Impacts of spatial explanatory variables on surface urban heat island intensity between urban and suburban regions in China

ABSTRACT

The intensified thermal environment in suburban areas is raising wide concerns for human society and public health due to rapid urbanization. Although the satellite-derived surface urban heat island intensity (SUHII) is a commonly used indicator, it still needs to be determined the SUHII between urban and suburban areas due to the challenges in delineating their boundaries with changes. Thus, a comprehensive analysis of the spatial explanatory variables (SEVs) and SUHII among urban and suburban areas is highly needed. Here, using the long-term satellite observations, we analyzed the spatiotemporal patterns of SUHII in different temporal intervals (i.e. seasonal and diurnal) and the contribution of SEVs in urban and suburban areas. Our results indicate that SUHII shows predominantly increasing trend from 2012–2021 in cities of China. Despite the trends of SEVs (i.e. increasing/decreasing) being relatively consistent in both urban and suburban, the latter shows a higher proportion regarding the trends in various SEVs. Besides, the partial least squares regression (PLSR) model shows that the major contributors to SUHII in urban areas are landscape shape index (LSI), patch density (PD), and the digital elevation model (DEM), while in suburban areas, those critical SEVs are LSI, normalized difference built-up index (NDBI), and DEM. These findings can facilitate the sustainable design of urban planning in a nature-based solution.

KEYWORDS:

• Land surface temperature
• suburban areas
• urban landscape
• urban heat island
• time series

1. Introduction

China’s rapid urbanization process has caused the urban heat island (UHI) effect by altering the urban land surface (Guo et al. 2020; Xu et al. 2022; Li, Liu, and Yu 2014). The UHI is widely seen in the urban ecosystem, where the temperature in urban areas is hotter than nearby suburban areas (Chatterjee and Majumdar 2022; Amir Siddique et al. 2021; Moazzam, Doh, and Lee 2022). As a commonly used indicator to characterize the UHI, the surface UHI intensity (SUHII) derived from the satellite observations of land surface temperature (LST) data, has been widely used in diverse applications (Moazzam, Doh, and Lee 2022; Yang, Huang, and Tang 2019; Gage and Cooper 2017; Shi et al. 2023). The intensified UHI effect has raised wide concerns on human beings, such as heatwave and energy consumption, and water-atmosphere interactions, particularly under future development goals with sustainable cities (Chatterjee and Majumdar 2022; Moazzam, Doh, and Lee 2022; Amir Siddique et al. 2021; Macintyre et al. 2021; Hsu et al. 2021; Mu et al. 2022).

The spatiotemporal pattern of SUHII is closely related to diverse direct and indirect factors such as socioeconomic level, urban landscape, and urban form (Li et al. 2020; Ramírez-Aguilar and Souza 2019; Adelia et al. 2019; He et al. 2023). Large amounts of socio-economic activities (e.g. population density, building usage patterns, traffic flow and industrial activity) during urbanization can increase the heat capacity of the city, leading to the change of the UHI (Guo et al. 2020; Liu et al. 2022; Liang et al. 2020). In addition, the landscape pattern indexes could also be utilized to analyze the distribution and spatial organization characteristics of various landscape types within cities, facilitating studies relating to the formation and evolution of UHI. However, different landscape indexes can yield varied impacts on SUHII due to their distinct spatial patterns of land use and land cover (Shao et al. 2023; Guo et al. 2020; Liu et al. 2021). Moreover, the natural environment plays a significant role in influencing UHI through various mechanisms. For instance, topographic features such as mountains and rivers can alter the climate and temperature distribution within cities, thereby impacting the intensity of UHI (Liu et al. 2023; Hu, Dai, and Guldmann 2020). Besides, a series building environment (e.g. building height, building density and building materials) can impact SUHII by obstructing wind flow and absorbing and storing solar energy (Shao et al. 2023; Hu, Dai, and Guldmann 2020; Yin et al. 2018; Zhou et al. 2022). However, existing studies mainly focused on examining the impact factors of SUHII only from the perspective of urban areas. Meanwhile, there is a lack of long-term investigation into the multidimensional drivers of SUHII.

