Different explanations for surface and canopy urban heat island effects in relation to background climate

Summary The background climatic conditions and urban morphology greatly influence urban heat island effects (UHIs), but one-size-fits-all solutions are frequently employed to mitigate UHIs. Here, attribution models for surface UHIs (SUHIs) and canopy UHIs (CUHIs) were developed to describe UHI formation. The contribution of factors to SUHIs and CUHIs shows similar dependencies on background climate and urban morphology. Furthermore, the factors that mainly contributed to CUHIs were more complex, and anthropogenic heat was the more critical factor. Influence from urban morphology also highlights that there is no one-size-fit-all solution for heat mitigation at the neighborhood. In particular, maintaining a low building density should be prioritized, especially mitigating CUHIs. Moreover, it is more effective to prioritize urban irrigation maintenance over increasing green cover in arid regions but the opposite in humid regions. The work can provide scientific evidence to support developing general and regional guidelines for urban heat mitigation.


INTRODUCTION
Urban development changes the energy balance in peri-urban areas, which typically have higher air and surface temperatures than their surrounding rural areas. 1,2][10] UHIs are also a threat to human health, and a warming environment will increase the risk of morbidity and mortality, 3,11 especially when synergistic interactions with heat waves occur. 12,13According to the World Cities Report 2022: Envisaging the Future of Cities by UN-Habitat, China's urban population is expected to exceed 1 billion by 2035, with an estimated increase of 180 million compared with 2020. 14China's rapid urbanization will exacerbate the thermal risks faced by urban populations. 15,16However, one-size-fits-all solutions are frequently employed to mitigate UHIs; another tendency is the conflation between the mitigation strategies of surface UHIs (SUHIs) and canopy UHIs (CUHIs).For instance, shading has a more pronounced cooling effect on surface temperatures than on air temperatures.Therefore, understanding the mechanisms responsible for the UHIs is important for identifying general guidelines to mitigate problems related to heat.
Howard observed higher air temperatures in London compared with surrounding rural areas to first identify the UHIs 17 and correctly hypothesized most of the causes that are now considered responsible. 1However, selecting representative measurement sites for studies of the CUHIs is still a problem that needs to be addressed. 18,19Due to the proliferation of satellite observations in the land surface temperature (LST) field, several studies have attempted to explain the mechanisms responsible for SUHIs. 20,21The same sensor used in a single satellite can provide complete global coverage observation data to obtain large-scale surface temperature observations for applications in comparative studies of cities under different background climatic conditions.However, considering that most vertically oriented building faces are not observed by a nadir-viewing remote imaging radiometer, satellite-based LST data cannot fully represent all of the urban surface temperatures in areas with complex three-dimensional geometry. 22,23In recent studies, the attribution analysis method was applied at city scale to show that the urban-rural contrast in evapotranspiration (ET) and/or thermal convection efficiency are the main determinants of summer SUHIs, 24,25 and the background climatic conditions affect the relative contributions of these factors. 2 On a relatively long-period, SUHIs exhibit a tightly connection with CUHIs, with large SUHIs will experience a large CUHIs. 26However, observational studies have shown that SUHIs and CUHIs differ in terms of their frequency and characteristics. 27,28Some similar biophysical drivers affect the development of SUHIs and CUHIs, but they may differ in terms of magnitude. 21Therefore, the quantitative study on SUHIs in the previous study does not provide a comprehensive Figures 1A and 1B show the attributions to the summer SUHIs and CUHIs intensities for 990 cities across China.Clearly, the main contributors to the summer SUHIs and CUHIs intensities are the urban-rural differences in the evapotranspiration (DET) and convection efficiency (DCV).In particular, when with a low urban irrigation index (I r,u = 0), the largest contributor to the SUHI intensities was DET, which is consistent with the findings obtained by Li et al., 24 and it can be explained by the lower soil moisture and vegetation in urban areas.However, the contribution of DET was significantly reduced when the idealized urban irrigation index (I r,u = 1) was adopted.Higher I r,u will make the evapotranspiration from vegetation in urban areas not be limited by the soil moisture, even with the lower green cover in urban areas, resulting in negative contribution of DET, which is more obvious in arid regions.In fact, DET also exhibits a large standard deviation (SD), and it is considered that the differences in heat mitigation achieved by urban irrigation are great under different background climatic conditions.Unlike the SUHIs, the difference in the contributions of DET and DCV to the CUHI intensities was not significant.In the present study, the convection efficiency included the convective heat exchange between the canopy air and urban underlying surface, and the convective heat exchange between the canopy air and underlying atmosphere, which is more pronounced for a fluid such as air due to flow.Given the impact of the geometric configuration of an urban canyon on radiation exchanges, the urban canyon receives more net radiation (R*) than rural areas, 33 and the contribution of DR* is positive.It is important to note that the difference between urban and rural areas is small, so the contribution of the net radiation is not significant.Meanwhile, the heat stored (G) during the daytime was released at nighttime, and the value of G was close to 0 when time scales exceed a day, i.e., the contribution of DG was also small.In addition, the contributions of both DR* and DG to the SUHI intensities were slightly higher than those to the CUHI intensities.It is considered that if the heat stored in air is ignored, R*and G have direct effects on the urban surfaces and an indirect effect on the urban canopy air, and thus the effect on the CUHI intensities is weaker.
According to the mean annual precipitation (P), cities in arid regions (P < 400 mm yr À1 ) and humid regions (P R 800 mm yr À1 ) were selected to analyze the contributions to the summer SUHI intensities and summer CUHI intensities (see Figures 1C-1F).DT s ranged from À2.9 C to À0.4 C in arid regions (mean: À0.6 C), and from 0.1 C to 3.2 C in humid regions (mean: 0.9 C).The DT s values were much smaller for cities in arid regions than humid regions, and the specific contributions to DT s also differed significantly between arid and humid regions.The average contribution of DR* was 0.5 C in arid regions, which was higher than the contribution in humid regions (0.2 C), mainly because the solar radiation is generally higher in arid regions (see Figure S1B).The contributions of DET and DCV were very different in arid and humid regions, and the contribution of DCV was much higher than that of DET in arid regions.Given the short vegetation found in arid regions, cities are generally aerodynamically rougher than the surrounding rural areas and the heat dissipation by convection could be more efficient, 34 and thus DCV made a negative contribution to DT s .Urban areas had lower vegetation cover in arid regions, and water limitations reduced the ET by vegetation in arid rural areas, thereby limiting the contribution of DET to DT s .However, the largest contributor to DT s was DET in urban areas where higher precipitation resulted in a higher soil water content, thereby reducing the restriction on ET by vegetation, and the lower vegetation cover in urban areas led to higher DET values, and the contribution to DT s was positive.Cities in humid regions are generally surrounded by tall vegetation, and there was no significant difference in heat dissipation by convection between urban and rural areas, thereby reducing the contribution of DCV to DT s .The contributions to DT c were similar to those to DT s , but the differences in the contribution of each factor to DT c were much smaller compared with DT s .In addition, DCV provided similar contributions to DT s and DT c in humid regions, with values of 0.2 C and 0.3 C under I r,u = 0 (0.2 C and 0.2 C under I r,u = 1), respectively.Whereas, DCV provided a significantly greater contribution to DT s than DT c in arid regions, with values of À1.1 C and À0.3 C under I r,u = 0 and À1.0 C and À0.2 C under I r,u = 1, respectively.The contribution of anthropogenic heat (AH) was high in both arid and humid regions.Consistent with previous studies, the thermal mitigation effect achieved by increasing the urban irrigation index was influenced by the background climatic conditions. 26,35Increasing the urban irrigation index reduced the contribution of DET to DT s from 0.3 C to À1.6 C in arid regions, but only from 1 C to 0.4 C in humid regions.In addition, the effect of this measure on DT c was relatively weak, where the contribution of DET only decreased from 0.1 C to À0.5 C in arid regions.

