Changes in urban heat island intensity during heatwaves in China

With rising occurrence of heatwaves and ongoing urban expansion, urban residents are facing severer heat-related stress under the combined effects of urban heat island (UHI) and heatwaves. Controversial results, however, have been reported regarding whether the UHI is exacerbated during heatwaves. In this study, we used fused ground and satellite daily maximum air temperature data to evaluate the variation of UHI intensities under heatwaves across 225 major cities in mainland China during 2003–2020. Overall, urban areas showed an enhanced UHI intensity of up to 0.94 °C during heatwaves, nearly double compared to normal periods. The interaction between UHIs and heatwaves was sensitive to local background precipitation. Under the similar urbanization and vegetation greenness, the amplified warming in urban areas during heatwaves was more pronounced in wet climates. In megapolitan regions characterized by continuous urban development, the UHI intensified much stronger during heatwaves due to the heat accumulation within urbanized areas and the advection of heat from neighboring cities. Our results contribute to understanding the interactions between UHIs and heatwaves which may strongly increase heat risk in cities. Further work on the variations of this interaction under future warming and consequent impacts on human health and energy use is needed.


Introduction
Heatwave is among the most damaging climate extremes characterized by a prolonged period of hot weather, spanning several days to several weeks, and has garnered increasing attention in science and policy (Meehl and Tebaldi 2004, Zscheischler et al 2018, Christidis et al 2020).More frequent and severe heatwaves can cause serious damage to human society and natural ecosystem under warming climate, with impacts on human mortality (Estoque et al 2020, Yang et al 2021), agricultural yield (Vogel et al 2019), and energy demand (Flores-Larsen and Filippín 2021).Urban areas are more vulnerable to heatwaves because of the preexisting urban heat island (UHI) (Oke 1982).Such compound effects in urban areas will likely be intensified as urban expansion continues throughout the world in the future, especially in developing countries (United Nations 2022), and as climate warming intensifies (IPCC 2023).In recognition of knowledge gap on this combination of heatwaves and UHIs, in-depth investigation is needed to better understand the heterogeneous response of urban temperature to heatwave.
Considerable scientific interest has recently focused on the interaction between large-scale heatwaves and the local-scale UHIs.There is still controversy regarding the assessment of changes in UHI intensity during heatwaves.Contrary to the reported intensified UHI intensity during heatwaves (Ramamurthy and Bou-Zeid 2017, Miao et al 2022, Cui et al 2023), some studies observed an absence of interaction between them (Chew et al 2021, Richard et al 2021), or even a reduction in UHI intensity during heatwaves (Scott et al 2018, Kumar andMishra 2019).Urban expansion, causing reduced evapotranspiration and weakened wind speed, coupled with anthropogenic heat and aerosol emissions, leads to the UHIs, thus magnifying heat extremes (Liao et al 2018, Shi et al 2021, Tuholske et al 2021).
Heatwaves changed the disparity of temperature between urban and rural areas.The amplified warming impact of heatwaves on urban areas was primarily attributed to the reduced evapotranspiration caused by less vegetation and moisture in urban areas (Li andBou-Zeid 2013, Ao et al 2019).Other factors, such as decreased wind speed (Founda and Santamouris 2017), increased anthropogenic heat emissions (He et al 2020), and increased radiative input (Hardin et al 2018) during heatwaves, also contribute to the intensification of UHI intensity.On the contrary, the investigation of a large ensemble of cities in the United States found that UHI intensity tended to decrease during heatwaves in most cities (38 out of 54 cities) (Scott et al 2018).A similar weakening of UHI intensity during heatwaves was also demonstrated in India (Kumar and Mishra 2019).The interaction between heatwaves and UHIs was found be sensitive to local hydroclimate, with the increase of UHI intensity during heatwaves being higher in humid regions than in arid regions in the United States (Zhao et al 2018).Meanwhile, the selection of rural reference stations in UHI calculations also influenced the assessment of the combined effects (Founda and Santamouris 2017, Jiang et al 2019).
Despite the progress made in understanding of urban temperature variations under heatwaves, major controversy remains across regions, largely.Recent studies were generally based on observations from a few representative stations or numerical model simulations in single or a few cities.There is still a lack of quantitative analysis on the combined effects between UHIs and heatwaves across different climates to get large scale picture.
Considering the ongoing debates regarding the changes of UHI intensities during heatwaves, as well as major knowledge gap on their combined effects across different climates, here, we evaluate the intensity of UHIs during heatwaves in 225 cities across China using fused ground and satellite daily maximum air temperature data.The objectives of this study are to (a) quantify the UHI variation during heatwaves and normal periods, (b) explore the controlling factors of spatial heterogeneity in the combined effect between heatwaves and UHIs, and (c) investigate the accumulation of UHIs during heatwaves in urban agglomerations.