Suburban areas, as sensitive zones of urbanization, have a gap in studies of the long-term spatial explanatory variables (SEVs)’ impacts on SUHII due to the difficulty of identifying their boundaries and their distinct differences from urban area (Moazzam, Doh, and Lee 2022; Ming et al. 2023; Zhang, Li, and Han 2020). The suburban areas are seen as a ‘bridge’ connecting urban and rural areas. As a transition area between urban and rural areas, suburban areas undergo complex and dynamic demographic, landscape, and socioeconomic processes (Tian and Qian 2021; Liu, Shi, and Wu 2022). This dynamic process has significant implications for SUHII formation and diffusion. Different from urban or rural areas, suburban areas are associated with a high degree of mixed land uses, making it difficult to define their extent (Tian and Qian 2021; Koch et al. 2019; Peng et al. 2018). Although existing studies about SUHII have delineated simple buffer zones for suburban areas (Zhou et al. 2014; Yang, Huang, and Tang 2019), such a manner may underestimate or overestimate the SUHII for large cities and small cities (Yao et al. 2018; Liu et al. 2023), due to the heterogeneity of China's urban scale and development. Although several studies that focus on identifying suburban areas have been carried out (Tian and Qian 2021; Koch et al. 2019; Peng et al. 2018), they are limited to individual cities or provinces and lack accurate identification of suburban areas over a long temporal span. Examining SUHII based on suburban and urban extent not only excludes the influence of natural factors on SUHII in rural areas but also avoids the misestimation of SUHII in small and medium-sized cities caused by the extent of consistent buffer zones in the rural areas, allowing us to better understand the contribution of urbanization to temperature (Yao et al. 2018; Zhou et al. 2014; Yao et al. 2019; Li et al. 2020).

The novelties of this study include (1) calculating SUHII based on the identification of long time-series suburban regions, and (2) comparing the impacts of long time-series urban and suburban SEVs on SUHII. The general objective for this study is to quantitatively assess SUHII by creating a comprehensive long-term dataset specifically focused on suburban areas. Meanwhile, it aims to analyze and compare the similarities and disparities regarding the impacts of urban and suburban SEVs on SUHII. Based on this, we explored the impacts of long-term (2012–2021) multi-dimensional urban and suburban SEVs on daytime and nighttime SUHII. We first calculated SUHII based on long time-series urban and suburban boundaries. Then, we obtained multidimensional SEVs within the urban and suburban boundaries and explored the effect of multidimensional SEVs on SUHII in urban and suburban areas. The findings will enhance our understanding of the correlation between SEVs and SUHII, providing valuable insights to inform and guide evidence-based practices in urban planning and design.

2. Study area and data source

2.1. Study area

In this study, we selected prefecture-level cities in China as our basic research unit. We excluded cities with missing data, primarily due to the unavailability of MODIS LST and other ancillary data (e.g. statistics). In total, 201 prefecture-level cities in China were finally selected, mainly distributed in central and eastern China, where the urbanization levels and the consequent UHI are distinct (Liu et al. 2023; Li et al. 2020; Gong, Li, and Zhang 2019). Besides, we divided these cities into four groups (i.e. 15 megacities, 102 large cities, 55 medium-sized cities, and 29 small cities) for analysis in our study (Figure 1, Table S7), according to standards from the State Council.

Figure 1. Selected 201 prefecture-level cities in China.

