1. Introduction
Air and surface temperatures in urban areas are often higher than in surrounding rural areas. This phenomenon is known as the urban heat island (UHI) effect [
1] which is caused due to higher shares of impervious surfaces and the intensity of usage in urban environments. Asphalt, cement, and roofing tiles, among other urban building materials, have a substantially higher heat capacity than other natural components [
2]. UHIs are formed as a result of higher anthropogenic heat emissions, less evaporative cooling, increased surface roughness, lower surface albedos, and narrow urban canyon geometries as a result of urbanization. The UHIs have caused negative effects such as increased energy consumption due to air conditioning [
3], air pollution [
4], and water shortages as evapotranspiration increases and precipitation decreases in some desert cities. Moreover, it threatens the health of urban residents [
1] and human comfortability [
5,
6]. For example, the continuous high temperature in summer in Arizona led to the state’s highest temperature mortality rate in the United States from 1993 to 2002, and the excessive heat event in France defined the deaths of approximately 15,000 people in the summer of 2003. While traditional approaches for evaluating urban climate relate to air urban heat islands, an increased number of studies relate to the effects of certain land cover types to describe local microclimate phenomena under the term surface urban heat island (SUHI) [
7,
8,
9,
10,
11]. Especially in times of climate change and increased excessive heat events, there exists an urgent need to better understand UHIs and thus derive a larger information basis for developing mitigation strategies.
To derive area-wide data on land surface temperature (LST), earth observation satellites incorporating thermal sensors have been utilized for SUHI research (e.g., [
8,
9,
10]). When compared to air temperatures gathered from urban weather stations, thermal imagery provides current, explicit, and area-wide data at multiple spatial and temporal scales [
1,
9,
11], which are important prerequisites for studying SUHI. LST, which is influenced by land cover/land use type and the spatial urban structure, is one of the most obvious responses to the urban thermal environment [
12,
13] making these data crucial to understanding how urbanization affects the urban thermal environment. However, the mechanisms and complex interactions behind the surface temperature in urban areas are not yet fully understood.
Numerous related studies have reported that the SUHI generated by the specific urban structure, such as landscape compositions [
14] and three-dimensional (3D) building morphological layout [
15,
16], can be related to specific land use/land cover patterns (LUCP) characteristics, making it crucial to understand how LUCP affects LST. In related studies, the normalized difference vegetation index (NDVI) and the normalized difference built-up index (NDBI) have shown strong relations with LST [
6,
17,
18,
19]. Further, the building height (BH), building density (BD), and floor area ratio (FAR) could be significantly related with LST [
20,
21].
Urban heat islands are not static in terms of the seasonal or even daily dynamics because the thermal response of certain surface types, such as vegetation or surface water, react with delay to changing air temperature [
18,
22]. The temperature differences are observed especially in summer nights [
23], and it was shown that nighttime temperatures, especially, have a major effect on the UHI and the health of urban residents [
24].
Previous research on the SUHI concentrated more on the phenomenon during the daytime in summer because high and moderate resolution images were widely available for this point in time. In addition, the mechanisms which drive LST are also stated to differ between day and night in cities [
25,
26]: various land use types are measured with varying LSTs at daytime compared to nighttime [
27]. While many researchers have examined the effects of biophysical factors on the SUHI, multi-temporal examinations on season and time of day have not been systematically documented. In recent years, an increasing number of scholars have aimed at the relation of urban 3D morphology to LST [
20,
28]. Buildings, their physical characteristics, and their pattern modify solar radiation reflection and absorption; likewise, the roughness of the urban surface, affects surface ventilation and heat exchange. To add to the current related literature, the purpose of this study aims to statistically quantify how the LUCP indicators affect urban LST characteristics on seasonal and diurnal scales. We selected Germany’s capital, Berlin, as a case study, as it is the largest city in Germany. incorporates a broad variety of land use/cover types, and it is situated in a climate zone with high variations in daytime and nighttime temperatures as well as well high differences between summer and winter temperatures.
