Using station-pair to elucidate the cooling effect of vegetation in megacities Authors: Zixuan Tang, Jing Tian, Yongqiang Zhang, Junlong Zhang, Xuanze Zhang, Ning Ma Published: 2026 ABSTRACT AbstractThe air urban heat island (UHI) poses significant challenges to urban environments. While vegetation is a recognized mitigation strategy, the in-situ cooling effects of different vegetation types, such as forests and grasslands, across diverse climatic regions remain poorly quantified. This study addresses this gap using a station-pair regression approach to evaluate vegetation’s cooling efficacy in three major Chinese cities: Beijing, Shanghai, and the Mega-city Cluster in the Pearl River Delta (MCPRD). We developed multiple linear regression models using summer data from 2016 to 2020. These models predict the air temperature difference between paired meteorological sites based on spatial differences in vegetation conditions, derived from high-resolution Sentinel-2 imagery, distance to large water body, wind speed, and elevation. The model performed well, achieving coefficient of determination of 0.71, 0.58 and 0.82 for Beijing, Shanghai and MCPRD, respectively. Crucially, our results show that compared to completely urbanized land, existing vegetation significantly lowers air temperature by 2.43 °C in Beijing, 0.58 °C in Shanghai, and 2.74 °C in the MCPRD. This research quantifies the variable cooling benefits of urban vegetation across different climate zones, providing an empirical basis for optimizing green infrastructure in urban planning to enhance thermal comfort.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-55645-z.Keywords: Sentinel-2, Urban heat island, Vegetation cooling effect, Air temperatureSubject terms: Urban ecology, Forest ecology Introduction An urban heat island (UHI) is an urban or metropolitan area that is significantly warmer than its surrounding rural areas due to human activities1. Elevated temperatures from heat islands affect a community’s environment and quality of life in multiple ways, including increased energy consumption2,3, elevated emissions of air pollutants and greenhouse gases, compromised human health and comfort4, impaired water quality and secondary impacts on weather and climate5. As one of the most pronounced surface climate changes caused by human activities6, UHIs are usually expressed as a difference in air temperature or surface temperature between urban areas and surrounding rural areas7–9. In recent decades, China has experienced the most rapid urbanization. Severe UHI intensity in China, especially in megacities with developed economies and concentrated populations such as Beijing, Shanghai and Guangzhou, was observed10–12. China is facing serious challenges in terms of mitigating the UHI effect. Urban vegetation is recognized as a crucial approach for mitigating UHIs due to its cooling effect3,13–16. Vegetation reduces a large amount of solar radiation through light absorption and reflection. Generally, only 10% of visible light and 30% of infrared radiation can penetrate trees to reach the Earth’s surface17. Vegetation also plays an important role in regulating the thermal environment via transpiration18–22. Forest, grassland and cropland are the main landscapes with significant cooling effects23–27. Researches on UHI are divided into two main categories, air UHI and surface UHI, which characterize UHI by air temperature and land surface temperature (LST), respectively1,28–30. The cooling effect of vegetation on LST has garnered significant attention, With the emergence of large-scale LST datasets31,32, the cooling effect of vegetation on LST has attracted great attention and has been widely studied33–38. However, LST amplifies UHI more than air temperature, particularly during summer39,40. Air temperature directly stems from land‒atmosphere energy exchanges and closely aligns with the human experience of climate. Therefore, it is more important to study the cooling effect of vegetation on air temperature for human comfort, and all references to UHI in the following are to air UHI. Alkama and Cescatti41 discovered that deforestation elevates air temperature by approximately 1 °C in temperate and tropical zones, and over 2 °C in arid zones. Zeng et al.42 found that since 1982, greening has mitigated the rise in global air temperature by 0.09 ± 0.02 °C, according to global climate models. In Wisconsin, Ziter et al.43 demonstrated that canopy covers exceeding 75% could lower temperatures by more than 2.5 °C. Skelhorn44 showed, using the ENVI-met model, that in Manchester, UK, a 5% increase in adult deciduous forest coverage could reduce air temperature by 1 °C. Further, Duarte et al.45, Srivanit and Hokao46, and Herath et al.47 utilized the ENVI-met model to achieve significant temperature reductions in São Paulo, Brazil (0.3–1.5 °C), at Saga University in Japan (2.27 °C), and in the Colombo Metropolitan Region, Sri Lanka (1.87 °C), respectively. Zhao et al.48 reported that trees have reduced air temperature by 1.5 ± 1.3 °C in global tree-covered urban areas from 2000 to 2015, according to ERA5-Land reanalysis. Additionally, Cao et al.49 noted that vegetation greening led to a cooling effect of -0.08 °C on air temperature from 2001 to 2018, using China Meteorological Forcing Dataset50. However, the observation of air temperature is frequently limited to the confines of individual cities, resulting in a constrained spatial coverage. Large-scale air temperature analyses usually rely on reanalysis products, models, and semi-empirical formulas derived from LSTs, which are