An investigation into urban heat mitigation by adopting local climate zones and land surface temperatures in the Tokyo prefecture

Abstract

This study aimed to investigate urban forms susceptible to heightened heat intensities in the Tokyo Prefecture in Japan. Adopting Landsat 8 data at a pixel resolution of 100 m, local climate zones (LCZ) were identified. LCZs contain urban forms which are primarily defined by building compactness and height. Daytime spatial distribution of land surface temperatures (LST) was provided by MODIS 100 m resolution data from 2013 to 2021. Median LSTs for compact and super high‐rise, high‐rise, mid‐rise, and low‐rise LCZs were 34.4, 35.5, 37.3, and 38.1°C, respectively. Additionally, LSTs for open and super high‐rise, high‐rise, and mid‐rise LCZs were 37.4, 37.5, and 37.1°C, respectively. Therefore, this suggests lower‐rise and open LCZs are prone to increased urban heat intensities and higher‐rise and compact LCZs are an urban heat mitigation strategy. Open mid‐rise also offers heat reduction capabilities. Compact low‐rise and open mid‐rise spatial analysis also confirmed this trend with vegetation indices validating urban configuration as significantly influencing LSTs. Furthermore, due to LSTs constituting heat health risks, 11 municipalities comprised of predominantly compact low‐rise LCZs were identified as a priority for urban heat mitigation. Among these, Nakano, Nerima, and Suginami posed the greatest heat risks.

Introduction

The urban heat island (UHI) is a phenomenon that occurs when urban areas experience higher temperatures than rural.¹ UHIs give rise to increased public health risks with global climate change intensifying such side effects.² Furthermore, by 2050, urban areas are expected to accommodate 68% of the world's population³ and a growing number of elderly people vulnerable to heat.⁴ In addition, urban areas are continuing to expand⁵ and, according to the IPCC, by the end of the 21st century, the temperature of the Earth's surface will increase by 0.3–4.8°C.⁶ These facts imply that heat issues will persist and intensify in urban areas by means of UHIs. As a result, there is an increasing need for recommendations for the development of strategies for UHI mitigation.⁷

Factors which contribute to the development of UHIs comprise urban expansion, reduction in green and blue areas, depletion of airflow, storage of large amounts of heat by urban materials,⁸ and building forms.⁹ Two urban morphological parameters which strongly influence UHI variation include building height and density.¹⁰ In addition, although the characteristics of UHIs share similarities throughout global cities, they can be highly localized.¹¹ Furthermore, the need for site-specific and actionable UHI mitigation recommendations has also been shown.¹² Therefore, there is a strong relationship between building heights and densities and UHI development, and the need for site-specific and actionable recommendations for UHI reduction is also clear.

One place presently experiencing the harmful consequences of UHI development is the Tokyo Metropolitan Area, in Japan. The Tokyo Metropolitan Area has a population of 38 million,¹³ is among the largest urban masses on the planet, and has doubled in size since the 1970’s.¹⁴ The Tokyo Metropolitan Area is also subject to increases in heat stroke incidence rates² and accommodates a society which is aging rapidly.¹⁵ Furthermore, a limited number of studies have specifically investigated the relationship between building height and density and UHI intensity in the Tokyo Metropolitan Area.¹⁶ Nor have the implications for urban heat mitigation been extensively considered. Adopting the wet bulb globe temperature, Wang, Gao, Zhou, Kammen, & Peng¹⁷ found that higher-rise and medium density buildings have comparatively lower heat stress levels in the Tokyo Metropolitan Area. However, the main aim of the paper is to introduce a new urban structure classification system. Furthermore, research outcomes are based on the physical urban boundaries present in the Tokyo Metropolitan Area. Therefore, any recommendations are less actionable from a political standpoint with priority areas for urban heat mitigation also not being identified. Finally, the cooling effect of vegetation for each LCZ and the super high-rise nature of many buildings in Tokyo's city center are also not considered. Therefore, with the Tokyo Metropolitan Area as a priority, urban heat associated with building height and density needs to be investigated further to facilitate the development of UHI mitigation strategies.

To achieve this, traditionally weather stations have been adopted to gather meteorological data pertaining to urban heat.¹⁸ For example, Oliveira, Lopes, Niza, & Soares¹⁹ utilized air temperature data collected from a weather station to assist in investigating heat storage in various urban forms. Also, Bassani et al.²⁰ adopted pairs of weather stations to calculate mean temperature difference data to investigate the UHI effect in the Italian city of Turin. However, such data is spatially limited due to stations being located far from each other, which is also the case in Tokyo.²¹ Given the heterogeneous nature of Tokyo's urban landscape, this limitation is further heightened when studying UHI intensity for various urban configurations within close proximity to each other.²²

Alternatively, land surface temperatures (LST) combined with remotely sensed images classified by land types have been widely adopted for urban heat studies.²³ Machine learning algorithms are trained with remotely sensed images to classify the land types in an area. These are then combined with LSTs to identify heat intensities of different urban configurations.²⁴ Moreover, the Moderate Resolution Imaging Spectroradiometer (MODIS) LST datasets have been widely utilized for such investigations²⁵ and are available in Japan. Therefore, making this a feasible method for urban heat investigations in the Tokyo Prefecture.