Given the above background, three aspects of this study are: (1) to investigate the spatial variability of LST patterns seasonally and diurnally for the city of Berlin; (2) to analyze and compare the performance of land surface temperature on various land surfaces at different times and identify its impacts on land surface temperature spatial distribution; (3) to explore quantitative associations between seasonal and diurnal LST variations and LUCP variables which determine LST intensity including land use/land cover factors and urban morphology indices; and (4) to provide valuable insight and scientific guidance for urban planners to mitigate the SUHI effects to effectively improve the thermal environment.
2. Data and Methods
2.1. Study Area
The research area is Berlin, the capital city of Germany (52.34°–52.68° N, 13.10°–13.77° E). It is located in northeastern Germany and covers an area of around 900 km
2. With an annual mean temperature of 9.5 °C and annual precipitation of 591 mm, Berlin is deeply affected by the prevailing westerlies in summer and characterized by a transition from maritime temperate climate to the continental climates of the interior of Europe, according to the Köppen climate classification. Berlin has a population of 3.6 million residents, with one-third of them dwelling in the city center. In spite of the temperate climate, mortality rates in Berlin are up to 67.2% higher during intense heat waves [
29], most notably due to increased risk of heat stress in the central city [
30] because of a microclimate which is characterized by urban heat island effects. This is substantial to exhibit the diversity of land cover such as different densities of built-up areas, vegetation areas, water, etc. (
Figure 1). The numerous lakes in the western and southeastern areas are the most prominent physical attributes.
2.2. Data Sets
This study relies on three different input data types: (1) surface temperature from thermal remote sensing, (2) land use/land cover classification, and (3) urban morphologic indicators.
For thermal data, we rely on multi-temporal data from the Landsat 8 Thermal Infrared Sensor (TIRS) and Operational Land Imager (OLI) data, which are publicly available (level-1-products
http://earthexplorer.usgs.gov/ accessed 20 May 2020). The spatial resolution of Landsat 8 image bands is 30 m except for band 8 (panchromatic, 15 m spatial resolution) and the two thermal bands 10–11 (100 m). Conditions for images to be selected for the study, are: (1) cloud-free; (2) coverage of the entire study area; (3) four-season division covering spring (March–May), summer (June–August), autumn (September–November), and winter (December–February); and (4) daytime (local overpass time before noon) and nighttime (local overpass time after 8 p.m.). Based on these criteria, the following eight images between 2018 and 2019 were selected to retrieve LSTs in spring, summer, autumn, and winter as well as day and nighttime of Berlin (
Table 1). The time of data acquisition was approximately 12:00 a.m. and 10:30 p.m. Greenwich Mean Time (GMT) in Berlin.
The land use/land cover (LU/LC) data (10 m) of Berlin (
Figure 2) in 2018 were obtained from the European Urban Atlas (
https://land.copernicus.eu/local/urban-atlas accessed 18 February 2020), which provides reliable, inter-comparable, high-resolution data. The LU/LC classification includes six primary types (croplands, woodlands, grasslands, water areas, built-up lands, and unused lands) and 25 secondary types. We identified seven important classes (
Table 2): Transportation (15.2%), Commercial and Industrial (6.7%), Residential (35.2%), Sports and Leisure (4.3%), Vegetation (30.2%), Agriculture (2.9%), Wetlands (5.5%) (the percentage of the corresponding class within the study area in 2018).
Berlin is characterized by high shares of impervious land and a generally low to moderate pace of urban growth between 2018 and 2019. Thus, the LU/LC changes in Berlin over these two years are fewer than the inaccuracies in the mapping results, whereas the shift in time between the LU/LC map and the building survey data can be considered negligible.
2.3. Research Framework
The following conceptual framework, as shown in
Figure 3, provides an overview on the three essential analysis steps:
(1) Evaluating of the pattern of the urban thermal environment by retrieving the LST from Landsat images and analyzing the corresponding spatial patterns for multiple points in time at the inner-city scale;
(2) Obtaining 2D metrics (i.e., NDVI, NDBI, and albedo) as well as 3D indicators (i.e., height, volume, and density), respectively, from Landsat images, land use/land cover data, and building survey data; and
(3) Conducting statistical analysis (correlation and regression) to estimate the seasonal and diurnal associations between LST and these indicators.
