Introduction

The process of urbanization exacerbates the deterioration of climatic conditions [1] and global warming [2]. Global climate change such as heat waves is also changing urban agglomeration patterns [1], exerting pressure on urban infrastructure and social institutions [3]. Excessive urban temperatures negatively affect residents’ physical and mental health, air quality, and resource consumption [4, 5]. Owing to the interdependence between the thermal environment and residents’ well-being, research on the urban thermal environment has attracted increasing interest from the scientific community and policymakers in recent years. The high incidence of heat stroke is a typical example of how climate change affects human well-being, and a policy response is strongly needed. However, despite considerable research on urban thermal environments, no effective measures have been found.

Extreme heat leads to increased morbidity and mortality from heat-related illnesses such as respiratory and cardiovascular issues and heat stroke [6, 7]. Heat stroke, which disrupts the functioning of various organs of the body and can lead to death within hours [8], is one of the most threatening health outcomes of extreme temperatures. Additionally, it threatens occupational health and labor productivity [9,10,11]. A study covering more than 1500 locations in 77 countries found that temperature has a significant impact on productivity levels, with an increase in global average surface temperature of about 3.5 °C reducing global output by 7–14% in 2100 [12]. When the daily maximum temperature exceeds the daily average maximum temperature by approximately 1.5 °C, the number of emergency transport cases due to heat-related illnesses increases by 2.4–8.9 times [13]. Compared with residents living in cold urban areas, residents in hot regions face a nearly 6% higher risk of heat-related mortality [14]. A study in California indicated that for every 10 °F increase in average ambient temperature, hospitalizations due to heat exposure, ischemic stroke, and acute renal failure rose by 393%, 3%, and 15%, respectively [15]. A study in New York also found that for each 1 °C increase in the temperature-health effect curve threshold on a day, the number of respiratory admissions on that day increased by 2.7–3.1% [16]. Heat-related illnesses impose a significant financial burden on societal economic productivity, public health, and energy consumption. Between 1991 and 2004, there were 100 annual hospitalizations in New York State due to extreme heat-induced respiratory diseases, with direct hospitalization costs amounting to $64,069 per case. Furthermore, it is estimated that between 2080 and 2099, the number of hospitalizations will increase by 206–607 individuals annually, with hospitalization costs rising between $26 million and $76 million [17]. The economic cost of specific health impacts of heatwaves in France reached €25.5 billion, primarily attributed to mortality (€23.2 billion), days with mild activity restrictions (€2.3 billion), and morbidity (€0.031 billion) from 2015 to 2019 [18]. The risk of filing workers’ compensation claims due to heat exposure increases with the rise in Wet Bulb Globe Temperature [19]. From 2013 to 2019, it was estimated that the average number of premature deaths in Wuhan due to high temperatures was 77,369, resulting in an economic loss of 156.1-billion-yuan, accounting for 1.81% of Wuhan’s annual GDP [20]. In addition, the high temperature has a serious impact on the electricity consumption of the construction industry, and the peak electricity consumption and total electricity consumption increase significantly. For every 1-degree Celsius increase in temperature, peak electricity demand increases by 0.45–4.6%, resulting in an average increase of about 21 W (± 10.4) per person. The actual electricity consumption per degree Celsius rise in temperature increases by 0.5–8.5% [21]. In addition, public expenditure on heat stroke countermeasures is increasing rapidly in Japan, with ministries and agencies contributing a total of approximately 98 billion yen to the relevant budget in 2023[22]. However, effective measures are required to break the deadlock.

The economy, society, and environment are all significantly influenced by land-use patterns [23]. Land reforms are of paramount importance to ensure sustained productivity, food security, poverty alleviation, nature conservation, and environmental preservation [24]. It is important to note that policies frequently shape land-use changes [25]. Consequently, effective land-use policies play a pivotal role in realizing the Millennium Development Goals [26]. Many countries have formulated a series of land-use policies for the sustainable development of cities. The United States has established a national park system to protect natural ecosystems, while Sydney has adopted “green landscape” measures to protect wetlands and rivers. China has implemented an ecological compensation mechanism to encourage the protection and improvement of its ecosystems. Moreover, Japan began enacting the Law on Productive Green Areas in 1992, which has remained unchanged for 30 years. By utilizing tax incentives, unused and vacant land within cities is used for agriculture and greening, providing more green spaces for urban residents and increasing the cities’ self-sufficiency. In Japan, this policy underwent a phased change in 2022, and various cities implemented the corresponding policy. For instance, in Kobe City, it was stipulated that once land was designated for special productive green spaces, this policy would remain effective for a period of 10 years. The Other Effective area-based Conservation Measures (OECMs) underscore the significance of informal nature conservation areas for global biodiversity and ecosystem preservation.

