Abstract

The impact of urbanization on temperature trends in China was investigated with emphasis on two aspects of urbanization, land cover change, and human activity. A new station classification scheme was developed to incorporate these two aspects by utilizing land cover and energy consumption data. Observation temperature data of 274 stations and National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis temperature from 1979 to 2010 were used in conducting the observation minus reanalysis (OMR) method to detect urban influence. Results indicated that nearly half of the stations in the study area have been converted from nonurban to urban stations as a result of land cover change associated with urban expansion. It was determined that both land cover change and human activity play important roles in temperature change and contribute to the observed warming, particularly in urbanized stations, where the highest amount of warming was detected. Urbanized stations showed higher OMR temperature trends than those of unchanged stations. In addition, a statistically significant positive relationship was detected between human activity and temperature trends, which suggests that the observed warming is closely related to the intensity and spatial extent of human activity. In fact, the urbanization effect is strongly affected by specific characteristics of urbanization in local and regional scales.

1. Introduction

Urbanization is a very important anthropogenic factor in local and regional climate change and has been extensively studied. The urban heat island (UHI) induced by urbanization warms the air temperature in urban areas relative to the surrounding rural areas. The related warming strongly affects local and regional temperature changes while the magnitude is small from a global perspective. Several recent studies have examined the significant impact of the UHI on temperature in urban areas from a static perspective by analyzing temperature differences between urban and nonurban areas with surface temperatures derived from remote sensing (Peng et al. 2012; Ren and Ren 2011; Xiao and Weng 2007) and numerical simulations (Lin et al. 2007). Because many meteorological stations are located in or near cities, it is highly possible that UHI would introduce excessive warming signals to the observation temperature. However, the extent to which urbanization can impact local temperature change remains unclear. It is a rather complicated issue and usually leads to ambiguous and controversial conclusions. Some studies have determined that the urbanization impact on temperature trends is not obvious and that no significant differences appear between urban and rural stations (Jones et al. 2008; Jones et al. 1990; Li et al. 2004b; Peterson et al. 1999). However, other studies argued that urban warming in the observation records is significant and detectable (Englehart and Douglas 2003; Kalnay and Cai 2003; Karl et al. 1988; Ren et al. 2008). Many factors are responsible for the existing discrepancies of the urbanization impact including data inhomogeneity, spatial variations, analysis methods, study period, proxies of urbanization, and criteria for defining urban and rural stations.

Approaches that compare the temperatures between urban and rural stations and attribute their differences to the urbanization impact are the most popular in UHI studies. The underlying assumption is that the urban impact is negligible in rural areas. However, this assumption has generally not been proven under actual conditions. It is difficult to find stations that are completely free from the urbanization impact, particularly in regions under rapid urbanization. Since alternative approaches are limited, the urban-minus-rural approach is still preferred in many studies. In this case, the designation of urban and rural stations is of primary importance in estimating the urbanization influence. Population and night-light data are frequently used as proxies of urbanization to facilitate the classification of urban and rural stations. However, such proxies have been recently criticized (Hale et al. 2008; Ren and Ren 2011). Population varies significantly in regions with different economic conditions and cultures. In addition, the population criteria are usually subjective and outdated in the classification. Moreover, the definitions of urban and rural differ considerably among countries. For example, the population criteria for rural stations in China are comparable or even larger than that for urban criteria in the United States (Englehart and Douglas 2003; Karl et al. 1988; Li et al. 2010; Peterson et al. 1999; Ren et al. 2008). As a measure of urbanization, such inconsistences of criteria based on population are apparently problematic. In the 1990s, as an improvement to population, satellite nightlights data were introduced to represent urbanization (Peterson 2003; Peterson et al. 1999). Although such data can provide information on the urbanization process (Yang et al. 2011), problems similar to those reported for the population criterion have been reported. Light intensity varies significantly among regions in various stages of economic development and depends highly on the way of usage (Peterson et al. 1999; Ren and Ren 2011). To overcome such limitations, some researchers have attempted to use combined data from various sources to improve the classification results. The combined use of nightlight and population data, in addition to land use and land cover (LULC) type (Gallo et al. 1999; Peterson et al. 1999); surface brightness temperature and Google Earth imagery (Ren and Ren 2011); spatial land use and population gridded data (Fujibe 2009); and satellite data such as Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper (ETM) (He et al. 2007) have provided more information for station classification. However, to better characterize the urbanization process and to develop a more suitable station classification scheme, researchers should concentrate on the mechanisms of urbanization influence on temperature change.

