Articles
An Updated Estimate of the Urban Heat Island Effect on Observed Local Warming Trends in Mainland China's 45 Urban Stations
2020 Volume 98 Issue 4 Pages 787-799
Details
2020 Volume 98 Issue 4 Pages 787-799
Observed surface air temperature (SAT) warming at urban stations often contains both the signal of global warming and that of local urban heat island (UHI) effects; these signals are difficult to separate. In this study, an urban impact indicator (Uii) developed by the authors was modified to represent the extent to which the observed temperature from a given station was influenced by UHI effects. The Uii of a city was calculated by simplifying the city's shape to a circle. In addition, a modified Uii (MUii) was calculated by considering the realistic horizontal distribution of the urban land. We selected 45 urban stations in mainland China, along with an adjacent station for each to give a station pair. Background climate changes across each pair were near-homogeneous. Thus, differences in the trends of annual averaged daily mean SAT (Trendmean), maximum SAT (Trendmax), and minimum SAT (Trendmin) between the urban and adjacent stations (ΔTrend) could be mainly attributed to differences in MUii changes between the urban and adjacent stations (ΔMUii). Several linear regressions between ΔTrend and ΔMUii for the 45 station pairs were calculated to estimate UHI effects on Trendmean (UTmean), Trendmax (UTmax), and Trendmin (UTmin). The results showed that the mean MUii of the 45 urban stations increased from 0.06 to 0.35 during 1992–2013. Positive correlations between ΔMUii and ΔTrend for the 45 station pairs were significant at the 0.001 significance level (except for Trendmax). The average UTmean and UTmin of the 45 urban stations during 1954–2013 were approximately 0.05 and 0.11°C decade−1, respectively, accounting for 18 % and 31 % of the overall warming trends, respectively. The UTmin estimated in this study is about twice that of previous results based on regression equations between Uii and SAT trends.
In the past few decades, several countries have experienced rapid urbanization and the dramatic growth of urban constructed land, energy consumption, and city populations (Grimm et al. 2008; Parker 2010). Surface energy balances within urbanized areas are often impacted by these changes, resulting in severe urban heat island (UHI) effects (Zhou et al. 2014; Li et al. 2018). In company with large-scale background climate change, this local warming is generally recorded by meteorological stations located in or close to cities (i.e., urban stations). Because few stations in populated regions such as eastern China are free from UHI effects (i.e., rural stations), regional warming is generally overestimated when based on all stations (Jones et al. 2008; Wen et al. 2019). Therefore, it is critical to accurately detect the biases that UHIs contribute to the surface air temperature (SAT) series for urban and suburban stations. This will allow us to better analyze and evaluate large-scale changes in climate.
The effects of UHI on long-term SAT trends are usually investigated by comparing the SAT series of urban stations with those of rural stations (i.e., the UMR method) (Jones et al. 2008; Ge et al. 2013; Jin et al. 2018a). However, the UMR method often contain some uncertainties owing to the varied criteria for station classification, the uneven distribution of stations, and the limited number of rural stations in some regions, e.g., populated areas (Ren et al. 2008; Wang and Ge 2012; Liao et al. 2016). In addition, some studies have estimated UHI effects by subtracting reanalysis data from observation data (i.e., the OMR method) (e.g., Kalnay and Cai 2003). However, the reanalysis data normally retain a degree of uncertainty owing to the varied assimilation systems and datasets used in the process of data generation (Wang and Yan 2016). Meanwhile, reanalysis data often show an obvious bias against observations in some regions, e.g., mountainous areas with complicated topography (Bao and Zhang 2013). The limitations of the UMR and OMR methods can lead to certain divergences in the estimates of UHI effects. For example, Wang and Yan (2015) detected a very slight urban warming in the Beijing–Tianjin–Hebei metropolitan area of China during 1979–2009 based on the ECMWF reanalysis, while contributions of urban warming to overall warming trends were estimated to be 18.77, 51.19, and 25.69 % based on the NCEP/NCAR reanalysis, NCEP/DOE reanalysis, and UMR methods, respectively.