Simultaneously, for the investigation of urban heat, local climate zones (LCZ) are widely adopted.²⁶ LCZs are a system of land-type classifications which indicate regions of uniform surface covers and structures. The parameters for classification include surface albedo, surface admittance, terrain roughness, height and roughness of elements, pervious surface fraction, impervious surface fraction, building surface fraction, aspect ratio, and sky view factor.²⁷ However, as a result of urban LCZs being dependent on them,²⁸ the most significant factors are building height and density.²⁹ Additionally, remotely sensed Landsat 8 data is available globally and widely adopted for LCZ classifications.³⁰ Therefore, the LCZ system is a suitable land-type classification methodology for investigations into urban heat intensity for various building height and density configurations in the Tokyo Prefecture.

Additionally, remotely sensed data has also been adopted for mapping the cooling effects of vegetation on LSTs.³¹ The normalized difference vegetation index (NDVI) has been widely adopted for urban heat investigations.³² Higher NDVI values equate to increased vegetation and cooling effects which mitigate increased urban heat.³³ For example, Aboelata³⁴ investigated the best vegetation scenarios for various urban aspect ratios and Tuczek et al.³⁵ examined the optimal distribution of vegetation and buildings in Brisbane, Australia. However, no studies have investigated NDVI values and the cooling effect for LCZs in the Tokyo Prefecture. MODIS NDVI data is also available globally making it a suitable dataset for investigation in the Tokyo Prefecture.

Therefore, this study aimed to utilize LSTs and NDVI combined with LCZs in the Tokyo Prefecture to identify urban configurations prone to increased urban heat intensities. While, also providing actionable recommendations for implementing UHI mitigation strategies. This was achieved by carrying out LCZ land classification and accuracy assessment. Followed by spatial and box-plot trend analysis for summertime LSTs and NDVI of LCZs on a political boundary scale of wards, cities, towns, and villages (WCTV). The aim of the study was also ascertained by a comparative spatial trend analysis of the LCZs compart low-rise (CLR), open mid-rise (OMR), and NDVI with LSTs, also on a WCTV scale. Finally, research aims were achieved through the identification of WCTVs with LSTs hazardous to health and trend analysis of their LCZ composition. Therefore, identifying priority WCTVs for UHI mitigation. It is expected that this study will assist in identifying urban configurations that are prone to increased heat intensity and facilitate the development and implementation of UHI mitigation strategies. In addition, it is predicted that this research will contribute to the existing knowledge base used by academics, authorities, designers, and city planners seeking to investigate the causes and management of increased urban heat intensity.

Materials and Methods Study area

Tokyo is a cultural and economic central point of the world³⁶ while also being the capital of a country with the world's third largest economy.³⁷ The Tokyo Prefecture is situated within the Greater Tokyo Area and constitutes 23 special wads, 26 cities, 5 towns, and 8 villages.³⁸ Cities, towns, and villages are districts with local governments while the 23 special wards are a special type of municipality located in the most eastern area of the prefecture (Figure 2, regions 31–53). The Tokyo Prefecture is also composed of a diverse selection of land types, ranging from mountainous vegetative areas in the west to a dense urban area in the east. As a result of the availability of remotely sensed data, 2013–2021 was gauged as a suitable period of analysis, and therefore, selected for evaluation. Also, a 100 m pixel resolution was adopted for LCZ analysis. A 10-year timescale is often adopted in urban heat related studies,² and therefore, the 9-year period, adopted in this study, was deemed sufficiently close for use.

Local climate zones (LCZ)

The LCZ classification system is used to specify urban and rural land types and structures in an area. For LCZ studies, remotely sensed data is frequently utilized.²⁸ The LCZ methodology consists of remotely sensed data being pre-processed, training polygon selection with a remotely sensed image, training of a machine learning algorithm, and finally classification of land types.³⁹ Utilizing the Google Earth Engine platform⁴⁰ LCZ classification was carried out for this study.

The Google Earth Engine team have pre-processed remotely sensed data which is available to the public in the form of atmospherically corrected surface reflectance images. For the selection of training polygons, the Landsat 8 data of USGS Landsat 8 Surface Reflectance Tier 1 images with a 30 m pixel resolution was selected. Combining a Google Earth 3D image with a 9-year (2013–2021) summer cloud-masked median image in the Tokyo Metropolitan Area⁴⁰ LCZ training polygons were selected at a local scale of >100 m².³⁹ In the Tokyo Prefecture, a total of 12 LCZs were identified. Compact super high-rise (CSHR), compact high-rise (CHR), compact mid-rise (CMR), CLR, open super high-rise (OSHR), open high-rise (OHR), OMR and large low-rise (LLR) comprised the urban LCZs, whereas forest, grassland, cultivated land, and water constituted the rural LCZs. These can be seen in Figure 1. As a result of comparatively higher classification accuracies, a summer median image was used for the selection, as opposed to other seasons or annual images, of training polygons and classification. When selecting training polygons, the most significant factors were building height and density. Table 1 shows the categories which were used to account for the building heights in the Tokyo Prefecture. The floor numbers at an interval of 4 m were used to calculate the building height for the selection of training polygons.⁴¹

