1. Introduction
The urban heat island (UHI), a phenomenon that air temperatures or land surface temperatures (LSTs) are generally higher in urban areas than its surrounding rural areas, is a significant anthropogenic alteration to the Earth’s environment [
1]. The acceleration of global urbanization exacerbates the UHI effect, resulting in a variety of consequences, such as atmospheric environment intervention [
2], regional climate change [
3], enhanced vegetation growth [
4,
5], and air and water pollution by disrupting the surface-energy balance [
6]. Meanwhile, UHI can increase energy consumption [
7], endanger human health [
8], and considerably obstruct the development of environmentally sustainable cities [
9].
In response to these issues, an early study assessed the UHI effect by considering multiple influential factors, such as canyon radiative geometry, thermal characteristics of buildings and vegetation, anthropogenic heat emission, and turbulent transmission [
10]. According to one study, natural terrain (e.g., soil, vegetation, and water bodies) was gradually being replaced by impervious artificial surfaces (e.g., asphalt, and concrete) that reduced evapotranspiration and increased the sensitive heat, which was especially noticeable in high-density areas [
11]. Another study found that rapid building construction, as well as the emergence and spread of skyscrapers, exacerbated the UHI effect [
12].
The above studies imply that the spatial configuration of buildings and the thermal environment are the root causes of UHI. On this basis, numerous indicators have been proposed to perform correlation analysis with the urban heat magnitude (UHM) defined by the temperature difference between urban and rural areas, which can be classified into three causative categories: (i) land use and land cover (LULC) that can be described by normalized difference vegetation index (NDVI) [
13], normalized differential build-up index (NDBI) [
14], and tree cover ratio and shrub cover ratio [
15], (ii) urban morphology (UM) that can be depicted by floor area ratio and building density [
15], sky view factor (SVF) [
16], and building surface fraction and fabric density ratio [
17], and (iii) thermal and radiative property (TRP) that can be represented by solar irradiation [
18], surface albedo [
17], and wind velocity [
19,
20].
It is discovered that these indicators have varying effects on UHMs. For example, UHMs are increasing in proportion to the amount of thermal mass absorbed by cities [
20,
21,
22,
23]. While the UHI effect will be mitigated when the vegetation cover ratio is high because vegetation increases latent heat transfer in the air via transpiration and evaporation [
4,
5,
24,
25], and UHMs are decreasing in areas where there are more high rises buildings casting significant shadows [
20,
23,
26,
27]. However, these studies did not account for the regional effects of the urban landscape, which can create a variety of microclimates and thus have a significant impact on the urban thermal environment.
Since it has been demonstrated that the UHMs can be largely explained by background climate in over 30,000 global cities [
28] and urban form in 1288 Chinese urban clusters [
29], it is reasonable to develop a model that integrates microclimate with the above three causative categories for a better understanding of the UHI effect. This is critical for climatic modelling and may even be useful for urban planning and weather forecasting [
2], which is a growing concern for urban planners [
30]. Therefore, there is a trend to deepen from tracking spatio-temporal dynamic behaviour of UHIs [
31,
32] and gaining an in-depth understanding of the UHI formulation [
28,
29] to propose solutions to mitigate the UHI phenomenon, such as optimizing locations of green space [
33].
The new trend necessitates the development of a robust model capable of evaluating and even mitigating the UHI effect through urban design, which has not been well established yet. To address this issue, this study aims to: (i) propose and determine UHI related indicators by considering comprehensive effects from microclimates and the three categorized indicators, (ii) investigate spatial and seasonal variations of UHMs in a densely urban area and build spatial regression models to estimate seasonal changes of UHMs, and (iii) predict UHM distribution in different seasons based on a master plan of buildings.
6. Building Spatial Regression Model
Based on the analysis in
Section 5.3 that the correlation between UHM and each variable is significant and there is no multicollinearity, it presents established multivariate regressions used to estimate UHMs in built-up areas (denoted by UHM
B) and land-cover areas (denoted by UHM
L) in four seasons (i.e., Equations (16)–(19) for spring, Equations (20)–(23) for summer, Equations (24)–(27) for autumn, and Equations (28)–(31) for winter), where X
1 (Equation (14)) and X
2 (Equation (15)) are two matrices for land-use LCZs. The correlation trends revealed by PR in
Table 5 and
Table 6 are consistent with the coefficient matrices of C
1 and C
2. The regressions show that NDBI and NDVI are two indicators that significantly increase UHMs in the built-up LCZs, as the two corresponding coefficients are significantly larger than the others. The results also reveal that SVF and ASR are two important factors that cause UHIs during the daytime, which is different from UHIs formed by longwave emission from the ground at night [
52,
73]. Notably, the NDVI coefficient for UHM
B is positive but the coefficient for UHM
L is negative. This is because NDVI values are low in built-up LCZs (
Figure 8) that represent barren soil but high in land-cover LCZs that correspond to vegetation, resulting in an inverse influence on UHMs.
