of areas with low building density and a low floor area ratio revealed that the wind speed within these sites was relatively high. Airflow could enter the interior of the building complex smoothly, forming a good ventilation environment. Simulation of areas with low building densities and high floor area ratios showed that airflow could still enter the interior of the site relatively smoothly. Therefore, the above two forms of sites had a good ventilation environment and the floor area ratio did not have a significant effect on the low building density of the sites (Figure 16).

3.5.4. High Building Density Scenario

The sites with high building densities were divided into two types of patterns: low or high floor area ratios. When simulating the sites with high building densities and low floor area ratios, it was found that the wind speeds in the sites were close to 0, showing a dark blue color. When simulating the sites with high building densities and high floor area ratios, it was found that the wind speeds in the sites were smaller, except for the ventilation corridor area, and a larger area of dark blue appeared. This suggests that the effect of the floor area ratio on wind speed is not significant enough (Figure 17).

4. Discussion

4.1. Urban Development Elements and the Urban Heat Environment

According to the results of the geographic weighted regression and BP neural network, it is known that the vegetation and water coverage and plot ratio are negatively correlated with that land surface temperature (LST), whereas building density is positively correlated with the LST. Previous studies have confirmed that in tropical and subtropical cities with similar climate conditions to Shanghai, areas with high vegetation and water coverage have a smaller impermeable surface area on the ground, and both impermeable surface and urban vegetation significantly affect the LST and heat balance components of the city [57]. This is because the higher the percentage of impermeable surface in the urban heat island, the stronger the heat performance. Meanwhile, when there are fewer impermeable surfaces and urban vegetation is more representative in the land cover/land use categories, the urban heat island effect will weaken [58]. The same conclusion has also been proven in temperate cities, where studies suggest that the reason for higher urban temperatures compared with suburban areas in cold seasons is due to the heat release and loss from heating systems; in the absence of snow cover, the surface heat radiation in the city enhances the anthropogenic impact of these sources [59].Therefore, in order to reduce the urban surface temperature, measures should be taken to increase vegetation and green coverage, reduce building density, and control plot ratio. These measures are not only applicable to Shanghai but have also been proven effective in other tropical, subtropical, and temperate cities.

In terms of ranking the relevance of urban development elements to the heat environment, the greening and water coverage had the greatest effect on the urban land surface temperature, with building density having the next greatest effect; the floor area ratio had the smallest effect. Studies have been conducted to investigate the effects of the occupancy factor, floor area index, and vegetation cover index on urban microclimate temperature. They have also concluded that the range of values for the vegetation factor can lead to more significant temperature differences [60]. One study has been conducted using the unsteady Reynolds-averaged Navier–Stokes (URANS) equations for computational fluid dynamics (CFD) simulations to analyze the cooling effect of street trees during heatwaves in urban environments. It was found that vegetation can effectively lower the surrounding temperature, thereby reducing energy consumption and promoting thermal comfort conditions [61]. Furthermore, in terms of human-perceived thermal comfort, pedestrians generally tend to choose walking and sitting positions with high thermal comfort, such as shaded areas on sidewalks. There is also a correlation between the level of thermal comfort and pedestrian behavior patterns. Additionally, sky view factor (SVF) and tree canopy coverage (TCR) are also correlated with pedestrian thermal comfort [62]. Therefore, to improve the urban thermal environment, it is important to focus on increasing vegetation and water coverage, as well as creating large parks, road greening, and street parks.

To optimize the urban thermal environment, control of urban development should be managed at three levels: macro, meso, and micro. At the macro level, spatial planning strategies that limit urban sprawl would significantly help to constrain the further increase in urban heat island intensity [63]. At the meso level, the thermal comfort index shows non-monotonic variation with the change of urban density due to the offset effect of changes in wind speed and radiation pattern on thermal comfort. On the one hand, to achieve the required thermal comfort in hot climates, the strategy for enhancing urban ventilation is preferred in high-density areas, whereas a minimizing radiation strategy is required for lower building densities [64]. On the other hand, green roofs have no negative impact on the urban heat island surface in winter and cool the environment in summer while helping to insulate buildings, thereby saving building energy and making them a suitable urban greening solution [65]. At the micro level, the most effective way to reduce the average radiant temperature and human heat sensation in summer is to create large shaded spaces by constructing artificial shading devices such as pergolas and planting trees with high canopy density [66]. Furthermore, the heat environment within cities is better with than without watersheds. Watersheds with larger surface areas show greater cooling effects and the average radiant and physiological temperatures are optimized in watersheds parallel to prevailing winds [67].