Generate image description

FIGURE 1. Local climate zones (LCZ) training polygon examples in Tokyo, Japan acquired from Google Earth images

TABLE 1 Building heights for classification of urban local climate zones

Local climate zone	Number of floors	Height range (m)
Super high-rise	15+	60+
High-rise	9–14	36–56
Mid-rise	4–8	16–32
Low-rise	1–3	4–12

When training the machine learning algorithm for this project, the iterative process of training polygon selection followed by running of the classifier and using a Google Earth image to confirm classification accuracy was adopted.⁴² For land classification studies, random forest classifiers have been utilized often and was therefore selected for this study.^43,44 Random forest classifiers consist of various decision trees and adopt a majority vote approach for allocating classifications.⁴⁵ The decision tree number for the random forest classifier was defined as the first peak, all other adjustable parameters were set to default. The first peak being the first number of decision trees which is followed by two consecutive decreases in the classifier's overall accuracy. As part of the classifier's accuracy assessment, producer, consumer, overall and kappa accuracies were calculated.

To carry out the LCZ classification, a 9-year Landsat 8 image of pixel resolution 30 m was rescaled to the LCZ pixel resolution of 100 m²⁷ and clipped to the boundary of the Tokyo Prefecture. However, when classifying the LCZs of rock, sand, or bare soil and open low-rise convoluted misclassification occurred in urban and rural areas. Therefore, these LCZs were excluded from the classification process. For the LCZ spatial distribution analysis of the WCTVs, values were calculated as the percentage of the total number of urban pixels in each WCTV. For the LCZ composition analysis of priority WCTVs, all percentages are calculated as a portion of the total number of all pixels in each WCTV.

Normalized difference vegetation index (NDVI)

NDVI has been utilized in multiple studies⁴⁶ as a vegetation index. The NDVI dataset adopted for this study is the MOD13A1.061 Terra Vegetation Indices 16-Day Global 500 m resampled to a 100 m pixel resolution. Vegetation either reflects near-infrared radiation or absorbs red radiation from a ray of light, depending on its position in the electromagnetic spectrum. Equation (1),⁴⁷ adopted in this study uses this fact for calculation of NDVI values. A pixel's photosynthetic activity, or NDVI, is measured on a scale ranging from −1 to +1. Generally, the presence of liquid water or snow is shown by negative values, exposed ground by zero values and vegetation by positive values.[Image Omitted. See PDF]

In this study, NDVI was selected as an indicator of the degree of vegetation present in a LCZ and WCTV. It was adopted to facilitate understanding the extent to which different LCZs influence LST variation. For the NDVI analysis, initially a median value of all the pixels in the Tokyo Prefecture from dates 2013–2021 in June, July, and August were extracted from the Google Earth Engine platform.⁴⁰ This data was used for all NDVI analysis. For the NDVI box-plot analysis, the NDVI data for each LCZ was calculated using data pertaining to all urban pixels in the Tokyo Prefecture. For the NDVI spatial analysis, a median was calculated of the NDVI pixel data for all urban pixels in each WCTV.

Land surface temperature (LST)

LSTs have been adopted in numerous studies investigating heat in urban built environments.⁴⁸ Therefore, LSTs were utilized in this study as a thermal environment index and indicator of UHI intensity. The Google Earth Engine platform⁴⁰ was utilized for extraction of the MOD11A1.061 Terra Land Surface Temperature and Emissivity Daily Global 1 km dataset which was adopted as an LST representative. This dataset is provided by the MODIS instrument which collects remotely sensed diurnal and nocturnal LSTs and is carried aboard the NASA Terra satellite. The Terra satellite, at 1:30 a.m. and 1:30 p.m., passes over the equator from north to south.⁴⁹

There are two frequent issues associated with remotely sensed datasets: cloudy pixels⁵⁰ and anomalous data.⁵¹ To mitigate these, a 9-year median diurnal LST image in June, July, and August from 2013 to 2021 was adopted. This image was then resampled to 100 m pixel resolution for combining with LCZ land classification data to determine LSTs of each LCZ. In total, in the Tokyo Prefecture 178 177 100 m² pixels of LST data were collected. This was deemed a suitable dataset size for the scope and range of the data analysis in this study. For all LST analysis, median pixel data from the 9-year diurnal LST image in June, July, and August from 2013 to 2021 was used. For the LST spatial distribution analysis of the WCTVs in the Tokyo Prefecture, each WCTV's LST represents a median of urban pixels in that WCTV. For the LCZ box-plot analysis, the LST data for each LCZ was calculated using data pertaining to all urban pixels in the Tokyo Prefecture. Adoption of box-plot analysis also enabled comparison of compact and open LCZ categories, which spatial analysis does not permit. In addition, for the WCTV health heat risk box-plot analysis, the LST for each WCTV was calculated using data pertaining to all urban pixels in the each WCTV. Estoque et al.⁵² stated there was a strong connection between heat health risks and LSTs. The threshold of 38.3°C as an indicator of heat health risk was based on MODIS Terra LSTs and mortality data in Philippine cities. In this study, the LSTs of WCTVs were deemed as hazardous if they exceeded 38.3°C, this LST has also been previously adopted for heat risk studies.⁵³

