SpatiotemporalCharacteristicsofUrbanSurfaceTemperatureand Its Relationship with Landscape Metrics and Vegetation Cover in Rapid Urbanization Region

Key Research Institute of Yellow River Civilization and Sustainable Development & Collaborative Innovation Center on Yellow River Civilization of Henan Province, Henan University, Kaifeng 475001, Henan, China School of Geography, Geomatics and Planning, Jiangsu Normal University, Xuzhou 221116, China Key Laboratory of Wetland Ecology and Environment, Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, Changchun, Jilin 130102, China Department of Environmental Sciences, College of the Coast and Environment, Louisiana State University, Baton Rouge, LA 70803, USA


Introduction
e process of rapid urbanization deeply affects biological diversity, ecosystem services, and regional environment [1][2][3]. Because of unprecedented urbanization, the issues of urban eco-environment have been attracting increased attention from the public, scientists, and urban managers [4,5]. One of the acute problems is the UHI effect, where urban surfaces and atmospheric temperatures are warmer than the surrounding nonurban areas owing to transformation of land-cover types and anthropogenic-based heat [6,7]. e UHI effect is a substantial issue, as it has several negative influences on urban environments and human health. e existing research has indicated that UHI effect markedly increases energy consumption [8], raises risks of heat and pollution-related mortality [9], deteriorates air pollution [10], and worsens living environment of city dwellers [11]. Based on these impacts, the spatiotemporal patterns of UHI and the factors affecting its impacts have become popular topics and are being broadly researched across multiple disciplines [12][13][14]. Better understanding the UHI effect is vital to offer the scientific foundation for taking effective measures to mitigate urban thermal environment and promoting sustainable development of urban areas.
In the previous literature, the studies of UHI effect have been implemented by contrasting temperature observations of meteorological network within urban and rural areas surrounding the city [15][16][17]. However, these ground-based observations are too scattered, and some observation data are commonly unrepresentative of complex thermal environments, and it is hard to depend on field data alone to achieve accurate and enough information concerning the UHI [18,19]. Consequently, with the fast development of RS technology in recent years, satellite image data have been applied universally in urban environment and climate research across the world. Satellite image data have become a strong means for UHI research because they provide a convenient and consistent mechanism for detecting the thermal features of urban surfaces [20][21][22]. e satellite remote-sensing images with a thermal infrared band can be transformed into land surface temperature (LST) to characterize an UHI [23,24]. For example, Landsat TM/ETM+/OLI data have been widely applied to calculate the LST values for UHI effect studies [25,26]. As a result, the SUHI is addressed in this study, and thermal band images of remote sensing are applied to determine LST.
In urban thermal environment studies, many documents have illustrated associations between LST and urban expansion, particularly changes in land-use types [27][28][29].
ese studies have found that urban heat effect is substantially influenced by LULC change. Each type of urban land has its own thermal features and radiation characteristics that make different contributions to UHI effect [30,31]. Urban impervious surfaces (IS), including asphalt roads and cement buildings, usually emit more heat and lead to higher temperatures, while areas with water and vegetation have lower temperatures [32]. Accordingly, like urban expansion and transformations in land-cover types, the urban thermal environment will change correspondingly. Numerous studies have found that natural landscapes, such as forests and croplands, are transformed into impervious surfaces during the process of urbanization [33,34] and thus contribute to UHI effects. It has been suggested that