In general, the urbanization impact on temperature depends on two main aspects including changes in land surface properties, such as changes in surface albedo, roughness, and heat capacity, and human activity within a city, which releases anthropogenic heat and other pollutants. Land cover change modifies existing components of energy balance while human activity is considered a new energy source that changes the surface energy balance, thus affecting surface temperature. Both aspects play important roles in the urban climate and should not be neglected. It has been proved that LULC changes near a station likely to increase air temperature and intensify UHI (Hale et al. 2008, 2006; He et al. 2007). However, LULC changes are poorly reflected in the frequently used population or nightlight data. Although such data may be correlated to the factors of urbanization that affect temperature, neither are direct measures of land cover change associated with urbanization (Hale et al. 2008). Further, the normal classification schemes are relatively static and fail to characterize the evolving urbanization process. In fact, urbanization is a dynamic process that involves urban expansion and landscape changes. It is highly likely that several sites previously defined as rural have been transformed to urban stations during the study period. These urbanized rural stations are not true rural that receive urban impacts and exhibit strong warming. Such bias in station designation may lead to overestimation of regional averaged temperature trends, which has attracted substantial attention (Peterson et al. 1999; Ren et al. 2008).

In addition to LULC change, human activity within a city is an important aspect of urbanization. Human activity serves as major anthropogenic radiative forcings in urban areas through the release of excessive heat, greenhouse gases, and other pollutants. Anthropogenic heat released from energy consumption aggravates the formation of UHIs (Fan and Sailor 2005; Ichinose et al. 1999; Narumi et al. 2009) and thus affects temperature change. However, population does not have a direct impact on UHI intensity as do other factors related to energy release or fractions of various land cover. (Böhm 1998). Urban warming is more closely related to internal changes such as business activity and vehicle numbers (Fujibe 2009). Moreover, due to the burning of fossil fuels in industrial processes and the transportation sector, urban regions are major producers of greenhouse gas emissions and are thus potential contributors to global warming (Grimmond 2007). These facts underline the importance of the human activity effect in local temperature change.

In this study, we developed a new classification scheme to address the aforementioned issues, and we examined the urbanization impact on temperature trends. The classification scheme is designed to focus on the two major processes related to urbanization, changes in land surface and human activity. China was selected as the study region because the country has undergone fast economic growth and rapid urbanization processes since the 1980s, making it an ideal region for analyzing the relationship between urbanization and temperature trends. The main objective of this paper is to examine the impacts and extent of land cover change and human activity in urban areas on temperature trends. In addition, we demonstrated that the station conversion occurring in China was a consequence of urbanization, and we explored its possible influence on the analysis.

2. Methodology

The station classification scheme is based on the recognition that urbanization effect on temperature mainly results from the physical changes of land surface and the human activity within a city. These two processes are represented by land cover and energy consumption data, respectively. Energy consumption data are preferred over other data as a proxy for human activity because energy consumption is a direct source of anthropogenic heat and is a primary emission source of greenhouse gas. In addition, energy consumption has similar role as population if treated as an indirect proxy of urbanization.