Considering the limitations of the UMR and OMR methods, a few studies have used mathematical methods based on regression models to assess UHI effects. For instance, Karl et al. (1988) and Chung et al. (2004) used a mathematical formula that related urban population to the urban biases of stations to investigate UHI effects in the USA and Korea, respectively. He et al. (2013) developed a regression equation between urbanization rate and temperature trend to detect the urban warming signals in Beijing, Tianjin, and Hebei Province, China. Using regression models based on urban-related indexes can avoid the dependency on rural stations (such as exist in the UMR method). However, the SAT records used in mathematical methods generally include the effects of background climate change, which may vary over large-scale regions and obscure the warming caused by UHI (Pitman et al. 2011; Zhao et al. 2014). Recently, Wang et al. (2017) reduced the signal of background climate change using the OMR method before applying a regression model to estimate urbanization effects in eastern China; nevertheless, the uncertainties of the OMR method (as mentioned above) may have remained in that study. Reducing the impact of background climate change on the assessment of UHI effects before applying a mathematical method to estimate urbanization effects remains a challenge.
Previous studies based on the mathematical method have rarely taken specific determinants of UHI into consideration over large-scale regions. It has been acknowledged that UHI intensity is significantly correlated with urban area or fraction (Yuan and Bauer 2007), population density (Karl et al. 1988), surface albedo (Peng et al. 2012), and vegetation index (Zhou et al. 2014). However, population information does not objectively and precisely reflect spatial changes in the UHI-related environments of surrounding stations (Peterson and Owen 2005). Surface albedo and vegetation index are not suitable for large-scale studies because they vary according to land cover, climatic conditions, and geographical environment (Zhai et al. 2014; Jin et al. 2018b). In addition, it has been reported that land surface winds can influence the horizontal distribution of the UHI (Du et al. 2016), and that UHI intensity is negatively correlated with city–observation site distance (Knight et al. 2010). Taking into account urban area, station location relative to a city, and the impact of wind direction on UHI distribution, Jin et al. (2015) proposed a composite index (urban impact indicator, hereinafter Uii) to reflect the extent of UHI effects on the observed temperature at urban stations. The Uii can provide more information about UHI effects than can the simple parameters, such as population, used in previous studies. However, the effects of UHI on the SAT trends estimated by Jin et al. (2015), based on the regression models between SAT trends and Uii, were lower than many previous estimates for urban stations (e.g., Wang and Ge 2012). Two flaws in the method of Jin et al. (2015) may contribute to this underestimation: (i) the local background climate change mentioned above was not removed from the SAT trends and (ii) the city shape was assumed to be circular and include all urbanized lands when calculating the Uii. As a result, the calculated Uii based on this assumption may have included a large bias against the actual conditions when reflecting UHI effects for those stations close to non-circular cities (e.g., strip cities). Therefore, developing a more effective index that reflects the impacts of the UHI on station data is very important prior to applying the abovementioned mathematical methods.
Owing to a severe lack of rural stations in populated regions and the relatively low reliability of reanalysis data in mountainous regions in China, the UMR and OMR methods may be unsuitable for examining the effects of UHI on temperature change in these regions. Therefore, we attempted to advance knowledge in this field on the basis of the mathematical method of Jin et al. (2015) by solving its two abovementioned flaws. To this end, a modified Uii (MUii) was calculated based on the objective distribution of urban lands surrounding selected urban and adjacent stations to better reflect the impact of UHI on these stations. The difference in UHI effect between an urban station and its adjacent station, which had similar background climate change effects, was calculated by comparing differences in their SAT trends. Observed differences were thus attributed to their difference in the effects of urbanization. Finally, we explored the effects of UHI on temperature changes during 1954–2013 at 45 urban stations in mainland China.