Influence of background climate
We quantified the contributions to DT from different biophysical factors comprising R*, ET, CV, G, and AH (see STAR Methods).The distributions of the contributions to DT in China from different biophysical factors are shown in Figure 2. Without urban irrigation (I r,u = 0), the distribution of the contribution from DET to DT s was similar to the distribution for mean annual precipitation in China (see Figure S2).In cities where P < 400 mm yr À1 , the contribution of DET was low and generally below 20%, and the contribution to DT s was dominated by DCV with greater than À60%, where ''-'' implies an inverse effect on DT s .DET made the dominant contribution to DT s in humid regions, especially in cities where P > 1200 mm yr À1 , and the contribution of DET exceeded 90%.As I ru increased, the contribution of ET also increased in arid regions, but with the reverse effect on DT s , and this influence gradually expanded from northwest China (arid regions) to southeast China (humid regions).However, in southeast China, ET still made a high positive contribution to DT s , especially in areas where P > 1600 mm yr À1 , and DET still contributed more than 80%, thereby implying that improving urban irrigation levels alone had a limited impact on mitigating UHIs.It was also found that under idealized urban irrigation (I r,u = 1), the contribution of DT s was more complex in some cities where 800 < P < 1200 mm yr À1 , and both DET and DCV did not make major contributors, and their contributions were similar.As mentioned earlier, the composition of  the contributors to DT c was more complex compared with DT s , and it was difficult to identify the main contributors to DT c , which also suggests that it may be more complex to mitigate DT c compared with DT s .The dominant contributor was more likely to be identified in arid regions (P < 400 mm yr À1 ), where DCV and DET accounted for >60% of the inverse contributions under I r,u = 0 and I r,u = 1, respectively.Given the influence of latitude, the contribution of DAH was higher in cities at low latitudes than high latitudes during the summer.A nonlinear relationship between DT (DT s and DT c ) and mean annual precipitation was found (see Figure 3A), which is consistent with the results obtained on a global scale by Manoli et al. 2 However, compared with their findings, DT s was slightly lower in the present study because cities in arid regions in China contain high buildings and urban areas have a stronger convective efficiency compared with suburban areas with short vegetation, as shown in Figure 3B.In low precipitation regions, DT s increased in a linear manner with precipitation, and DT s saturated at high precipitation values exceeding P z 1,000 mm yr À1 DT c shows a similar trend to DT s but reached saturation at P z 500 mm yr À1 .It was also found that the magnitude of the change in DT s was much larger than that in DT c , where the latter was influenced less by the background climatic conditions.The P-DT relationship was mainly controlled by DET and DCV, especially DCV, which had a similar shape to the P-DT relationship.In humid regions with high relative air humidity, the vapor pressure deficit in the air was low, thereby restricting the ET by vegetation in rural areas.Thus, an upper bound for DET was defined between urban and rural environments.Conversely, water limitations reduced the magnitude of ET in rural arid regions, thereby limiting the contribution of DET to DT (see Figures 3B and 3C).Considering the larger ''deficit'' of water in the air, supplementing the water budget of urban vegetation with irrigation increased the ET in urban areas, and it made a negative contribution to DT, thereby creating an ''oasis'' effect 36 (see Figures 3E and 3F).As discussed earlier, the height of natural vegetation increases with precipitation, and thus the heat dissipation efficiency by convection decreased in rural areas as the precipitation increased.Therefore, the urban-rural differences in the convection efficiency also contributed to cooling cities in arid regions. 24,25By contrast, the tall vegetation that surrounded cities in humid regions exchanged heat more efficiently than dense building blocks, and DCV made a positive contribution to DT.
The DET-DT and DCV-DT relationships had similar shapes for DT s and DT c .According to Equation 6 and Equation 10, the energy redistribution factors f s and f c represent the sensitivities of DT s and DT c to energy forcing of 1 W m À2 .As shown in Figure S3, the energy redistribution factor f c for DT c was higher than f s for DT s , thereby damping the influence of urban-rural differences in energy forcing on the magnitude of DT c .In addition, to calculate the contribution of DCV to DT c , the convective heat exchange between the urban surface and canopy air was also considered, which generally had a positive influence.Therefore, this reduced the negative influence of DCV on DT c in arid regions, whereas DCV had a greater positive influence on DT c than DT s in humid regions.It was also found that f s and f c varied with P, which tended to be higher in humid regions (see Figure S3).Thus, temperatures were more stable in humid areas and less susceptible to energy forcing, which may also explain the difficulty mitigating UHIs in these cities because it is necessary to eliminate more heat fluxes.