Moreover, nature-based solutions are considered effective measures for alleviating urban overheating and improving the well-being of residents [27]. The cooling benefits exhibited by urban green–blue spaces during hot summers can save energy and reduce mortality rates from heat-related illnesses. Forest coverage helps alleviate vulnerability to the negative effects of heat, with a significant negative correlation observed between forest coverage and heat-related mortality rates [28]. Research on traditional gardens in Beijing indicates that the total cooling effect caused by water bodies and the entire garden is 4.7 × 108 J and 8.6 × 108 J, respectively [29]. Compared with cities with a low green-space value, cities with a high-green space value have a relatively lower risk of heat-related mortality [30]. In environments hotter than 29.61 °C, for each 1 °C rise in temperature, mortality rates increase by 5.7%, 5.4%, and 4.6% for exposure to low, medium, and high-quality green spaces, respectively, thus confirming the fact that high-quality green spaces reduce mortality from heat-related illnesses [31]. As the key natural elements in the city, blue–green spaces have a particularly positive effect on cooling the thermal environment [32]. Factors such as environmental indices and landscape pattern changes can alter the LST. The cooling effect of green spaces on a city is impacted by the city’s area, shape, pattern, and surroundings [33, 34]. Blue-space cooling and transport during the day can be substantial, and nocturnal warming is likely when the conditions are the most oppressive. Thus, the combination of blue–green spaces provides numerous additional ecosystem benefits [35].

However, a systematic comparison of the cooling effects and influencing factors of blue–green spaces in cities with humid subtropical climates at the patch class and landscape levels has not yet been conducted [32]. Numerous studies have attempted to quantify the temperature-reduction values of green and blue spaces individually; however, there is limited literature assessing their combined cooling effect [36]. In recent years, the importance of the combined effects of blue–green space planning has been identified [37]. Some studies have shown that the combined cooling effects of blue–green spaces are clear in the 7–12 m surrounding waterfront areas, where the mean air temperature reduction is 3.3 °C higher than the sum cooling effect of standalone water and forest [38]. A few studies have confirmed the synergistic cooling effect of blue–green spaces on the microclimate at a small scale, including the neighborhood and community scales, through field investigation or numerical simulation [39]. However, there is limited research on the combined effects of blue–green spaces at different moderate scales. Therefore, it is necessary to explore more accurate and quantitative results at moderate scales. In addition, because drastic land use land cover (LULC) changes bring about more changes in LST, most previous research on the thermal environment has focused on rapidly developing regions, while research on developed regions is relatively limited. However, it is worth noting that, owing to the stabilization of built-up areas in developed regions, land-use change is relatively gentle, which eliminates the influence of other factors when studying the combined effect of blue–green spaces on the thermal environment. The immense economic burden imposed by extreme heat environments renders studies on the economic benefits of cooling from blue–green spaces increasingly valuable. However, in-depth research is limited due to factors such as the difficulty of data acquisition, high data volume requirements, and fuzzy calculation range. On this point, this study discusses the cooling effect of urban blue–green spaces and the heat-related disease and energy consumption while considering economic land policy, specifically the production green space policy. This study mainly used remote-sensing images to evaluate the combined effect of blue–green spaces on urban thermal environments. This study aimed to (1) examine the spatial–temporal conditions of urban thermal environments in blue–green spaces of different land-use types, (2) examine the effects of landscape metrics on urban thermal environments at different grid scales, (3) use regression analysis to explore the combined effects of blue–green spaces, and (4) analyze the economic impact of special productive green space.

Study area and datasets

Study area

Kobe, located in western Japan, is the capital of Hyogo Prefecture and covers an area of 557.02 km2. Kobe is an alluvial fan backed by Mount Rokko (with the highest altitude of 931 m) and faces Osaka Bay. Kobe has a humid subtropical climate, with mild weather throughout the year. In the 30 years from 1991 to 2020, its annual average temperature was 17.0 ℃, and the annual precipitation was 1277.8 mm. In addition, its annual average temperature was 27.1 ℃ with an average of 55.3 days exceeding 30 ℃ and 4.6 days exceeding 35 ℃ during the period from July to September. Over the past decade, there have been an average of 632 heat stroke emergencies in Kobe annually (975 cases in 2018), and 59% of the cases were of patients aged over 60. This paradoxical climate of mild temperatures but with increasing heat waves in recent years makes Kobe an ideal area for studying thermal environments (see Fig. 1).

Fig. 1
figure 1

Source: Geospatial Information Authority of Japan

Study area map (Kobe, Japan).

Full size image

Datasets

In this study, summer, which has a prominent thermal environment, was selected as the research season. Two images from the Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS; path 110, row 36) were obtained from the United States Geological Survey (USGS) for LST retrieval. Cloud cover also affects the accuracy of surface temperature retrieval. Cloud cover in the images was close to zero for the study area. The images for September 9, 2014, and September 7, 2019, were adopted to eliminate any error caused by the date. LULC data with a spatial resolution of 30 m were collected from the Japan Aerospace Exploration Agency and further reclassified into six categories: water bodies, built-up, forest, farmland, grass, and bare land.