a positive correlation exists between impervious surfaces (IS) and LST [35,36]. In addition, as vegetated landscapes generally are beneficial for decreasing LST [37], some documents have stressed on comprehending the relationship among NDVI, FVC, and LST [38,39]. ese studies demonstrated that there is a negative relationship between NDVI, FVC, and LST at different spatial resolutions [32,40], and the LST-NDVI relationship was more adapted to analyzing UHI effect in the summer [41]. In general, these previous documents have quantitatively explored the relationship between LULC changes and LST based on RS technology. However, urban LULC patches have various shapes and spatial configurations, and these factors may also influence the urban thermal environment.
ere are several UHI documents that have researched the impacts of landscape configuration and composition on LST from the perspective of landscape ecology. However, this approach fails to consider influence of other elements of landscape patterns, like patch shape, patch structure, and patch density, which may hide important details related to spatial distribution and add the uncertainties of analyzing the relationship between UHI and landscape patterns [42,43]. erefore, a more reasonable approach that studies the influence of landscapes of various sizes, spatial arrangements, and shapes on UHI effects would be needed.
To better comprehend urban landscape processes and dynamics, one should be able to precisely quantify its spatiotemporal variations [44,45]. Landscape metrics, which are landscape ecological variations that can precisely describe spatial features of land surfaces [46,47], have been employed to describe spatial patterns of urban landscapes and to clarify impacts of LULC patterns on urban heat effect in this study. erefore, compared with the existing literature, this study has the following contributions. First, we investigate the relationship between NDVI, FVC, and LST, respectively, and discuss the correlation between different biophysical indicators and urban thermal environment. According to the research results, we can determine a suitable indicator for this study, making the research results more accurate. Second, we use the method of comprehensive analysis to investigate the relationship between landscape pattern and LST based on RS data, which provides an effective way to reveal the change process of urban ecology and thermal environment. e main purpose of this research is to explore the impacts of landscape (LULC) patterns on UHI effects more accurately based on landscape metrics analysis and to investigate the landscape patterns (including composition and spatial configuration) of different LST zones to preferably comprehend the relationship between LULC and LST at a regional scale. We center on investigating the intensity of LST based on LULC using Landsat images and estimating the correlation among LST, fractional vegetation cover (FVC), NDVI, and urban landscape patterns to elaborate the related impacts of urbanization on urban thermal environment. Furthermore, remote-sensing data and landscape metrics provide new perspectives for scientists to research UHI effects, and the findings from this study can be used to recommend important policies for urban planning to mitigate UHI effects, which provides an effective way to reveal the process of urban ecological change.
us, the main objectives of this research are as follows: (1) information regarding the spatiotemporal variation of LST and LULC of Zhengzhou City was investigated from 2000 to 2014; (2) a comparative analysis was implemented to explore the impacts of NDVI and FVC on LST using quantitative models; (3) the relationship between landscape patterns and LST was explored by statistical analysis. e sketch map of this study is given in Figure 1. 4°C [48]. Like many other megacities in China, Zhengzhou City has been experiencing rapid urbanization and population growth in the past few decades. Based on local statistical data, from 2000 to 2014, Zhengzhou City's urbanization rate raised from 55.1% to 68.3%. is rapid land use process unavoidably led to urban environment degradation caused by UHI effect.

