A Three-Dimensional Investigation of Spatial Relationship between Building Composition and Surface Urban Heat Island

Abstract

Urban heat island (UHI) phenomena are on the increase and are responsible for excessive energy use, environmental harm, and detrimental effects on human health in many parts of the world. Case studies of individual cities imply that wise planning and policymaking might lessen the consequences of UHI by altering aspects of urban settings such as land use/cover (LULC). Determining the influence of LULC planning on UHIs has proven difficult due to the diversity of elements that can alter urban surroundings. This study evaluated building functions and the three-dimensional distribution of structures on land surface temperatures (LSTs) in Tsukuba, a science city in Japan, to estimate the influence of landscape composition on UHIs. We used multiple regression analysis to examine the relationship between LST and LULC, building height, building function, and other variables (e.g., distance to/from roads). Our results showed that management and control of the distribution of buildings, population density, and energy consumption could significantly reduce LSTs, realize sustainable development, and ensure a healthy environment in a planned city. Furthermore, a better theoretical understanding of what makes cities sustainable can enhance the ecological awareness of urbanites and help cities evolve in a sustainable way.

1. Introduction

The urban population has grown year after year as the pace of urbanization and industrialization has accelerated. According to the UNDP (2020), over 50% of the people in the world have been living in urban areas since 2007, and that proportion is predicted to reach 60% by 2030 [1]. It is foreseeable that rapid urbanization will intensify the conflicts and competition for environmental resources and aggravate problems associated with urbanization, including urban climate [2], bioenergy [3], and deforestation [4]. One of the most notable consequences has been the increase in local temperatures as the proportion of impervious surfaces increases during rapid urbanization [5]. The result has been the urban heat island (UHI) phenomenon, which is characterized by higher temperatures in urban than rural areas.

The UHI phenomenon has increasingly captured the attention of city planners and scientists in recent years [6]. UHI includes two types: a surface UHI (SUHI) that has been quantified in terms of land surface temperature (LST) and an atmospheric UHI (AUHI), which can be quantified in terms of air temperature. Replacement of vegetation with impervious surfaces in urban areas changes surface radiation and thermal properties in ways that result in heat-trapping and the creation of UHIs [7]. UNDP (2020) projects that 5 billion people will live in urban areas by 2030. With the expansion of the urban population, the environmental issues related to UHIs are projected to increase. The UHI phenomenon is one of the facets of urban climate with the greatest potential to cause harm [8]. The increase in the frequency and intensity of extreme weather in the most recent decade has been attributed to climate change [9]. UHIs will exacerbate the adverse effects of climate change by creating centers of high temperature in urban areas. These UHIs can adversely affect human health by stressing the digestive and nervous systems of people and increasing the risks of heat-related death [10,11]. Moreover, one of the characteristics of UHIs is that absorption of infrared radiation emitted by hot, impervious surfaces creates warm air masses that can exacerbate air pollution problems by hindering the dispersion of soot and ash [12].

UHIs are complex phenomena that are affected by changes in natural and social environments. Previous studies have proven that LST is closely related to land use and land cover (LULC) [13,14]. The unique thermal properties, moisture-holding capacity, and radiation absorption/emission characteristics of each type of land cover affect the local thermal environment [13]. The UHI effect is influenced by the quantity of each type of LULC (e.g., buildings, vegetation) and the composition of the LULC [15]. Previous studies have proposed several LULC scenarios to reduce LSTs and mitigate UHI impacts by changing the composition of LULC [16]. The traditional strategy has been to increase the coverage of vegetation in the central urban area [17,18]. However, little attention has been given to a three-dimensional counterpart that could impact LST. Rational LULC planning should be systematically practiced to mitigate LST effects. Most scenarios have tended to focus on the spatial composition of LULC. Anthropogenic factors such as building density, building function, road networks, and energy use can impact LST but have not been thoroughly considered in spatial analyses mostly due to an insufficiency of accurate data. Accordingly, a multivariate study that combines spatial and statistical analyses is required to identify strategies for reducing the LST effect.

Significant progress has been made in studies of the UHI effect over the last two decades. Essential variables such as LULC composition, road density, green cover, and building density, have always been used in UHI research [13,19,20,21]. Numerous studies have also shown that “built-up area” is a critical determinant of LST [22,23]. Whereas the results of these studies have been informative, implementation of the results to lower LST in city planning has been problematic for several reasons. To address the implementation problem, we carried out a fine-structure analysis that combined information on building characteristics and other anthropogenic variables to clearly reveal the distribution of LST.