Methods

LST retrieval

The study adopted the Radiative Transfer Equation (RTE) algorithm to retrieve LST. The RTE algorithm is one of the primary methods for retrieving LST and offers higher accuracy [40]. The RTE algorithm is based on the thermal radiation exchange between the Earth's surface and the atmosphere. It considers radiation from the Sun and the Earth, as well as absorption, scattering, and transmission within the atmosphere. By observing the radiation characteristics of the atmosphere and the Earth's surface, it is possible to infer the surface temperature. The basic principle of the RTE algorithm is to first estimate the influence of the atmosphere on the thermal radiation emitted from the Earth's surface. This atmospheric influence is then subtracted from the total thermal radiation observed by satellite sensors to obtain the thermal radiation intensity of the Earth’s surface. Finally, the thermal radiation intensity is converted into the corresponding surface temperature.

The Landsat 8 OLI/TIRS provided two thermal bands: 10 and 12. Thermal band 10, which has a better inversion effect, was adopted to retrieve the LST. The Landsat remote-sensing images were preprocessed before LST retrieval. This involved atmospheric correction, radiometric correction, layer stacking of different bands, and extraction of the area of interest. A Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes FLAASH) model based on the moderate resolution atmospheric transmission (MODTRAN) program was used to implement atmospheric and radiometric correction on the Landsat 8 OLI/TIRS images [41]. The FLAASH model was adopted in this study. This processing procedure was implemented in the ENVI 5.3 software platform. The detailed calculation process is provided in the Appendix [42].

Classification of LULC

In view of the inconsistent classification of land-use types obtained from JAXA in 2014 and 2019, the land-use images were reclassified into six categories using ArcGIS 10.2 software for comparative research: built-up, water body, forest, farmland, grass, and bare land. In addition, the land-use transfer matrix (image difference) technique was applied to detect the change in LULC pattern for each LULC class, that is, to determine which land use/cover class has been converted into what [43].

Landscape pattern metrics

Landscape pattern metrics are considered the main tools for quantitatively evaluating landscape patterns and can measure heterogeneity and characterize its composition and distribution. Landscape measurements are divided into three levels based on the level of the described object: patch, class, and landscape levels [44]. Common landscape metrics were selected to quantify landscape patterns at class levels. Landscape pattern metrics provide an objective and quantitative measurement of landscape structure. They can be analyzed at different spatial scales to reveal landscape diversity and complexity and monitor landscape changes such as urbanization and land use changes. The four selected indices in this study were analyzed at multiple scales, reflecting various characteristics of landscape structure such as quantity, aggregation, fragmentation, and complexity. Additional details are presented in Table 1. Landscape metrics were calculated using FRAGSTATS 5.3 to explain the structure of the entire landscape, and the patch-type level and landscape level indexes were typically used.

Table 1 Landscape pattern metrics used in the study
Full size table

Grid analysis method

Landscape patterns are spatially correlated and scale-dependent; therefore, it is necessary to employ multiscale analysis to adequately characterize and monitor landscape structures and functions [45]. Considering the scale effect of the landscape pattern metrics and resolution of the remote-sensing data, the study area was divided into 600 m × 600 m, 900 m × 900 m, 1200 m × 1200 m, and 1500 m × 1500 m grid size. Moreover, grids with incomplete landscape types at the boundary of the study area were eliminated, resulting in 1365, 567, 309, and 182 grids, respectively. Grid analysis was based on the equidistance sampling function in ArcGIS 10.2.

Statistical analysis

The statistical area was delimited according to certain standards, and the correlation or regression relationship between temperature and all types of landscape pattern metrics in the area was analyzed, which is considered the primary method to explore the relationship between landscape pattern metrics and temperature [46]. The data obtained from September 9, 2014, and September 7, 2019 have been used in the analysis of this study. First, the Pearson correlation analysis was performed to evaluate the correlation between landscape metrics and temperature for various land types based on grid data at different scales. Second, Ordinary Least Squares (OLS) analysis was conducted to analyze the relationship between blue–green spaces and LST.

Results

Spatial distribution and changes in LULC

Multi-year land cover mapping

The land cover map revealed that forests dominated most of the area and were primarily concentrated in the northern region of the study area. Built-up areas were predominantly situated in the southern sector, whereas farmland was distributed in the northeast and northwest. Bare land was sporadically interspersed within the built-up areas, and water bodies were situated between the forests and scattered within the built-up regions (Fig. 2). The layout of land-use types exhibited no significant alterations, implying a relatively stable land-planning arrangement that has not undergone substantial expansion. Subsequent discussions delve into the transitions in land cover from 2014 to 2019, shedding light on the discernible change trends.