Influence of urban morphology
We selected a building density (r b ) range of 0.2-0.6 to analyze the influence of differences in urban morphology on the UHIs in arid and humid regions.In contrast to urban green cover, the influence of urban morphology on the UHIs was affected less by the background climatic conditions.In arid and humid regions, the DT-r b relationship had a similar nonlinear shape, and DT c was influenced more significantly by the building density compared with DT s (see Figure 4).The attribution of UHI intensities for building densities of 0.2 and 0.6 were analyzed (see Figure 5).Under a building density of 0.2, DT s was slightly higher than DT c , whereas the opposite was found with the building density of 0.6.The dominant contributor to DT s was DET in both the cases, and the different contribution of DET to DT s was also influenced by background climatic conditions, which shows more significance in high precipitation regions (see Figure S7A).DET also shows a similar contribution trend to DT c , but the contributions of DET, DCV, and DAH to DT c were all around 0.15 C, with no significant dominant contributor (r b = 0.2, see Figure 5A).The contribution of DCV to DT s and DT c still cannot be ignored and increases significantly with increasing building density, especially for DT c .In contrast to DET, the different contribution of DCV to DT s or DT c is more significant in arid regions, whereas the contribution to DT c is more susceptible to the influence of building density compared with DT s (see Figure S7).It is suggested that as the building density increased, higher building densities restricted the efficiency of convective heat transfer from the urban surface and urban canopy to the atmosphere, thereby resulting in poorer heat dissipation with a negative impact on the UHIs.The higher building density also led to more anthropogenic heat 37,38 and when coupled with the poor heat dissipation, the contribution of DAH to DT c was significant at a building density of 0.6 (see Figure 5B).Therefore, maintaining lower building densities is beneficial for urban heat mitigation, which provides greater heat dissipation and space for urban greening.It needs to be clarified that as the urban population is growing, choosing higher buildings or building densities to accommodate urban residents still requires further research, because the exact relation between aerodynamic resistance and urban morphology is complicated. 24The urban morphology should be able to enhance heat dissipation performance.