4.2. Urban Development Elements and the Urban Wind Environment

Correlation analysis between urban development elements and the urban wind environment in the study area confirmed that building density had a significant attenuating effect on wind speed and the effect of the floor area ratio on the wind environment was not significant. Related studies have demonstrated that: wind speed has a significant effect on urban heat island characteristics and intensity, as well as humidity and air mass temperature daily variability [68]; wind speed has a more profound effect on the heat environment in hot and humid subtropical environments; urban wind fields can effectively mitigate urban heat island effects; and ventilation driven by wind and thermal buoyancy can dissipate heat islands and remove heat from urban areas [69]. In Mediterranean climates, enhanced wind speeds are considered positive in summer but unpleasant in winter [70]. Therefore, increasing wind speed to improve urban environmental comfort is more suitable for tropical and subtropical regions. Countries with colder climates may consider high wind speeds as having negative impacts, especially in winter.

As building density has a significant weakening effect on wind speed, adjusting building density can affect urban wind speed and improve urban wind environment comfort. Regarding the effects of building density and floor area ratio on the wind environment, some studies have concluded that the perimeter densification of high-rise buildings and internal grid discontinuous streets indicates poorer ventilation conditions and that winds from large perimeter roads are blocked from reaching internal areas, creating an enclosure effect [71]. Stepped building configurations can significantly increase canyon wind speeds, whereas higher building spacing ratios are associated with lower average temperatures in urban canyons [72]. The effect of the floor area ratio on the wind environment is reflected in the fact that the presence of tall buildings significantly alters the prevailing wind environment around buildings in terms of time-averaged and instantaneous flow characteristics, due to the complex interaction between flows around buildings and in and around the surrounding street canyons. Tall buildings create a large gust factor area where downwash and street flows collide [73]. To minimize the challenges to safety and pedestrian comfort posed by strong winds, roadside green spaces with surface areas ranging from 0.7 to 1.4 can reduce wind speed in the nearby urban environment by 0.5–1.5 RTV. Therefore, wind speed should be regulated in the planning and design of the tree structure and greenery coverage of green spaces to maintain effective cooling and avoid affecting the range of pedestrian comfort [74]. Therefore, optimizing the urban wind environment involves reducing building density and height to avoid affecting urban ventilation and increasing roadside trees to ensure a comfortable commuting experience for urban residents in strong winds.

4.3. Inspiration and Limitations

Research on wind and thermal environment optimization using methods such as wavelet analysis, geographically weighted regression, a BP neural network, and wind environment simulation, compared with traditional research methods that use statistical data for linear modeling, is innovative. In this study, wavelet analysis was used to investigate the periodic variations of the temperature and wind speed in the study area. This method provides a more intuitive illustration of climate change trends, proving the necessity and urgency of improving the urban wind and thermal environment. It also quantifies the patterns of meteorological changes, enabling predictions of future climate.

The study employed geographically weighted regression to examine the correlation between surface temperature, vegetation coverage, water coverage, and other data obtained through remote sensing image inversion. The impact of urban development elements on surface temperature was the main focus, and BP neural network analysis was used for comparison and validation. The consistency between the geographically weighted regression and BP neural network results indicates high goodness of fit for both models. They effectively demonstrate the role of urban development elements in surface temperature from spatial heterogeneity and nonlinear regression perspectives.

Furthermore, the study utilized DEEPUD as an intelligent urban research platform for wind environment simulation. This platform enables real-time and fast simulation of urban wind environments at different scales and levels of detail. By controlling variables and forming four control groups, it provides new research approaches to investigating the correlation between different urban development elements and wind environments.