Fig. 2
figure 2

Land use and land cover in 2014 and 2019

Full size image

Multi-temporal land cover transition assessment

Land-use change matrix analysis indicated that all classes experienced both gains and losses in different years, and the general change tendency of land use was as follows (Table 2). Forest and built-up were two main land-use types in Kobe, accounting for 74% of the total area. Grass and built-up expanded rapidly over the past 5 years. Forests, bare land, water, and farmland areas experienced significant losses between 2014 and 2019. Accuracy assessment of the land cover classified maps for 2014 and 2019 showed that bare land areas occupied 20.96 km2 (3.77%) in 2014, which decreased to 15.8 km2 by 2019, with a 0.93% change. The water body covered a total area of 9.77 km2 (1.76%) in 2014, which decreased to 5.7 km2 (1.02%) in 2019, with a 0.73% change (Table 2).

Table 2 Land-use change transition matrix from 2014 to 2019 in Kobe (km2)
Full size table

The dynamic patterns of urban land-use change are discernible through the land-use transfer matrix. Grass increased by 2.23%, primarily because of conversions of farmland and forests, accounting for 9.41 km2 and 5.94 km2, respectively. The built-up area expanded by 7.14% and mainly transformed from farmland and bare land, with areas of 10.98 km2 and 10.46 km2, respectively (Table 2). Forests decreased by 7.86%, occupying 1.41% of the study area. Most diminishing forested land was converted into grass (11.13 km2), farmland (5.94 km2), or built-ups (4.9 km2). Bare land exhibited a 4.07 km2 decrease, corresponding to 0.93% of the total area, transitioning primarily into built-ups (10.46%) and farmland (2.29%). Water body experienced a decrease of 4.07 km2, accounting for 0.93% of the total water bodies (Table 2). Some forests and water bodies were sacrificed for urban development, but some farmlands were returned to forests. The increase in grassland compensated for the loss of forest resources to a certain extent. In addition, although the overall quantity of built-up land within the study area remained stable, its spatial arrangement appeared to resemble a pattern akin to forest expansion. In the process of urban development, a portion of forest resources and water bodies were sacrificed, leading to the conversion of some farmland into forests. This suggests that transitions between green spaces were prevalent, with an increase in grasslands compensating for the loss of forest resources. While the total amount of construction land in the city remained relatively stable, its distribution was partially expanded into forested areas.

Spatial distribution and changes in LST

In this section, we examined LST from 2014 to 2019. Subsequently, a comprehensive investigation of the relationship between LST and LULC was conducted.

The spatial pattern map of LST elucidates that the predominant portion of the study area exhibited temperatures ranging from 26 to 32 °C, with temperatures exceeding 38 °C primarily in the built-up areas, indicating an expanding trend. Notably, the temperature levels in built-up regions interspersed within forests and in the southwestern part of the study area increased in 2019 compared to those in 2014 (Fig. 3). To a certain extent, the expansion of built-up areas and the corresponding reduction in forest resources contributed to the deterioration of the surface temperature thermal environment.

Fig. 3
figure 3

Spatial distribution of land surface temperature (LST) in the study area

Full size image

The area with temperatures below 32°C accounted for 62.78% of the total area in 2014, which decreased to 49.56% in 2019. Moreover, 37.31% of the land area exceeded LST of 32 °C in 2014, and this increased to 50.8% in 2019. In 2014, 18.69% of the total area had LST less than 20 °C, with 9.87% of the area having the same range of LST in 2019 (Table 3). The percentage of LST indicated a significant increase in the extent of heat intrusion in the study area. This increase can be predominantly attributed to population growth and the accelerated pace of development.

Table 3 Distribution of land surface temperature (LST) among different ranges for 2014 and 2019
Full size table

To provide further insight into the variations in LST among distinct LULC categories, we conducted statistical analyses of the highest and average surface temperatures within these different land types (Fig. 4). The average LST exhibited the following hierarchy: built-up had the highest average temperature, followed by bare land, water bodies, and farmland; forest had the lowest average temperature.

Fig. 4
figure 4

Maximum and mean land surface temperature (LST) for different land use and land cover (LULC)

Full size image

Furthermore, when comparing the surface temperatures between 2014 and 2019 across various regions, the temperatures in 2019 were generally higher than those in 2014. At the highest temperature, a pattern consistent with the average temperature was observed. A notable difference is that the highest surface temperatures of water bodies and bare land were lower in 2019 than in 2014, whereas construction land exhibited relative stability. Conversely, farmland, grassland, and forests had higher temperatures in 2019 than in 2014. Changes and differences in land-use types caused differences in the LST, and blue–green spaces had a certain improvement effect on the surface thermal environment.

Correlation analysis between LST and landscape metrics

There were significant differences in the correlations between the pattern indices of the different landscape types and LST. The results showed that the average surface temperature was significantly correlated with the landscape pattern indices for bare land, construction land, forest land, and agricultural land over the two years, whereas the average temperature of water bodies and grasslands was significantly correlated with the individual landscape pattern indices.