2.2. Data source

We adopted LST data and the China suburban extent datasets as our primary datasets for SUHII analysis and other ancillary datasets (e.g. population, statistics, nighttime light, and land use/cover). Here, we derived the LST data in 2012–2021 from the Moderate Resolution Imaging Spectroradiometer (MODIS) 8-day composite product (MOD11A2 product version 006, Terra), around 10:30 am and 10:30 pm (Yao et al. 2019; Si et al. 2022; Wan, Hook, and Hulley 2015). This data has good performance in characterizing LST due to its agree with in situ measured LSTs well within ±1 K in the range from 263 to 322 K (Wan et al. 2002). The China suburban extent data were derived in our previous study using nighttime light data (Liu, Shi, and Wu 2022), which contains urban and suburban boundaries over the past decade. In addition, we used other ancillary datasets to analyze the relationships between the SEVs and SUHII. We derived population density (Pop) and per capita road density (Road) data from the LandScan (Dobson et al. 2000) and statistic yearbook, respectively, to reflect the socioeconomic activities in cities. The nighttime data from satellites was also included in our study. Meanwhile, we calculated three landscape indexes, including patch density (PD), landscape shape index (LSI), and Splitting Index (SPLIT), respectively, using the land use/cover data (Friedl and Sulla-Menashe 2022). These indexes were widely used for the SUHII analysis (Guo et al. 2020). In addition, we used the digital elevation model (DEM) and green coverage (Green) (including forests, shrublands, Savannas, and grasslands) to characterize the nature environments, which were derived from the Shuttle Radar Topography Mission (STRM) DEM (Reuter, Nelson, and Jarvis 2007) and land use/cover data, respectively. Finally, we extracted building height (Height) and normalized difference built-up index (NDBI) to indicate the built-up environment, using the Chinese building height (CNBH) (Wu et al. 2023) and the MODIS Nadir Bidirectional Reflectance Distribution Function Adjusted Reflectance (NBAR) product (MCD43A4 V6) (Schaaf and Wang 2021), respectively. We calculated these variables for urban and suburban areas separately. Details about these variables and their information can be referred to in Table S1.

3. Methods

We developed the analysis framework below (Figure 2) to explore urban and suburban SEV’s impacts on the SUHII. First, we calculated SUHII from MODIS data using derived urban and suburban boundaries from NTL satellite observations (Figure 2(a)). Second, we derived multiple SEVs from different aspects, such as the socio-economic activities, landscape indexes, natural and built-up environments (Figure 2(b)). Finally, we explored urban and suburban SEV’s impacts on the daytime and nighttime SUHII, across different temporal scales (e.g. annual, seasonal, and diurnal) (Figure 2(c)).

Figure 2. The proposed overall framework in our study, including the calculation of SUHII using urban and suburban boundaries (a), the derived SEVs from multiple aspects (b), and the exploration of the SEV’s impacts on SUHII (c).

3.1. The calculation of SUHII using urban and suburban boundaries

We calculated the SUHII using the temperature difference between urban and suburban areas based on LST data (see Eq. 1(1) $\mathrm{SUHII}=\mathrm{LS}{\mathrm{T}}_{\mathrm{urban}}-\mathrm{LS}{\mathrm{T}}_{\mathrm{sub}}$(1) ). We removed unreliable and cloud effects pixels of LST data and derived the complete monthly LST of daytime and nighttime through the 8-day LST data monthly compositions, during the whole year and specific seasons (i.e. summer (June to August) and winter (December to February)). We processed the LST data from 2012 to 2021, using the Google Earth Engine (GEE) platform. It is worthy to note that cities with substantial missing pixels mainly located in western region, which have been excluded in our study. Additionally, we only selected cities with both urban and suburban pixels when calculating the SUHII. (1) $\mathrm{SUHII}=\mathrm{LS}{\mathrm{T}}_{\mathrm{urban}}-\mathrm{LS}{\mathrm{T}}_{\mathrm{sub}}$(1) We also derived urban and suburban boundaries from NTL datasets, spanning from 2012-2021. Satellite-derived NTL data have been widely used to characterize diverse socio-economic activities, such as population, electricity consumption, and building density (Li et al. 2020; Shi et al. 2014; Zhao et al. 2020; Mu et al. 2022; Li and Zhou 2017). Here, we used the NTL data to identify urban and suburban areas (Liu, Shi, and Wu 2022; Tian and Qian 2021). First, we derived rough urban and suburban boundaries from the NTL data utilizing the K-means approach (Figure 3(a)). Then, we implemented a temporal consistency approach to ensure the rationality of identified urban and suburban pixels regarding their changes in consecutive years (e.g. from non-urbanized to urbanized pixels monotonously) (Li, Gong, and Liang 2015; Li and Gong 2016) (Figure 3(b)). Finally, we optimized derived urban and suburban boundaries using three rules, including filling the ‘hole’, excluding independently emerged suburban areas, and removing suburban areas smaller than 2km² (Figure 3(c)). Our derived urban and suburban boundaries contained more details, in particularly, the identified suburban areas are consistent with suburban population, vegetation, and landscape characteristics (Fig. S1) (Liu, Shi, and Wu 2022). Details about the urban and suburban boundary can be referred to in Liu et al. (2022). The derived urban and suburban boundaries can used to calculate the SUHII and analyze the urban and suburban socio-economic and environmental applications (Liu et al. 2023).