not validated and introduces significant uncertainties. Expanding air temperature observations at larger scales could improve our understanding of vegetation’s cooling effects across different climates. Despite extensive research on the cooling effect of vegetation in mitigating UHI affecting air temperatures, the reliability of temperature data and the scarcity of observations present significant challenges. The cooling effect of different types of vegetation on UHI in cities with different climates based on in situ observed air temperatures remains unclear. Sentinel-2 satellite imagery, with its 10 m resolution, offers an unprecedented opportunity to detail urban land characteristics more precisely. Concurrently, the comprehensive air temperature records maintained by the China Meteorological Administration (CMA) furnish a robust dataset for such studies. Yet, the utility of these resources in assessing UHIs has not been thoroughly examined. Based on the above knowledge gaps, this paper proposed a station-pair regression method to quantify the cooling effect of different vegetation types (forest and grassland) in the three megacities in China with Sentinel-2 high-resolution remote sensing data. While inspired by the logic of paired-catchment studies used in hydrology to control for background variability51, our approach applies this concept in a cross-sectional spatial context. The UHI during the summer period is the focus because of its larger intensity and magnitude at that time. We use 10 m-resolution fractional vegetation cover (FVC) retrieved from Sentinel-2 to portray the vegetation abundance distribution. Air temperature observed at meteorological stations is used to quantify UHIs. The multiple linear regression model based on station-pair regression method is built according to the difference in air temperature (dependent variable), elevation, distance to large water body, wind speed, and abundance of forest and grassland (independent variables) among meteorological stations. Based on these results, the cooling effects of forests and grasslands in different cities were quantified. Materials Study area In this study, we select Beijing, Shanghai and the Mega-city Cluster in the Pearl River Delta (MCPRD) surrounding Guangzhou as the study area, which are the most developed cities in China. MCPRD includes Foshan, Dongguan, Shenzhen and Guangzhou. According to the Köppen–Geiger climate classification52, Beijing has a cold climate (classification code Dwa). Shanghai and MCPRD are temperate climates (classification code Cfa and Cwa, respectively). They are all located in monsoon-influenced regions, with high air temperatures during summer. The summer period is from June to August in Beijing, from June to September in Shanghai and from April to October in the MCPRD (Selection of months with a multi-year average monthly temperature more than 22 °C according to https://www.climatestotravel.com/). The areas of Beijing, Shanghai and MCPRD are 16,411 km2, 6,431 km2 and 15,518 km2, respectively. Beijing has an average altitude of 368.0 m. Mountains are mainly distributed in the west and the north of the city. Cropland and forest are the main vegetation types. Shanghai has an average elevation of 4.5 m. The vegetation type is dominated by cropland, accounting for 1/2 of the area. The average elevation of the MCPRD is 73.1 m. Forest is widely distributed around the city. The areas of forest and cropland account for 1/3 of the city. Figure 1 shows the location, land cover, FVC and altitude of the three regions. The maps in Fig. 1 were created and visualized using ArcGIS 10.2 software (https://resources.arcgis.com/en/help/install-guides/arcgis-desktop/10.2/). Figure 1 (a) shows the boundaries of China, Beijing, Shanghai, and the MCPRD. Figure 1 (b)-(d) show the results of cropping the Land Cover, FVC, and Elevation layers using the boundaries of Beijing, Shanghai, and the MCPRD, respectively. Altitude data are obtained from Shuttle Radar Topography Mission (SRTM) Digital Elevation Data Version 4 which are downloaded from Google Earth Engine (GEE). Basic information of the study area. (a). Locations of the study areas; (b). Land cover classification; (c). Fractional vegetation cover and meteorological stations; (d). Elevation. Meteorological data The observed daily average, maximum, minimum air temperature, and average wind speed during summer from 2016 to 2020 was collected from 20 meteorological stations in Beijing, 11 stations in Shanghai and 10 stations in the MCPRD administrative area. The elevations of these meteorological stations are also obtained from in situ observations. The elevations range from 19.8 ~ 1224.7 m, 4 ~ 8.1 m and 12.2 ~ 70.7 m in Beijing, Shanghai and the MCPRD, respectively. Figure 1 (c) shows the spatial distribution of meteorological stations. Sentinel-2 TOA reflectance In this study, Sentinel-2 satellite data are used to retrieve FVC. It provides 13 spectral band data and is available at the Copernicus Open Access Hub, ESA (https://scihub.copernicus.eu/). The top-of-atmosphere reflectance data (TOA) have been available since November 2015 and bottom-of-atmosphere (BOA) reflectance data have been available since January 2017 (November 2018 in China). To obtain BOA reflectance covering the study period (2016–2020), atmospheric corrections for the Sentinel-2 TOA reflectance were performed by means of the Py6S model on the GEE platform. We also use built-in cloud quality control for declouding operations. Cloud-free Sentinel-2 BOA reflectance data of 159, 206 and 354 days during the summer period from 2016 