Figure 3.
The conceptual framework to investigate statistical relationships between LUCP indices and LST characteristics.
Figure 3.
The conceptual framework to investigate statistical relationships between LUCP indices and LST characteristics.
2.4. Retrieving Land Surface Temperature
Considering that LST is sensitive to the atmospheric effects, and in order to obtain accurate and consistent spectral information [
31], atmospheric correction is necessary to convert the top-of-atmosphere reflectance to surface reflectance. The FLAASH (Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes) model provided by the ENVI software was used for the atmospheric correction, which is based on MODTRAN (Moderate Spectral Resolution Atmospheric Transmittance Algorithm and Computer Model). The computation of at-sensor spectral radiance is required to convert image data from multiple sensors and platforms into a physically appropriate common radiometric scale. The following equations were used to convert the digital numbers for both reflective and thermal bands to top-of-atmosphere radiance [
32]:
where
is top-of-atmosphere radiance at the sensor’s aperture in W/(m
2·sr·μm),
is the band-specific multiplicative rescaling factor from the satellite metadata equal to 0.0003342,
is the pixel digital number for thermal band 10,
is band-specific additive rescaling factor from the satellite metadata.
The following equation is used to convert the spectral radiance
to the at-sensor brightness temperature [
32]:
where,
is the effective at-sensor brightness temperature in Kelvin,
is the spectrum radiance at the sensor’s aperture in W/(m
2·sr·µm), and
and
are the pre-launch calibration constants. For Landsat 8 OLI,
= 774.89 W/(m
2·sr·µm) and
= 1321.08 K.
The temperature values were derived using a black body, which has properties that are different from that of real objects. The land surface temperature was calculated after correction for spectral emissivity (ε) of a grey body was implemented [
33,
34]:
where
is the land surface temperature,
is the black body temperature in Kelvin, λ is the wavelength of radiance emitted (which for Landsat 8, band 10 is 10.8, and band 11 is 12), α = hc/b (1.438 × 10
−2 mK), h = the Planck’s constant (6.626 × 10
−34 J/s), c = velocity of light (2.998 × 10
8 m/s), b = Boltzmann constant (1.38 × 10
−23 J/K), and ε = surface emissivity.
Land surface emissivity is a key parameter in the measurement of LST. An accurate estimate of surface emissivity is crucial for the reliable derivation of the surface temperature. Water, vegetation, and roughness are a few aspects of the variables that affect a surface’s emissivity [
35]. Empirical approaches [
36,
37] were used to discover a relationship between emissivity and the value of NDVI in the scientific literature. This emissivity estimating method was utilized in TM/ETM+ band 6, however, because of their approximate spectral range, it can also be applied to Landsat 8 band 10 [
38]:
where ε is land surface emissivity and
is, according to Sobrino, Jimenez-Munoz, and Paolini [
39], the vegetation proportion obtained:
where
and
are the maximum and minimum vegetation index in the study area, respectively.
2.5. Selected LUCP Indicators
Different land cover types reflect various heat capacities and evapotranspiration ratios. The roughness of the urban surface and urban air ventilation conditions are reflected in several morphological indices. Nonetheless, each of these types of indicators have different effects on urban LST, which together moderate the urban surface thermal climate [
16].
For vegetation, we apply the commonly used NDVI [
40], which relates to the density of vegetation. For the built landscape, we apply the NDBI [
41] which is a measure for the intensity of the built-up area. For water, we apply the modified normalized difference water index (MNDWI), which represents the amount of water state of vegetation [
42,
43,
44]. MNDWI has been identified to be preferable compared to the normalized difference water index (NDWI) for characterization of the original biophysical water index since the latter tends to extract water cover combined with built-up area characteristics [
19]. Further, albedo is an essential property of the land surface heat budget [
45].