Heat mitigation strategies
Urban greenery is used as a strategy for mitigating the UHIs, and it plays an important role in promoting the urban outdoor thermal environment as well as enhancing the thermal comfort of residents. 39,40It shows that the g c,u -DT relationship was strongly influenced by the background climatic conditions in urban areas (see Figures 6A and 6B).In humid regions, the ET by vegetation is less water limited, and it is a dominant component of the surface energy balance in rural areas. 41The urban-rural ET difference was significant due to the lack of vegetation in cities, and thus increased vegetation cover is needed to reduce DT s and DT c .This means that it is difficult to reduce DT s and DT c by only focusing on vegetation strategies in higher precipitation regions (P > 1000 mm yr À1 ), because the growth of vegetation is strongly dependent on the precipitation and humidity. 42Indeed, to achieve DT s < 1 C, almost the entire city area would need to be replaced with vegetation, and it is almost impossible to achieve DT s < 0.5 C without urban irrigation management (I r,u = 0).The g c,u -DT relationship had a similar shape for DT s and DT c , but with a lower value of DT c , it is more feasible to achieve a weak DT c by applying urban greenery strategies.In addition to increasing urban green cover, improving urban irrigation is an effective method for achieving urban cooling (see Figures 6C and 6D).In particular, the low soil moisture content in arid regions will limit the evapotranspiration by rural vegetation to a certain extent. 2 Therefore, increasing urban irrigation can effectively increase the soil moisture content and reduce the limitation on vegetation evapotranspiration, achieving high heat mitigation even with low urban green cover (g c,u = 0.3).A recent study has shown that urban irrigation in more arid areas can improve the potential cooling performance. 43By contrast, in humid regions with high soil moisture contents, the evapotranspiration by vegetation is less restricted by water, and thus the heat mitigation effect achieved by increasing urban irrigation is weaker (see Figure S10).Clearly, increasing urban green cover and urban irrigation are both feasible strategies for mitigating the heat island effect. 44,45In addition, according to further analysis, the effects achieved by these two measures also differed among different background climatic conditions.Urban green cover of 0 was set as a base case, and it was considered the UHI mitigation effect is achieved by increasing urban green cover under different urban irrigation indexes (see Figure S8).Under a low urban irrigation index (I r,u = 0.3), the heat island mitigation effect achieved by increasing urban green cover was more significant in humid regions, and this effect saturated at high precipitation values exceeding P z 1000 mm yr À1 (see Figure S8B).Interestingly, the UHI mitigation effect decreased with increasing precipitation in arid regions (P < 400 mm yr À1 ).It is suggested that the difference between the potential evapotranspiration (PET) and mean annual precipitation decreased rapidly in this range (P < 400 mm yr À1 ) (see Figure S9), and this difference indicates a contradiction between the positive effect of ambient air moisture deficit on vegetation evapotranspiration and the limitation imposed on vegetation evapotranspiration by the lack of available water.Therefore, the results suggested that urban irrigation management was more important in arid regions, even with a low urban irrigation index (I r,u = 0.3), where the water provided by irrigation could promote the evapotranspiration by vegetation, and the UHI mitigation effect decreased as the precipitation increased.By contrast, with no urban irrigation management (I r,u = 0), the evapotranspiration by vegetation was affected by the availability of water, and the situation mentioned earlier did not occur, and the UHI mitigation effect increased with the precipitation.In addition, significant differences in the heat island mitigation effects were found as the urban green cover increased in humid regions (see Figure S8A).The opposite situation occurred under a higher urban irrigation index (I r,u = 1), where the heat island mitigation effect of greater urban green cover increased in a nonlinear manner as the precipitation increased, before saturating at precipitation of P z 500 mm yr À1 , which is consistent with the results shown in Figure S9 to further illustrate the importance of urban irrigation management in arid regions (see Figure S10).The mitigation effect was much lower for DT c than DT s , i.e., about half of that for DT s , and both exhibited similar trends.This can be explained by the higher energy redistribution factor of DT c than DT s .
The effectiveness of heat mitigation strategies is influenced not only by background climatic conditions but also by urban morphologies.The Mann-Whitney U test was used to analyze whether the heat mitigation with different building density had a significant difference, as shown in Figures S11A and S11B.The heat mitigation strategy obtains more significant effects at the higher building density (p < 0.01) and shows more difference in reducing DT c than DT s .At I r,u = 0.2, the reduction of DT s and DT c by increasing urban green cover from 0 to 0.4 was more significant in humid regions, even at a building density of 0.6 (see Figure S11C).It needs to be clarified that although the heat mitigation effect of increasing urban green cover is weaker at r b = 0.2 than at r b = 0.6, more space can be provided for urban greening at r b = 0.2, which can further reduce DT s and DT c .In arid regions, increasing the urban irrigation index to reduce DT s and DT c was more favorable, and the urban morphology still shows a significant effect on reducing DT c compared with DT s (see Figure S11D).Fortunately, heat mitigation strategies can still have significant effects at higher building densities under different background climatic conditions, which of course requires a combination of urban greening and management of urban irrigation.It needs to be clarified that heat mitigation strategies will increase air humidity, especially in humid regions, which may further increase heat risk. 46