In this study, we used the observation minus reanalysis (OMR) method proposed by Kalnay and Cai (2003) to investigate the urbanization impact, which takes advantage of different features of observation and reanalysis temperature data. In National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis data, little or no observation data are used in the data assimilation process. The surface temperature in reanalysis is estimated from upper-atmospheric layers, which enables the reanalysis temperature to capture the atmospheric variation signal; however, these data are insensitive to land surface processes. In contrast, surface observation temperature well preserves information of all radiative forcings. Therefore, the temperature differences between observation and reanalysis data could be attributed the impacts of local factors such as urbanization and land use changes because OMR removes the signals of large-scale temperature variations induced by greenhouse gases (Cai and Kalnay 2004) while retaining the local signals. This OMR method has been widely used to estimate the urbanization influence on temperature change in a number of studies (Kalnay and Cai 2003; Yang et al. 2011; Zhou et al. 2004).

a. Data

The observation temperature data were derived from a surface monthly temperature data product of the Chinese meteorological data sharing system (http://cdc.cma.gov.cn/), which includes climate data of 752 meteorological stations over mainland China. The height for surface air temperature measurement has been 1.5 m above ground since 1957 (Liu and Li 2003). Quality control procedures were conducted by the data provider and include the checking of continuous erroneous data, temporal and spatial checking of outliers, and manual advanced identification of continuous suspicious data and outliers. Further details on the processing methods have been reported by Ren et al. (2007). We selected stations with complete records of surface monthly mean temperature from January 1979 to December 2010 with site elevations of less than 500 m MSL. Noting that six stations with elevations lower than 500 m in Xinjiang province, northwestern China was removed in the analysis for spatial consistency. Stations with land area lower than 50% within a 30-km-radius circular zone of each station were also excluded to avoid ocean influence. Therefore, 274 stations meeting these conditions were used for analysis. These stations are distributed mainly in eastern and southern China where more than 90% of the Chinese population resides. For reanalysis data, we used NCEP–NCAR reanalysis (hereafter reanalysis) 2-m monthly mean temperature (www.esrl.noaa.gov/psd/data/reanalysis/reanalysis.shtml) from January 1979 to December 2010, which has an approximate spatial resolution of 1.9°. Reanalysis temperature was linearly interpolated to each station site. Temperature anomalies of observation and reanalysis were then produced by removing their monthly mean annual cycles of the entire period to reduce the systematic differences between them owing to deficiencies of reanalysis in the model forecast or the method of assimilation, particularly near the surface (Kalnay and Cai 2003). The constraint of station elevation lower than 500 m reduced the interpolation error induced by different surface heights between reanalysis and those of the actual locations (Kalnay and Cai 2003). Such measures were necessary to ensure that the reanalysis temperature is comparable to observation. The mean correlation coefficients between observation and reanalysis temperature anomalies were 0.84. The good agreement between these two data demonstrates that the reanalysis temperature effectively captured the interannual variability of the observations; therefore, the quality of the reanalysis temperature was satisfactory for performing the OMR method. Linear temperature trends from 1979 to 2010 of each station for observation, reanalysis, and OMR were estimated by the least squares method.

Land cover data used in China were extracted from two global land cover datasets. The first is the Advanced Very High Resolution Radiometer (AVHRR) 1-km global land classification data obtained from the Global Land Cover Facility (http://glcf.umd.edu/data/landcover/). These data have a record length from 1981 to 1994, which represents the general conditions of land cover in the 1980s. The second is the 1-km Moderate Resolution Imaging Spectroradiometer (MODIS) global land cover data of 2003 (University of Maryland scheme, http://webmap.ornl.gov/wcsdown/wcsdown.jsp?dg_id=10004_12). Although both land cover datasets consist of 14 land cover types, only urban land cover was extracted. In addition, we used the 1-km spatial energy consumption data of China in 2003 developed by Li and Zhao (2012). These data are the total energy consumed by sectors including agriculture, industry, power production and supply, construction, transportation, storage, postal and telecommunication services as well as residential consumption and others. Energy consumption from various sources such as coal, oil, natural gas, and electricity was converted into tons of standard coal equivalent (tce).