Data and methods
We selected 225 Chinese cities with an area greater than 100 km 2 based on the Global Urban Boundary (GUB) dataset in 2018 to examine the changes of UHI intensities during heatwaves (figure 1).The countryscale distribution of these cities, with diverse temperature, precipitation, and size, enables us to understand how heatwaves interact with UHIs under different geographical locations and climate backgrounds (figure S1).In order to obtain enough rural pixels for analyzing the UHI, after extracting the urban polygons of each city from the GUB dataset, we further established a 10 km buffer zone around them.The following analyses were conducted within the regions established after building the buffer zones.The 10 km buffer distance was chosen based on the general buffer definition in the widely used Global Urban Heat Island Dataset, primarily considering that larger buffer distances (e.g. over 10 km) would result in overlaps between the buffers and other nearby urban areas (Center for International Earth Science Information Network 2016).
The latest released all-sky daily maximum air temperature data (https://doi.org/10.25380/iastate.c.6005185.v1)fused from satellite-based land surface temperature (LST) and in situ meteorological observations with a spatial resolution of 1 km was used to examine the combined effects between heatwaves and UHIs during warm seasons (May to September) in 2003-2020.This gridded daily maximum air temperature data was developed using the Spatially Varying Coefficient Models with Sign Preservation, which used gap-filled MODIS LST at mid-daytime (1:30 PM local time) and elevation as explanatory variables to establish its relationship with station-observed air temperature (Zhang et al 2022a(Zhang et al , 2022b)).It was widely used in UHI analyses owing to its acceptable accuracy and fine resolution (Du et al 2023, Prihandrijanti andAzzizi 2023).Note that the UHI type focused on in this study was the canopy layer UHI (Oke et al 2017), which was calculated based on the gridded 2 m air temperature dataset mentioned above.
Meteorological data from 2474 national stations were collected from the China Meteorological Data Service Center.It includes observations of daily maximum temperature (T mx , • C), mean temperature (T mn , • C), precipitation (P, mm) and wind speed (V, m s −1 ).Stations with missing data for five or more days were excluded, resulting in a total of 2206 stations being retained.The observed daily T mx was used to further validate the gridded temperature data.Specifically, for each day, we compared the observed T mx at meteorological stations with the corresponding estimated pixel values to evaluate the accuracy of the gridded air temperature data.Overall, the observed and gridded T mx generally aligned along the 1:1 line with an average R 2 value of 0.86 during 2003-2020 (figure S2).The observed T mn , P, and V were respectively calculated to obtain annual mean values in 2003-2020 and used to examine the controlling factors of the interactions between heatwaves and UHIs.The annual precipitation was the accumulation of daily precipitation over the year.
The overall workflow in analyzing the changes of UHI intensities during heatwaves was showed in figure 2. It included three major components: (1) extraction of heatwaves, (2) random selection of normal days, and (3) calculation of UHI intensity.The delineation of periods into heatwaves and normal phases was one of the key steps in evaluation of the combined effects between heatwaves and UHIs.We defined a heatwave as a period spanning at least three consecutive days with the daily maximum temperature surpassing its respective historical 90th percentile threshold (Perkins and Alexander 2012).The determination of the 90th percentile threshold involved ranking the 15 d temperature samples surrounding a calendar day over an 18 year period (2003-2020).These thresholds were calculated individually for each city.Considering the comparability of the sample sizes between heatwaves and normal periods, we proposed a random sampling method to define normal periods.Firstly, we determined the number of days for heatwaves based on the previously mentioned heatwave extraction method, which served as the sample size for defining normal periods.Secondly, the remaining days after excluding heatwave days were selected as the random sample population for nonheatwave days.In order to make the UHI intensities during heatwaves and non-heatwave periods comparable, we excluded the influence of precipitation, with the sample population for non-heatwave days including only days with daily