to 2020 were finally used for Beijing, Shanghai and the MCPRD, respectively. Land cover To quantify the cooling effect of forest and grassland separately, a 30 m annual China land cover dataset (CLCD) from 2016 to 2020 was used in this study, which is freely available at 10.5281/zenodo.441781053. In the CLCD, land cover is classified into nine classes, including cropland, forest, shrub, grassland, water, snow or ice, barren, impervious surface and wetland. Because mapping errors in cropland in land cover dataset may originate from the spectral similarity between cropland and grassland, which has been reported in previous studies54–56, 57. After comparison with the Sentinel 2 true-color imagery, we also found that CLCD classified large grasslands in parks in the study area as cropland. So, cropland and grassland were combined as grassland in the study. Because of the low ratio of shrubland cover in the study area, the cooling effect of shrubland was not analyzed and was ignored. Figure 1a shows the land cover classifications in the three study areas. Based on land cover data, this study extracts water bodies with an area greater than 1 km2 within a 10 km buffer zone near each meteorological station as large water bodies, and then extracts the distance between each meteorological station and the large water bodies. Methods Station-pair regression In the study, we proposed a station-pair regression method to quantify the cooling effect of forests and grasslands in mitigating UHI. Conceptually, this method refers to the logic of the paired catchment method which has been widely used in hydrological studies58–60. However, unlike traditional paired-catchment designs that rely on temporal comparisons (before and after intervention), our approach utilizes spatial differencing between concurrently monitored stations. Here, station-pair regression assumes that in two paired meteorological sites, similar in climate, the correlation between their air temperature will remain the same if vegetation conditions and elevation remain the same. Suppose vegetation conditions and elevation in one meteorological station are changed and the second meteorological station is unchanged. This study assumes a linear relationship between vegetation cover and cooling effect. The difference in their air temperatures () would be mainly controlled by the differences in vegetation conditions, distance to large water body, wind speed, and elevation. Here we use forests abundance and grassland abundance to represent vegetation conditions, then could be quantified as the following: where , , , , , and are the differences in air temperature, forest abundance, grassland abundance, elevation, distance to large water body, and wind speed between the two meteorological stations. , , , , , and are regression coefficients. is the residual. Because the vegetation conditions and elevation are different, Eq. (1) could be used to estimate air temperature difference due to vegetation difference and elevation difference. , , , , , and required in the Eq. (1) are obtained using observations as follows. Taking Beijing as an example, equals the air temperature at the ith station (i is in a loop from the 1st to the 20th station) minus the air temperature at other 19 stations. In this way, all the 19 pairs can be compared and used for regression. and are quantified by the vegetation abundance difference averaged from 5 km radius buffer area around each pair of two meteorological stations (detailed in Sect. 3.2). The reason for choosing 5 km in this study is as follows: Firstly, parks and other green spaces have a significant cooling effect within a small radius (usually a few hundred meters)15, but this does not mean that green spaces have no impact beyond this range. Their cooling effect can affect the entire city through atmospheric turbulence. Secondly, this study aims to explore the cooling effect of vegetation on the entire city, rather than its local cooling effect. When studying UHI, some literature used a 2 km buffer zone radius to distinguish between urban and non-urban sites61,62. To fully cover the impact of vegetation around the site, this study chose a larger radius of 5 km to explore the cooling effect of vegetation. , , and is quantified from elevation difference (detailed in Sect. 2.2, 2.4, and 2.2, respectively) from each pair of two meteorological stations. To avoid duplication, when station A is paired with station B, only is used for regression and would be excluded. To further minimize the impact of cloud contamination, image samples with less than 99% (90% in MCRPD due to the high frequency of cloud days) of valid pixels in the buffer will be excluded. In the end, Beijing obtained 541 valid samples covering 87 days; Shanghai obtained 82 valid samples covering 35 days; and MCPRD obtained 265 valid samples covering 61 days. Only valid samples from different sites on the same day can be considered paired and used in the regression, and the final number of paired samples for Beijing, Shanghai, and MCPRD were 1799, 70, and 527, respectively (in Fig. 2). This study attempted to add the proportion of impermeable surfaces (representing the degree of urbanization) as an independent variable to the regression equation, but it showed strong multicollinearity with vegetation abundance (see in Table S4 and S5). Therefore, this study excluded this variable. Correlations between the response variables (, , , , and ) and explanatory variable (). Vegetation abundance is divided into forest abundance and grassland abundance, which are defined as the ratio of the