The focus of this study was the composition of LULC and the distribution of LST in the center of Tsukuba, Japan. We examined the spatial interaction between LST and anthropogenic factors to determine whether urban planning could help alleviate the UHI effect in Tsukuba. We had three sub-objectives: (1) to measure urban landscape patterns, urban functions, and building density; (2) to quantify the impact of the distribution of LULC on LSTs; and (3) to investigate the effect of the functional composition of urban areas on the formation of UHIs. Based on an examination of the degree of influence of these anthropogenic variables, we created some proposals for planning urban environments.

2. Materials and Methods

2.1. Study Area

Tsukuba is located about 50 km northeast of Tokyo and has a total area of 284 km² [24]. The population increased from 165,978 in 2000 to 245,958 in 2021 [24]. Tsukuba has a humid, subtropical climate with four distinct seasons. The annual average temperature of Tsukuba is 13.8 °C, and the annual average precipitation is 1283 mm [25]. Since 2005, a new urban rail named the Tsukuba Express has connected Tsukuba with Akihabara Station (Tokyo) and has greatly improved commuting between Tsukuba and Tokyo. Since then, much urban infrastructure, including tall buildings and shopping malls, has appeared in the central area of Tsukuba. We focused this study on the area within a 2.5-km radius of the center of Tsukuba because of its rapidly developing characteristics (Figure 1) [26].

Tsukuba is a national strategic city established in 1963 to relieve socioeconomic pressures in Tokyo. At the beginning of city construction, improving environmental quality and promoting sustainable development were the primary targets of urban planning [27]. Accordingly, the orderly arrangement of large quantities of high-quality trees and shrubs was well planned in the central area. However, Tsukuba evolved in a way that allowed the UHI phenomenon to gradually emerge in the 1970s [28]. From 1976 to 2014, the built-up area increased threefold. However, as the pace of urbanization slowed and the population aged, the rate of building construction slowed, resulting in a change in urban planning [29].

In accordance with plans to reduce the government bureaucracy, the number of civil service dormitories significantly declined after 2014 (City of Tsukuba, 2021b). Civil service dormitories, particularly in Tsukuba’s central region, are therefore no longer in use. To avoid the influence of empty buildings, we chose the year 2014 as our study year.

2.2. Data Used and Preparation

2.2.1. Data Source

We selected Landsat-8 Operational Land Imager and Thermal Infrared Sensor (Landsat-8 OLI/TIRS) images as the main source of data for our analysis of the distributions of LST and LULC (

Table 1. Data used in this study.

Data Type	Date Acquired	Resolution	Data Source
Landsat 8 OLI/TIRS	16 June 2014	30 m	USGS
Building information	2013–2014	Vector	Zenrin company
Building location	2014	Vector	GSI
Fieldwork data	-	-	Fieldwork
Google Earth	2014	-	-
Main roads	2014	Vector	National land numerical information
Residential population density	2014	100 m	World pop
Slope	2014	30 m	USGS (derived from DEM)

). Because the satellite had a 16-day repeat cycle, 22 imageries were collected in one year. Because the accuracy of LST estimates is affected by weather and cloud cover, we chose the Landsat imagery on 16 June 2014 as the best source image. The weather conditions on that day were the best; the cloud cover was only 20%. According to the data report of the Japan Meteorological Agency, the precipitation was 0 mm, and the temperature ranged from 19.4 °C to 29.1 °C on 16 June 2014 [25].

Building information in the form of vector data was provided by the Zenrin© company and included building location, height, and function. Because there were a large number of buildings, the collection work took two years. We analyzed spatial information for 9593 buildings collected from 2013 to 2014 to delineate the impact of buildings on the LST pattern. Building polygons made by the Geospatial Information Authority of Japan (GSI) in 2014 (477 buildings) were used as additional data. We also collected information on building height and function for 477 buildings via fieldwork and visual interpretation of images from Google Earth. Information on main roads, residential population density, and slope were treated as independent variables to estimate their influence on the pattern of LST.

2.2.2. Data Preparation

It is difficult and challenging to collect data on the function of each building. Data updating is labor-intensive and time-consuming. In this study, we tried to rearrange the building function data in the Zenrin© database. We assigned the building characteristics relevant to LST to one of six categories: apartment, stand-alone house, business, public service, under construction, and others (

Table 2. Description of building functional categories.