precipitation less than 0.1 mm (Jiang et al 2019).The daily precipitation for each city was calculated as the average of the observed precipitation from all meteorological stations within the study area.The percentage of remaining days after excluding the influence of precipitation, calculated by dividing the number of days with precipitation less than 0.1 mm by the total number of days in the warm season, exhibited a decreasing trend from north to south and from inland to coastal areas (figure S3).Overall, the proportion of days retained after removing the influence of precipitation in all cities was 63%, approximately 95 d.Finally, we randomly selected a sample from the non-heatwave populations with the same number as the heatwave days for the subsequent analysis of UHI intensities during normal periods.This experiment was repeated 100 times.The classification of heatwaves and normal days was conducted for each city during every year when a heatwave occurred.Taking Beijing in 2019 as an example, a total of three heatwaves were extracted, with a cumulative duration of 11 d.The sample pool for non-heatwave days consisted of 69 d, and 11 d were randomly sampled as normal periods (figures 2(a) and (b)).
To understand the response of UHIs to heatwaves under different urbanization levels, we further stratified each city into three zones dynamically based on the ISA fraction.Starting from the lowest to the highest ISA fraction in a city, these three zones were rural (ISA ⩽ 20%), suburban (20% < ISA ⩽ 60%) and urban (60% < ISA ⩽ 100%) (Phinn et al 2002, Lu andWeng 2007).We also examined the changes of UHIs during heatwaves in three major urban agglomerations distributed from north to south in China, namely, Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Pearl River Delta (PRD).To avoid topographical effects within a city, we used the 1 km GTOPO30 dataset (Harding et al 1999) to calculate the mean elevation of urban and suburban areas (pixels with ISA > 20%) and excluded from our analysis all pixels whose elevation was outside a ±50 m window from this mean elevation (Imhoff et al 2010, Yao et al 2018).The UHI quantification approach made the continuous measure of urban-rural gradient possible, reducing the impact of selecting urban and corresponding reference rural stations.Then, we separately calculated the UHI intensities under heatwaves and normal days using the following equations: where T represented the average temperature, subscripts h and n represented heatwaves and normal periods, and subscripts u and r represented urban and rural areas respectively.The ∆UHI was defined as the difference of UHI intensity between heatwaves and normal periods.
The nonparametric Wilcoxon rank-sum test (Bethea et al 1995) which was applicable for wider range of data because of no or limited assumptions was employed to assess the differences in the changes of UHI intensity under heatwaves between urban and suburban areas.The null hypothesis posited that the two sets of scores were drawn from the same population, implying that the changes in UHI intensity during heatwaves do not differ systematically between the two areas.
Recent studies showed that the urban intensity and greenspace regulated the temperature variation along urban-rural gradients (Jia and Zhao 2020), and the UHI intensity increased with annual background precipitation (Zhao et al 2014).Motivated by these, we further hypothesized that the spatial variability of heatwave-normal period contrasts of UHI intensity could be explained by urban-rural differences in surface attributes (i.e.ISA and normalized difference vegetation index, NDVI) and background climate factors (i.e.precipitation, temperature and wind speed).To test this hypothesis, we explored the relation between heatwave-normal period contrasts of UHI intensity and theses potential factors using the Pearson correlation coefficients.Specifically, the ∆ISA and ∆NDVI were calculated as the annual mean differences between the corresponding urban and rural averages.Here, the NDVI (MOD13A3, https://e4ftl01.cr.usgs.gov/MOLT/MOD13A3.061/)at the 1 km resolution was collected during 2003-2020.Considering the similar background climate conditions within a given city, the annual mean values of T mn , P, and V were calculated from all meteorological stations within the city.It is worth noting that, in order to avoid the influence of spatial autocorrelation between adjacent cities and to ensure a reasonable number of independent samples, we averaged the values of intersecting cities for further analysis.The number of independent samples after correction was 173.