forest/grassland pixels to the total pixels in the 5 km radius buffer area multiplied by the average FVC in the 5 km radius buffer area, as shown in Eq. (2) and Eq. (3). where For and Gra are forest abundance and grassland abundance, respectively. and are the pixel numbers of forest and grassland, respectively. N is the total pixels in the 5 km radius area. and are the mean values of FVC on forest pixels and grassland pixels in the buffer area, respectively. To evaluate the collinearity between the three explanatory variables in Eq. (1) and avoid the effects of collinearity on the method, the variance inflation factor (VIF) is used here to measure the collinearity and independence between the three explanatory variables, which is calculated as follows: where is the VIF of the explanatory variable and is the determined coefficient of the explanatory variable regressed on the other explanatory variables. A large VIF indicates severe collinearity. A rule of thumb is that if VIF > 10, then collinearity is high63(a cutoff of 5 is also commonly used64. Quantification of the cooling effect of forests and grasslands The partial regression coefficients in Eq. (1) indicate that the degree of increases or decreases with each unit change in the explanatory variable. Therefore, the cooling effect of forest and grassland can be quantified as follows: where and are the cooling effects of forest and grassland, respectively; and are the mean values of For and Gra in the summer of 2016–2020, respectively. To illustrate the change in the cooling effect due to the change of 10% in FVC or the area ratio of vegetation, and are calculated as: where and are the cooling effect enhancement by increasing the 10% FVC and 10% area ratio of vegetation, respectively; is the partial regression coefficient of forest or grassland (or ); and and are the average area ratio of vegetation and FVC value of vegetation, respectively. Calculation of fractional vegetation cover based on Sentinel-2 BOA reflectance We calculate the NDVI from Sentinel-2 BOA reflectance by: where and are the BOA reflectances in the near-infrared band (773 ~ 907 nm) and red band (646 ~ 684 nm), respectively. Based on NDVI, a dimidiate pixel model was used to calculate FVC65,66: where and are the NDVI values of the bare soil pixel and fully covered vegetation pixel, respectively. In the study, and were set as the mean values of the pixels that are less than 1% quantile and more than 99% quantile of the NDVI, respectively. Results Correlation analysis and multiple regression model building Figure 2 shows the correlations between the response variable () and explanatory variables (, , , , and ). There is a significant (the significance level is set to 0.01) negative correlation (R=-0.59, p < 0.01) between and in Beijing, which is consistent with the vertical thermal gradient law. However, a weak correlation (|R| ≤ 0.3) for MCPRD (R=-0.21, p < 0.01) and even a positive correlation for Shanghai (R = 0.13, p = 0.14) are obtained. The reason for these incorrect results is because of the small differences in elevation in the two regions (Fig. 1). and show significant negative correlations in Beijing (R=-0.56, p < 0.01), MCPRD (R=-0.68, p < 0.01) and Shanghai (R=-0.33, p < 0.01). Similarly, and also show significant negative correlations in Beijing (R=-0.44, p < 0.01), MCPRD (R=-0.49, p < 0.01) and Shanghai (R=-0.68, p < 0.01). and show a negative correlation in all three cities; and show weak positive correlation in Beijing and MCPRD, and significant negative correlation in Shanghai. The above illustrates that when the forest abundance or the grassland abundance increases, the air temperature decreases. The simulated by Eq. (1) and the observed are shown in Fig. 3. The entire dataset is given in Fig. 3 (a)-(c) and the results of ten-fold cross validation are given in Fig. 3 (d)-(f). Adjusted R2 equal to 0.71, 0.58 and 0.82 and RMSE is equal to 0.987 °C, 0.587 °C and 0.8 °C for Beijing, Shanghai and MCPRD, respectively. This indicates that Eq. (1) performs well in the air temperature difference analysis and can be applied to quantify the cooling effect of forests and grasslands. The model also presented similar statistics in ten-fold cross validation, with adjusted R2 values of 0.63, 0.58 and 0.52, and RMSE values of 1.125 °C, 0.631 °C and 1.342 °C for Beijing, Shanghai and MCPRD, respectively. The performances on the entire dataset and the cross-validation dataset are very close, indicating that the multiple linear model performs consistently and robustly in the three cities. Density scatterplots of the simulated against the observed . (a)-(c). Regression results for the entire dataset; (d)-(f). Regression results for 10-fold cross validation. Table 1 shows the partial regression coefficients for the three cities. All coefficients are significant at the 0.001 level (2-tailed), indicating the high precision and low variance of the parameter estimation. The proportion of forest in Shanghai is very small (approximately 0.001, see Fig. 1), which may lead to large uncertainty. Therefore, we ignore the cooling effect of the forest in Shanghai and do not consider it when building Eq. (1). The partial regression coefficients of are − 4.01 × 10− 3 °C/m and − 2.86 × 10− 3 °C/m in Beijing and the MCPRD, respectively, which is consistent with the vertical thermal gradient law. However, the partial regression coefficients of have a value of -1.54 × 10− 2 in Shanghai. The slopes of are − 3.06 and − 4.62 in Beijing and the MCPRD, respectively. This suggests the significant cooling effect of forests in the two