Figure 3. Illustration of deriving urban and suburban boundaries from NTL datasets, the drived rough urban and suburban boundaries (a), the implementation of temporal consistency (b), and the rules of post-processing (c).

3.2. The derived SEVs from multiple aspects

We selected ten SEVs (i.e. socio-economic activities, landscape indexes, natural and built-up environments) to comprehensively explore the relationship between SEVs and SUHII (Table 1). These variables are frequently used in analyzing the drivers of SUHII as well as being characterized long time series, dynamics, and rich spatial information (Guo et al. 2020; Ming et al. 2023; Shao et al. 2023). And we averaged the different variables within the city units without changing the resolution of the original data. Socio-economic factors have a significant impact on urban climate since they directly drive urban space expansion physically (Guo et al. 2020). Here, we selected three socio-economic indicators, i.e. Pop, mean nighttime light (Light), and Road, to represent different impacts from socio-economic factors. Meanwhile, SUHII is also related to land cover, as well as its spatial configuration and landscape characteristics (Gage and Cooper 2017; Chatterjee and Majumdar 2022). Here, we selected PD, LSI, and SPLIT for a quantitative evaluation across cities, using the FRAGSTATS 4.3 software. Besides, natural conditions (e.g. terrain and vegetation) also contribute to the spatial distribution of LST. For instance, green space is important in mitigating the urban heat island effect through evaporation and shade provision (Liu et al. 2023; Guo et al. 2020). Thus, we selected the DEM and Green as two explanatory variables. In addition, we also included two indicators of Height and NDBI to characterize the built-up environments, both of them closely related to human activities and anthropogenic heat flux (Li et al. 2020; Hu, Dai, and Guldmann 2020; Yuan et al. 2023; Li et al. 2020).

Table 1. Description and calculation process of spatial explanatory variables in this study.

Note: A_it, Pop_it, L_it, represents the area, sum of population density and sum of NTL values per urban or suburban i in year t, respectively; N_it, P_it, NC_it represents the number of patches, sum perimeter of patches and the number of different types of patches per urban or suburban i in year t, respectively; D_i indicates the sum of DEM per urban or suburban i and the G_it represents the sum of forest and grassland area per urban or suburban i in year t; BH_i is the sum of building height per urban or suburban i and ρNIR, ρSWIR denote reflectance in the near-infrared, and mid-infrared bands, respectively.

3.3. The exploration of the SEVs’ effects on SUHII

We explored the spatiotemporal trends of SUHII and SEVs for 201 cities from 2012 to 2021, using the Mann-Kendall (MK) test and Sen's slope approach (Mann 1945). The MK test and Sen's slope approaches were considered reliable for trend detection, which has been widely used to analyze climate and environmental variables (Blöschl et al. 2019; Yang, Huang, and Tang 2019; Siddiqui et al. 2021). Here, we determined the significance according to the MK test (with a threshold of 0.1), while the temporal increasing / decreasing trend was estimated using Sen’s slope approach.