Figure 9 presents the predicted and observed UHMs for the whole study area with 3048 samples. It demonstrates that the red linear regression lines are almost on the diagonal with R
2 values equaling 0.522, 0.603, 0.636, and 0.549 from spring to winter, indicating a reasonable predicting accuracy. The corresponding RMSE equals 0.912, 0.851, 0.621, and 0.836, which are also limited to 1 degree Celsius.
Figure 10 depicts predicted UHM distributions that are quite similar to the observation as shown in
Figure 5. Overall, they are spatially consistent and quantitatively close with each other across the entire Kowloon area, with a clear seasonal variation on the four days. Several UHM hotspots in specific urban areas, such as Kowloon East, Kowloon Tang, and Hung Hom, are successfully predicted in summer in
Figure 10b. This means that, based on the proposed and determined indicators, the spatial regression models can estimate UHM distributions with reasonable accuracy in different seasons.
Based on the new LCZs in
Figure 4a, SVF in
Figure 4b, and ASR in
Figure 6 that will occur in a transformed urban area and the established spatial regression models adaptive to four different seasons as presented in Equations (14)–(31), the distribution of UHMs with the same spatial resolution can be predicted on the four corresponding days in the Kowloon East area. In the future, the UHI effect will be most pronounced in summer in
Figure 11b, followed by autumn in
Figure 11c and winter in
Figure 11d, and least pronounced in spring in
Figure 11a. It is worth mentioning that the observed UHM on 14 November 2019 is overall 0.5 °C higher than on 20 August 2017, which makes the predicted UHMs in winter slightly larger than that in autumn. Nonetheless, in Kowloon East, the future UHMs based on the proposed master plan will be significantly lower across the four seasons when compared to either the observed UHMs (
Figure 5) or estimated UHMs (
Figure 10) based on the current urban landscape, resulting in obvious mitigation of the UHI effect. For instance, there will be only 12 grid cells with UHMs larger than 4 °C in summer in
Figure 11b; while in winter, there will be several grid cells with UHMs lower than 0 °C in
Figure 10d, revealing an urban heat sink phenomenon. The predicted UHM distribution suggests that the Government’s urban reform initiative will aid in the creation of a thermal-friendly urban environment.
7. Discussion and Conclusions
The forecasting of the UHM distributions based on a master plan of new buildings established by an urban reforming initiative in Kowloon East suggests that the UHI effect will be significantly mitigated compared to current UHMs. The main reason could be that this area will be transformed into an open space fulfilled with low plants that can effectively cool down LSTs, according to the future local climate zones. It is crucial to reveal that NDVI and ASR are the two most important indicators for explaining the daytime UHIs in both the built-up LCZs and land-cover LCZs. Furthermore, different LCZs have different effects on UHMs in the built-up area, indicating that microclimate can also be used to explain the UHI phenomenon. Although the values of R2 are not significantly high, the established models are accurate and reliable based on the evidence that the linear regressions between over 3000 observed and predicted samples have high consistency with the diagonal lines and the correlation is significant without multicollinearity.
This study established a framework for predicting UHMs based on a future urban landscape, which can effectively aid in urban planning to create a liveable urban environment. However, three limitations could have an impact on the estimation results. First, the spatial regression models are constantly based on a grid resolution of 100 m. This may result in a portion of the gird cells containing a mixture of LCZs, which challenges an accurate prediction of UHMs, particularly in densely populated areas. Second, heat generated by vehicle flows has not been incorporated into the model, which is supposed to be an unignorable heat source of UHIs during the daytime in Hong Kong [
41]. Third, as the study area of Kowloon is a peninsula along the coast, cold wind from the sea may move along the ventilation corridors between buildings, thus creating a cooling effect on the UHI.
Future research can be conducted in four areas based on the proposed framework. First, both solar irradiation and traffic flow can be modelled as the heat source of UHI, and wind directions and intensities can be modelled by using computational fluid dynamic estimations and integrated into the proposed model as a new influential factor, which is especially important for Hong Kong, a city close to the coast. Second, with the availability of the thermal images captured during the nighttime, the nighttime UHI effect can be thoroughly investigated. Because the UHI formulation mechanism differs during the day and at night, this may result in different findings and effects from the indicators. Third, instead of building multivariate regression models, advanced machine learning (e.g., neural network regression and random forest) or deep learning methods (e.g., deep neural networks) can be used to construct a robust regression model to achieve high prediction accuracy. Fourth, further study can be directed toward optimizing the urban landscape to mitigate the UHI effect based on a throughout understanding of the UHI phenomenon and well-designed objective functions.
In conclusion, this study believes that solar irradiation, land use and land cover, urban morphology, and local climate zones can all have a significant impact on the daytime UHI phenomenon significantly. To estimate the future UHI effect in a reformed urban area, spatial multivariate regression models adaptive to different seasons are developed and tested. The prediction of the UHMs based on a master plan of new buildings suggests that a thermo-friendly urban environment can be expected in Kowloon East, Hong Kong. The proposed framework is useful for evaluating the UHI effect in urban planning and can be used by urban planners to improve the model for optimizing the urban landscape to further mitigate the UHI effect.