Specifically, in terms of the grid scale, the landscape pattern metrics exhibited a highly significant correlation with the mean LST for built-up, bare land, farmland, and forest at the grid of 600 m, 900 m, and 1200 m, with the highest correlation coefficient observed at a grid size of 1200 m. However, no significant correlation was observed at the grid scale of 1500 m. The correlation between the landscape pattern metrics and mean LST for grasslands was only partially reflected at a grid scale of 600 m (Table 4). The observed correlation in water across the various grid scales was not significant and displayed a degree of complexity.

Table 4 Correlation coefficients between landscape pattern metrics of land use/land cover and mean land surface temperature
Full size table

There was a highly significant positive correlation between PLAND, LPI, AI, and the mean LST for built-up areas in 2014 and 2019. The correlation coefficient between the mean LST and PLAND was greater than 0.7, especially in 2019, when it exceeded 0.8. However, a limited negative correlation was observed between SPLIT and mean LST (Table 4a). A significant positive correlation was observed between PLAND, LPI, AI, and the mean LST of bare land over the two-year period. The correlation coefficient between PLAND and mean LST exceeded 0.3, with a particularly strong association of over 0.36 in 2014. Similarly, the correlation coefficient between AI and mean LST in 2014 surpassed 0.55. Conversely, a highly significant negative correlation was identified between SPLIT and the mean LST over the two years (Table 4b). PLAND, LPI, and AI exhibited a highly significant negative correlation with the mean LST of the forest over the two-year period. The correlations of both PLAND and LPI with mean LST individually exceeded 0.8. Similarly, the correlation coefficients between AI and mean LST exceeded 0.6, and those of PLAND and LPI with the mean LST exceeded 0.8. However, correlation coefficients between SPLIT and mean LST were consistently below 0.2 (Table 4c). A highly significant positive correlation was found between PLAND, LPI, AI, and the mean LST for farmland in 2019. In contrast, SPLIT exhibited a highly significant negative correlation with mean LST in 2019. However, this correlation was nearly absent in 2014 (Table 4d). In different years and at varying grid scales, SPLIT did not exhibit a significant correlation with mean LST for the water body, whereas PLAND, LPI, and AI showed highly significant negative correlations only at the 900-m grid scale. At other grid scales, these variables did not demonstrate any correlation with the mean LST (Table 4e). Only AI exhibited a statistically significant negative correlation with the mean LST for grassland, with a correlation coefficient below 0.2, whereas PLAND and LPI showed no statistically significant correlations with the mean LST (Table 4f). The results showed that built-up land had the strongest warming effect, which was strengthened by an increase in the area and degree of agglomeration. With the increase in forest area, the low disturbance from human activities was conducive to the cooling effect. However, the cooling effect of small areas of water and grassland on the LST is complicated and cannot be generalized.

Combined effects of blue–green space on LST

Considering that the previous analysis found no significant correlation between landscape pattern metrics and mean LST at a grid scale of 1500 m, the regression analysis in this section focused exclusively on grid scales of 600 m, 900 m, and 1200 m using OLS. Considering that the area is the most influential factor affecting the cooling effect in various regions and holds practical significance for medium-scale land planning, based on the results of the previous section, the independent variables of water, forest, and built-up were selected and analyzed for classification. The independent variable in the regression analysis is the area proportion (%) of each land type in the grid. Two primary regression analyses were conducted. The first investigated the principal determinants of LST, while the second introduced interaction terms to examine the combined impact of blue–green spaces. To mitigate this issue, we first included only the three primary independent factors in the regression analysis. The descriptive statistics are presented in Table 5. The basic model is given in Eq. (1).

Table 5 Descriptive statistics (1)
Full size table
$$LST=\alpha +{\beta }_{1}\text{Water}+{\beta }_{2}\text{Built}-\text{up}+{\beta }_{3}\text{Forest}+\varepsilon $$
(1)

The regression outcomes (Table 6) exhibit variability with respect to grid size. At the 600 m grid scale, water bodies and forests exhibited a significant cooling effect on LST, whereas built-up demonstrated a marked warming effect. With an enlargement of the grid, the significance of the cooling effect attributable to water bodies decreases. At the 1200 m grid, no significant cooling benefit was observed. Nevertheless, the significance of the impact of both forests and built-up on LST remained consistently high. This indicates that the cooling effect of water bodies, particularly in small areas, is constrained by scale.

Table 6 Regression results of the main influencing factors on land surface temperature
Full size table

The land-use map (Fig. 2) of Kobe reveals that water bodies are predominantly located in forests, farmlands, and built-ups. Moreover, within built-up areas, the coexistence of water bodies and green spaces was sparse at the grid scale of the study, with the preponderance of water bodies concentrated in the forest and farmland regions. Interaction terms were introduced in the regression analysis to delve deeper into the combined effects of blue–green spaces. The descriptive statistics are presented in Table 7. The regression model is described in Eq. (2).