b. Station classification scheme

On the basis of Li and Zhao's (2012) previous work, a new station classification scheme was developed by utilizing the land cover and energy consumption gridded data described above. Figure 1 explains the roles of land cover and energy consumption data in the classification scheme. As reported by Li and Zhao (2012) (and shown in Fig. A1), energy consumption has an overall positive relationship with OMR trends, which suggests the importance of human activity in analyzing temperature change in urban areas. Therefore, we used the threshold technique reported by De Laat and Maurellis (2006) to determine an effective energy consumption value above which human activity would have an apparent influence on temperature trends. According to threshold results, pixels with energy consumption higher than 700 tce were defined as effective energy consumption pixels (EECPs); on this basis together with another identified value (4000 tce), energy consumption can be divided into three groups including high, middle, and low. The definition of EECP enables the transformation of energy consumption from continuous to binary variables, which are similar and comparable to urban land pixels. Further details on the threshold technique and the results are given in the  appendix.

Fig. 1.

Diagram of the station classification scheme. EECP denotes effective energy consumption pixel. Circle surrounding each station in classification examples denotes the 30-km buffer zone; 20- and 10-km buffer zones are not shown here.

Fig. 1.

Diagram of the station classification scheme. EECP denotes effective energy consumption pixel. Circle surrounding each station in classification examples denotes the 30-km buffer zone; 20- and 10-km buffer zones are not shown here.

A basic assumption for the classification scheme is that the temperature trends are influenced by land cover change and human activity near the station, thus it was developed by considering the spatial characteristics of urban land cover and EECP near the station to represent the two aspects of urbanization. For example, a station located in the center of a large city would receive a much stronger urbanization impact than that of a station in the suburbs of a small city.

Here, we describe the spatial characteristics of urbanization near a station from the perspectives of structure and extent. Different structure and extent types generally reflect the different magnitudes of urbanization impacts at various stations. Structure refers to the specific spatial relationship of urban land cover and EECP pattern near a station, which is classified as core, exposed, and edge. Core-type stations are those located in built-up areas (urban land cover), which are considered to be the “core” of a city and experience the strongest anthropogenic impact. Exposed-type stations are those located in only EECP (nonurban land cover), which in areas with intense human activity rather than built-up areas and are thus “exposed” to strong anthropogenic impact, although less than that at core stations. Stations located in nonurban land or non-EECP are classified as edge type, which is far from the city and experiences a moderate or small anthropogenic impact. It should be noted that each structure type is excludable. If a station is located in both urban land and EECP, it is designated as core station. The system is designed such that urban land has a higher priority than EECP, based on the knowledge that the anthropogenic impact is significantly more intense in urban areas than that in nonurban areas. The definition for extent category is quite simple. Extent refers the size of spatial extent of EECP where a station is situated, reflecting the spatial range of human activities. Extent category is classified into large, medium, and small on the basis of the EECP percentage over circular zones centered at each station. Therefore, structure and extent emphasize different aspects of urbanization (i.e., land cover and human activity).

As an improvement over the methods of Li and Zhao (2012), who performed the classification manually at each station, all of the classifications in this research were executed quantitatively with prescribed criteria. Prior to the implementation, a 3-km × 3-km analyzing raster window was created at each station to reduce the inherent error induced by coarse spatial information of stations with minimum units of one arc minute (equivalent to 1.85 km), in addition to errors induced by the inaccuracy of land cover and energy consumption data at fine scale. In this manner, the introduction of an analysis window that contains 9 pixels complicates the simple binary classification principles of the structure types illustrated above. That is, the determination of structure type would rely on the number of urban land and EECP pixels in the analysis window rather than being simply based on whether a station is located in a given pixel. Here, we define stations with urban land fractions higher than 3/9 as core type, and stations with EECP fraction higher than 6/9 as expose type. Assuming that urban land and EECP are nearly continuously distributed in space, the threshold of urban land fraction for the core type (>3/9) can ensure that a station is located in urban land or is half surrounded by urban land in the neighborhood. The corresponding threshold (>6/9) for exposed type ensures that a station is not only located in EECP but is also nearly fully surrounded by EECP. Different thresholds for core- and exposed-type stations ensure there are relatively large number of stations contained in each class that are representative. It should be noted that a station meeting the criteria for both core and exposed type is defined as core type because of its higher priority. Edge-type stations meet none of the above criteria. Given the land cover data in the 1980s and 2003, a station with a structure type in the two periods can be identified, as well as the associated conversion by urban expansion.