Then, we explored the relationship between urban and suburban SEVs and SUHII using the partial least squares regression (PLSR) model. The PLSR model is robust and advantageous regarding its capacity to analyze both X- and Y-variables with noisy and collinear observations, or incomplete multivariate in standardized data matrices (Yao et al. 2022; Zhang, Li, and Han 2020). It performs iterative regression to estimate the principal components that represent both the X and Y variations, and the algorithm terminates when the regression equations reach satisfactory accuracy through the cross-validation. Specifically, we established the linear regression model to connect the SUHII (Dependent variable Y) and the spatial explanatory variables (Independent variable X) using the PLSR model (see Eq. 2(2) $\mathrm{Y}={\beta }_0+\sum _{\mathrm{t}=1}^{\mathrm{n}}{\beta }_{\mathrm{p}}{\mathrm{X}}_{\mathrm{tp}}+\epsilon$(2) ). (2) $\mathrm{Y}={\beta }_0+\sum _{\mathrm{t}=1}^{\mathrm{n}}{\beta }_{\mathrm{p}}{\mathrm{X}}_{\mathrm{tp}}+\epsilon$(2) where P is the number of X; t = 1, 2, … , n is the size of observation (sample) for X and Y; β_p is the regression coefficient of X_tp, indicating the relative contribution of X_tp to Y.

In addition, we measured the variable importance in the projection (VIP) of urban and suburban SEVs using PLSR model. In general, the greater VIP value is, the more significant the variable is in explaining the model, i.e. the VIP greater than one indicates that the X is significant in explaining Y (Yang et al. 2022; Yao et al. 2022). The formula for calculating VIP is as follows (see Eq. 3(3) $\mathrm{VIP}={\left\{{\displaystyle \frac{\mathrm{P}{\sum }_{\mathrm{h}=1}^{\mathrm{m}}{\sum }_{\mathrm{k}}{\mathrm{R}}^2({\mathrm{y}}_{\mathrm{k}},{\mathrm{t}}_{\mathrm{h}}{)\mathrm{W}}_{\mathrm{hj}}^2}{{\sum }_{\mathrm{h}=1}^{\mathrm{m}}{\sum }_{\mathrm{k}}{\mathrm{R}}^2({\mathrm{y}}_{\mathrm{k}},{\mathrm{t}}_{\mathrm{h}})}}\right\}}^{{\displaystyle \frac{1}{2}}}$(3) ). (3) $\mathrm{VIP}={\left\{{\displaystyle \frac{\mathrm{P}{\sum }_{\mathrm{h}=1}^{\mathrm{m}}{\sum }_{\mathrm{k}}{\mathrm{R}}^2({\mathrm{y}}_{\mathrm{k}},{\mathrm{t}}_{\mathrm{h}}{)\mathrm{W}}_{\mathrm{hj}}^2}{{\sum }_{\mathrm{h}=1}^{\mathrm{m}}{\sum }_{\mathrm{k}}{\mathrm{R}}^2({\mathrm{y}}_{\mathrm{k}},{\mathrm{t}}_{\mathrm{h}})}}\right\}}^{{\displaystyle \frac{1}{2}}}$(3) where P is the number of independent variables; m is the number of extracted components; k is dependent variable; t_h is the hth component of the independent variables; R²(y_k, t_h) is the square of the correlation between y_k and t_h; ${\mathrm{W}}_{\mathrm{hj}}^2$ is the weight of the contribution of the independent variable X_j to the construction of the t_h component.

4. Results and discussion

4.1. Spatiotemporal trends in SUHII

The increasing trends of SUHII (i.e. annual and seasonal) were widely observed in Chinese cities due to the rapid urbanization (Liu et al. 2023; Liang et al. 2020) (Figure 4, Table S8), similar to the findings of Yao et al. (2019) and Yang, Huang, and Tang (2019). The increase/decrease trend of SUHII was more significant at nighttime than daytime. Annual and summer SUHII was dominated by increasing trends (Figure 4(a, b)), i.e. more than 50% of cities exhibit increasing trends in the nighttime, and such a trend is more evident than that in the daytime (Figure 4(c)). However, the winter SUHII showed predominantly decreasing trends both in the daytime (5.97%) and nighttime (18.41%) (Figure 4). The fact of urban areas experience higher temperatures at night is likely due to the absorption and storage of solar radiation by built-up surfaces and buildings, which is notably different compared to suburban areas, where built-up surfaces and buildings are sparsely distributed. This leads to an increase in the nighttime SUHII in many cities (Zhang, Murray, and Turner Ii 2017; Zhou et al. 2014). Besides, from the seasonal aspect, the SUHII was more significant in summer, whereas it is relatively consistent in winter (Figure 4(a,b)), consistent with findings in Yang et al. (2019). The cities with a significant increasing trend of SUHII in summer were approximately 50%, while the proportions of cities without noticeable changes of SUHII in winter were 90.05% and 79.10% in daytime and nighttime, respectively (Figure 4(c)). Increasing human activities, including the use of air conditioning and traffic flow, along with relatively high solar radiation intensity and atmospheric stability, are attributable to the increase of summer SUHII (Zhou et al. 2014).