Urban population data from China's primary cities between 1992 and 2013 were compiled from the China City Statistical Yearbook of the National Bureau of Statistics of China (http://www.stats.gov.cn). DMSP-OLS nighttime stable light (NSL) data from 1992 and 2013 were used to extract urban constructed lands (Liu et al. 2012; Jin et al. 2018b). NSL data with a spatial resolution of 1 km were collected by the US Air Force Weather Agency and obtained from the NOAA's National Geophysical Data Center (https://ngdc.noaa.gov/eog/download.html). The NSL values, which indicate nighttime light intensities, ranged from 0 to 63.
Climate data were obtained from the China Meteorological Data Service Center (http://data.cma.cn/site/index.html). Daily wind directions during 1992–2013 were used in this study. Monthly mean air temperature data were derived from the China Homogenized Historical Temperature (CHHT) (1951–2004) dataset, which includes 731 meteorological stations. The CHHT dataset is homogenized for the purposes of quality control, i.e., in relation to inhomogeneities mainly caused by station relocation. This homogenization is conducted using the Easterling–Peterson (EP) technique and by considering actual Chinese meteorological observations (Li et al. 2004). Previous studies have demonstrated that temperature series before and after homogeneity adjustment can show contrasting trends due to station relocation (Cao et al. 2016). Given the impact of the noncontinuity of temperature series caused by station relocation on analysis results, it is necessary to study the effects of UHI on temperature trends using homogenized temperature data (Jones et al. 2008). Inhomogeneous discontinuities in the temperature series of a target station were generally adjusted based on data from its neighboring station(s) (Li et al. 2009). The inter-annual variations in the adjusted temperature series of a target station before and after discontinuity adjustment were rarely affected by the data from neighboring station(s). Therefore, the adjusted temperature series of a target station and the temperature series of its neighboring station(s) were considered independent. The detailed processes relating to the adjustment of inhomogeneous discontinuities are described in the study of Li et al. (2004). The CHHT dataset, which is highly reliable, has been frequently used to study the effects of UHI on temperature changes (Li et al. 2009, 2010). Following the study by Ren and Zhou (2014), we updated the CHHT data to 2013 using historical SAT records. Then, the updated data were adjusted for inhomogeneities caused by station relocation after 2004 following the study of Li et al. (2004). A detailed explanation of this procedure is shown in the study of Jin et al. (2018a). Finally, temporal trends in annual averaged daily mean SAT (Trendmean), maximum SAT (Trendmax), and minimum SAT (Trendmin) were calculated to assess SAT changes.
2.2 Methods a. Selection of meteorological stationsIn this study, an adjacent station was selected for each urban station to give a station pair. The purpose of this pairing was to compare the difference in SAT trends and MUii between the urban station and its adjacent station (rather than applying the UMR method). Therefore, the selection of adjacent stations was not limited to rural stations. Moreover, it was considered critically important that the background climatic parameters at the urban station and its adjacent station should be very similar to allow comparison analysis. Therefore, the distance and the difference in altitude between an urban station and its adjacent station were required to be within reasonable ranges. To this end, when applying the UMR method, Wang et al. (2015) selected 194 rural stations and 413 urban stations for which the average distance between the urban and rural stations was 145 km; Hua et al. (2008) selected 190 station pairs (consisting of an urban station and a rural station) for which the largest distance between urban and adjacent stations exceeding 200 km; and Gallo and Owen (1999) only analyzed those cities with meteorological stations that exhibited elevation differences of less than 500 m to eliminate topographical effects. Based on these studies, we selected meteorological stations in mainland China according to the following criteria.