Data Sources and Preprocessing. In this study, available
Landsat-7ETM+ (2000), Landsat-5TM (2006), and Landsat-8 OLI/TIRS (2014) images were collected from the USGS website (http://www.earthexplorer.usgs.gov), and these images were applied to establish time series data to acquire LULC information and to describe the heat environment of the entire area of Zhengzhou City. Additionally, each image was subjected to a series of preprocessing before LST retrieval, and these procedures mainly contained the radiometric calibration, atmospheric correction and geometrical distortions correction. Subsequently, all these images were resampled to 30 m to ensure correspondence to the resolution of multispectral bands [49]. e flowchart of this study is given in Figure 3.

Classification of LULC.
e LULC maps of Zhengzhou City in 2000, 2006, and 2014 were obtained from Landsat imagery. e supervised classification and human-computer interactive interpretation were applied in ENVI 5.1. e urban landscape was classified into six categories: IS, water, bareland, farmland, forest, and grassland. e accuracy assessment of the classified LULC was implemented with 400 reference points that were generated to represent different LULC types of Zhengzhou City [50]. erefore, the overall accuracy of LULC category was approximately over 86.6%, which met the requirements for this study.

LST Retrieval and Delineation of LST Intensity Zones.
To date, several good retrieval algorithms of LST based on different data sources have been developed [51,52]. In the research, the image-based method (IB) is applied to retrieve LST values because of its validity and simplicity [49]. Before retrieval of LST, a quadratic model was adopted to transform the digital number (DN) of Landsat TM/ETM + thermal infrared (TIR) band into radiant temperatures [53]: where L λ denotes the radiance of the thermal band pixel in W/(m 2 · sr · μm); gain denotes the slope of the radiance/DN conversion function; offset denotes the slope of the radiance/ DN conversion; QCAL denotes the quantized calibrated pixel value in DN [21]. en, the spectral radiance was transformed into the at-satellite brightness temperature [53], and equation (2) is as follows: where T B denotes the effective at-satellite temperature in K; both K 1 and K 2 are prelaunch calibration constants (for Landsat-7, K 1 is 666.09 W/(m 2 · sr · μm)and K 2 is 1282.71 K; for Landsat-5, K1 is 607.76 W/(m 2 · sr · μm) and K 2 is 1260.56 K; for Landsat-8, K 1 is 774.89 W/(m 2 · sr · μm) and K 2 is 1321.08 K).
According to the above information, it is inappropriate to directly compare the LST values or examine the variances of LST values over multiple years periods on account of the interannual variability of atmospheric conditions. Hence, the LST values was standardized in ArcGIS 10.2 software according to the data range (equation (4)), and the mean standard deviation (STD) means was used to separate the LST values into five zones (Table 1). ese zones can be employed to analyze the spatial pattern and relationships of urban landscape metrics: where T s denotes the standardized LST value of a pixel, T i denotes the initial LST value of pixel I, T max denotes the maximum LST value, and T min denote the minimum LST value.  e NDVI and FVC were applied as the key biophysical indicators. e NDVI was calculated as follows [59]: where NIR denotes the reflectance of the near-infrared and RED denotes the reflectance of red bands.
e FVC depicts the amount and nature of vegetation cover, and it depends on the NDVI as follows [60]:   LST zones Classification Low Note: T S denotes the standardized LST value; T mean denotes the mean standardized LST value; std denotes the standard deviation.

Calculation of Landscape Metrics.
To quantify the relationship between urban landscape patterns and LST, a series of landscape metrics (including composition, shape, and configuration) was computed based on 5 × 5 pixel moving window in FRAGSTATS 4.2 software package. e metrics were chosen according to the principles of minimal redundancy [63][64][65]. Overall, six landscape-based metrics and eight class-based metrics were chosen to describe characteristics of landscape patterns (Table 2). In addition, the Pearson correlations between mean LST values and landscape metrics were calculated and employed to determine the impacts of land-use landscape patterns on the UHI. In addition, this research calculated the mean LST values for each LULC category between 2000 and 2014 ( Figure 5). e results reflected that mean LSTs of the different types of landuse/landcover had clear differences. According to the mean LST values, it can be observed that impervious surfaces had a higher LST value, whereas forest and water had lower LST values during this study period. e impervious surfaces were linked to roads, buildings and dense populations, and thermal properties of these building materials, and anthropogenic heat sources made impervious surfaces become the hottest regions. Bareland and farmland had relatively higher mean LST values because the bareland class usually had little vegetation cover, while the LST of farmland was always influenced by its associated agricultural activities [67]. e forest class with high vegetation coverage has a lower LST because of the shading and evapotranspiration that occur within this class. Overall, the LULC categories with different thermal characteristics influenced the distribution of LST.

Results
As shown in Figure 6, the spatial dynamics revealed that the LST values in urban areas were higher than that in the surrounding areas in 2000-2014.
e sub-high and high zones are mainly distributed in the commercial districts of the central urban areas and in the industrial districts' northwest and south of Zhengzhou City. In comparison, the sub-low and low zones were found inside the urban areas and gradually had a scattered distribution. Meanwhile, the area of sub-high and high zones increased substantially, while the areas of the sub-low and low zones decreased sharply, which was due to the dramatic changes that occurred in the LULC patterns that greatly influenced the thermal environment. A large amount of farmland and other vegetation cover types have been converted into impervious surfaces and artificial buildings due to rapid urban expansion during this study period. Table 3 shows the proportion of LULC categories of different LST zones from 2000 to 2014. e magnitude of the differences in zonal proportions associated with impervious surfaces was more pronounced in the medium zone and high zone, with an obvious increase between 2000 and 2014. However, the proportions declined in sub-high zone. One reason for this result may be the variations in the urban land use types, characterized by concrete surfaces interspersed with spotty patches of green parks and grasslands [45]. For the same reason, the proportion of grassland increased in all the zones from 2000 to 2014. However, the proportions of forest clearly declined in the low zone and sub-low zone, which was probably due to impervious surfaces replacing forest patches during rapid urbanization. In addition, it appeared that the proportion of farmland declined in the low and sub-low zones, but it increased in the high zones, which was on account of a combination of agricultural activities and seasonal weather impacts on LST.