Results and discussion
The following subsections would elaborate on the quantitative evaluation results of UHI intensity changes between heatwaves and normal periods, analyze the factors influencing the changes of UHI intensity, and explore the accumulation of UHIs in the YRD urban agglomeration.

Enhanced UHIs
Most Chinese cities experienced stronger UHI intensity during heatwaves compared to normal days, particularly in urban areas (figure 3(a)).Overall, the enhancement of UHI intensity during heatwaves reached 0.50 • C in urban areas, with mean UHI intensities during heatwave and non-heatwave periods of 0.94 • C and 0.43 • C, respectively.This coincided with the empirical cumulative distribution function (ECDF) of the UHI intensity shifting toward a larger mean value (figure 3(c)).The upper tail of the ECDF shifted to the right, indicating stronger UHI intensity during heatwaves.This was similar in suburban areas, albeit with a smaller increase of UHI intensities during heatwaves compared to urban areas, at 0.24 • C (figure S4).The intensification of the UHI varied with precipitation conditions.In humid areas where annual mean precipitation larger than 620 mm, the excess UHI intensity during heatwaves was 0.57 • C, while in drier regions, it was 0.46 • C.Moreover, from the latitudinal variation of the enhanced UHI, we found that around 30 • N there was a cluster of high values, which coincided with the location of YRD urban agglomeration.The comparison among three major urban agglomerations showed that YRD suffered severest intensification of UHI during heatwaves, reached 0.84 • C, followed by the PRD (0.51 • C) and the BTH (0.37 • C).Although the spatial gradients of the three urban agglomerations were generally consistent with Jiang et al (2019), which indicated that the UHI intensified most strongly in the YRD (0.90 • C), there were differences in specific quantitative values.This discrepancy was mainly due to the following two reasons.Firstly, Jiang's study was sensitive to the selection of reference rural stations.The selection of coastal stations as reference rural stations would lead to an overestimation of the UHI intensity compared to inland stations because of the influence of sea-land breezes, which would increase the cooling mechanism through interactions between sea breezes and synoptic winds at coastal stations (Founda and Santamouris 2017, Ao et al 2019).Secondly, the study periods were different, with their study period being from 2013 to 2015.
To further evaluate whether the UHI intensification was affected by the selection of normal days, we randomly selected days as long as the duration of the heatwaves to serve as non-heatwave days from the constructed sample population, and recalculated the changes in UHI intensity.This experiment was repeated 100 times.The intensification of UHI during heatwaves were within the range of 0.24 • C-0.26 • C in suburban areas and 0.49 • C-0.53 • C in urban areas (figure S5).These results demonstrated a robust enhanced UHI during heatwaves in Chinese cities.However, the nationwide evaluations in the India (Kumar and Mishra 2019) indicated a decline in the UHI during heatwaves which contrasts with our findings.During the pre-monsoon season (March-May) in India, 60%-70% of cities exhibited the urban cool island effect, attributed to limited availability of moisture and vegetation in the surrounding nonurban areas due to the drought climate and the harvesting of crops by the end of March (Kumar et al 2017).This period coincided with frequent heatwaves in India, the unusually high temperatures further elevated warming in rural areas above urban areas.Therefore, the national-scale assessment results indicated a weakening of the UHI intensity during heatwaves.Compared to studies based on meteorological observations, gridded temperature data could help us obtain continuous air temperature variations along urban-rural gradients, avoiding the influence of selecting reference rural stations.However, some uncertainties may be introduced during the fusion process of satellite and station observations.The complexity of topography (Heynen et al 2016), the distribution density of meteorological stations (Li et al 2018), and the time lag between station-observed T mx and mid-daytime LST (Vancutsem et al 2010) can all affect the accuracy of gridded air temperature estimation.Future works focused on improving the accuracy of gridded temperature datasets can help further deepen understanding of the impacts and adaptation to extreme temperatures at the city scale.