We initially selected the meteorological station nearest to each city with an urban population of more than 1 million in 2013 (i.e., urban station). Then, 61 urban stations for which the climate records covered the period 1954–2013 were selected from more than 700 national meteorological stations in mainland China. Next, we selected a meteorological station near each urban station and defined it as the adjacent station. The adjacent stations were selected from the remaining national meteorological stations, excluding the selected urban stations. The difference in altitude between an urban station and its adjacent station was less than 500 m and the distance between them was less than 140 km. These requirements were relatively strict compared with the previous studies mentioned above. The background climate change was expected to be nearly homogenous for a given station pair. The climate records from the adjacent stations also covered the period 1954–2013. Finally, 45 station pairs (45 urban stations and 42 adjacent stations) were selected (Fig. 1). The remaining 16 urban stations without an adjacent station were not considered further in this study. For the 45 station pairs, the mean difference in altitude between the urban station and adjacent station was 105 m and their mean distance was 102 km (Supplement 1). Further information on the selected meteorological stations, such as the distances from each station to the nearest water body and the coast can be found in Supplement 1. Because of the limited number of meteorological stations, three adjacent stations were each shared by two urban stations.
Location of the 45 selected station pairs in mainland China. Nighttime light intensity indicates the value of nighttime stable light data, which ranges from 0 to 63.
Before calculating the Uii, Jin et al. (2015) assumed that all urban lands in a city were concentrated in a circle. Taking a meteorological station located in the simplified city circle as an example, the impact of urban lands on the station along a given wind direction was calculated by multiplying the distance from the city boundary to the station along this wind direction by the corresponding wind frequency. Then, Uii was calculated by standardizing and summing all of the impacts of urban lands along the different wind directions (detailed processes can be found in the study of Jin et al. (2015)). However, we found that the shapes of over half of the cities near the 45 selected urban stations were obviously not circular due to topographical and environmental factors, especially in the mountainous and coastal areas of China (Supplement 2). Therefore, the calculation of Uii was improved by weighting the areas of urban lands in different buffer areas surrounding a given station. The procedures used to calculate the MUii were as follows.
First, urban constructed lands for 1992 and 2013 were automatically extracted from the NSL images by the empirical threshold technique (Elvidge et al. 1997). Due to the “light diffuse” phenomena in densely populated areas (Huang et al. 2014), different thresholds for extracting urban constructed lands should be adopted according to the regional economic or urbanization levels (Liu et al. 2012). Based on the study of Jin et al. (2018a), the extraction threshold for cities with a population of less than 2 million was determined to be 45; otherwise, it was determined to be 50 in the present study.
Second, urban land areas in different subzones around each station were measured. Wang and Ge (2012) studied the correlations between temperature trend and urban area change at 5-km increments around stations and found high correlation coefficients at 10–20 km. To cover this area, we drew three circular buffer areas around each station, with the meteorological station as the center and distances of 10, 20, and 30 km as the radii (Li et al. 2013) (Fig. 2). The three buffer areas, 0–10, 10–20, and 20–30 km, were denoted as a, b, and c, respectively. Then, we overlaid the boundaries of 16 wind directions (denoted by i, equaling 1, 2, …, 16) on the three buffer areas. The 30-km-radius circular buffer area (hereafter 30-km area) around each station was divided into 48 subzones (denoted as ai, bi, and ci) (Fig. 2). The locations of subzones a1, b1, c1, a5, b5, and c5 are shown in Fig. 2 as examples. The area of urban constructed land in each subzone (Sai, Sbi, and Sci) was measured by ArcGIS 10.2 software.
Sketch of the urban constructed lands in the 48 subzones around Zhengzhou station in 2013. Whole numbers from 1 to 16 represent different wind directions; a1, b1, and c1 indicate the three subzones located along the first wind direction (inner to outer); a5, b5, and c5 indicate the three subzones located along the fifth wind direction (inner to outer). The latitude and longitude of Zhengzhou station are 34°25′48″N and 113°23′24″E, respectively.