cities. The regression coefficients of are − 4.92, -1.61 and − 8.16 in Beijing, Shanghai and the MCPRD, respectively, which also reflect the significant cooling effect of grassland. Table 1Partial regression coefficients for Beijing, Shanghai and MCPRD.CityΔForΔGraΔEle (°C/m)ΔDistance (°C/m)ΔU2 (°C/m/s)Beijing-3.06***-4.92***-4.01e-03***-3.24e-060.347***Shanghai——-1.61***-1.54e-02-8.01e-06-0.266*MCPRD-4.62***-8.16***-2.86e-03-1.49e-05***0.246****, **, and *** denote that coefficient estimation is significant at the 0.05, 0.01, and 0.001 level (2-tailed), respectively. Partial regression coefficients for Beijing, Shanghai and MCPRD. *, **, and *** denote that coefficient estimation is significant at the 0.05, 0.01, and 0.001 level (2-tailed), respectively. The VIF values, representing collinearity, among , , , , and are given in Table 2. The VIFs are 2.58, 1.43, 1.98, 1.73, and 1.10 for , , , , and in Beijing, respectively. The VIFs in Shanghai are 2.17, 2.70, 2.27, and 2.06 for , , , and , respectively. The VIFs in MCPRD are 1.79, 1.32, 1.86, 2.01, and 1.07 for , , , , and , respectively. The higher VIFs in Beijing relative to Shanghai and MCPRD are because the higher the elevation is, the larger the forest cover. Nevertheless, the VIFs in the three cities are far lower than the threshold of 10 or 5, representing low collinearity. This indicates that the three explanatory variables are completely independent of each other, which means that the partial regression coefficients can be used to explain the effects of the change in , , , , and on . Table 2VIF for the Beijing, Shanghai and MCPRD city clusters.CityΔForΔGraΔEleΔDistanceΔU2Beijing2.581.431.981.731.10Shanghai——2.172.702.272.06MCPRD1.791.321.862.011.07 VIF for the Beijing, Shanghai and MCPRD city clusters. The cooling effect of vegetation in the three cities The cooling effect of vegetation is closely related to the area and the density of vegetation in the city. As shown in Table 3, there are large variabilities in the FVC and area ratio of forest and grassland among the three regions. Beijing has the highest ratio (0.487) of forest area to FVC (0.942 ± 0.088), suggesting abundant forest resources. MCPRD also has a high forest area ratio of 0.334 and forest FVC of 0.854 ± 0.136. Although Shanghai has little forest cover, more than half of the area is covered by grassland (0.521) and the FVC of grassland is 0.687 ± 0.203. The percentage of grassland area ratio in the MCPRD is 0.306 and the FVC of grassland is 0.567 ± 0.215. Beijing has a grassland area ratio of 0.280 and has the highest FVC (0.745 ± 0.166) of grassland. This indicates that vegetation in Beijing is denser than that in Shanghai and the MCPRD in terms of both forest and grassland. Table 3The multiyear average value of FVC and ratio of forest and grassland in the three cities.CityFVC of forestFVC of grasslandRatio of forestRatio of grasslandBeijing0.942 ± 0.0880.745 ± 0.1660.4870.280MCPRD0.854 ± 0.1360.567 ± 0.2150.3340.306Shanghai——0.687 ± 0.203——0.521The number before ± is the mean value and the number following ± is one standard deviation of FVC. The multiyear average value of FVC and ratio of forest and grassland in the three cities. The number before ± is the mean value and the number following ± is one standard deviation of FVC. Figures 4 and 5 illustrate the multiyear average cooling effect of vegetation in Beijing, Shanghai and the MCPRD in the summer from 2016 to 2020. The cooling effects of forests are 1.40 ± 0.13 °C and 1.32 ± 0.21 °C in Beijing and the MCPRD, respectively. Grassland was 1.03 ± 0.23 °C, 0.58 ± 0.17 °C and 1.42 ± 0.54 °C in Beijing, Shanghai and the MCPRD, respectively. These results illustrate that the cooling effect of vegetation is the largest in the MCPRD, followed by Beijing and the least is Shanghai. The vegetation cooling magnitude in the MCPRD also shows the greatest variability. In Beijing, the cooling effect of the forest is greater than that of grassland, but this phenomenon contrasts with MCPRD. Nevertheless, the cooling effect of forests in the MCPRD and Beijing is comparable to that of grassland. Multiyear average cooling effect of forest and grassland in Beijing, Shanghai and the MCPRD in the summers of 2016 to 2020. Time series of the cooling effect, FVC and area ratio of forest and grassland in Beijing, Shanghai and the MCPRD in the summers of 2016 to 2020. (a)-(b). The area ratio of forest and grassland; (c)-(d). FVC; (e)-(f). cooling effect of vegetation. As shown in Fig. 5, in the summer of 2016 to 2020, the area ratio of forest and the area ratio of grassland were almost constant in the three regions. Comparatively, FVC has larger changes. The average FVC of forest ranges from 0.918 ~ 0.959 and from 0.814 ~ 0.881 in Beijing and the MCPRD, respectively. The mean FVC of grassland ranges from 0.719 ~ 0.771, 0.548 ~ 0.578 and 0.661 ~ 0.733 in Beijing, MCPRD and Shanghai, respectively. The average annual cooling effect of forests ranges from 1.38 ~ 1.43 °C and 1.27 ~ 1.36 °C in Beijing and the MCPRD, respectively. The cooling effect of grassland is 1.00 ~ 1.06 °C, 1.38 ~ 1.44 °C and 0.56 ~ 0.61 °C for Beijing, MCPRD and Shanghai, respectively. These indicate that the cooling effect of vegetation in summer has a small range of variation and is relatively stable over the five years. In practice, it is important to know how much the air temperature will change with the change in vegetation. Based on the established model, we further estimated the air temperature change with the change in the vegetation area and the FVC by using Eq. (7) and Eq. (8). Table 4 displays the