Results Local climate zone (LCZ) classification

Figure 2 shows the LCZ land-type classification for the Tokyo Prefecture, spatially distributed at a pixel resolution of 100 m. As seen in Figure 3, the Tokyo Prefecture predominantly consists of two urban and rural LCZs: CLR at 36.5% and forest at 34.5%. Urban LCZs of OMR, CMR, OHR, and CHR represent smaller portions of 10.8%, 3.6%, 3.2%, and 2.9%. While grassland constitutes the second largest rural LCZ with 5.6%. All remaining LCZs are less than 1.5%. Urban and rural LCZs are divided at an approximate longitude of 139.30° E. The WCTVs in the western area at longitudes of 138.95°–139.30° E and latitudes of 35.60°–35.90°N constitute mostly forest LCZs. Such as Okutama and Hinohara. However, Ome, Hinode, Akiruno, and Hachioji consist of a combination of urban and rural LCZs. The WCTVs in the central area at longitudes of 139.30°–139.60°E and latitudes of 35.50°–35.80°N are comprised of mostly urban with clusters of rural LCZs. For example, Chofu and Mitaka consist of mostly CLR with clusters of OHR and OMR. While Inagi and Tama have denser clusters of OMR, forest, and grassland.

Generate image description

FIGURE 2. Local climate zones (LCZ) classification of the Tokyo Prefecture with 100 m pixel resolution

Generate image description

FIGURE 3. Proportions of local climate zones (LCZ) classification of the Tokyo Prefecture with 100 m pixel resolution

The WCTVs in the eastern area at longitudes of 139.60°–139.90°E and latitudes of 35.55°–35.80°N consist of mostly urban LCZs. For example, in the city center Chiyoda and Chuo consist of dense clusters of higher-rise compact LCZs such as CSHR, CHR, and CMR. Chiyoda also constitutes small areas of grassland and forest. Minato and Koto include CLR but are predominantly a mixture of CHR, CMR, OHR, and OMR. Perimeter WCTVs such as Adachi, Nerima and Setagaya are mostly CLR with smaller clusters of higher-rise compact and open LCZs.

With regard to accuracy assessment, the LCZ classifier's overall and kappa accuracies were 96% and 95%, respectively. The vast majority of user and producer accuracies were above 70% with several over 90%. The random forest's optimum number of decision tress was also calculated as 75. In the subsequent sections, a comparative spatial and box-plot analysis is conducted for LSTs and NDVIs classified with LCZs to determine LCZs prone to increased urban heat intensity. WCTVs with a majority of rural pixels have been excluded from analysis. These include Okutama, Hinohara, Ome, Hinode, Akiruno, and Hachioji.

Spatial analysis of land surface temperatures (LST) for wards, cities, towns, and villages (WCTV)

Figure 4 shows daytime spatially distributed LSTs for the WCTVs in the Tokyo Prefecture, with median values of urban pixels for June, July, and August from 2013 to 2021 being used. The LSTs of WCTVs in the central area at longitudes of 139.30°–139.60°E and latitudes of 35.50°–35.80°N range from Inagi and Tama's 36.0°C and 36.1°C to Nishitokyo's 38.9°C. According to Figure 2, Inagi and Tama consist of mostly OMR and Nishitokyo consists of mostly CLR. Also, the majority of WCTVs in this area are comprised of predominantly CLR with LSTs above 37.0°C. This shows CLR to be prone to increased LSTs and OMR to offer urban heat reduction capabilities. Furthermore, this suggests construction of OMR LCZ to be an urban heat mitigation strategy. However, Inagi and Tama are in the close vicinity of larger clusters of forest and grassland pixels. This may contribute to reduced LSTs in urban pixels neighboring rural pixels which lowers the median urban LST of the WCTV.