Relationship among LST, NDVI, and FVC. NDVI and
FVC have been regarded as two key biophysical parameters in thermal remote-sensing analyses. Previous studies have demonstrated that NDVI could be considered a representative indicator of land surface vegetation coverage [21,68], whereas FVC is often applied to represent vegetation abundance. To better comprehend the effects of vegetation cover on UHI, we applied Regression function and Pearson correlation analysis to estimate the relationships among the LST, NDVI, and FVC. e results are shown in Table 4. A strong and negative correlation between FVC and LST can be observed, but there is a weak and negative correlation with NDVI across the entire range of values.
is result shows that FVC is a better reflection of LST characteristics than the NDVI, and several studies also identified similar relationships among LST, NDVI, and FVC [61,69]. Hence, in this study, FVC is more suitable for assessing the correlation between vegetation abundance and LST. Figure 7 shows the spatial patterns of FVC in Zhengzhou City between 2000 and 2014. e city center exhibited lower FVC and higher LST, while the surrounding areas with a relatively higher FVC exhibited lower LST. is indicates that inverse trends in FVC and LST were observed. To further validate the results, we obtained the mean and SD zonal FVC shown in Table 5. Obviously, there was a decrease in the FVC value as the higher LST zone increased, which illustrated a clear correlation between FVC and LST.  6 Complexity erefore, the changes in FVC well explained the impact of biophysical features on the SUHI pattern. Tables 6 and 7, clearly, bareland and water changed insignificantly on account of their scattered distributions and small patch sizes. e other four main LULC types, including farmland, forest, and grassland and impervious surface, accounted for the major components of the urban surface and had important effects on the UHI in Zhengzhou City. us, the landscape patterns of these four LULC types were considered and discussed in the statistical analysis of class-level metrics.

Relationship between Landscape Metrics and LST. As shown in
is research demonstrated that LST is not only related to landscape composition but also to spatial configurations at the landscape level. e LST is positively correlated with PD and CONTAG, indicating that landscapes with higher patch density and fewer contiguous patches will lead to high values of contagion, which results in enhancing the UHI        Additionally, the correlation of LST with AI and LPI is unstable because of the changes in the shape and proportion of the urban landscape patterns. In general, these findings indicated that abundance of LULC categories and spatial configurations will contribute to mitigating the UHI effect. As shown in Table 7, the LST was negatively correlated with AI, DIVISION, ED, FRAC MN, LSI, PD, PLAND, and SHAPE MN for farmland, forest, and grassland. However, the landscape metrics for impervious surfaces mainly had positive impacts on LST because of their thermal conductivity [62]. For the spatial configuration variables, it was indicated that higher aggregations of farmland, forest, and grassland patches could reduce UHI effect, while the aggregation of impervious surfaces could enhance the UHI effect. For the composition variables, it was implied that LST would be reduced with the increase of vegetation patches' percentage and density, while the increase of impervious surface patches proportion would intensify the thermal environment. For shape variables, it was demonstrated that the increase of vegetation patches in shape irregularity would enhance cooling effects, while the increase of impervious surface patches in the shape irregularity would intensify UHI effect. In addition, FRAC_MN of vegetation patches was negatively correlated with LST, which indicated better cooling effects of vegetation with improvement of shape complexity. However, the FRAC_MN of impervious surface patches was positively correlated with LST, which implied a reduction in UHI effects for impervious surface patches with less complex shapes.