Category	Description
Apartment	Including apartment, mansion, and tower mansion
stand-alone house	single-family home
Business	Restaurant, company, shopping mall, supermarket
Public service	Hospital, school, museum
Under construction	Buildings under construction
Others	Includes storage, parking space with roof, and 3D-parkade

).

2.3. Workflow

Figure 2 shows the logical chain of the workflow linking the data, methods, and final results. We used the data from the Landsat 8 Operational Land Imager (OLI) Thermal Infrared Sensor (TIRS) and building information data as the primary sources. We derived the LULC and LST maps from Landsat 8 data using maximum likelihood classification and band calculation. We then manipulated the building data to obtain various attributes, including the location, height, and function of each building. Finally, we conducted spatial and statistical analyses to estimate the effect of the building on the pattern of LST.

2.4. LST Retrieval

LST is an essential metric of the distribution of thermal energy [19]. We used the radiative transfer equation (RTE) to derive LST from Landsat images as follows. First, we carried out a radiometric calibration and atmospheric correction using TerrSet software for all images [15,30,31]. The digital number (DN) of the thermal band (band 10 in Landsat OLI/TIRS) and the top-of-the-atmosphere brightness temperature were then used to express temperature in degrees Kelvin [26,32]. We used land surface emissivity ($\epsilon$) to calculate LST based on the Normalized Difference Vegetation Index (NDVI). The following formula is used to compute the land surface emissivity:

$$\epsilon =m\ast P_v+n$$

(1)

where $m=\left({\epsilon }_v-{\epsilon }_s\right)-\left(1-{\epsilon }_s\right)F{\epsilon }_v$ and $n={\epsilon }_s+\left(1-{\epsilon }_s\right)F{\epsilon }_v$; ${\epsilon }_s$ is soil emissivity and ${\epsilon }_v$ is vegetation emissivity. In this study, we used the results from previous studies, which showed that m is 0.004 and n is 0.986 [33].

We expressed the proportion of vegetation ($P_v$) as follows:

$$P_v={\left(\frac{NDVI-NDV{I}_{min}}{NDV{I}_{max}-NDV{I}_{min}}\right)}^2$$

(2)

We calculated the NDVI from the surface reflectance of band red (RED) and near-infrared (NIR) bands with the following equation:

$$NDVI=(NIR-RED/\left(NIR+RED\right)$$

(3)

Minimum and maximum NDVI values were designated $NDV{I}_{min}$ and $NDV{I}_{max}$, respectively.

Finally, we computed the LST as follows [34]:

$$LST=T_B/\left[1+\left(\lambda \ast T_B/\rho \right)\ast ln\epsilon \right]$$

(4)

where $T_B$ is the brightness temperature of band 10 in Landsat OLI/TIRS; $\lambda$ is the wavelength of the emitted radiance (10.8 µm for Landsat OLI/TIRS band 10) [35]; $\rho =h\times c/\sigma \left(1.438\times {10}^{-2}\mathrm{mK}\right)$; $\sigma$ is the Boltzmann constant $\left(1.38\times {10}^{-23}\mathrm{J}/\mathrm{K}\right)$; $h$ is Planck’s constant $\left(6.626\times {10}^{-34}\mathrm{Js}\right)$; $c$ is the velocity of light $\left(2.998\times {10}^8\mathrm{m}/\mathrm{s}\right)$; and $\epsilon$ is the land surface emissivity, which can be obtained from Equation (1).

2.5. LULC Classification

The maximum likelihood supervised classification method was used to determine the LULC composition in the study area because that method enabled us to incorporate local information to pick training samples. This approach to classification was based on Bayes’ classification. We assigned each pixel to the most appropriate LULC class and then assigned the pixels to categories. We used ArcGIS 10.8 for the LULC classification into five categories (built-up, cropland, grass, trees, and water) (Figure 3 and

Table 3. Description of LULC categories.

Category	Description
Built-up	Urban and rural settlements, industrial areas, transportation, and other human-constructed areas
Cropland	Cultivated areas, paddy fields, and horizontal terraced fields
Grass	Grass cover, including bare land (seasonal effect)
Tree	Tree space including shrubs and bushes
Water	Lake and artificial pools

). The LST of these land-use types varied greatly. After classification, we carried out an accuracy assessment by comparing the map classifications and the ground truth information from Google Earth. Five hundred random points were selected from among the five categories in the LULC map and compared to the ground truth in Google Earth.