Generate image description

FIGURE 4. Ward, city, town, and village (WCTV) spatial distribution of median daytime LSTs of urban pixels for June, July, and August 2013–2021 in the Tokyo Prefecture

The LSTs of WCTVs in the eastern area at longitudes of 139.60°–139.90°E and latitudes of 35.55°–35.80°N range from Koto, Chuo, Chiyoda, and Minato's 34.2, 34.3, 34.6, and 34.7°C to Nakano's 39.1°C. As shown in Figure 2, Chiyoda and Chuo consist of dense clusters of higher-rise compact LCZs such as CSHR, CHR, and CMR. While Minato and Koto include CLR but are predominantly a mixture of CHR, CMR, OHR, and OMR. Nakano, similar to many other perimeter wards such as Adachi, Nerima and Setagaya with higher LSTs, consists of mostly CLR. This shows that compact and open higher-rise LCZs are prone to reduced LSTs. Also, compact and lower-rise LCZs contribute to increased LSTs which can constitute a heat health risk. In addition, this suggests the construction of compact and higher-rise LCZs to be urban heat mitigation strategies. Also, regarding Koto, the presence of water in the Tokyo Bay is likely contributing to cooler LSTs; however, further research is required to identify the extent of this cooling effect.

Box-plot analysis of summertime land surface temperatures (LST) and normalized difference vegetation index (NDVI) for each local climate zone (LCZ)

Figures 5 and 6 show the box-plot analysis of the summer LSTs and NDVIs of all pixels in the Tokyo Prefecture classified using compact and open LCZs. For the LSTs of compact LCZs (Figure 5), the difference between the highest (CLR) and lowest (CSHR) median LSTs is 3.7°C while for open LCZs, the difference between the highest (OHR) and lowest (OSHR) median LSTs is 0.1°C. The LSTs of the LCZs are CSHR, CHR, CMR, and CLR with 34.4, 35.5, 37.3, and 38.1°C, respectively, and OSHR, OHR, and OMR with 37.4, 37.5, and 37.1°C, respectively. Excluding OMR, these results show lower-rise LCZs to be prone to increased LSTs, higher-rise LCZs to contribute to decreased LSTs, and compact LCZs to have lower LSTs than equivalent height open LCZs. Furthermore, among all LCZs, CLR has the most pixels equal or close to a heat health risk LST. In addition, OMR offers reduced LSTs. Also, for compact LCZs, the large variation in LSTs shows potential for significant reductions in LST depending on building height. These trends suggest higher-rise and compact or OMR LCZs to be a mitigation strategy for urban heat. Also, these trends imply that CLR contributes to a heat health risk increase for residents. Moreover, CLR and OMR have LSTs of 38.1°C and 37.1°C, therefore OMR being 1°C cooler. This supports the finding in section 3.2 of Inagi and Tama having lower LSTs than Nishitokyo.

Generate image description

FIGURE 5. Box-plot of median summertime LST classified with local climate zones (LCZ) from 2013 to 2019. The center lines show the LCZ median values, the colored boxes represent the upper and lower data quartiles, the highest and lowest LSTs are shown by the maximum and minimum extents and data outliers are not shown. CSHR: compact super high-rise, CMR: compact mid-rise, CHR: compact high-rise, and CLR: compact low-rise

Generate image description

FIGURE 6. Box-plot of median summertime NDVI classified with local climate zones (LCZ) from 2013 to 2019. The boxes and whiskers represent the same data as Figure 5, except pertaining to NDVI. CSHR: compact super high-rise, CMR: compact mid-rise, CHR: compact high-rise, and CLR: compact low-rise

For the median NDVIs of compact LCZs (Figure 6), CSHR, CHR, and CMR are 0.21, 0.20, and 0.21, respectively, while CLR's is 0.30. For open LCZs, OSHR, OHR, and OMR are 0.26 and 0.26 while OMR's is 0.44. NDVI and LST results for LCZs show CSHR, CHR, and CMR to have similar NDVI values and significantly varying LSTs. Also, higher NDVI values combined with CLR have the highest LSTs among all LCZs. Finally, higher NDVI values combined with OMR result in a reduction in LST.

Vegetation has been shown to have a cooling effect on urban areas.⁵⁴ Therefore, NDVI and LSTs for CSHR, CHR, and CMR suggest that although vegetation causes a cooling effect, building height and density are more influential on urban heat intensity variation. Also, NDVI and LSTs for CLR suggest that CLR's LST, free of the cooling effect of vegetation, is even higher than that shown in Figure 5. This implies that CLR constitutes a heat health risk and is prone to, more than previously suggested, increased urban heat intensities. This again indicates the significant effect of building height and density for urban heat variation and supports the previously stated outcome of compact and higher-rise LCZs being an UHI mitigation strategy. In addition, NDVI and LSTs for OMR imply that vegetation combined with OMR is effective for urban heat mitigation. Spatial and box-plot analysis identified CLR and OMR as key LCZs affecting LSTs. Therefore, they were selected for further WCTV spatial analysis with LSTs and NDVIs to confirm the trends identified in this section.