Table 7 Descriptive statistics (2)
Full size table
$$LST=\alpha +{\beta }_{1}\text{Water}+{\beta }_{2}\text{Forest}+{\beta }_{3}\text{Farmland}+{\beta }_{4}\text{Water}\times \text{Forest}$$
$$+{{\beta }_{5}\text{Water}*\text{Farmland}+\beta }_{6}\text{Forest}\times \text{Farmland}+\upvarepsilon .$$
(2)

Based on these findings, a certain warming effect was observed between water and forest and between farmland and forest (Table 8), which was contrary to previous studies that indicated that blue–green spaces have a certain synergistic cooling effect [38]. The reason for this difference lies in the scales used in the studies. On a larger scale, owing to the exceptionally high cooling effect of large-scale forests, the presence of water bodies encroaches on the feasibility of forest existence. Nevertheless, upon evaluating the magnitude of the regression coefficients, the combined warming effect exhibited a pronounced degree of weakness that was not notable. Additionally, there were no discernible combined effects of water and farmland (Table 6). This was attributed to the relatively minor disparity in the cooling effects between water and farmland, compared to the substantial cooling impact exerted by forests. Within a certain area, the replacement of water and farmland had little effect on the overall LST.

Table 8 Regression results of the combined effect of blue–green spaces
Full size table

There was indeed a certain synergistic cooling effect in blue–green spaces at the neighborhood scale. However, compared with the combined effect, when the scale was large enough, the independent cooling effect of the blue–green space dominated the LST. Specifically, the robust cooling effect of expansive forest regions resulted in subtle warming when combined with water.

The economic impact of the special production green space policy

LST variations are a critical determinant of the risk associated with heat-related health events. Research conducted in Philadelphia concerning heat events indicated that remote sensing-based LST had more predictive significance for mortality rates than socio-demographic information [47]. To thoroughly examine the economic impact of special production green spaces, this study assumes that after the expiration of the special productive green space policy, all existing special productive green spaces are converted to other types of land use. An analysis was conducted on the overall change in LST in Kobe and its effects on heatstroke incidence and energy consumption. Constrained by the scale and data availability of this study, previous research findings are utilized to analyze the overall variation in LST in Kobe and its impact on heatstroke incidence and energy consumption resulting from the special productive green space policy. O’Malley et al. conducted a quantitative analysis examining the relationship between LST in Tokyo from May to September and the incidence of heat-related illnesses [48]. A study in Beijing indicated that when the urban environmental temperature exceeded 26 °C, for each 1 ℃ rise in temperature, the city’s electricity demand increased by 3.97 × 108 W [49]. Using the findings of O’Malley et al. and Zhang et al., we calculated the changes in heatstroke cases for every 100,000 individuals and total energy consumption when converting special production green spaces to other land types. The changes in LST in Kobe after the transformation were obtained using the weighted average method based on the LST data in this study. Because the data in this study is based on LST in September, which are lower than average temperatures in summer, it may underestimate the occurrence of heatstroke. In addition, Zhang et al. used data based on air temperature for estimating energy consumption, whereas this study was based on LST, which is generally higher than air temperature. Therefore, it may overestimate the variations in energy consumption. Nonetheless, the estimations still intuitively demonstrate the economic impact of special productive green space policy to a certain extent.

The results of the policy impact analysis (Table 9) show that if special productive green spaces are converted into built-ups and bare land, the average LST in Kobe will increase. For every 100,000 people, heatstroke cases will increase by approximately 1–2 people. Electricity burden will increase by 1.27–2.62 × 106 W, and the increase of LST after conversion to built-ups will be 0.0034 °C higher than that of bare land. However, when special productive green spaces are converted into water bodies, grasslands, and forests, the mean LST will tend to decrease (0.0034 °C, 0.0007 °C, and 0.0068 °C, respectively). Heatstroke cases decrease by 1–2 people when converted into forests. The conversion of original special production green spaces into built-ups or forests has caused two extreme impacts. According to data from the Kobe City Hall, there were 786 cases of heatstroke in Kobe in 2022, with a population of approximately 1.51 million (Year 2020). Therefore, there were 52 cases of heatstroke per 100,000 people. If all built-ups for special production green space were avoided and converted to forest, there could be a reduction of 4 cases of heatstroke per 100,000 people (7.7%). Despite an extreme assumption, positive policy guidance can improve the well-being of residents. Moreover, after the conversion of built-ups into forests, the cooling effect is most significant, saving the most energy at 2.69 × 106 W. The positive impact after the conversion of built-ups into water bodies is larger than that into grasslands. This indicates that after the expiration of the special productive green space policy, the type of land that special production green spaces are converted into is crucial in determining the positive or negative effects on the urban heat environment and economy. In general, the conversion of special productive green space into forest can offset the negative impact of that into built-ups, while the conversion of special productive green space into water bodies can offset the negative impact of that into bare land.