UHI controlling factors
The heatwave-normal period contrasts of UHI intensity in both urban and suburban areas were well correlated with background precipitation (p < 0.05), suggesting that the precipitation played an important role in modulating the intensified UHI during heatwaves (figures 4 and S6).The differences of UHI intensity between heatwaves and normal periods were further compared in high and low precipitation areas, using the median precipitation (620 mm) among all cities as the criterion.Besides the obvious difference in precipitation, the ∆ISA and ∆NDVI were comparable between these two areas, indicating a similar level of built-up and vegetation greenness (figure S7).The median UHI intensities in low precipitation areas were 0.33 • C and 0.72 • C during heatwaves and normal periods, respectively (figure 5(a)), with corresponding values increasing to 0.54 • C and 1.07 • C in high precipitation areas (figure 5(b)).This coincided with recent studies that found the UHI intensity increased with the annual mean precipitation (Manoli et al 2019).In humid regions, rural vegetation would effectively cool the environment through stronger evapotranspiration due to the higher water availability, resulting in a larger urban-rural temperature difference (Zhao et al 2014).In addition to the stronger UHI intensity, the interactions between heatwaves and UHI were also more pronounced in humid areas.This could be further illustrated by the rightward shift of the ECDF tail for changes of UHI intensities during heatwaves in regions with high precipitation (figure 5(c)).We found that the change in the probability of the median value for enhanced UHI intensities during heatwaves between high and low precipitation areas was 0.17.Cities in humid regions was found to have higher urban-rural aerodynamic resistance and lower convective efficiency compared to cities in arid regions (Fitria et al 2019), which might be one of the possible reasons for the more pronounced interaction between heatwaves and UHIs in humid areas.Heatwaves associated with stationary and highpressure weather systems (Ackerman and Knox 2012) would result in lower wind speeds than normal periods, further reducing the advection cooling effect from surrounding rural areas, and intensifying the

UHI agglomeration
The YRD urban agglomeration experienced rapid urbanization during 2003-2018, with sprawling impervious surface extending from original city centers, blurring the division between urban and rural areas (figures 6(a) and (b)).The regional mean ISA fraction increased from 13% to 36%, nearly tripled in extent (figure 6(c)).To further investigate the heat accumulation in denser urbanized areas, we divided the period into two phases: the early period (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011) and the recent period (2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020).The intensified UHIs during heatwaves were observed in both the early period (figures 6(d) and (e)) and the recent period (figures 6(g) and (h)).The comparison of mean UHI intensity between early and recent periods illustrated the accumulation of UHIs during heatwaves in densely urbanized areas.The excess UHI intensity induced by heatwaves in the early and recent periods was 0.27 • C and 0.33 • C, respectively (figures 6(f) and (i)).In addition, the spatial extent of UHIs expanded to the neighboring cities and raised temperature over large regional swaths in 2012-2020, while it only strengthened around the individual cities during 2003-2011.Taking into account the potential impact of heatwave intensity on the UHIs, we further compared the relative intensity of heatwaves between these two periods, which calculated as the temperature anomaly of heatwaves exceeding the historical 90th percentile threshold (figure S8).The Wilcoxon rank-sum test showed that there was no significant difference in the relative heatwave intensity between the two periods, with the p = 0.14.Though our results did not indicate the impact of heatwave intensity on UHI changes, the frequent occurrence of heatwaves against the warming climate would exacerbate the compound effects of heatwaves and UHIs.In addition to heatwaves, the frequency of humid heatwaves was also projected to increase (Domeisen et al 2023), which would make the impacts of urban excessive heat worse.
The UHI can not only influence local climates within the urbanized area but also have an impact the regional climate through the advection of heat.The strengthening of UHI resulted in heat being transferred from the urban canopy layer to the boundary layer, affecting convection and circulation patterns, which could impact surrounding regions (Patz et al 2005, Smith et al 2011).Previous study indicated that the UHI could impact temperatures in an area as great as 2-4 times that of the city itself (Zhou et al 2015).The potentially large urban heat 'dome' formed by the clustering of large urban areas together during heatwaves would accumulate more heat and then be transported to the region beyond.The prevailing wind direction further influenced the transport of heat in densely urbanized clusters.A case study during a heatwave from 8 to 12 July in 2013 in Shanghai indicated that more than 63% of the winds originating from neighboring hot cities resulted in a significant increase in daytime UHI intensity (Jiang et al 2019).A 'heat plume' would form around downwind cities due to the advection of urban heat, leading to significant heating observed 70 km downwind of Chicago city (Cosgrove and Berkelhammer 2018).As the frequency of heatwaves increases and urban expansion continues, combined with the interaction between heatwaves and UHIs, effective mitigation strategy for urban heat would be needed, especially in megapolitan regions.