2.6. Statistical Analysis

2.6.1. Multiple Linear Regression

We used a multiple linear regression (MLR) model to investigate the relationship between LST and multiple anthropogenic factors (e.g., building density, population density, and building volume). The MLR model could be used to estimate the performance and contribution of each independent variable. The MLR equation was:

$$Y={\beta }_0+{{\displaystyle \sum }}_{i=1}^n{\beta }_i{X}_i+\epsilon$$

(5)

where $Y$ is the normalized value of LST; $X$ is the normalized value of an independent variable; n is the number of independent variables; ${\beta }_0$ is the intercept; ${\beta }_i$ is the regression coefficient for $X_i$; and $\epsilon$ is the error or residual.

Figure 4 shows a sample workflow illustrating the process of multiple linear regression. We used stepwise regression to eliminate redundant variables. We used a type I error rate of 0.05 as a criterion for the inclusion of an independent variable in the model. All anthropogenic factors were correlated with one other. The value of the variance inflation factor (VIF) was calculated to eliminate the effects of multicollinearity among the independent variables. If the VIF exceeded 5 for a variable, that variable was removed from the MLR model. After the elimination of multicollinearity effects, 11 variables remained for examination of their relationship with LST. The MLR analysis was performed by using the Statistical Package for the Social Sciences (SPSS) Version 27.

2.6.2. Processing of Dependent and Independent Variables

The dependent variable was the average LST of each buffer, which is a 45-m radius set by each building and includes the building and its surroundings. The LST map retrieval and buffer establishment were completed prior to computing the dependent variable. The LST was determined by both the NDVI and the emitted radiance at band 10, Landsat 8. The resolution of the LST map was 30 m × 30 m. The LST and buffer maps were overlaid to calculate the average value of LST in each buffer since the unit of the LST map (30 m × 30 m) and buffer size (45 m) were not the same.

There were mainly two types of building distributions (Figure 5). One type was a building located within a single pixel (30 m × 30 m). In that instance, the LST of the grid coincided with the LST of the building. The average value of the LST for the central building (one grid) and its environments (around eight grids) were used to compute the dependent variable. The other was a big building that occupied a number of pixels. In that instance, the LST of the building was equated to the LST of those pixels. The dependent variable was calculated using the average of the big building’s LST (several grids) and its surrounding environments (on the remaining grids).

At the stage of model preparation, we considered a total of 14 variables (Figure 4). We considered the basic building information, productive activities, city proximity, and five other variables (

Table 4. Potential variables for LST formulation.

	Independent Variables	Purpose and Relevance
1	Building density	Basic building information
2	Building volume
3	Population density	Productive activities
4	Perimeter (floor circumference)	Size of each building
5	Area
6	Distance from/to road	City proximity and the importance of location
7	Slope	Natural geographical information
8	Built-up proportion	Proportion of LULC categories in each building buffer (45-m radius)
9	Cropland proportion
10	Grass proportion
11	Tree proportion
12	Water proportion
13	Degree of mixing LULC	Complexity of LULC in each building buffer (45-m radius)
14	Energy consumption	Building energy based on a different function

). Because the LULC data were raster data and the building data were vector data, the analysis was performed for building units. In other words, each building was the smallest unit in the analysis process. The building location was extracted using the “feature to point” tool in ArcGIS. The pattern of the LULC surrounding each building was calculated from the LULC map. Because the resolution of the LULC map, the size of a grid was 30 m, each building covered at least nine pixels. To determine the composition of the LULC, a buffer with a 45-m radius was established around each building. The building density was calculated by dividing the total floor space of all buildings in a pixel by the pixel area (Figure 6b). The main road is an essential part of urban development for a planned city. Road accessibility affects the locations of buildings and the ecological environment (Figure 6c) [36,37]. We obtained the population density of each building by extracting the pixel value from the population density layer (Figure 6d). The distance to/from a road is the straight distance from a building to the road. Because the central street is built on a slope that connects shopping malls, stations, and green parks, we selected the slope as another variable. The slope was a natural factor extracted from the vector layer (Figure 6e). The building volume was calculated from the footprint area and number of floors (Figure 6f). The LULC proportion was equated to the area of each LULC type in the buffer zone (Figure 7). The degree of mixing LULCs was equated to the number of LULC types in the buffer zone. The Building Energy Consumption Report from 2014, which was utilized to represent energy in the various building functions, was used to compute energy consumption [38]. Next, the LST and the 14 independent variables were used as the input data to run the MLR model. Three variables (perimeter, area, and built-up proportion) were filtered out when the model was run because of their multiple collinear relationships with other variables. The 11 other variables remained in the MLR model.