DISCUSSION
In the UHI biophysical mechanism model, the difference between the urban and rural surface temperature (DT s ) and air temperature (DT c ) is attributed to different biophysical factors, thereby improving the understanding of the impacts of energy balance changes caused by iScience Article urbanization on the LST and air temperature.In the present study, it is demonstrated that the urban-rural differences in evapotranspiration (DET) and the convection efficiency (DCV) made the two main contributions to DT s .Considering that anthropogenic heat release will directly cause air temperature increases, anthropogenic heat flux also had an important effect on DT c , thereby making the mitigation of DT c more complicated.It needs to be clarified that the air temperature rise due to anthropogenic heat release will also play an indirect effect on the surface temperature by limiting the release of heat from the urban surface.However, it has been shown to play a negligibly small role in contributing to DT s 2 , so we do not consider it as a factor affecting DT s .Urban background climatic conditions significantly influenced the components that contributed to DT s and DT c , and the relationship between DT and mean annual precipitation was nonlinear.Anthropogenic heat release will directly cause air temperature increases, which also had an important effect on DT c , especially at the higher building density.Furthermore, the energy redistribution factor of DT c (f c ) is generally higher than that of DT s (f s ), which means it is necessary to eliminate more heat fluxes to reducing DT c .Therefore, the factors that contribute to DT c are more complex than DT s , and it may be more challenging to mitigate DT c .Maintaining a low building density should be prioritized, especially when mitigating DT c , which can provide more space for urban greenery and further reduce anthropogenic heat emissions.
Increasing urban green cover is an important strategy for mitigating the UHIs, 44,47 and the effect of this strategy was influenced by the background climatic conditions. 35,47Without urban irrigation, the effect of increasing urban green cover on mitigating UHIs was more pronounced in humid regions, but mitigation was difficult to achieve with a lower DT s and DT c .In arid regions, it is particularly important to improve urban irrigation considering the limitations imposed by soil water deficit on evapotranspiration.It is important for mitigating the UHIs, and a highly significant mitigation effect can be obtained by increasing urban irrigation even with low urban green cover.However, the irrigation strategy might jeopardize scarce water resources when considering the urban heat mitigation. 48Meanwhile, urban greening and urban irrigation may lead to elevated humidity levels, which is also an important contributor to human heat stress. 46Although urban greenery can effectively reduce DT c , its mitigation effect was much lower than DT s , about half of that for DT s .
The contributions of urban-rural heat differences in the convection efficiency to DT s and DT c are more complicated to assess.Because the exact relation between aerodynamic resistance and urban morphology is complicated, it is difficult to artificially change the efficiency of urban convection.Generally, the positive impact of DCV on DT c was higher than that on DT s , at different building densities.Thus, improving urban convective heat exchange efficiency plays more significant role in mitigating DT c than mitigating DT s , which means maintaining a low building density is most important for mitigating DT c .Improving urban convective heat exchange efficiency may be more important in those cities with water limitations by changing the urban morphology. 21Unfortunately, this heat mitigation strategy will face more challenges, as their exact relations are complicated. 24This heat mitigation strategy was difficult to use in built-up neighborhoods where buildings already exist.For new neighborhoods, rational urban morphology design can be considered to improve urban convective heat exchange efficiency.

Limitations of the study
Generally, improving the urban evapotranspiration capacity and urban convective heat exchange efficiency plays an important role in mitigating DT s and DT c . 2,24In arid regions, it is more important to improve urban irrigation than increasing the urban green cover.By contrast, in humid areas, it is quite important to maintain high urban green cover in urban areas to reduce the negative impacts of urban-rural differences in evapotranspiration on DT s and DT c considering the strong evapotranspiration capacity in rural areas. 2 However, the present model does not consider the effects of impervious surfaces on the urban water balance (the soil moisture may be lower in urban areas than suburban areas) but ensuring adequate urban irrigation is important.Urban morphology also affects the effectiveness of heat mitigation strategies.Targeted solutions should be proposed for different neighborhoods of the same city, especially in humid regions, where the impact of differences in urban morphology was more obvious.Furthermore, our discussion is limited to the summer-averaged SUHIs and CUHIs; some specific climatic conditions such as large-scale and local circulation, especially for the synergistic effect with heat waves in summer, are not considered.
These factors can influence physical mechanisms that drive the SUHIs and CUHIs, as well as the efficacy of heat mitigation strategies. 49,50Indepth and extensive studies are required to elucidate the general principles.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Lead contact
Requests for further information and resources should be directed to the lead contact, Liu Yan (liuyan@xauat.edu.cn).

Materials availability
This study did not generate new materials.

Data and code availability
All data can be obtained from the lead contact, provided the request is reasonable.
The code related to the attribution model can be accessed by reaching out to the lead contact.