Spatial distribution of the local climate zone (LCZ) compact low-rise (CLR) for wards, cities, towns, and villages (WCTV)

Figure 7 shows the spatially distributed WCTV CLRs as a percentage of WCTV total urban pixels in the Tokyo Prefecture. As shown, there is a general correlation between higher WCTV CLR percentages and the higher WCTV LSTs in Figure 4. The percentage of CLR in WCTVs in the central area at longitudes of 139.30°–139.60°E and latitudes of 35.50°–35.80°N range from Inagi and Tama's 32.6% and 29.5% to Kokubunji's 81.4%. Figure 4 shows Inagi and Tama's LSTs to be 36.0°C and 36.1°C and Kokubunji's to be 38.5°C. The majority of WCTVs in this area have CLR percentages above 65% and LSTs above 37.5°C. The percentage of CLR in WCTVs in the eastern area at longitudes of 139.60°–139.90° E and latitudes of 35.55°–35.80°N range from Chiyoda and Chuo's 34.6% and 34.3% to Meguro, Nerima, and Suginami's 88.4%, 83.5%, and 81.0%. As shown in Figure 4, Chiyoda and Chuo's LSTs are 34.6°C and 34.3°C and Meguro, Nerima, and Suginami's 38.2, 39.0, and 39.0°C. These results from the eastern and central areas further show, on a WCTV scale, CLR to be prone to increased LSTs. And, aligning with previous sections, indirectly suggests construction of higher-rise LCZs or OMR to be urban heat mitigation strategies. In subsequent sections, a spatial comparative analysis is also conducted between Figure 7’s CLR and spatially distributed NDVIs of WCTVs. Aiming to further ascertain the extent to which LCZs are influencing LST variation.

Generate image description

FIGURE 7. Ward, city, town, and village (WCTV) spatial distribution of compact low-rise (CLR) as a percentage of WCTV total urban pixels in the Tokyo Prefecture

Spatial distribution of the local climate zone (LCZ) open mid-rise (OMR) for wards, cities, towns, and villages (WCTV)

Figure 8 shows the spatially distributed WCTV OMRs as a percentage of WCTV total urban pixels in the Tokyo Prefecture. As shown, in a general sense, there is no correlation between higher WCTV OMR percentages and higher WCTV LSTs in Figure 4. However, the percentage of OMR in WCTVs in the central area at longitudes of 139.30°–139.60°E and latitudes of 35.50°–35.80°N provide some interesting observations. For example, Inagi and Tama have the highest percentages of OMR of 65.4% and 67.1%, whereas Nishitokyo, Kokubuni, and Higashiyamato have the lowest with 17.0%, 16.6%, and 16.1%. Figure 4 shows Inagi and Tama's LSTs to be 36.0°C and 36.1°C and Nishitokyo, Kokubuni, and Higashiyamato's to be 38.9, 38.5, and 38.1°C. These results further show, on a WCTV scale, OMR as prone to decreased LSTs. And, aligning with previous sections, suggests construction of OMR to be an urban heat mitigation strategy. In the following section, a spatial comparative analysis is conducted between Figure 8's OMR and spatially distributed NDVIs of WCTVs. This will allow for a greater understanding of the impact LCZs are having on LSTs in the Tokyo Prefecture.

Generate image description

FIGURE 8. Ward, city, town, and village (WCTV) spatial distribution of open mid-rise (OMR) as a percentage of WCTV total urban pixels in the Tokyo Prefecture

Spatial distribution of the normalized difference vegetation index (NDVI) for wards, cities, towns, and villages (WCTV)

Figure 9 shows the spatially distributed NDVI values for the WCTVs in the Tokyo Prefecture, with median values of urban pixels for June, July, and August from 2013 to 2021 being used. As shown, generally there is no correlation between higher WCTV NDVI value percentages and lower WCTV LSTs in Figure 4. However, in the central area at longitudes of 139.30°–139.60°E and latitudes of 35.50°–35.80°N, Machida, Mizuho, Tama, and Inagi have NDVI values of 0.45, 0.45, 0.49, and 0.47. According to Figure 4, their LSTs are 37.6, 37.0, 36.1, and 36.0°C. Furthermore, their percentages of OMR are 40.1%, 43.5%, 67.0%, and 65.4%. The similarity in NDVI values combined with declining LSTs suggest that forest and grassland LCZs or vegetation do not significantly influence LST reduction. This combined with the trend of increasing OMR percentages matching decreasing LSTs suggests that building configuration is more strongly influencing LST variation. This supports the trends identified in previous sections showing OMR as prone to decreased LSTs, and therefore, also suggesting OMR as an urban heat mitigation strategy.

Generate image description

FIGURE 9. Ward, city, town, and village (WCTV) spatial distribution of the median normalized difference vegetation index (NDVI) for WCTV urban pixels in the Tokyo Prefecture

In the central area at longitudes of 139.30°–139.60°E and latitudes of 35.50°–35.80°N, Toshima and Taito have similar NDVI values of 0.19 and 0.14 and LSTs of 38.9°C and 36.8°C. As seen in Figure 7, Toshima and Taito also have CLR percentages of 80.8% and 23.9%. The similarity in NDVI values combined with declining LSTs suggests that forest and grassland LCZs or vegetation do not significantly affect LSTs. This combined with the trend of higher LSTs matching higher CLR percentages suggests that building configuration is more significantly influencing LST variation. This also supports the previously identified trend of higher CLR percentages equaling higher LSTs. This, therefore, also indirectly suggesting construction of higher-rise LCZs or OMR to be urban heat mitigation strategies. Spatial and box-plot analysis has identified LCZ types which elevate or alleviate urban heat. In the following section, WCTVs with LSTs hazardous to health are identified and analysis of the individual makeup of LCZs for each is conducted. Therefore, aiming to further investigate trends identified in this and previous sections and identify priority WCTVs for heat mitigation.