Third, using the wind frequency (Fi) and the distance from the station to the subzone (da, db, or dc) as weight coefficients, the impact of the UHI on each selected station (K) was quantified by weighting all urban constructed lands for the station. Based on the study of Jin et al. (2015), the wind direction with the maximum speed on a particular day was regarded as the dominant wind direction of that day. The frequency of a given dominant wind direction was represented by the ratio of the number of days it appeared to the total number of days during 1992–2013. Finally, the MUiis of 1992 and 2013 were calculated by standardization. MUii was defined as the extent of the UHI effects on the urban stations owing to winds traversing urban areas within a certain distance (Jin et al. 2015). The equations for calculating MUii were as follows.
 |
 |
 |
Considering the limitations of the OMR and UMR methods, this study adopted a mathematical method based on a regression model to estimate the impacts of UHI on observed SAT trends. Initially, we assumed that the MUii of each station in 1954 was extremely small and could be ignored. Taking Guangzhou station as an example, the MUii in 1954 was estimated to be 0.01, which is very small when compared with the MUiis in 1992 (i.e., 0.33) and 2013 (i.e., 1.00), indicating that the abovementioned assumption was reasonable (details in Supplement 3). Thus, the magnitude of MUii change during 1954–2013 was approximately equal to the MUii value in 2013. The difference in the magnitude of MUii changes between urban and adjacent stations (ΔMUii) was calculated using Eq. (4). Next, the differences in Trendmax (ΔTrendmax), Trendmean(ΔTrendmean), and Trendmin (ΔTrendmin) between urban and adjacent stations during 1954–2013 were calculated using Eq. (5).
 |
 |
Differences in SAT trends between urban and adjacent stations (ΔTrend) are primarily caused by the different UHI effects on urban and adjacent stations. Therefore, ΔTrend is likely to be proportional to ΔMUii. Next, we calculated three linear regression equations relating ΔMUii to ΔTrendmax, ΔTrendmean, and ΔTrendmin based on the 45 station pairs. Based on the calculated linear regression coefficients and the MUiis of the 45 selected urban stations in 2013, UHI effects on Trendmax (UTmax), Trendmean (UTmean), and Trendmin (UTmin) during 1954–2013 were estimated. Then, the contributions of UHI effects to overall warming trends were calculated based on the ratios of UHI warming to the overall warming trends (Wang and Ge 2012).
On average, the Trendmean, Trendmax, and Trendmin of the 45 urban stations during 1954–2013 were 0.28, 0.19, and 0.35°C decade−1, respectively (Table 1). By comparing the differences in SAT trends between urban and adjacent stations, the average ΔTrendmean, ΔTrendmax, and ΔTrendmin of the 45 station pairs were calculated as 0.08, 0.03, and 0.10°C decade−1, respectively. This indicates that urban stations have generally undergone faster warming than have adjacent stations.
The mean MUii of the 45 urban stations increased from 0.06 to 0.35 during 1992–2013 (Table 1). However, there were large differences in MUii values among the 45 urban stations in 2013. In 2013, the MUiis of 17 urban stations were lower than 0.20, whereas nine urban stations had MUiis of greater than 0.60 (Fig. 3a). Urban stations with relatively large MUiis in 2013 were mainly located in eastern China. In contrast, the 42 adjacent stations showed small differences in MUii in 2013 (Fig. 3b), and their mean MUii increased from 0.01 to 0.05 in the past 22 years (Table 1). The average ΔMUii of the 45 station pairs was 0.30, indicating that urban stations showed more severe UHI effects than adjacent stations. This is consistent with the observation that urban stations show faster warming than adjacent stations. However, the ΔMUiis were markedly different among the 45 station pairs (Fig. 3c). The standard deviation of the ΔMUii was 0.25. Some station pairs located in the same city showed large differences in MUii. For example, the ΔMUii between the Tianjin and Tanggu stations was > 0.4 despite the fact that they were both located in Tianjin City.