Figure 4. Annual and seasonal trends of daytime (a) and nighttime (b) SUHII in China’s 201 cities, as well as their proportions of each trend category (c).

4.2. Spatiotemporal trends in SEVs of urban and suburban areas

In general, the SEV’s spatiotemporal trends (e.g. increasing/decreasing) were relatively consistent in urban and suburban regions, with a more pronounced trend observed in suburban areas (Figure 5). In both urban and suburban areas, variables of Road, PD, LSI, and SPLIT exhibited a significant increasing trend, while Pop, Light, Height showed an opposite pattern (Figure 5(a, b)). Specifically, the percentage of cities with a increase in Road, PD, LSI, and SPLIT in urban was 89.05%, 56.72%, 99% and 82.08%, respectively. A similar result was also observed in suburban areas regarding these variables (Figure 5(c)). The growing economy and improved living standards are driving an increasing demand for accessibility and ecological considerations, resulting in higher per capita road area (e.g. Road) and landscape indexes (e.g. PD, LSI, and SPLIT) (Guo et al. 2020; Liang et al. 2020; Liu et al. 2021). The suburban areas exhibited greater increasing/decreasing trends in various SEVs compared to urban areas (Figure 5(a, b)). The proportions of variables in suburban areas such as Road, PD, LSI, and SPLIT were higher than in urban areas, with increments of 4.48%, 23.38%, 1%, and 14.93%, respectively. In addition, variables of Pop, Light, and Height in suburban areas exhibited a decrease of 10.95%, 25.88%, and 8.45%, respectively, compared to that in urban areas (Figure 5(c)). This phenomenon may be attributable to the comprehensive impacts of urban sprawl, suburban land use transformation, and land development, as well as environmental protection and land-use management (Foelske et al. 2019). For some specific indicators, their urban and suburban areas showed opposite trends. For instance, DEM in urban areas exhibited a prominent decreasing trend (50.75%). In contrast, it showed an increasing trend (63.19%) in suburban areas (Figure 5(c)). Green in urban areas was relatively consistent. In comparison, suburban areas exhibited a substantial increasing trend of 83.09% (Figure 5(c)). The urban areas were characterized by relatively low elevation, with extensive built-up areas for residential, commercial, and infrastructure purposes. However, the suburban areas were located at the city's periphery, with relatively lower population density, ample land resources, and distinct green coverage (Liu, Shi, and Wu 2022; Tian and Qian 2021). It was worth noting that NDBI didn’t show any significant temporal variations in urban or suburban areas (Figure 5).

Figure 5. Trends of urban and suburban SEVs in China’s 201 cities. Including urban SEV trends (a), suburban SEV trends (b), and the proportion of different trends (the left is urban areas and the right is suburban areas) (c).

4.3. The VIP values of the urban and suburban spatial explanatory variables

Across different SUHII scenarios, those SEVs with relatively high (i.e. above one) VIP values were distinctly different. This finding is also demonstrated by Shao et al.(2023) regarding the importance of the contribution of individual characteristics to the annual mean SUHII. Still, this characteristic remained relatively consistent across different levels of urbanization (Figure 6). SEVs showed significant variability regarding their explanatory power for LST across different conditions (e.g. urban and suburban at daytime and nighttime) (Figure 6(a)). PD was contributable to explain the SUHII pattern at nighttime (i.e. both in urban and suburban). In contrast, LSI showed a close relationship to the SUHII pattern during daytime (Figure 6(a)). Besides, it was worth noting that those SEVs were different between daytime and nighttime, even in the same category (e.g. urban and suburban). Specifically, LSI, PD, NDBI, and PD were the key indexes that significantly influenced the variation in LST from urban daytime to suburban nighttime, with VIP values of 1.36, 2.00, 1.75, and 1.51 (Figure 6(a), Table S2). Also, SEVs in different sizes cities exhibited similar patterns in terms of their explanatory ability to LST (Figure 6(b)). Among them, natural environment SEVs (VIP value: 1.08) in urban daytime and landscape SEVs (VIP value: 1.26) in urban nighttime had the strongest explanatory power for LST, while landscape SEVs (VIP value: 1.13) in suburban daytime and natural environment SEVs (VIP value: 1.14) in suburban nighttime also played a significant role (Figure 6(b), Table S3-6).