3. Results

3.1. Building Distribution

Figure 8a shows the building function map. Apartments were concentrated mainly in the central and northern parts of Tsukuba. After the development law, the area of the footprint of an apartment was larger in central Tsukuba than in peripheral regions. Stand-alone houses were distributed mainly in peripheral areas, and their layout was compact. Business buildings were located primarily in the central area and along the main roads. The density of public service buildings gradually decreased from the center to the periphery. Most buildings under construction were located in suburban areas; only four of them were in central Tsukuba.

Figure 8b is a map of the 3D-building volume in central Tsukuba. The building volume was calculated from the footprint area and number of building floors. Most buildings had fewer than four floors; buildings with 5–8 floors were spaced evenly; buildings with 9–12 floors were located around central Tsukuba along two main roads (Higashi Odori and Nishi Odori); and buildings with 13–16 floors were near central Tsukuba (Figure 8c). There were two buildings with more than 16 floors. One was in central Tsukuba; the other was west of the study area. Japanese building standards require that the height of each floor be 2.5 m. Tall buildings were located mainly in the areas surrounding two rail stations (Tsukuba and Kenkyugakuen stations). Throughout the whole area, the building volumes ranged from 0 to 69,696 m³.

3.2. LULC, LST, and NDVI

Figure 9 shows the composition of LULC, distribution of LST, and patterns of NDVI in the study area. The accuracy of the pattern of LULC was verified based on 500 random points. The overall accuracy was 85.60%. After the classification, a classified map from the Japan Aerospace Exploration Agency (JAXA) and fieldwork were used to improve the accuracy of the LULC map. Built-up areas were dispersed throughout Tsukuba. In the central area, built-up areas were intermixed with grass and trees. The area of cropland gradually increased from urban to rural areas. Trees were concentrated mainly in the northeastern part of the study area. Water accounted for a few areas (eight pixels) and was found in central Tsukuba.

The LST ranged from 25 to 36 °C, with a mean value of 31 °C. A comparison of the LST and LULC maps revealed that LSTs were always higher in the built-up areas than in other land-use types. The highest temperature was not in the central urban area but rather in the western regions and areas surrounding central Tsukuba. The lowest LST was in the northeastern areas, where the vegetation coverage was high. A comparison of the LST and NDVI maps in Figure 7 indicated that coverage by vegetation could help cool the city.

3.3. Relationship between LST, NDVI, and LULC

Five hundred random points were selected and used to examine the relationship between LST and NDVI (Figure 9d). Different colors were used to designate different LULC categories. As Figure 9d shows, the LST decreased as the NDVI increased. The LST was, therefore, negatively correlated with NDVI in the study area. The R² was 0.55.

3.4. LST and Relative Variables

Table 5. Results of multiple regression analysis.

Independent Variables	Standardized Coefficients	Unstandardized Coefficients		t Value	Sig.	95% Confidence Interval of EXP (β)
	Beta	B	Standard Error			Lower Bound	Upper Bound
Constant		0.548	0.004	141.986	***	0.541	0.556
Building density	0.364	0.265	0.005	54.146	***	0.255	0.274
Population density	0.277	0.175	0.005	37.399	***	0.165	0.184
Grass proportion	−0.271	−0.199	0.005	−38.69	***	−0.209	−0.189
Cropland proportion	−0.232	−0.175	0.006	−30.296	***	−0.187	−0.164
Tree proportion	−0.1	−0.484	0.03	−16.062	***	−0.543	−0.425
Degree of mixing LULC	−0.067	−0.043	0.006	−7.712	***	−0.054	−0.032
Distance to/from road	−0.055	−0.036	0.004	−8.192	***	−0.044	−0.027
Water proportion	−0.02	−1.266	0.371	−3.407	**	−1.994	−0.538
Building volume	−0.018	−0.129	0.043	−3.014	**	−0.214	−0.045
Slope	−0.017	−0.02	0.007	−2.754	**	−0.035	−0.006
Energy consumption	0.013	0.008	0.003	2.226	*	0.001	0.014

Dependent Variable: LST. R²: 0.660. Adjusted R²: 0.659. D-W value: 1.945. * p < 0.05 ** p < 0.01 *** p < 0.001.

shows the results of the multiple regression analysis. The adjusted R² indicated that all independent variables accounted for 65.9% of the variance of the LST. The multiple regression analysis indicated that the independent variables with the greatest impact on the LST were (from high to low) building density, population density, proportion of grass, proportion of cropland, proportion of trees, degree of mixing LULCs, distance to/from road, proportion of water, building volume, slope, and energy consumption. The model indicated that LST increased mainly with building density, population density, and energy consumption. The other eight variables had a cooling effect on LST. Among them, the proportions of water, trees, and grass had the greatest potential to reduce LST.