Urban morphology
The present study focused on the thermal environment in residential areas at the neighborhood scale (% 1 km).It is generally assumed that megacities contain more tall buildings but according to a recent study of 36 major cities in China, the mean building height at the city level is not related to the city size, and the urban morphology parameters (building height, building width, and street width) similar in residential areas. 51At the neighborhood scale, various urban morphologies are present in Chinese cities, and urban morphology also has a significant impact on the urban thermal environment. 52Therefore, when analyzing the impact of background climatic conditions on UHIs, the influence generated by changes in urban morphology was eliminated by selecting the common urban morphology as the base case for the model.Additionally, the impact of urban morphology on UHIs has also been studied, we selected combinations of different urban morphology parameters.According to the building height data set for China, the average building height is generally below 20 m. 51,53 Beijing City Lab (https://www.beijingcitylab.com/)provides data for 141375 street blocks in 63 Chinese cities during 2017, with threedimensional parameters at the street block scale for these Chinese cities.According to this data set, the average number of building stories is predominantly <7 in China, which is also consistent with the results obtained in a previous study, 51,53 and the building density is mainly medium and high density (r b > 0.25) in these street blocks.Therefore, the urban morphology parameters established for the base case in the study comprised: building height h b = 18 m (about 6 stories), building density r b = 0.3, building width w b = 20 m, 54 and g c,u = 0.3.We also considered the effects of differences in urban morphology by changing the building density, in the range of 0.2-0.6,based on the street block data from 63 Chinese cities provided by the Beijing City Lab, which shows most building densities in China ranged from 0.22 to 0.52.

Background climate
The climate conditions vary significantly among years and the background climate data for a given year do not reflect the general UHIs in a city.Thus, it is required a customized background climate data set that adequately reflected the typical background climatic conditions in cities, so we obtained typical meteorological year data from the China building energy efficiency design basic data platform (https:// buildingdata.xauat.edu.cn/) as background meteorological data.Typical meteorological year data for Macau and Taipei were obtained from EnergyPlus (https://energyplus.net/weather).Mean annual precipitation data were retrieved from the 1970-2017 monthly precipitation data set of the China Meteorological Data Network (http://data.cma.cn/).

Mathematical model
Multivariable functions of T s (surface temperature) and T c (air temperature) were derived from the energy balance in rural or urban areas.Next, based on the first-order Taylor series expansion, the mechanistic attribution model for attributing the UHI intensity to contributions from different factors was developed (radiation, evapotranspiration, convection efficiency, heat storage, and anthropogenic heat).

Mechanistic attribution model
Based on the first-order Taylor series expansion, we attributed the UHI intensity (including SUHIs and CUHIs) to contributions from different biophysical factors.For SUHIs, based on the surface energy balance equation: , where g c is the green cover, b is the water stress factor, l v is the latent heat of vaporization [J$kg -1 ], and q sat (T s ) is the saturated specific humidity at temperature T s [kg$kg -1 ], i.e., q sat ðT s Þ = 0:622 esatðTsÞ patm À 0:378esatðTsÞ , where p atm is the atmospheric pressure [kPa], q a is the specific humidity of air [kg$kg -1 ], and r s is the surface resistance [s$m -1 ].Q G is the heat storage [W$m -2 ], which can be modeled by the objective hysteresis model (OHM).In this surface energy balance equation, we considered the heat exchange between the urban surface and the atmosphere (Q H ), while excluding the impact of urban canopy air temperature.Subsequently, we obtain the surface temperature as a boundary condition for the canopy energy balance equation (Q s ).Meanwhile, the anthropogenic heat flux directly affects air in the urban canopy to increase the air temperature, which then indirectly affects the heat dissipation from the urban surface, so we considered the effect of anthropogenic heat flux in the canopy energy balance equation (see the next section).
To obtain an analytical form of T s , the outgoing long-wave radiation term and the saturated specific humidity term were linearized at the point T a : 2) 3) where q' sat,s (T a ) is the derivative of q sat (T a ) relative to T a .By substituting Equations 2 and 3 into Equation 1 and rearranging, built the following analytical solution for T s .4) In this study, urbanization was considered as a perturbation of the rural base state and only consider changes in LST due to changes in five biophysical factors (i.e., Da, Dε s , Dr a , Dr s , and DQ G ).According to Taylor's theorem, the urban-rural LST difference was expressed as: (Equation 5) 6) where f s -1 represents the sensitivity of LST to changes of 1 W$m -2 in energy forcing at the land surface. 2,57milar to SUHI, CUHI is based on the canopy energy balance equation: 7) where Q c denotes turbulent exchanges between the air temperature in the urban canyon and atmosphere [W$m -2 ], i.e., Q c = rcpðTc À TatmÞ ra