To reflect the size of the spatial extent of energy consumption near the station, three circular buffer zones with different radii of 10, 20, and 30 km at each station were created to facilitate the determination of extent type. The 30-km radius of the buffer zone was selected in accordance with the mean representative area of meteorological station in China (Li and Zhao 2012). Therefore, extent type can be determined by the weighted EECP ratio of the three circular buffer zones, which can be simply calculated by the following equation:

 
formula

where is weighted EECP ratio; , , and are the EECP numbers contained in the 10-, 20-, and 30-km-radius buffer zones, respectively; and , , and are the total pixel numbers or area for the 10-, 20-, and 30-km-radius buffer zones. The range of is from 0 to 3. This equation contains an assumption that the impact of onsite human activity on temperature declines with increasing distance from the station. We divided extent type into large, medium, and small with the value of EECP weighted ratio of 0.6 and 0.17 which was determined by a similar threshold method as that of energy consumption (the  appendix). Several examples of the station classification are given in Fig. 1.

3. Results

a. Station conversion due to urbanization in China

Figure 2 shows the spatial distributions of 274 stations of various structure and extent types in 2003. Extent type reflects the general economic and population patterns of China at a regional scale. Stations in large extent are located mainly in the North China Plain, the Bohai Sea region, and the coastal region of southern China, which are the most highly urbanized areas in the country. Stations in medium and small extents are located in northeastern and southern China, where the economy is less developed. In general, extent types and city size were in good agreement; however, stations of various structure types expressed more local characteristics with fewer uniform spatial patterns because structure emphasizes the urbanization process at a local scale.

Fig. 2.

Spatial distribution of stations including structure and extent types in 2003. The inset of the map depicts national boundary of China in South China Sea.

Fig. 2.

Spatial distribution of stations including structure and extent types in 2003. The inset of the map depicts national boundary of China in South China Sea.

Table 1 summarizes the classification results of the structure types in the 1980s and in 2003. During these periods, some stations remained unchanged while others underwent conversions in structure type as a result of urban expansion, thereby generating three unchanged station types including unchanged core, unchanged exposed, and unchanged edge and two conversion types including exposed to core and edge to core. It should be noted that no stations were converted from edge to exposed because exposed and edge types are determined by EECP, which only utilized energy consumption in 2003. Therefore, urban expansion (change in land cover) would not lead to such conversion. While there might be increase in energy consumption during the study period, it is a matter that not examined in this study. Table 1 indicates that the number of core-type stations increased dramatically from 21 in the 1980s to 146 in 2003 because of the ongoing urban expansion. In the 1980s, the number of core stations accounted for merely 7.7% of all stations; however, the proportion rose to 53.3% after. Nearly half (125 stations; 45.6%) of core stations in 2003 underwent such conversions.

Table 1.

Numbers of various station structure types in the 1980s and in 2003.

Numbers of various station structure types in the 1980s and in 2003.
Numbers of various station structure types in the 1980s and in 2003.

The substantial conversion in structure type due to urbanization may introduce additional bias to urban climate analysis if not properly considered. In fact, the assumption that the designation of station type remains unchanged or constant throughout the entire study period is poorly satisfied. Thus, it is better to utilize the latest data to classify stations. In this paper, if not specified, the classification results of 2003 were used in the following analysis.