4. Discussion

As the urban population grows globally, the structure and function of cities are becoming more sophisticated. While the economy has been growing, natural resources have been diminished. The urban environment and its constructed landscape have become serious, troublesome problems associated with urbanization and have adversely affected human health. Among the most significant impacts have been local climate changes that have led to more frequent extreme-weather events. Many researchers have analyzed the relationship between LST and related factors, especially the distribution of LULC, landscape composition, and the kinds of urban structures [39,40]. However, because most studies have used grids as the units for analysis of the relationships between types of LULCs and LSTs, the spatial dimensions of buildings have been ignored. Unlike previous studies that have used grids as the units to characterize the distribution of LSTs, we focused on individual buildings to investigate the relationship between LSTs and anthropogenic variables. The building characteristics we considered included the footprint area of each building, the height of the building (i.e., number of floors), building volume, and building category (e.g., apartment, business, public service, etc.) (Figure 6).

In this study, the Landsat data has been used for calculating the LST. In the process of computation, both soil emissivity and vegetation emissivity have been considered and calculated in this study. Despite the roofs and surfaces of buildings being made of various materials and having different emissivities, the surface temperature varies from building to building. However, this study focuses on the surface temperature of the actual building (grid) as determined by the Landsat thermal image rather than attempting to estimate the surface temperature by computing the radiation for each building. Our results showed that LST was affected mainly by building density and population density. Similar results have been reported in previous studies [41,42]. The regression results showed that if other variables were held constant, every increase of one standard deviation in building density increased the LST by 0.364 standard deviations. In contrast, an increase of one standard deviation in the proportion of grass decreased the LST by 0.271 standard deviations if other variables were held constant (Table 5). In addition, LST decreased with an increase in cover by vegetation. Unlike the study in Chinese megacities [43], our study revealed that the LST in central Tsukuba did not increase with building volume. The maps of LULC and 3D building volume revealed that built-up land use was not concentrated in central Tsukuba. Instead, built-up areas were intermixed with ample green space that benefitted air circulation. This intermingling of green space and built-up areas was presumably the result of urban planning that called for buildings and green space to be interspersed with each other in the study area.

Because different LULCs affected LSTs in different ways, the pattern of LSTs was a function of the pattern of LULCs [44,45]. Like previous results, our analysis revealed that the highest LSTs in Tsukuba were associated with the built-up areas among all types of LULC. Types of LULC associated with vegetation reduced LST and contributed positively to urban ecology. Local urban warming effects in central Tsukuba were partially offset by the presence of green spaces. Furthermore, those green spaces facilitated central air ventilation, which may have helped to alleviate the warming that would otherwise have occurred in built-up regions. According to the regression results, the cooling effect of grass was the strongest among all variables (Table 5). The main reason was that the grass in central Tsukuba covered a large area and was well organized. Compared with grass, trees had a relatively small cooling effect. Considering that the number of trees in the study area is relatively small and randomly distributed, more preferred landscape patterns are recommended to increase the cooling performance of trees. Water, widely considered as a cooling resource during daytime [46], failed to show a stronger cooling effect than green spaces, which is common in several city-level case studies [15,47,48]. Contrary to expectations, building energy had little impact on LST. We hypothesized that two reasons might account for this small impact. First, central Tsukuba is well planned, and buildings are not arranged in a compact manner. Urban planners created a pedestrian network based on variations of the natural elevations to connect universities, hospitals, libraries, and other important facilities. Second, Tsukuba has had distinct targets and policies for energy conservation and reduction of emissions. For example, the government has announced that until 2030, the per-capita carbon emissions should be 26% less than the emissions in 2013 (Ministry of the Environment Government of Japan, 2016). Furthermore, after April 2010, newly constructed and renovated houses with footprints larger than 300 m² were forced to be eco-friendly and energy-saving. Another goal was to decrease the LST in central Tsukuba by separating people and vehicles. Urban planning resulted in the construction of ~7 km of pedestrian walkways over vehicle roads. The pedestrian network of walkways connected four green parks, which helped to reduce LST and increased coverage by vegetation in central Tsukuba.