results of the air temperature change with a 10% increase in the vegetation area ratio and a 10% increase in FVC. For forests, air temperature decreases of 0.15 °C and 0.15 °C in terms of a 10% increase in FVC and decreases of 0.29 °C and 0.39 °C in terms of a 10% increase in area ratio are obtained for Beijing and the MCPRD, respectively. For grassland, air temperature would decrease by 0.14 °C, 0.25 °C and 0.08 °C if FVC increases by 10% for Beijing, MCPRD and Shanghai, respectively. It would have decreased by 0.37 °C, 0.46 °C and 0.11 °C if there was a 10% increase in the area ratio for the three cities. Table 4The cooling effect induced by changes in vegetation abundance.CityForestGrassland10%FVC10%Area ratio10%FVC10%Area ratioBeijing0.130.260.140.38MCPRD0.170.420.280.51Shanghai————0.120.16 The cooling effect induced by changes in vegetation abundance. Discussion Comparison of the cooling effect of vegetation with other studies Compared with previous studies, the novelty of this study lies in the following: firstly, the cooling effect of vegetation on LST has been widely studied due to the global coverage, high spatial resolution and low cost of remote sensing data. Based on Landsat-8 satellite data, Khan and Li67 showed that the cooling effect on LST of green space in Beijing was 5.01 °C. Satellite images also show that forest and other vegetation cooling LST by 3.42 °C and 1.96 °C in Shanghai, respectively68. Zhou noted that vegetation in Guangzhou cooled daytime LST by over 4 °C69. But air temperature is most relevant to public health, yet the cooling effect of vegetation on air temperature has not received enough attention. This study focuses on the cooling effect of vegetation on air temperature to quantify the improvement in human comfort due to the cooling effect of vegetation. LST amplifies UHI more than air temperature has been proved39,40. Our results show that vegetation reduced the air temperature by an average of 2.43 °C, 0.58 °C and 2.74 °C in Beijing, Shanghai and the MCPRD, respectively. The results based on observational data reveal that the cooling effect of vegetation on air temperature is lower than that on LST. This also confirms that studies of LST may amplify the cooling effect of vegetation. Secondly, different from conventional studies which mainly focused on the cooling effects of urban parks, this study examined the cooling effect of not only urban parks but also other vegetation across the whole city were considered. Regarding the magnitude of the vegetation cooling effect, compared to completely urbanized land, our study yielded 2.43 °C, 0.58 °C and 2.74 °C in Beijing, Shanghai and the MCPRD, respectively. According to some previous studies, greening squares could decrease temperature by 0.3 ~ 3.3 °C (average of 1.9 °C)70 and green areas could reduce summer air temperature by 0.2 ~ 2.0 °C71 in Beijing. The Olympic Park of Beijing could cool air by 0.48 ~ 1.12 °C on average72. Urban green spaces could reduce the physiological equivalent temperature by approximately 2 °C in Beijing73. Based on Landsat-8 Thermal Infrared Sensor images on 20 July 2016 and 16 August 2020, grassland had cooling effects of 0.34 °C74 and 1.96 °C68 compared to the mean LST in Shanghai. The local cooling effect of pocket green spaces could reach 1.2 °C on average in Shanghai75. A study of 10 parks in Guangzhou showed a cooling effect of 1.82 ~ 4.62 °C76. Vertical green facades could reduce the wet bulb globe temperature up to 2.7 °C in Guangzhou77. Above all, the results obtained in this study are comparable with these existing studies. The cooling effect of urban vegetation is influenced by various factors. First, it is related to the area of green spaces44,78–80. In this study, the vegetation areas of Beijing, Shanghai and MCPRD are 12,587 km2, 3351 km2 and 9932 km2, respectively and the area ratios are 76.7%, 52.1% and 64%, respectively. Second, the characteristics of green space affect the cooling effect80–82, such as green space shape83–87 and the degree of patch aggregation81,82. Third, the density of vegetation is also a fundamental factor affecting the cooling effect. Some studies have noted that the vegetation cooling effect usually increases with increasing leaf area index (LAI)88,89. In this study, we also obtained similar results: the higher the FVC was, the stronger the cooling effect. Fourth, vegetation type has an obvious influence on the cooling effect23,90. Urban forests are the most effective type in mitigating UHIs because of their provided shade91. Irrigated grassland also plays an important role in controlling the cooling effect72,92. According to the results in this study, the abundance of vegetation is the largest in Beijing, followed by MCPRD and Shanghai. However, the cooling effect of vegetation is the largest in the MCPRD, followed by Beijing and Shanghai. Fifthly, this study assumes a linear relationship between vegetation cover and cooling effect, while some studies reported non-linear vegetation cooling effects43,48. The local background climate and the above mentioned five factors together may lead to this result. Cooling effects of different vegetation types in different climatic contexts It is known that the local background climate has a strong contribution to UHIs8. Namely, there is spatial variability in the cooling effect of vegetation due to differences in latitude, climate, environment and vegetation conditions32,93,94. Even in the same city, different vegetation abundances could lead to different levels of UHI mitigation. Vegetation’s cooling effect on air temperature varies