Heat health risk assessment of land surface temperatures (LST) for wards, cities, towns, and villages (WCTV)

Figure 10 shows the box-plot analysis of the summer LSTs classified using WCTVs identified with median LSTs hazardous to health. The difference between the highest (Nakano) and lowest (Kodaira) median LSTs is 0.7°C. The three WCTVs with the highest median LSTs are Nakano, Nerima, and Suginami with the vast majority of their pixels above 38.3°C. The range in LST also shows that even among the highest temperature WCTVs LSTs can vary substantially. These results suggest that all 11 WCTVs in Figure 10 should be targeted for urban heat mitigation to reduce potential threats to public health with Nakano, Nerima, and Suginami being a priority. Figure 11 shows the stacked bar chart analysis of the LCZ makeup for the WCTVs identified with median LSTs hazardous to health. Across all WCTVs the average percentage of CLR, OMR, and rural pixels is 73.6%, 13.1%, and 4.0%. This shows that increased numbers of CLR LCZ pixels in a WCTV are likely to result in increased LSTs, potentially hazardous to public health. This supports the trends identified in previous sections showing CLR to be prone to increased LSTs and suggesting construction of higher-rise LCZs or OMR to be urban heat mitigation strategies.

Generate image description

FIGURE 10. Box-plot of wards, cities, towns, and villages (WCTV) identified with median urban pixel summertime LSTs constituting heat health risks. The boxes and whiskers represent the same data as Figure 5

Generate image description

FIGURE 11. Box-plot of wards, cities, towns, and villages (WCTV) identified with median summertime LSTs constituting heat health risk

Discussion

Building height and density were the most significant parameters for urban LCZ definition in this project. However, LCZs can also be defined by a multitude of other geometric and surface cover parameters, as well as thermal, radiative, and metabolic properties. These include sky view factor, aspect ratio, surface admittance, surface albedo, anthropogenic heat, and pervious surface fraction. More specifically, compact and higher-rise LCZs, when contrasted with open and lower-rise LCZs, have comparatively reduced sky view factors, pervious surface fractions, and surface albedos and increased aspect ratios, surface admittances, and anthropogenic heat.²⁷

Aspect ratios and sky view factors have been shown to be correlated. Larger urban aspect ratios correspond to lower sky view factors.¹ Lower aspect ratios and higher sky view factors have more intense diurnal UHIs.⁵⁵ The opposite trend was observed for higher aspect ratios and lower sky view factors.⁵⁶ Surface admittance and albedo are also correlated²⁷ with increased surface albedo reducing urban temperatures.⁵⁷ However, such reductions can be offset by increases in solar reflection, resulting in intensified UHI effects.⁵⁸ Therefore, the effect of surface albedo on heat storage is complex and requires further investigation. Moreover, urban morphology significantly affects anthropogenic heat emissions, which are a key factor contributing to UHI development.⁵⁹ Anthropogenic heat in compact high-rise urban areas intensifies the UHI effects.^27,60 However, despite this, the LSTs in Tokyo's compact and higher-rise city center remains cooler than surrounding compact or open and lower-rise areas. This suggests that building height and density are more influential on urban LST variation than anthropogenic heat gains and supports selection of these urban parameters for investigation in this study. Quantification of the effects of anthropogenic heat for urban heat intensification in the Tokyo Prefecture remains an area of future research. In addition, an increased urban pervious surface fraction is also an effective UHI mitigation strategy.⁶¹ As shown in Figures 2, 9, and 4, higher-rise compact LCZs have lower NDVI values and LSTs than lower-rise compact LCZs. This suggests that, for compact LCZs, building height and density are more influential on LST variation that vegetation. Regarding open LCZs, OMR's reduced LST suggests that vegetation is the most effective for urban heat reduction in OMR urban configurations. Overall, this discussion suggests that higher-rise and compact buildings have reduced LSTs compared with lower-rise and open buildings. Furthermore, the significant effect of building height and density on LST variation was also highlighted.

As highlighted in section 3.3, the LSTs of CLR and OMR are 38.1°C and 37.1°C, respectively. Therefore, OMR offers a 1°C cooling effect when compared to CLR. Also, Table 1 shows the number of floors for LCZs, with CLR being 1–3 and OMR being 4–8. Considering the compact and open density nature of CLR and OMR, it is reasonable to suggest that switching from CLR to OMR would not result in a huge change in floor area ratio. This suggests that a 1°C reduction in LST could be achieved as a result of changing from CLR to OMR while maintaining the same number of residential units. However, additional research is required to explore this further.