Figure 6. VIP values of each SEV across different seasons in urban and suburban regions, including those in all cities (a) and different-sized cities (b).

4.4. The effects of urban and suburban spatial explanatory variables on seasonal SUHII

The SEV’s contribution to the SUHII varies significantly under different conditions (Figure 7). We used the standardized partial regression coefficients to explain the correlation of SEVs on SUHII and its importance (Zhang, Li, and Han 2020) (Figure 7, Table S2). The PLSR model is generally significant (p-value is infinitely close to 0). When keeping other variables constant, during the daytime, urban and suburban models accounted for 14.8% to 21.3% and 15.9% to 37.3% of the variation in SUHII respectively (Table S2). However, at nighttime, urban and suburban models accounted for 2.4% to 32.9% and 5.9% to 11.9% of the variation in SUHII, respectively (Table S2). The diurnal and nocturnal SUHII varied in their trends and intensities due to changed SEVs. During the daytime, several variables, including Pop, Light, LSI, and Height, showed positive correlations with the SUHII, whereas Road, NDBI, and DEM variables showed negative correlations. Among these variables, LSI had the strongest positive correlation with SUHII, while NDBI had the strongest negative correlation. During the nighttime, SUHII increased with variations in Road, LSI, and DEM, whereas it decreased with variations in PD, SPLIT, Green, and Height. DEM showed the strongest positive correlation with SUHII, while PD showed the strongest negative correlation. Urban and suburban SEV’s impacts on the SUHII were varying (Figure 7). In urban areas, there were significant positive correlations between daytime and nighttime SUHII and Pop, Light, and LSI. Conversely, we observed significant negative correlations between SUHII, PD, SPLIT, and NDBI. Among these variables, LSI showed the strongest positive correlation with SUHII, while PD had the strongest negative correlation. Only LSI showed a predominantly positive and significant correlation with daytime and nighttime SUHII in suburban areas.

Figure 7. Matrix of the standardized regression coefficient in the PLSR model.

4.5. Impacts of SEVs on seasonal SUHII in different city size

There was a distinct difference regarding the impacts of SEVs on SUHII, both in terms of intensity and trend (Figure 8). Notably, we found that the intensity of these SEVs’ impacts weakened as city size decreased. Specifically, the average intensity of SEVs’ impacts on SUHII ranged from 0.15 in megacities to 0.09 in small cities (Figure 8, Table S3-6). This observation can also be seen in the model's explanation of variations in SUHII, for example, in megacities (Figure 8(a)), during the daytime and nighttime, urban and suburban models accounted for 40.7% to 61.1% and 20.1% to 57.9% of the variation in SUHII respectively (Table S3). Whereas in small cities (Figure 8(d)), during the daytime and nighttime, urban and suburban models only accounted for 16.9% to 35.3% and 0% to 27.6% of the variation in SUHII (Table S6). There were varying mechanisms by which urban and suburban SEVs affect the SUHII at different scales of cities. Specifically, within urban areas, Pop and PD played a key role in driving SUHII, and their impacts were primarily positive (except for the SUHII in nighttime and megacities), while PD was predominantly negatively correlated. In suburban areas, Pop and Height had the highest contribution to SUHII. In megacities, large cities, and medium-sized cities, the effects of Pop and Height on SUHII were mainly negatively correlated, whereas in small cities, they were predominantly positively correlated. In addition, we also observed distinct differences during daytime and nighttime in explaining the SUHII at different city sizes. Specifically, Pop and Height had the strongest influence on daytime SUHII, from megacities to small cities. Pop exhibited a positive correlation, while Height showed a negative correlation (except for small cities). However, during nighttime, DEM and PD emerged as key factors impacting SUHII. DEM demonstrated a positive correlation, while PD demonstrated a negative correlation. These findings highlight the complex and varied nature of SUHII dynamics across different urban contexts. There were still some consistent patterns in the relationship between SEVs and SUHII across different city sizes. For example, the correlation coefficient matrices of large cities and medium-sized cities also exhibited a similar distribution pattern, although medium-sized cities generally displayed weaker correlations compared to large cities. Moreover, the correlation coefficient values at nighttime in small cities were relatively low.