Based on the correlation between urban development elements and the wind and thermal environment, the study proposes optimization pathways and establishes new goals for urban planning. In terms of the urban thermal environment, increasing public green spaces and landscape areas, improving vegetation and water coverage, and reducing surface temperatures are recommended. Planting trees and shrubs on both sides of roads to create shaded areas can reduce the direct solar radiation on pedestrians and building surfaces, enhancing thermal comfort. Increasing green roof coverage effectively improves building insulation and cooling. Gradually reducing urban building density can lower the surface temperature, improve urban ventilation, and enhance the livability of the city. Rational urban land planning should avoid excessively dense city areas to ensure ventilation and nighttime heat dissipation, alleviating the urban heat island effect. Controlling building height and coverage can minimize the obstruction of surrounding environments by buildings, whereas adopting low plot ratios in urban development increases public spaces and green areas, reducing the urban heat island effect.

In terms of the urban wind environment, increasing urban ventilation pathways and open spaces such as parks, squares, and waterways to provide airflow channels and promote air movement to improve the urban wind environment is recommended. Additionally, increasing greenery, such as street trees and lawns, can block wind currents while maintaining the aesthetic appeal of the city. Adjusting building layout and height to avoid the formation of wind tunnels between buildings and preventing excessive wind speeds or strong wind areas is crucial. The use of appropriate wind direction indicators and shielding structures can minimize the impact of wind on pedestrian walkways, squares, and other areas. Furthermore, thermal environment simulation and wind environment assessment can be conducted to develop corresponding urban planning measures based on scientific data.

These research findings can provide useful reference values for other cities to optimize the layout of urban development factors and improve their urban wind and thermal environments. Especially in today’s world, where we are faced with climate change and increasingly extreme weather events, this study is significant and valuable for achieving sustainable urban development, protecting the lives and health of urban residents, and promoting spatial justice and well-being (given that the majority of low-income groups reside in areas with higher building densities).

In addition, research also has certain limitations. On one hand, combining geographically weighted regression with BP neural networks into one model for nonlinear regression modeling based on spatial matrices can fully utilize the advantages of these two models, resulting in more accurate prediction results that may better explain the correlation between urban development factors and the urban thermal environment. On the other hand, expanding the scope of wind environment simulation can explore the correlation between urban development factors and the urban wind environment on a larger scale while avoiding the occurrence of the canyon effect.

5. Conclusions

In this study, the effects of urban development elements such as the greening and water coverage, building density, and floor area ratio on the urban wind–heat environment were studied by means of land surface temperature inversion and greening and water coverage extraction, geographically weighted regression, the BP neural network, and CFD numerical simulation, using the high-density urban area of Shanghai as the research scope. The results were as follows:

(1)

In terms of the meteorological characteristics, the annual average temperature in Shanghai from 1950 to 2020 showed an overall increasing trend, with strong significant cycles of 44 and 32 years, sub-cycles of 30 and 15 years, and interannual scale variation cycles of 6 years, with both the main and sub-cycles showing a shortening trend. There was a significant main cycle of 22 years, a sub-cycle of 12 years, and an interannual scale variation cycle of 5–7 years.

(2)

Regarding the spatial pattern of urban development elements and the urban heat environment, the greening and water coverage in the study area showed a spatial pattern of high in the west and low in the east and the building density and floor area ratio showed a spatial pattern of high in the middle along the road and low in the surrounding area. The land surface temperature in the study area in August 2022 ranged from 35.62 to 65.10 °C, with an overall distribution trend of high in the west and low in the east.

(3)

As for the correlation between urban development elements and the heat environment, the greening and water coverage and floor area ratio had significant negative effects on the land surface temperature, whereas building density was positively correlated with the land surface temperature; greening and water coverage had the greatest effect on the urban land surface temperature, followed by building density; the floor area ratio had the smallest effect.

(4)

Regarding the correlation between urban development elements and the wind environment, building density has a significant attenuating effect on wind speed and the effect of floor area ratio on the wind environment is not significant.