significantly with the background climate. Su et al.95 highlighted that this cooling is more pronounced in warm seasons within low latitude zones, particularly during the leaf growth stage. Ibsen et al.96 further observed that in eight American cities, the cooling effect on nighttime air temperature increases with aridity. Conversely, Peng et al.37 noted that in dry regions, nighttime warming often counteracts the daytime cooling provided by vegetation. When examining biophysical effects, Li et al.97 found that tropical forests exert a strong cooling influence throughout the year, temperate forests contribute to net annual cooling, and boreal forests result in net annual warming. In a specific study of California’s arid regions, Shiflett et al.31 discovered that mid-day cooling efficiency of vegetation significantly increases in the most arid areas. Comparing two cities, Zhou et al.32 found trees to be more efficient in cooling in Baltimore than in the hotter and drier Sacramento. Additionally, a study by Schwaab et al.98 across European cities determined that trees’ cooling effect strengthens with increasing latitude. Due to observational constraints, previous studies were limited to a single city and were unable to analyze differences in the cooling effect of vegetation in different climatic contexts; for larger-scale studies, temperatures are usually calculated based on reanalysis data or empirical formulas, which are unvalidated and subject to large uncertainties. This paper elucidates the cooling effects of vegetation in megacities located in cold with dry winter and hot summer climate (Beijing), temperate with dry winter and hot summer climate (Shanghai), and temperate without dry season and with hot summer climate (MCPRD), respectively, based on observational data. Forests and grasslands exhibit different cooling effects. The cooling impacts of green spaces significantly depend on the type of vegetation and the density of their canopies, as outlined by Xiao et al.80. This underlines the importance of exploring the cooling effects of various plant species. Under the same climatic conditions, since there is no significant difference in annual evaporation between forests and grasslands99, the cooling mechanisms of the two types of vegetation differ. The beneficial influence of forests on temperature reduction has been extensively documented33,37,38,41,43,48,97,98. The cooling mechanism of forests comes from the shading of solar radiation by tall canopies and their own transpiration. Similarly, grasslands play a crucial role as greening landscapes, contributing significantly to cooling, especially as grasslands are often irrigated, which increases the available water for evaporation, thereby enhancing the cooling effect100,101. Yu et al.101 developed a framework demonstrating that enhanced rainfall and irrigation amplify grasslands’ cooling effects. Jiang et al.102 observed that increased grassland coverage leads to a net annual cooling effect in arid regions. Moreover, Shen et al.103 found that grassland greening markedly lowers the annual mean LST for areas below the 50°N latitude from 2003 to 2017. Zhu et al.104 reported that grassland restoration typically results in significant cooling, with temperatures dropping more than − 0.01 K in China’s arid and semi-arid regions. Amani-Beni et al.72 demonstrated in a park study that adequate irrigation enhances the cooling effects of grasslands. We analyzed cooling effect of vegetation to the daily maximum (ΔTmax) and minimum (ΔTmin) temperatures. The simulated ΔTmax and ΔTmin, as well as the observed ΔTmax and ΔTmin, are presented in Fig S2 and Fig S3. Tables S2 and S3 show the partial regression coefficients. It can be observed that the partial regression coefficients are sensitive to time. When the partial regression coefficients of ΔFor and ΔGra are smaller, it indicates that the cooling effect of vegetation is more pronounced. Vegetation has the strongest cooling effect on daily minimum temperature, followed by daily average temperature, and finally daily maximum temperature. This is consistent with the fact that vegetation has the most pronounced cooling effect at night43. Uncertainty, limitations and future work There are some limitations and uncertainties in this study. Due to data quality issues (e.g., chance errors in observation data, remote sensing data error from atmospheric correction) and systematic errors in building Eq. (1), there may exist some uncertainties in the quantification of the vegetation cooling effect in the study. A limitation of this study is that the influence of the shape characteristics of vegetation and patch aggregation was not considered, which may lead to some uncertainties in the cooling effect estimation. Finally, it is important to acknowledge the inferential limitations of our study design. Unlike traditional paired-catchment experiments that utilize temporal baselines (pre- and post-treatment) to establish causality, our station-pair regression is a cross-sectional analysis. Therefore, the relationships identified between vegetation cover and temperature differences represent correlational associations rather than strictly proven causal effects. Future studies incorporating temporal replication or intervention monitoring would be required to strengthen causal inference. According to the analysis results of different radius, when the radius is less than 5 km, the performance of the regression equation decreases in Shanghai and MCPRD (see in Table S1, Fig S1, S2, and S3), which will reduce the explanatory power of the model. Indeed, different