Figure 8. Matrix of Standardized regression coefficients for different city sizes in PLSR models, Megacities (a); Large cities (b); Medium-sized cities (c); Small cities (d).

5. Conclusions

In this study, we explored the urban and suburban SEV’s effects on SUHII in daytime and nighttime. First, we calculated SUHII based on long time-series urban and suburban boundaries and LST data. Then, we derived SEVs from four dimensions (socio-economic, landscape indexes, nature, and building environments) within urban and suburban boundaries. Finally, we explored the urban and suburban SEV’s effects on SUHII using the PLSR model. Our main findings are as follows: (1) Suburban areas showed a higher intensity of increasing/decreasing trends in various SEVs than urban areas. (2) Although SEVs showed variations in explaining LST under different conditions, this variability remained consistent across different levels of urbanization. (3) The SEV’s strongest positive/negative correlations on SUHII were found to be for LSI/NDBI (daytime), DEM/PD (nighttime), and LSI/PD (urban areas). (4) At different city sizes, although SEV’s impacts on SUHII showed significant heterogeneity, the SEV’s impacts on SUHII showed a generally consistent direction and intensity in large cities and medium-sized cities. This study provides a fresh perspective to help us understand the effect of urban development on UHI.

Based on our findings, policies aim to mitigating the UHI are recommended for future urban planning. The temperature both in urban and suburban region should be considered, rather than focusing on urban areas only (Tian and Qian 2021). Also, the natural landscape (e.g. green spaces and their distributions) should be considered to maximize the cooling effects in urban and suburban regions (Liu et al. 2023). Besides, polices should be made according to different city sizes (e.g. megacities and small cities), with considerations of future climate change and urbanization (Yuan et al. 2023).

Several limitations should be addressed in the future. The expanding urban and suburban areas mainly represented the temporal trends of DEM and Height due to the need for national-scale long-term time series data. Also, only four dimensions (socio-economic activities, landscape indexes, natural and built-up environments) comprising ten SEVs were selected for this study, potentially overlooking other variables in different dimensions that may impact the SUHII. Finally, our study ignored the SUHII effects by SEVs in different climatic contexts because the commonly used climate variables (e.g. ERA5) are associated with relatively coarse spatial resolutions (e.g. > 25km) (Muñoz-Sabater et al. 2021). In the future, we will explore more dimensional (such as climatic factors) SEV's effects on SUHII and their intrinsic influence mechanisms in a comprehensive and detailed manner by integrating different climatic context.

Acknowledgments

The authors would like to thank two anonymous reviewers for their constructive comments and suggestions, which greatly improved the quality of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability

The China Suburban Extent Dataset are available in National Earth System Science Data Center at http://nnu.geodata.cn/data/datadetails.html?dataguid=232582871367490&docid=1. The Moderate Resolution Imaging Spectroradiometer (MODIS) 8-day composite products (MOD11A2 product version 006, Terra) were collected at https://lpdaac.usgs.gov/products/mod11a2v006/.

Additional information

Funding

This study was funded by the Key Laboratory of Territorial Spatial Planning and Development-Protection of the Ministry of Natural Resources of PRC and CAUPD Beijing Planning & Design Consultants LTD [grant number: TSPDP23/03], the open fund of State Key Laboratory of Satellite Ocean Environment Dynamics, Second Institute of Oceanography, MNR (No. QNHX2202), the National Natural Science Foundation of China (42101418), the NSFC Excellent Young Scientists Fund (Overseas), and the Chinese Universities Scientific Fund.