This study also has limitations. For example, for the calculation of LSTs for LCZs, a 1 km MODIS data were resampled to 100 m. The adoption of a finer resolution LST dataset would enable a more accurate analysis of the spatial distribution of urban heat in Tokyo. In addition, for the selection of polygons for LCZ classification building height and density were the most important factors. But, LCZs can also be defined by impervious surface fraction, pervious surface fraction, height and roughness of elements, terrain roughness, surface admittance, surface albedo, sky view factor, aspect ratio, and building surface fraction.²⁷ Inclusion of these factors when selecting training polygons for LCZ classification will produce more accurate classifications. Furthermore, LSTs were shown to vary significantly among LCZs, for example, CLR in sections 3.2 and 3.4. Further research is required to establish the causes for this. Such research will allow for a greater understanding of the impact that the built environment is having on urban heat variation in the Tokyo Prefecture and improve the usefulness of recommendations. Moreover, spatial and box-plot trend analysis for LSTs classified by LCZs were conducted in the summer season. Further research is required to test the identified trend's temporal validity across all seasons and assess their applicability for urban heat mitigation.

Conclusion

To identify urban forms prone to increased heat intensity in the Tokyo Prefecture, we adopted LCZs, NDVIs, and LSTs. For the LCZ calculation, Landsat 8 remotely sensed data at a pixel resolution of 100 m was adopted. NDVI data was provided by the MODIS Terra Vegetation Indices dataset at a resolution of 100 m from 2013 to 2021. For LST data, daytime MODIS 100 m resolution LSTs from 2013 to 2021 were utilized. The urban heat intensity of LCZs were evaluated both on an individual LCZ and WCTV scale to allow for more actionable recommendations for urban heat reduction. We performed summertime LST and NDVI spatial and box-plot trend analysis for each LCZ on a WCTV and prefectural scale. This was followed by spatial analysis for CLR, OMR, NDVI, and LST, also on the WCTV scale. Subsequent identification of WCTVs subject to heat health risks and trend analysis of their LCZ composition was also conducted. Utilizing these results, the LCZs and urban morphologies prone to increased heat intensity were ascertained.

Spatial analysis of total LCZ and LST identified trends of CSHR, CHR, CMR, and OMR with lower median LSTs than CLR. Adopting prefectural data, the LST box-plot analysis identified trends of, excluding OMR's 37.1°C, higher-rise and compact LCZs having lower median LSTs than lower-rise and open LCZs. This is shown by LST of CSHR, CHR, CMR, and CLR of 34.4, 35.5, 37.3, and 38.1°C, respectively, and also by LST of OSHR and OHR of 37.4°C, 37.5°C, respectively. Therefore, OMR also offers reduced LSTs. The NDVI box-plot analysis suggested building height and density to be more influential on LST variation than vegetation. Therefore, spatial and box-plot analyses showed that higher-rise and compact LCZs are less prone to increased urban heat intensity than lower-rise and open LCZs. OMR's comparatively low LST shows this urban configuration is also less prone to increased urban heat intensity. Such LST trends among the LCZs also imply that the construction of higher-rise and compact or OMR LCZs are mitigation strategies for UHI development. The primary aim of this study was to identify UHI mitigation strategies.

Spatial analysis of CLR and LST identified trends of WCTVs with higher percentages of CLR pixels having increased LSTs. Spatial analysis of OMR and LST determined trends of WCTVs with larger percentages of OMR pixels having decreased LSTs. This confirms the previously identified LCZ WCTV spatial and box-plot-based trends of lower-rise LCZs with higher LSTs and OMR with disproportionately low LSTs. In addition, spatial analysis of LCZ and LST with NDVI further suggested urban configuration to be significantly influencing LST variation. Therefore, this suggests that CLR and lower-rise LCZs are prone to increased daytime urban heat intensities and higher-rise compact LCZs and OMR to be an UHI mitigation strategy. Finally, 11 WCTVs were identified with LSTs which posed heat health risks. Among these are Nakano, Nerima, and Suginami which have the highest median LSTs in the whole prefecture. These 11 WCTVs are comprised of predominantly CLR pixels with very small numbers of rural pixels. This further showing the connection between CLR and lower-rise buildings and increased urban heat intensity. Therefore, these WCTVs were also identified as priority WCTVs for UHI mitigation.

Acknowledgments

Funding from organizations in the public, commercial, or not-for-profit sectors was not provided for this research project.

Disclosure

There are no conflicts of interest to declare.

Data Availability Statement

The data that support the findings of this study are openly available in “4TU.ResearchData” at https://doi.org/10.4121/21067720.v1, reference number 21067720. Also, the title of the dataset is ‘Data for paper titled "An investigation into urban heat mitigation by adopting local climate zones and land surface temperatures in the Tokyo Prefecture"’.