Distribution of MUii in 2013 for (a) the 45 urban stations (MUiiurban, 2013) and (b) the 42 adjacent stations (MUiiadjacent, 2013) in mainland China, and (c) the difference in MUii between urban and adjacent stations (ΔMUii).
Figure 4 shows that the positive correlation between ΔMUii and ΔTrendmean (ΔTrendmin) is significant at the 0.001 significance level based on the 45 selected station pairs. The linear regression coefficients between ΔMUii and ΔTrendmean and between ΔMUii and ΔTrendmin were 0.139°C decade−1 and 0.319 °C decade−1, respectively. The correlation coefficients (r) between ΔMUii and ΔTrendmean and between ΔMUii and ΔTrendmin were 0.503 and 0.639, respectively. The degree of fit of these equations was higher than that for the regression equations between Uii and the SAT trend that contained background climate change (e.g., the r value of the regression equation between Uii and Trendmin was 0.308), as calculated in Jin et al. (2015). Moreover, the correlation between ΔMUii and ΔTrendmax was relatively poor (r = 0.314, P < 0.05), indicating that environmental changes related to urbanization may have had only a weak influence on the warming of Trendmax.
Linear correlations between ΔTrend and ΔMUii of the 45 station pairs. ΔMUii indicates the difference in the magnitude of MUii changes during 1954–2013 between urban and adjacent stations. ΔTrendmax indicates the difference in annual averaged daily maximum temperature trends between urban and adjacent stations during 1954–2013. ΔTrendmean and ΔTrendmin are same as ΔTrendmax but for annual averaged daily mean and minimum temperature, respectively.
Additionally, we analyzed the relationships between MUii and Trendmean and between MUii and Trendmin (Fig. 5). The r value of the regression equation between MUii and Trendmean (Trendmin) was 0.447 (0.599), which was less than that between ΔMUii and ΔTrendmean (ΔTrendmin). This indicates that the fitted degree of the regression equation between ΔMUii and ΔTrendmean (ΔTrendmin) was greater than that between MUii and Trendmean (Trendmin).
Linear correlations between MUiis and SAT trends of the 45 selected urban stations. Red and blue dots indicate analyses for the annual averaged daily minimum and mean temperature trends, respectively.
Based on the mean MUii of the 45 urban stations in 2013 (i.e., 0.35) and linear regression coefficient between ΔTrendmean and ΔMUii, the mean UTmean was calculated to be approximately 0.05°C decade−1 during 1954–2013. Similarly, the mean UTmin of the 45 urban stations during 1954–2013 was calculated to be approximately 0.11°C decade−1. The mean contributions of the UHI effects to the overall warming trends of the 45 urban stations during 1954–2013 were estimated to be 18 % and 31 % for Trendmean and Trendmin, respectively. In contrast, based on the equations developed for the relationship between Uii and the SAT trend, Jin et al. (2015) found that the mean UTmean and UTminwere 0.046°C decade−1 and 0.054°C decade−1, respectively, during 1955–2012. The UTmin estimated in this study is about twice that reported by Jin et al. (2015).
In addition, we calculated UTmean and UTmin of the 45 urban stations during 1954–2013 according to their MUiis from 2013. There was a marked difference in the warming induced by the UHI among the 45 urban stations, especially for UTmin (see Supplement 4 and Fig. 6). The UTmin of 13 urban stations was less than 0.05°C decade−1, whereas nine urban stations had a UTmin greater than 0.20°C decade−1. Urban stations with relatively large UTmean and UTmin were mainly located in the metropolitan areas of eastern China where the economy was very strong. The largest UTmean and UTmin values during 1954–2013, calculated for the Guangzhou station, were 0.14°C decade−1 and 0.32°C decade−1, respectively. The Yibin station showed the smallest UTmean and UTmin values, both of which were close to zero.
Distribution of the impacts of UHI on the annual averaged daily (a) mean (UTmean) and (b) minimum (UTmin) temperature trends for the 45 urban stations during 1954–2013 in mainland China.