Discussion
In view of the results offered by the spatial dynamics of LULC and LST zones, we identified some important features of the impacts of LULC on UHI effects during the process of urbanization. First, Zhengzhou City underwent rapid urbanization, as was evident based on the significant increase in spatial scale of IS from 2000 to 2014. ese outcomes are consistent with findings of other previous literature [70,71]. As the core city of central Henan urban agglomeration, Zhengzhou City has played an important role in regional development. Its urban areas have continued to grow since 2000, but after 2006, the expansion rate was much larger due to growth stimulation from planning policies at multiple levels. Second, LULC changes could result in drastic changes to the urban thermal environment in Zhengzhou City. e observed increasing trend of LST in study area was affected by landscape changes. In general, most of the expansion of built-up lands was produced from cropland and grassland transformations. A substantial portion of the natural surface landscape was replaced by traffic facilities, buildings, and asphalt roads. According to spatial pattern of the LST zones, the high LST zones generally were located within IS classes, but low LST zones were located within vegetated land-cover types. In addition, the investigation of LST and vegetation cover indicated that FVC had a stronger impact on LST than did NDVI. Hence, FVC was used to offer a better understanding of the correlation between LST and vegetation abundance in Zhengzhou City. rough the above analysis, we found that not all study areas are suitable for NDVI to reflect the relationship between vegetation cover and LST.
rough comparative analysis, this study found FVC indicator that can accurately reflect the relationship between vegetation cover and LST, make up for the shortcomings of previous studies, and provide reference value for other areas.
is finding can make up for the lack of using a single biophysical index in the previous research and provide reference value for other regions.
We also analyzed the relationship between urban landscape patterns and LST through landscape metrics and statistical analysis, which offered a new perspective for comprehending the evolution of the UHI effect. Our findings showed that spatial configuration and composition metrics were obviously correlated with LST, positive for vegetation landscape patches (including grassland and forest) and negative for impervious surfaces at the class level.
ese results are consistent with those of the other relevant researches [59,62], for instance, Chen and Yu [72] identified obvious relationships between the LST and landscape patterns of impervious surfaces (positive) and vegetation cover (negative) in Guangzhou City of China. Overall, this research has indicated that integration methods of GIS, remote-sensing image classification technology, and landscape metrics are valid means to quantitatively analyze the relationship between landscape patterns and LST. ese approaches can offer precise characteristics of the spatiotemporal change of urban thermal environment and some implications for other similar studies.
In addition, this research has some implications for policymakers and urban planners to mitigate the negative UHI effect. Our results showed that a higher proportion of impervious surface areas results in higher LST, while vegetation helps decrease LST and generates cool island effects by providing shade. In this context, one way to mitigate the UHI effect is to increase vegetation coverage, and it is important for the urban planners to consider the supply of more forested areas and green spaces in future development of Zhengzhou City. For example, the green roofs, small green space, and cool pavements should be encouraged to build. In addition, the findings of this research could help land-use managers optimize the spatial arrangements of urban landscapes to control UHI effects. It is recommended that regulating the spatial structure of impervious surface patches is a useful method to reduce UHI effects in future urban development. Meanwhile, continuous vegetation cover patches can help to optimize the spatial configurations diversity and obviously decrease LST values.

Conclusions
In this study, RS technology and geospatial approaches were applied to explore the effects of LULC changes on UHI. Using Zhengzhou City as a case study, the spatial features of urban landscape and thermal environment were investigated using the FVC and landscape metrics obtained from LULC Complexity types and LST zones. (1) e results indicate that land-use changes of Zhengzhou City had a significant influence on its UHI effect, and the mean LST increased by 2.23°C from 2000 to 2014. Overall, this increasing trend of mean LST value is consistent with the rapid urbanization, which led to the sharp increase in impervious surfaces area and dramatic losses in vegetation coverage area. (2) e investigation of LST and FVC indicated that the FVC had an obvious negative relationship with urban thermal environment. (3) e relationships between LST and landscape patterns were investigated using landscape metrics and correlation analysis method. e findings indicated obviously different effects of LULC landscape patterns on LST, and it can be considered that impacts on LST is not only reflected in urban landscape composition but also in its shape and spatial configuration. Hence, it is noteworthy that the abundance of land-use categories and spatial configurations will contribute to mitigating UHI effects. In general, these findings suggest that some actions (e.g., increasing forests and grasslands, adjusting the spatial distribution of different landscape types, and optimizing the landscape configuration) could be taken to reduce the urban thermal environment to ensure a healthy and livable environment.
LST is one of the significant surface parameters applied for estimating the UHI effect, and this study indicated that different LULC types have complex relationship with LST. With the high-speed advancement of RS technology, we can investigate the effects of LULC on UHI at a finer spatial scale. In addition, there are some other factors including socioeconomic transformation, climate change, and seasonal variation that may influence UHI effects at different levels. In the future, more attention should be given to the research of relationship between LST and urban landscape by comprehensive conditions analysis.

Abbreviations
LST: Land surface temperature UHI: Urban heat island FVC: Fractional vegetation cover NDVI: Normalized difference vegetation index IS: Impervious surface LULC: Landuse/landcover.

Data Availability
In this study, (1) the available Landsat-7ETM+, Landsat-5TM, and Landsat-8OLI/TIRS images were collected from the US Geological Survey (USGS http://www.earthexplorer. usgs.gov) and were used to support the findings of this study; (2) the Biophysical Indices (NDVI and FVC) which were used to support the findings of this study were taken from Landsat images; (3) the Landscape Metrics is available from the LULC maps of Zhengzhou City.

Conflicts of Interest
e authors declare no conflicts of interest.