Table 9 Results of the impact analysis of the special production green space policy
Full size table

Discussion

Effects of LULC and landscape pattern metrics on LST

Land-use types have discernible cooling or warming effects on the LST. The cooling effect of green space and water bodies observed in this study is larger than that observed in previous studies [50]. Numerous studies have explored the cooling effects of blue–green spaces at urban or neighborhood scales [51]. However, research on the combined effect of blue–green spaces and whether their synergy is influenced by spatial scale is limited. This study selected a typical port city with a humid subtropical climate as the target and examined the correlation between LULC, along with the landscape pattern metrics, and LST at various grid scales. Subsequently, the study analyzed the influence of forests, water bodies, and built-up areas on LST and further investigated the combined effect of water bodies and forests on regional LST at a certain grid scale.

This study demonstrated that built-up land significantly contributed to pronounced warming effects on LST. Green spaces such as forests, farmland, and grass had significant cooling effects on LST, with large forested areas exhibiting the most notable cooling effect. The cooling effects of water bodies varied with the grid scale, with smaller water bodies showing less significant cooling effects at larger scales but exerting a more noticeable influence on microclimates [52]. Water bodies and forests exhibited a weak combined warming effect. However, this warming effect was not notable overall, which emphasizes the combined cooling effect and microclimate improvement in the blue–green spaces at the microscopic scale. Instead, this effect functioned as a supplement. Moreover, this study showed that area and proportion emerged as the primary factors influencing the cooling effects of blue–green spaces on LST, while the correlation between other landscape metrics and LST was more complex and challenging to verify consistently. The fragmentation of built-up and farmland was negatively correlated with LST. This result suggests that complex mosaic landscape patterns can include small-scale blue–green spaces, and such landscapes have a certain cooling effect in the urban context. Land-use planning should consider such spatial patterns of individual land-cover categories, in addition to the area sizes of each land conversion category, to maximize the cooling effects of various landscapes.

Analysis of the mean LST around special production green spaces showed that such spaces had a cooling effect. The more they were distributed, the larger the proportion of their area, and the better their effect on the thermal environment. Owing to the sample size, this negative correlation only indicates to a certain extent that effective and reasonable land-use policies improve the thermal environment. Land-use policy plays a pivotal role in shaping land-use patterns, which are influenced by both biophysical and socioeconomic environments [53]. This results in land-use policies that exert direct or indirect influences, leading to cooling or warming effects on the urban thermal environment. Thus, the integration of land-use, environmental, and transportation policies [54] is beneficial for comprehensive effects. Encouraging the collaborative participation of relevant stakeholders and enthusiasm for internal collective action is extremely important for policy implementation [55].

Impact of grid scale on the study

The correlation between the landscape pattern metrics of LULC and LST exhibited scale-dependent variations [56]. The analysis was conducted on four grid scales and revealed no significant correlations between any of the landscape pattern metrics and LST at a grid scale of 1500 m. The regression analysis demonstrated a suboptimal fit. In contrast, robust correlations and regression results were obtained at the scale of 1200 m. Furthermore, the correlations exhibited variability across different scales. Owing to the differences in forest and water body areas, significant variations in the correlation between landscape pattern metrics and LST existed at the same grid scale. Despite the influence of landscape pattern metrics and grid scales on research outcomes, numerous studies have not identified a universally applicable grid scale for different study areas [57]. Thus, when studying the urban thermal environment, it is challenging to partition appropriate grid scales for different landscape features. Owing to factors such as climate and land-use scale specific to the study area, this uncertainty and challenge may persist in future research.

Implications for public policy of blue–green space management

The results of this study show that forests demonstrate the most effective cooling effect, with a greater concentration leading to a more pronounced cooling effect. Kobe, with its natural framework formed by the Rokko mountain ranges, adopts a predominantly public transportation-focused approach to construct a compact urban structure. Concentrated population and service facility usage have reduced infrastructure construction and operation costs, as well as energy consumption and emissions. However, the concentration of built-ups also leads to insufficient internal blue–green spaces, resulting in negative impacts on the urban heat environment. The results of land use conversion in this study also show that forest resources are exploited to a certain extent. Therefore, in the process of resource development, especially before the development of natural resources such as forests and farmland, ecological assessment should be conducted to alleviate their adverse ecological impacts and implement ecological compensation policies.

Policies promoting urban green spaces are crucial for achieving sustainable development goals in urbanized areas [58]. Kobe has a conservation area for special green spaces covering 2,622.3 Ha, with 104.25 Ha designated as productive green space, which plays a significant role in urban cooling. After the expiry of the special productive green space policy, the government should provide incentives to encourage individuals to retain or convert such green spaces into those with higher biodiversity, while minimizing their conversion into bare or built land. In addition, Land use policies should be integrated with environmental and transportation policies. The European Commission (EC) introduced the concept of Sustainable Urban Mobility Plans (SUMPs), focusing on citizen and stakeholder involvement, as well as the coordination of cross-sector policies (transport, land use, health, energy, etc.).