Beyond the ratios from land cover compositions introduced above, we apply parameters further describing spatial morphology of the urban environment. To better define the urban morphological factors of LST from building survey data, we calculate the following indicators: Fraction of impervious surface (ISF), which quantifies the fraction of impervious surface in percent; BD, which quantifies the total area of building ground floors per reference unit in percent; and FAR, which quantifies the total area of building floors per reference unit [
46]. All factors selected in this study are presented in
Table 3.
2.6. Statistical Analyses Relating Spatial Indicators and LST
In this study, two common statistical approaches were used to quantify associations between LST and selected indicators on multi-temporal scales. For this, the optimal pixel size for simulating environmental characteristics between LST and LUCP indicators must be considered. The ideal geographical scale for examining LST and ISF relationship was determined to be 500 m, as proposed by relevant research [
14], while the optimal size of the green space that effects LST effectively was found to be 210 to 240 m [
48]. Generally, at coarser scales a stronger correlation has been already shown [
49]. Thus, in this study we used 240 m as the window size for LST and the land cover factors, while for LST and the urban morphologic indicators we applied 500 m. As advised by related studies [
8,
50], the pixel aggregation tool provided by ENVI was used to perform the resampling. The tool uses weighted average to aggregate the input cell values that contribute to the output grid values, removing the effect of nearby pixels.
Pearson correlation, a frequently employed statistical method, was utilized to evaluate the relationship between each selected factor and LST. Pearson’s correlation coefficient is a measurement of the linear correlation between two variables [
51]. The Pearson correlation coefficient is calculated as follows [
52]:
where
r is the correlation coefficient,
x and
y represent the selected factors and LST, and
and
are the mean values of
x and
y, respectively.
Ordinary least squares (OLS) regression model has been employed to examine how land cover factors affect the LST. The corresponding slope coefficients produced from the models are used to assess the localized contribution of each indicator to LST. Improved OLS regression models [
53,
54] use the equation:
where
Y denotes LST,
is the intercept value,
represents slope coefficient, which means how the LST changes linearly with each indicator,
represents each selected indicator,
n represents the number of indicators, and
ε is the random error term.
A multi-scale geo-weighted regression model (MGWR) is used to describe the non-stationary characteristics of spatial data [
55], which enables us to explore the relationship between building form layout parameters and LSTs. The MGWR models were built on the open-source platform GWR 4 (
https://sgsup.asu.edu/sparc/gwr4 accessed on 2 March 2022), which is expressed as:
where
denotes the coordinates of the
i-th point in space,
is the intercept value,
represents the ideal bandwidth for the association modeling between the
k-th indicator and LST,
is the
k-th indicator at observation
i,
is the random error term at point
i, and
n is the number of independent variables.
5. Conclusions
This study assessed the seasonal and diurnal variations of the severity of the SUHI in relation to the urban land use/cover and the urban morphology by employing the indicators for the sample city of Berlin. Based on statistical analysis, associations between LST and four land use/land cover factors and four urban morphology indicators were investigated. Four conclusions were drawn from this study:
Firstly, the LSTs in Berlin showed clear seasonal and diurnal differences in spatial distribution, with thermal hot spots primarily in districts of high shares of impervious surfaces during the daytime and, in contrast, in water bodies during nighttime. Secondly, within areas of high imperviousness, the diurnal temperature range in commercial and residential areas were lower than in transportation and industrial areas, and, naturally, higher than in vegetation areas, the water body was measured with the lowest range in any season. The temperature differences across the seven investigated land cover types were much higher at daytime than at nighttime. Thirdly, among the investigated land cover indices, NDBI, MNDWI, and NDVI related to built-up areas, water bodies, and greenery. These were the key variables determining different distributions of temperature. From the positive or negative correlation, water areas and green spaces play an extraordinary role in alleviating urban thermal environment. Finally, from the perspective of urban morphology, four indicators were found to explicitly have a heating effect on LST during daytime as well as during nighttime: ISF and BD revealed the most evident positive association. BH was found to have the least influence on LST among all indicators.
Based on the empirical results highlighted above, the study reveals how different indicators impact the urban thermal environment at different seasons and at different times of the day and night. The local distinction of LST mechanism for different periods may provide urban planners in Berlin with useful information and, in general, effective reference for implementing specific mitigation measures.