1. Introduction
During the process of urbanization, the natural land cover changes to artificial surfaces on a large scale. The increased artificial surfaces have different thermal capacities, reflection rates, aerodynamics, and levels of evaporation [
1,
2]. This leads to urban heat island (UHI) effects, which cause a series of ecological consequences, such as the formation of hazy weather, the deterioration of air quality, and extra urban energy consumption [
3,
4]. Therefore, mitigating UHI has become one of the most important topics in the fields of urban ecology, urban landscaping, urban geography, and urban meteorology.
Changes to the underlying surface are the main cause of UHI effects. Therefore, it is significant to explore the relationship between the underlying surface and the urban thermal environment. At present, there is a lot of research focusing on this field [
5,
6,
7]. The research of [
8] shows that land surface temperature (LST) is significantly positively correlated with the normalized difference buildup index (NDBI) and negatively correlated with the normalized difference vegetation index (NDVI). Previous studies also showed that underlying landscape patterns have an important influence on LST [
9,
10]. The authors of [
11] took the Aksu oasis in northwest China as an example to study the influence of green space on surface temperature. The research results show that the area ratio of green space patches in the landscape is the most important factor. Asgarian showed that increasing the connectivity and complexity of urban landscapes can increase LST, which is caused by the high energy exchange between different landscape units [
12]. The conclusion of [
13] shows that the loss of green space leads to a major LST increase, while green space expansion generates an LST decrease. However, most research in this field concentrates on the relationship between LST and land cover types or area. In-depth research of the relationship between LST and the landscape patterns of land coverage types still needs to be improved.
At present, the research methods of UHI effects include traditional ground observations [
14], numerical simulations [
15,
16], and remote sensing [
17]. Traditional ground observations involve the collection of data from meteorological observation stations or artificial ground observation points. Then, UHI effects are explored by climatology and statistics. This method has the advantage of continuity and controllability, but only rough spatial distribution of UHI can be obtained due to the limited number of observation points [
18]. Numerical simulation can be used to simulate the urban thermal environment. This method has the advantage of continuity, and the mechanism of thermal environmental change can be quantitatively analyzed. However, the disadvantage is uncertainty of model parameters [
19]. Remote sensing (RS) monitoring obtains surface temperatures through remote sensors on satellites. This method has continuity, integrity, and real-time data acquisition, overcoming the disadvantages of traditional ground observation and providing more scientific data support for study [
20]. At present, common RS image data sources include NOAA/AVHRR, Terra and Aqua/MODIS, Terra and Aqua/ASTER, Landsat/TM and ETM+ and OLI/TIRS. The thermal infrared band of Landsat TM/TM+/OLI/TIRS data has high spatial resolution and geometric accuracy, so it is widely applied in urban thermal environment analysis and served as a database for this research. UHI research based on RS images and LST retrieval uses surface UHI, which is different from air UHI. Unless otherwise mentioned, the UHI in this research indicates only surface UHI.
Shanghai is one of the most urbanized and developed cities in China. In recent years, along with fast urbanization, the thermal environment problem has become serious. The average temperature in summer rises continuously [
21,
22,
23]. The research of [
24] shows that, since the 1980s, the UHI area has increased by more than 700 square kilometers, and the average temperature has increased by 0.9 °C. The research of [
25] shows that the average maximum temperature in the Shanghai area fluctuates steadily in summer, but the UHI strength shows a rising trend. Combined with hazy, high-temperature weather, UHI threatens residential life. The study of [
26] indicates that UHI enhances the negative impact of a high-temperature climate on residential health and is one of the key factors of the death rate in Shanghai in summer. Improving the urban thermal environment is a common topic of concern for the government and citizens. It is important to accomplish this by optimizing urban planning according to the relationship between LST and underlying coverage.
Taking Shanghai as an example, this study first explores the patterns of spatial and temporal changes in LST rankings in the past 15 years. Then, a model of the areas and patterns of different land cover types and corresponding LST rankings is built. This study can provide a reference for Shanghai and other megacities to improve the living environment, mitigate the UHI effect, and optimize landscape planning.
3. Results
3.1. Characteristics of LST
Based on the method in
Section 2.2, the LSTs and temperature classifications of Shanghai in 2000, 2004, 2007, and 2015 were obtained (
Figure 4). The LSTs of different LCTs were also calculated, as shown in
Table 4.
According to
Table 1, the average temperature of Shanghai in July and August increased significantly, from 29.5 °C and 28.6 °C in 2000 to 30.5 °C and 31.0 °C in 2015, respectively.
The LSTs of different LCTs are quite different. The average LST ranking of all LCTs for all years is: CL > BL > GL > AL > WB. CL has the highest LST, while WB has the lowest. GL, AL, and WB are significantly lower, and CL and BL are significantly higher than the average LST of Shanghai.