58
; T atm is the temperature of the atmosphere above the urban canopy [K], and considering that these data were missing for many cities, which was assumed that it was equal to the air temperature measured at the rural weather station, i.e., T atm = T a ; T c is the canopy air temperature [K], Q s denotes turbulent exchanges between the air temperature in the urban canyon and urban surface [W$m -2 ], i.e., Q s = rc p C h ðT s À T c Þ, C h denotes the turbulent transfer coefficient for sensible heat [s$m -1 ], and Q ah is the anthropogenic heat flux [W$m -2 ] calculated based on the building energy model framework. 59In residential buildings, the heat load generated by indoor human metabolism and heat dissipation from equipment (electric lights, etc.) is relatively small relative to the overall anthropogenic heat, so only considered the heat load generated by heat transfer through the building structure and ventilation into the room 60 : , where H out and E out are the sensible and latent heat pumped out from the building, respectively, and COP represents the energy efficiency of heat source equipment.H out = ½hðT s À T in Þ + ð1 À b in Þrc p v a ðT c À T in Þ4 p , where h is the convective heat transfer coefficient [W$m -2 $K -1 ], T in is the indoor air temperature [K], which was selected according to the local air temperature, b in is the thermal efficiency of the total heat exchanger, v a is the total ventilation rate in the building [m 3 $h -1 ], and 4 p is the ratio of hourly occupants relative to the peak number of occupants per floor area (0 % 4 p % 1).E out = ð1 À b in Þl v rv a ðq a À q in Þ4 p , where q in is the specific humidity of the indoor air [kg$kg -1 ].
By rearranging the canopy energy balance equation, built the following analytical solution for T c .
T c = rc p T atm r a +rc p C h T s +Q ah rc p r a +rc p C h (Equation 8) Similar to SUHI, according to Taylor's theorem, the urban-rural air temperature difference was expressed as: ðT c À T atm ÞDr a + rc p ðT s À T c ÞDC h + DQ ah !(Equation 9) 10) where f c -1 represents the sensitivity of the air temperature to changes of 1 W$m -2 in energy forcing in the urban canyon.
The contributions to DT from different biophysical factors comprising R*, ET, CV, G, and AH were expressed as: 2 +ðDAHÞ 2 3 100% (Equation 11) where Contribution i denotes the contribution of a particular biophysical factor, i = {R*, ET, CV, G, AH}.
DR Ã = 8 > > > < > > > : (Equation 12) The urban surface albedo (a urban ) is influenced by the reflective properties of urban surface materials and the urban geometry (cavity effect). 61onsidering the radiation exchange process in the urban canopy, the derived a urban is more accurate when the reflection is calculated in the urban canopy more times.Numerical simulation studies have shown that it is sufficient to consider only two reflection processes in practice. 62he a urban was calculated by considering two reflection processes: 2SVF w , SVF wr = 1 2 ð1 À SVF r Þ. 63 The rural surface albedo (a rural ) was selected based on the typical surface albedo values provided by the local climate zone (LCZ) system.The change in albedo induced by urbanization was calculated as follows.Da = a urban À a rural (Equation 18)

Emissivity, ε
Considering the ''trapping'' effect of urban geometry on long-wave radiation (cavity effect), the effective emissivity of the urban canopy was calculated as 64 : 19) where ε u,0 is the emissivity of the urban fabric (assumed to be 0.9 65 ) and SVF is the sky view factor of the urban canopy, i.e., SVF = È cos Â arctan À H W ÁÃÉ 2 . 66tellite observations show that the emissivity varies little over mostly vegetated surfaces and it deviates only slightly from 0.95 among different land cover types. 57Therefore, for vegetation-covered suburban areas, assumed a surface emissivity of: ε rural = 0.95.
The change in emissivity induced by urbanization was calculated as follows.

Figure 1 .
Figure 1.Attributions of summer surface and canopy urban heat islands (UHIs) across China (A and B) Summer UHIs (surface UHIs, orange; canopy UHIs, blue) and components considering urban irrigation indexes of (A) I r,u = 0 and (B) I r,u = 1.(C and D) Modeling results considering an urban irrigation index of I r,u = 0 in arid regions (C) and (D) humid regions.(Eand F) Modeling results considering an urban irrigation of index I r,u = 1 in (E) arid regions and (F) humid regions.The urban irrigation index is a coefficient describing the level of irrigation, establishing the connection between actual and potential evapotranspiration.For I r,u = 0, it describes the actual evapotranspiration under natural background climatic conditions.For I r,u = 1, it means there is no water supply limitation and that actual and potential evapotranspiration are equal.Error bars indicate G1 standard deviation.

Figure 2 .
Figure 2. Distribution of the contribution of each component to summer surface and canopy urban heat islands (UHIs) under different urban irrigation indexes (A-D) Surface UHIs and (E-J) canopy UHIs, [(A-B) and (E-G)] under urban irrigation index I r,u = 0 and [(C-D) and (H-J)] under urban irrigation index I r,u = 1.

Figure 3 .
Figure 3. Effects of precipitation on summer surface and canopy urban heat island effects (UHIs) and components (A-C) Urban irrigation index I ru = 0. Modeled (markers) and fitted (lines) nonlinear relationships between DT (DT s and DT c ) and (A) mean annual precipitation (P), (B) components of DT s , and (C) components of DT c .(D-F) Urban irrigation index I ru = 1.Modeled (markers) and fitted (lines) nonlinear relationships between DT (DT s and DT c ) and (D) mean annual precipitation, (E) components of DT s , and (F) components of DT c .The line was obtained by fitting the mean values for the modeled data.Error bars indicate G1 standard deviation.