The difference in urban warming between urban stations was generally attributed to their various surrounding urbanization levels (Yao et al. 2017). However, we found that some station pairs located in the same city still showed large differences in MUii (Fig. 3). For example, the ΔMUii between the Tianjin and Tanggu stations was > 0.4 in 2013 despite the fact that they are both located in Tianjin City. Similarly, Jin et al. (2017) found that the warming induced by the UHI differed markedly between two urban stations located in the east and west area of Changsha City, China. Moreover, Zhao and Wu (2017) found that urban warming varied in different areas of Beijing City, China. These findings are related to the fact that some cities in China cover a very large area. Therefore, the consideration of UHI intensity on the basis of simple parameters only, such as population, makes it difficult to distinguish the extent of UHI effects on stations located in the same city (Karl et al. 1988). Although impervious surfaces and urban fractions are better parameters than is population for quantifying UHI intensity, the influence of wind direction on the location of UHI is often ignored (Wang et al. 2017; Li et al. 2018). Gedzelman et al. (2003) demonstrated that sea breezes commonly displace New York City's UHI approximately 10 km downwind. Therefore, the impacts of UHI on urban stations are determined not only by UHI intensity but also by UHI distribution and the relative locations of stations (Oke 1982; Knight et al. 2010).
In the study of Jin et al. (2015), Uii was calculated based on the assumption that all urbanized lands within a city are concentrated in a circle. However, in China, larger cities generally have more surrounding satellite towns than do small cities. This implies that urban constructed lands in the suburban areas of a larger city would cover a greater area than those in smaller cities (Jin et al. 2018a). Therefore, by using Uii, Jin et al. (2015) may have underestimated the impact of UHI on urban stations near large cities despite taking into account urban area, station location, and wind. When compared with previous indicators associated with UHI intensity, such as population and Uii, MUii is likely to be more accurate in reflecting UHI effects on stations.
This study also employed regression equations between ΔMUii and ΔTrend for the 45 station pairs to explore urban warming signals in the records of urban stations. This approach contrasts with that of Jin et al. (2015) who adopted equations relating Uii and SAT trend. In China, the warming trend has been more pronounced in northern regions than in southern regions during recent decades (Cao et al. 2016). Therefore, the SAT trends of stations with a slight UHI impact in northern China are even higher than those of stations with a severe UHI impact in southern China (Ge et al. 2013). Regression equations that directly relate Uii or MUii to the SAT trends of these stations would show a smaller regression coefficient, resulting in the lower estimate reported in the study of Jin et al. (2015).
Overall, the use of different urban impact indicators and different regression approaches could lead to a large divergence in the results. Taking 29 urban stations that appeared in both the present study and that of Jin et al. as an example (Supplement 4), the averaged UTmin was approximately 0.12°C decade−1 during 1954–2013 in the present study, which was markedly greater than the estimations given by Jin et al. (2015) (i.e., 0.06°C decade−1 during 1953–2012). The same problem exists in other studies based on the mathematical method. For instance, Karl et al. (1988) revealed that the urban bias was 0.06°C in the United States during the twentieth century, which is much smaller than the result reported by Kalnay and Cai (2003) based on the OMR method (i.e., 0.27°C). Therefore, subtracting large-scale climate change from the SAT trends before building the regression models has reduced the uncertainties to some extent in our study.