3.2. Characteristics of LCT
Table 5 shows the area proportions of different LCTs. It indicates that percent of agricultural land (PerAL), percent of wetland (PerWL), and percent of bare land (PerBL) decreased over the years from 37.95%, 25.19%, and 3.74% in 2000 to 21.71%, 12.36%, and 0.36% in 2015, respectively. Percent of green land (PerGL) and percent of construction land (PerCL) increased from 7.81% and 25.31% in 2000 to 24.87% and 40.70% in 2015, respectively.
3.3. Quantitative Relationship Between LST and LCT
The quantitative relationship between LST and LCT was studied based on data of 3 August 2015.
According to
Figure 5, there is a good linear relationship between LST and the area ratios of CL, GL, AL, and WB. The fitting equation passes the significance test of 0.05, which indicates that these four LCTs can well explain the change of average LST.
The linear fitting equation of average LST and PerCL is the best (y = 15.68x − 529.92, R2 = 0.81), indicating a significant positive correlation between them. CL has the greatest influence on LST. A large PerCL leads to high LST. Mean LST is significantly negatively correlated with PerAL (y = −12.44x + 491.11), PerBL (y = −0.93x + 37.71), and PerWL (y = −3.07x + 123.31). In other words, high PerAL, PerBL, or PerWL indicates low LST.
In order to obtain a complete overview of the relationship between LST and LCT, a multiple regression analysis of the relationship between the area ratios of different LCTs and LSTs was conducted (
Table 6). The results indicate the contribution of different LCTs to LST. The prediction model is Equation (2):
where
T stands for LST,
k0 is a constant, and
k1–
k5 represent the coefficients of variables.
According to
Table 6, the model is significant (R
2 = 92.8%). The model expression is shown in Equation (3):
The model can be applied to predict LST or adjust the area of various LCTs accordingly to achieve optimal land use planning.
3.4. Quantitative Study of LST and Landscape Patterns of LCTs
The quantitative relationship between LST and the landscape pattern of each LCT was studied based on data of 3 August 2015.
The proportion of BL is relatively small compared to the other types, so it is not sufficient to analyze the relationship between its landscape pattern index and LST at the type level. Therefore, only the relationships between LST and the landscape pattern indices of WL, GL, CL, and AL were analyzed.
The Pearson correlation analysis between the landscape pattern of each LCT and LST is shown in
Table 7. The regressive fitting curve of each LCT landscape pattern is shown in
Figure 6,
Figure 7,
Figure 8 and
Figure 9.
According to
Figure 6, the curve fitting between ENN_MN, LSI, NP, and LST of GL is significant.
According to
Figure 7, PLAND (y = 14.59x − 491.09) and LPI (y = 131x
2 − 79.68x + 1189.93) of CL are well fitted to the LST curve (R
2 > 0.80). NP (y = −2.65x + 104.95) and LSI (y = 0.34x − 8.57) have significant fitting with average LST.
According to
Figure 8, PLAND (y = −11.34x + 453.85), LPI (y = −0.17x
2 + 1.68x + 202.91), and LSI (y = −0.13x
2 + 9.29x − 162.12) of AL have significant fitting with their average LST.
Figure 9 indicates that PLAND (y = −3.50x + 140.64), NP (y = 1.84x − 57.65), LSI (y = −0.43x + 19.22), and ENN_MN (y = 33.25x
2 – 2331.89x + 41363.84) of WL have significant fitting with their average LST.
5. Conclusions
In this paper, multidisciplinary theories and methods were applied to study the impact of LCTs and their configurations on the LST of the megacity of Shanghai. The results show that the thermal environment of Shanghai has become worse over the years. The LSTs of different LCTs are significantly different. The average LST ranking of LCTs is CL > BL > GL > AL > WL. CL contributes the most to the UHI effect and forms heat island centers easily, while WL and GL have significant effects on relieving UHI effects and form cold island centers easily. For GL, WL, and AL, a large area, a small degree of fragmentation, concentrated distribution, and complex shape lead to low average LST rankings and strong mitigation of the UHI. For CL, the effect is the opposite, which means that a large area, a small degree of fragmentation, concentrated distribution, and a complex shape lead to a high average LST and obvious UHI effect. Therefore, to carry out effective landscape planning in order to mitigate UHI effects as much as possible, CL should be distributed in as many discrete pieces as possible, and large areas of CL should be broken up by GL or WL, which can effectively alleviate UHI effects. The study helps with the understanding of the UHI in Shanghai and provides a reference basis for authorities to formulate targeted strategies to alleviate UHI effects.
In further study, the following points still need to be addressed:
(1) The correlation analysis between LST and LCT landscape patterns can be improved by introducing multivariate analysis. Principal component analysis can also be introduced to avoid correlation of the variants.
(2) The most frequently applied landscape indices for LCT are used in this study without justification. More indices may influence the LST of the LCT and should be evaluated.