Conclusion
In this study, we used fused ground and satellite daily maximum air temperature data to evaluate the changes in UHI intensities during heatwaves across 225 major cities in mainland China from 2003 to 2020.We also explored the impacts of climate and surface factors on UHI intensity changes, and analyzed the accumulation of UHIs using the YRD urban agglomeration as an example.Our results showed that most cities experienced stronger UHI during heatwaves compared to normal days.The average difference in UHI intensity between heatwaves and normal periods reached 0.50 • C in urban areas across all selected cities.We found significant sensitivity of the interaction between UHIs and heatwaves to local background precipitation.Under the similar urbanization and vegetation greenness, the excess UHI intensity induced by heatwaves was more pronounced in wet climates than in arid climates.Meanwhile, the intensification of UHIs during heatwaves was stronger in urban agglomeration.The additional urban warming during heatwaves in the YRD urban agglomeration was up to 0.84 • C. The potentially large urban heat 'dome' formed by the clustering of large urban areas during heatwaves may accumulate more heat and then be transported to the surrounding urbanized areas, raising temperatures over large areas.
Heatwaves and their interaction with UHIs will have serious implications for energy use and human health under warming climate.This study emphasized the increasing importance of enhancing awareness and preparedness for heat risk to protect urban populations, especially in large urban agglomerations.Further examination of the mechanisms underlying the relationship between UHIs and heatwaves through high-resolution model simulations and dense station observations is crucial for developing local urban heat mitigation strategies.Additionally, studies investigating the changes in this interaction under future warming and urbanization scenarios, as well as its impacts on human health and energy use, are also needed.

Figure 1 .
Figure 1.Spatial distribution of the 225 selected cities across China.The 620 mm precipitation contour line was derived from the median value of the annual mean precipitation of all cities.

Figure 2 .
Figure 2. The workflow in this study taking Beijing as an example.(a) An example of the extraction of heatwaves.The left panel indicates the probability density distribution function of the 15 d temperature samples surrounding day 142 in 2003-2020.The right panel describes extraction of heatwaves in the year 2019.(b) An illustration of the random sampling method for normal days.(c) A diagram for UHI intensity calculation.

Figure 3 .
Figure 3. Analyses of the difference in UHI intensity between heatwaves and normal days: (a) spatial distribution, (b) latitudinal variation, and (c) empirical cumulative distribution function (ECDF).The shaded area in (b) represents the standard deviation.The boxes with dashed lines in (a) represent the BTH, YRD, and PRD urban agglomerations, from north to south, respectively.

Figure 4 .
Figure 4. Relationship of UHI differences between heatwaves and normal days with surface attributes and background climate factors.The values in the upper triangle represent correlation coefficients.The ' * ' and '-' in the lower triangle indicate a significant and a non-significant correlation at a 95% confidence level.

Figure 5 .
Figure 5.Comparison of UHI intensities in high and low precipitation areas: (a) probability distribution functions of UHI intensity in low precipitation area, (b) the same as (a) but in high precipitation area, and (c) the ECDF of the intensified UHI during heatwaves between two areas.The vertical lines in (a) and (b) denote the median values.The ∆P in (c) denotes the changes in the probability of the median value between the low and the high precipitation area.