Kobe has implemented demand management measures centered around parks to facilitate the use of public transportation, reducing energy consumption and increasing the utilization rate of large parks. Large parks have higher vegetation diversity, leading to better cooling effects and increased utilization rates, which contribute to improving residents’ well-being. Currently, Kobe has 18 large parks with a total area of approximately 856 Ha. Irregularly shaped green spaces and water bodies have a greater effect in alleviating the urban heat island effect compared to regular ones [59]. Simultaneously, fragmented green spaces and water bodies help reduce surface temperatures [59]. Therefore, more complex-shaped green spaces and water bodies should be developed. Some small parks in Kobe, such as citizen parks and community parks, have a relatively low ratio of blue–green space. In situations where space is limited, more attention should be paid to plant configuration to enhance the quality of blue–green spaces. In Kobe, some riverbanks are of regular design style, which can be moderately transformed into natural riverbanks by planting aquatic plants, enriching the landscape configuration along the riverbanks, and facilitating the construction of ecological corridors. Overall, while compact housing generally saves energy, it exacerbates the risk of environmental overheating. Roof greening and green parking lots have a certain cooling effect on buildings and can also save land resources. The government should take measures to encourage and guide residents in constructing rooftop gardens. Additionally, idle spaces in residential areas and major public transportation routes can be fully utilized to create green spaces connected to adjacent “green space,” guiding the formation of residential environments surrounded by extensive low-rise housing and greenery. In conclusion, through the mediation of public transportation, a multi-center urban structure should be formed, organically combining “compactness” and “dispersal” to achieve the integration of multiple functions. Avoiding the high temperatures in cities has become a serious policy challenge. While public policy has been adopted through ad hoc fiscal spending, the possibility of effective design through integration of the public policy with traditional land-use policy could be examined.

Limitations

First, although using the RTE algorithm to invert surface temperature has high accuracy, owing to limitations related to the cloud cover and satellite revisit cycles, only one-scene images were selected in the summer of each year. This resulted in the omission of seasonal, monthly, and daily variations in LST trends. Future research should employ multi-sensor and multi-temporal remote-sensing imagery to furnish more precise time-series datasets for LST analysis [60]. Second, because of the absence of complete grids, the edges of the study area were removed from the grid analysis, potentially resulting in lack of comprehensive details regarding the edge area. Supplementing this with field studies will enrich the edge features of the research area in future research. Finally, because the combined effect of blue–green spaces on a large scale is relatively complicated, further microclimate research on blue–green spaces should be conducted in the future to examine the robustness of this study’s findings. Regarding the study of urban thermal environments, the correlation between the LST of water bodies and landscape pattern metrics at different grid scales needs to be studied further by combining remote-sensing images with other methods, which poses challenges for future research.

Conclusion

We studied the correlations among LULC, landscape pattern metrics, and LST during summer at different grid scales. Furthermore, we explored the combined effects of blue–green spaces on the LST. Kobe was selected as the study area, and Landsat 8 remote-sensing images from two years of cloudless or less cloudy summer days were used. Subsequently, we applied a high-precision RTE method to derive LST values and scrutinized the spatial distribution characteristics of the LST for each year within the study area.

The principal conclusions drawn from our study are as follows: (1) The spatial distribution of LST within the study area exhibited similarities between 2014 and 2019. However, the area with LST above 32 ℃ in 2019 showed a significant trend of expansion compared with that in 2014. The high-temperature area was distributed in the built-up area, whereas the low-temperature area was concentrated in the blue–green spaces, especially in the forest area. (2) LST of the different LULC types showed significant differences. The LST in 2014 was ranked as follows: built-up (36.97 °C), bare land (35.31 °C), farmland (30.93 °C), water bodies (30.81 °C), grasslands (30.43 °C), and forests (26.75 °C). In 2019, the LST of built-up areas, grasslands, farmlands, and forests increased by 1.07–2.4 °C, while that of bare land and water bodies was relatively stable without significant changes. The expansion of built-up areas and reduction in forest areas have led to the deterioration of the urban thermal environment. (3) The impact of landscape pattern metrics on LST varied with the grid scale, showing a highly significant correlation at the grid scale of 600 m, 900 m, and 1200 m, and it was well validated at a grid scale of 1200 m. However, the correlation was not significant at a grid scale of 1500 m. (4) The higher the concentration and area of the forests, the better the cooling effect. Regression analysis indicates that water bodies provide cooling benefits to the LST at a certain scale. At a sufficiently large scale, water bodies and forests exhibited weak combined warming effects; however, their regression coefficients were quite small, with no notable overall combined effect. (5) The Law on Productive Green Areas protects green spaces in built-ups and ensures cooling effects on the surrounding environment. Thus, policy-driven green space planning has a long-term and stable impact on the thermal environment.