This study was underpinned by buffer analysis, which was used to derive the pattern of LULC and the degree of mixing of LULCs for each building. Buffer analysis exhibits three principal advantages. First, it is possible to obtain the central building’s own characteristics, which reflect the conditions surrounding the building. Second, a comprehensive analysis of the impact of buildings on the pattern of LST can be conducted. Finally, each building can be separated from its complex environment, and the surrounding conditions can be analyzed as individual conditions in the model. In this study, buffers for 9593 buildings were established in central Tsukuba. Results indicated that the greater the proportion of grass in each buffer, the greater the cooling effect. It is noteworthy that the LST associated with each building was sensitive to the composition of the LULC in the surrounding 45-m buffer zone. These discoveries suggested that the more complex the LULC (e.g., the northeast section of central Tsukuba), the greater the reduction of LST.

In addition, the city of Tsukuba will continue to attract people because of its proximity to Tokyo and its comfortable living environment. Urban development has been taking place around Kenkyugakuen Station, the station closest to Tsukuba Station on the rail line between Tsukuba Station and Akihabara Station. Some of the bare land and cropland around Kenkyugakuen Station have been transformed into built-up land use, and some old buildings (e.g., buildings for civil servants) in central Tsukuba have been bulldozed and rebuilt. As the population of Tsukuba increases, more and more large-scale, comprehensive residential structures will be built near Tsukuba Station. The result will be an increase in the LST in central Tsukuba if the government does not take any preventive measures and actions.

In this study, we examined the factors that influenced LSTs in a planned city with a population of about 250,000. In general, green spaces, including grasslands and trees, have the optimal cooling effects, compared to both buildings and croplands. In this context, to mitigate the UHI, stand-alone houses (especially in central Tsukuba) could be replaced by taller buildings, so that the built-up area, as well as the cropland around the houses, could be released for arranging cooling resources.

Several limitations of this study deserve mention. First, the LST was estimated from single daytime Landsat observations. Even if the highest quality data were obtained and the estimated temperatures were close to the true LSTs, estimates made on different dates with Landsat data may result in different estimated LSTs. For example, paddy fields are classified as wetlands in spring, croplands in summer, and bare lands in winter. Our results should therefore be considered reasonable explanations of the UHI phenomenon in summer. Another limitation was the resolution of LST. In this study, the Landsat 8 data, with a resolution of 30 m × 30 m, has been used for calculating LST of buildings. However, in most cases, the size of single buildings in Tsukuba did not perfectly meet the size of the gridd. In this context, the surface temperature of buildings measured in this study was in fact the mean value of the buildings and their surroundings that fell into the same grid. Therefore, the building buffer (45-m) has been set as a unit to calculate the LST of a building and its environment. In future studies, when more detailed LST data is available, the actual LST of each building is recommended to be used to derive more accurate results.

5. Conclusions

The goal of this study is to examine the relationship between LST and anthropogenic factors (e.g., LULC, building density, and building function) in Tsukuba, a planned city. The results showed that the spatial pattern of LST differed between types of land use. Building density was the most significant factor affecting the LST. The presence of grass reduced the LST, especially in central Tsukuba. The results of this study indicated that stand-alone houses had an obvious heating effect, whereas high-rise buildings (apartments) did not. To control the heat island effect, urban planners should therefore design more high-rise buildings instead of stand-alone houses and plan to set aside the saved space for trees, shrubs, and grass, especially in the center of urban areas.

Because this study used a single-building rather than traditional grid-based analysis, it is easier to consider the different properties of individual buildings and provide urban planners with a more detailed way to design each single building. Consideration of buffer zones revealed that LST was affected by both building attributes and conditions surrounding the building. In addition, indicators related to building fail to show a dominant impact on LST. Considering this, more detailed research on designing building indicators to better reflect the characteristics of thermal property is proposed. New indicators reflecting the cooling effect of green roofs and the configuration of buildings in a three-dimensional view, and the detection of seasonal differences and diurnal differences on the performance of these indicators are recommended in future research. In general, the findings of this study can help planners identify policies that will reduce the risks of high LSTs. The results can be applied to similar cities to facilitate planning for creating a comfortable and sustainable living environment.