Figure 4 .
Figure 4. Effects of different building densities on the urban heat island effect in arid (orange) and humid regions (blue) Solid lines (DT s ) and dashed lines (DT c ) indicate the ensemble means, and shaded areas represent the ensemble mean G1 standard deviation.

Figure 5 .
Figure 5. Attribution of summer surface and canopy urban heat island effects (UHIs) under different building densities Summer UHIs (surface UHIs, orange; canopy UHIs, blue) and their components at: (A) building density r b = 0.2 and (B) building density r b = 0.6.Error bars indicate G1 standard deviation.

Figure 6 .
Figure 6.Impact of urban background climate on the efficiency of heat mitigation by urban evapotranspiration (A and B) Mitigation effect of urban green cover change on (A) DT s and (B) DT c .(C and D) Mitigation effect of urban irrigation changes on (C) DT s and (D) DT c .Results are shown for I r,u = 0.2 in (A-B) and for constant green cover g c,u = 0.3 in (C-D).
, respectively.Similar to the surface energy balance model, the canopy energy balance model shows higher accuracy in rural areas (rural: RMSE = 0.1 C, urban: RMSE = 1.0 C).Considering the error caused by the first-order Taylor series expansion of the mechanistic attribution model, the mean values of the rural and urban variables were used as parameters in this model, the mechanistic attribution model validation as shown in FiguresS12C and S13C.

1 ) 7 ,
where Q* is the net surface radiation [W$m-2  ], S* is the net short-wave radiation [W$m -2 ], i.e., S Ã = S in ð1 À aÞ, S in is the incoming shortwave radiation [W$m -2 ], a is the surface albedo, and L* is the net long-wave radiation [W$m -2 ], i.e., L Ã = L in À L out , where L in is the incoming long-wave radiation [W$m-2  ].Considering that these data were missing for many areas, we modeledL in = ε a sT a 455, where ε a is the atmospheric emissivity, i.e., ε a = 1:723 e sat ðTaÞ RH 100 Ta 1 e sat (T a ) is the saturation vapor pressure at temperature T a [kPa], i.e., e sat ðT a Þ = 0:611 exp 17:27ðTa À 273:15Þ Ta À 35:85 56 , s is the Stefan-Boltzmann constant (s = 5.67 3 10 -8 W$m -2 $K -4 ), and T a is the air temperature which measurement at the rural weather station [K].In addition, L out is the upward long-wave radiation [W$m -2 ], i.e., L out = ε s sT s 4 + ð1 À ε s ÞL in , where ε s is the surface emissivity and T s is the LST [K].Q H is the sensible heat flux [W$m -2 ], i.e., Q H = rcpðTs À TaÞ ra , where r is the air density [kg$m -3 ], c p is the specific heat of air at constant pressure [J$kg -1 $K -1 ], and r a is the aerodynamic resistance [s$m -1 ].Q E is the latent heat flux [W$m -2 ], i.e., Q E = g c b rlv½qsatðTsÞ À qa ra+rs 2

a urban = 2h b a w SVF w +a w 2
SVF ww SVF w +a w a r SVF rw SVF r +w r ða r SVF r +2a r a w SVF wr SVF w Þ 2h b +w r (Equation17)where h b is the building height [m], w r is the road width [m], a w is the albedo of the wall, a r is the albedo of the road, and SVF denotes the sky view factors calculated as:SVF w = 1 2 h b= w r + 1 À ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi h b= w r 2 +1 r !O h b= w r = SVF rw , SVF r = ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi h b= w r 2 +1r À h b= w r , SVF ww = 1 À

TABLE
d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d METHODS DETAILS B Urban morphology B Background climate B Mechanistic attribution model B UHI components STAR+METHODS KEY RESOURCES TABLE RESOURCE AVAILABILITY 2,24,25The surface energy balance model validation was performed with the 2013 Global Urban Heat Island Dataset, which was provided by Manoli et al.2We selected cities with populations over 10 5 for model validation.The observed urban surface temperature was obtained from the observed rural surface temperature + urban heat island intensity (i.e.Ts_summer_2013 + dTs_summer_2013).Comparison of the surface temperature observed and simulated in rural and urban areas were shown in FiguresS12A and S12B, respectively.The surface energy balance model shows higher accuracy in rural areas (rural: RMSE = 1.0 C, urban: RMSE = 4.3 C), due to the more complex urban underlying surface in urban areas.The canopy energy balance model validation was performed with the observed data in Xi'an, China, in 2016, which available from China building energy efficiency design basic data platform (https://buildingdata.xauat.edu.cn/).The hourly meteorological data are averaged to obtain daily data for the model inputs.Comparison of the canopy air temperature observed and simulated in rural and urban areas were shown in FiguresS13A and S13B