buffer radius choices can lead to uncertainty in the results, but considering the above reasons described in Sect. 3.1 comprehensively, this study chose a radius of 5 km. In the analysis results of different radius, it can be found that the partial regression coefficients of and have changed. In Beijing, when the buffer zone radius is changed to 2000 m, 1000 m, and 500 m, the partial regression coefficients of are − 0.48, 0.17, and − 0.92, respectively, and the partial regression coefficients of are − 4.32, -1.82, and − 1.94, respectively. In MCPRD, the performance of the model sharply decreases when the analysis radius is less than 5000 m. The partial regression coefficients of are − 2.51, -3.18, -4.82, -2.69, and 0.28, respectively, and the partial regression coefficients of are − 3.58, -3.37, -0.41, 1.78, and 2.11, respectively in MCPRD. In Shanghai, when the radius decreases, the performance of the model decreases and lacks sufficient interpretability. According to this study, some suggestions could be given in future city planning. FVC in Beijing is already close to saturation. In contrast, Shanghai and MCPRD can probably increase the cooling effect of vegetation by increasing FVC. Shanghai can also regulate the thermal environment by increasing the forest area due to its small existing forest area. In addition to forest expansion, increasing grassland cover also serves as a viable strategy for Shanghai to regulate its thermal environment, given the cooling effects observed in our analysis. In future work, we will continue to study the cooling effect of vegetation in more cities and to investigate its spatial pattern across China. Meanwhile, we expect to further clarify the reasons that result in the difference in the vegetation cooling effect in different regions. Conclusions Based on the observed air temperature at meteorological stations and the vegetation fractional cover retrieved from high-resolution Sentinel-2 remote sensing data, this study analyzed the cooling effect of vegetation in the three megacities of China (Beijing, Shanghai and MCPRD) in the summer of 2016 to 2020. A station-pair regression approach was proposed to analyze the relationship between spatial differences in air temperature and vegetation abundance and to quantify the cooling effect of forest and grassland. The results indicate that (1) there are significant negative correlations between air temperature and grassland abundance in Beijing, the MCPRD and Shanghai. A negative correlation between air temperature and forest abundance was also obvious in Beijing and the MCPRD. (2) Differences in grassland abundance, forest abundance and elevation explained 71%, 82% and 58% of the variation in the spatial distribution of air temperature in Beijing, the MCPRD and Shanghai, respectively. (3) Compared to completely urbanized land, forest reduced the air temperature by an average of 1.40 °C and 1.32 °C in Beijing and the MCPRD, respectively. Grassland decreased by 1.03 °C, 0.58 °C and 1.42 °C in Beijing, Shanghai and the MCPRD, respectively. This suggests that urban vegetation has an obvious effect on regulating the urban thermal environment and mitigating UHIs in the three cities in summer. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements This work was supported by the National Natural Science Foundation of China (42071327), the State Key Program of National Natural Science of China (Grant No. 42330506). Author contributions Z.T.: formal analysis, investigation, methodology, software, validation, visualization, and writing - original draft preparation; J.T.: supervision, resources, project administration, fund acquisition, and writing – review & editing; Y.Z.: conceptualization, methodology, resources, and writing – review & editing; J.Z.: data, resources; N.M.: methodology; X.Z.: methodology, project administration. Data availability The Köppen–Geiger climate classification data is available at [https://www.gloh2o.org/koppen/](https:/www.gloh2o.org/koppen) . The SRTM data is available at [https://srtm.csi.cgiar.org/srtmdata/](https:/srtm.csi.cgiar.org/srtmdata) . Observed daily average air temperature data will be shared upon reasonable request, please contact the corresponding author via [tianj.04b@igsnrr.ac.cn](mailto: tianj.04b@igsnrr.ac.cn) . The Sentinel-2 reflectance data is available at [https://scihub.copernicus.eu/](https:/scihub.copernicus.eu) . Annual land cover data is available at [https://doi.org/10.5281/zenodo.4417810](https:/doi.org/10.5281/zenodo.4417810) . Code availability statement. The analysis software used in this study is python and R. The code used in this study can access in https://github.com/BastianT22/Cooling-effect-of-vegetation.git. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The Köppen–Geiger climate classification data is available at [https://www.gloh2o.org/koppen/](https:/www.gloh2o.org/koppen) . The SRTM data is available at [https://srtm.csi.cgiar.org/srtmdata/](https:/srtm.csi.cgiar.org/srtmdata) . Observed daily average air temperature data will be shared upon reasonable request, please contact the corresponding author via [tianj.04b@igsnrr.ac.cn](mailto: tianj.04b@igsnrr.ac.cn) . The Sentinel-2 reflectance data is available at [https://scihub.copernicus.eu/](https:/scihub.copernicus.eu) . Annual land cover data is available at [https://doi.org/10.5281/zenodo.4417810](https:/doi.org/10.5281/zenodo.4417810) . statement. The analysis software used in this study is python and R. The code used in this study can access in https://github.com/BastianT22/Cooling-effect-of-vegetation.git.