In this study, the UTmean of the 45 urban stations varied from zero to 0.14°C decade−1 during 1954–2013. This is consistent with the work of Ren et al. (2008) who demonstrated that urban warming increased from 0.07°C decade−1 in a small city station group to 0.16°C decade−1 in a large city station group in north China during 1961–2000 using the UMR method. Moreover, UTmean was 0.11°C decade−1 at the Tianjin station in this study (Supplement 4). These results were consistent with the studies of Wang and Yan (2015) based on the UMR method and Wang et al. (2017) based on a regression model. The maximum UTmean (i.e., 0.14°C decade−1) was found at the Guangzhou station in the present study. Based on the UMR method, Xiong et al. (2010) suggested that urban warming in Guangzhou City was 0.12 °C decade−1 during 1978–2007. However, we noted that one of the two rural stations selected by Xiong et al. (2010) was located in Zengcheng City, which may be influenced by the UHI. It is well known that selecting rural stations with insufficient representativeness due to urbanization may underestimate UHI warming (Jones et al. 2008; Ren and Ren 2011; Ren et al. 2015). Therefore, the result of Xiong et al. (2010) represents a relatively low estimate. Moreover, the result of Yang et al. (2011) showed a UHI warming of 0.398°C decade−1 during 1981–2007 for metropolis city stations in east China, which is even larger than our estimated maximum. Apart from the difference in study period, this divergence may be related to the quality of reanalysis data used by Yang et al. (2011); these data normally contain systematic biases in the multi-decadal variability of SAT (Wang and Yan 2016). Overall, uncertainties in selecting rural stations and using reanalysis data may lead to divergent results (Jones et al. 2008; Wang and Ge 2012; Ren and Zhou 2014; Liao et al. 2016). The current study, based on the regression equations between ΔMUii and ΔTrend, avoids the major limitations of the UMR and OMR methods, which could advance the assessment of UHI effects on temperature change.
It is reported that climatic conditions and topography can influence UHI intensities (Kassomenos and Katsoulis 2006; Zhou et al. 2016). Although the present study adopted stricter criteria for the selection of station pairs compared with some previous studies, the different distances and elevations between urban and adjacent stations may have still led to some uncertainty in the results (He et al. 2013). In order to understand the effect of topography on the results, we built a linear regression equation between ΔTrendmean (ΔTrendmin) and ΔMUii for 32 station pairs that exhibited less than 100-m elevation differences. The average UTmean and UTmin of the 45 urban stations during 1954–2013 based on the abovementioned equations were calculated to be 0.051°C decade−1 and 0.103°C decade−1, respectively. These are very close to the corresponding results mentioned in Section 3.3 (details in Supplement 5). Moreover, because this study focuses on 45 urban stations in a large-scale region, more detailed factors associated with UHI, such as building density and distance from station to water (ocean), were not considered in the calculation of MUii. Recently, Theeuwes et al. (2017) derived a diagnostic equation for the daily maximum UHI by considering solar radiation, diurnal temperature range, wind speed, and vertical temperature gradient; this could be used as a reference for future study.
In this study, an urban impact indicator (Uii), which combines the urban land area, wind directions, and station locations, was modified to represent the extent of UHI effect on meteorological stations. The modified Uii (MUii) was considered likely to be more accurate than Uii because it considered the realistic horizontal distribution of the urban lands, whereas Uii is calculated by simplifying the shape of a city to a circle. Moreover, in order to reduce the impact of background climate change, we built several linear regression equations between ΔMUii and ΔTrend to explore the impacts of UHI effects on SAT trends for 45 urban stations in mainland China. The main findings are as follows:
Supplement 1 shows information for the 45 selected urban stations and their adjacent stations in mainland China. Supplement 2 shows aerial views of the non-circular cities located near the selected urban stations. Supplement 3 shows validation of the assumption that the MUii of each station in 1954 could be ignored. Supplement 4 shows the effect of UHI on SAT trends of the 45 selected urban stations during 1954–2013 in mainland China. Supplement 5 shows an evaluation of the influence of topography on the results of the UHI effects.
The study was financially supported by the National Nature Science Foundation of China [41771558], the National Key Research and Development Program of China [2016YFC0501707], the External Cooperation Program of BIC, Chinese Academy of Sciences [16146KYSB20150001], and the European Commission Program Horizon 2020 project [635750]. We thank Professor Deliang Chen of the University of Gothenburg for his comments and suggestions on the paper.