b. Temperature trends for observation, reanalysis, and OMR

Figure 3 shows the spatial distribution of linear temperature trends for observation, reanalysis, and OMR during 1979–2010 over China. The observation temperature exhibited an overall warming trend of 0.380°C decade−1 with large spatial variations (standard deviation = 0.156°C decade−1). Strong warming was detected mainly in northeastern China, the lower reaches of the Yangtze River, Beijing, and Tianjin while moderate warming was observed in southern China. Reanalysis temperature captured the regional rather than local warming pattern of observation temperature over China, presenting a more uniform spatial distribution (standard deviation = 0.082°C decade−1) with an overall moderate warming of 0.237°C decade−1. The smaller standard deviation of reanalysis temperature can be attributed to its insensitive nature to local land factors, but its coarse spatial resolution may also have contribution. As a result, the OMR showed an overall warming of 0.142°C decade−1; 234 stations showed positive OMR trends. OMR trends were mostly positive in North China and in the lower reaches of the Yangtze River, which are attributed to the urbanization impact and land use effects (Kalnay and Cai 2003). Compared with reanalysis temperature, OMR exhibited more local features with large spatial variations (standard deviation = 0.151°C decade−1), which supports the principle that the influences of local factors are highlighted in OMR.

Fig. 3.

Spatial distribution of temperature trends for observation (OBS), reanalysis (RE), and observation minus reanalysis (OMR) over China (°C decade−1). The inset of the map depicts national boundary of China in South China Sea.

Fig. 3.

Spatial distribution of temperature trends for observation (OBS), reanalysis (RE), and observation minus reanalysis (OMR) over China (°C decade−1). The inset of the map depicts national boundary of China in South China Sea.

Table 2 gives mean OMR trends for stations of various structure and extent types. As expected, the mean OMR trends declined in structure types from core to exposed to edge at 0.169°, 0.135°, and 0.099°C decade−1, respectively, as well as in extent types from large to medium to small at 0.178°, 0.116°, and 0.109°C decade−1, which is consistent with our theoretical design.

Table 2.

Observation minus reanalysis (OMR) trends for stations of various structure and extent types in 2003 (°C decade−1). Figures in parentheses indicate the station number of each category. The mean OMR trends among both structure and extent types (last row and column of the table) are not equal (sig < 0.01), as determined by the one-way analysis of variance (ANOVA) test. Multiple comparison tests (Tukey's least significant difference; LSD) indicate OMR trends between core and edge, large and medium/small are significantly different at the 0.01 level. Values in italics denote temperature trends averaged with sample size smaller than 10, which may have larger uncertainty (same in Tables 5 and 6).

Observation minus reanalysis (OMR) trends for stations of various structure and extent types in 2003 (°C decade−1). Figures in parentheses indicate the station number of each category. The mean OMR trends among both structure and extent types (last row and column of the table) are not equal (sig < 0.01), as determined by the one-way analysis of variance (ANOVA) test. Multiple comparison tests (Tukey's least significant difference; LSD) indicate OMR trends between core and edge, large and medium/small are significantly different at the 0.01 level. Values in italics denote temperature trends averaged with sample size smaller than 10, which may have larger uncertainty (same in Tables 5 and 6).
Observation minus reanalysis (OMR) trends for stations of various structure and extent types in 2003 (°C decade−1). Figures in parentheses indicate the station number of each category. The mean OMR trends among both structure and extent types (last row and column of the table) are not equal (sig < 0.01), as determined by the one-way analysis of variance (ANOVA) test. Multiple comparison tests (Tukey's least significant difference; LSD) indicate OMR trends between core and edge, large and medium/small are significantly different at the 0.01 level. Values in italics denote temperature trends averaged with sample size smaller than 10, which may have larger uncertainty (same in Tables 5 and 6).

We performed the one-way analysis of variance (ANOVA) to the mean OMR trends in structure and extent types, which indicates a statistical test of whether the means of several groups are equal and therefore generalizes the t test to more than two groups. Results indicated that the OMR trends among various structures and extent types were not equal (sig < 0.01). Since ANOVA can only tell whether several groups of means are equal, therefore our main focus and interpretation are based on the followup tests such as multiple comparison in order to assess exactly which groups are different from which other groups. Multiple comparison tests by Tukey's least significant difference method (LSD in Matlab) indicated mean OMR trends between core and edge, lager and medium/small are significantly different at the 0.01 level. Results of the statistical tests were provided in the captions of each table or figure.

A declining pattern of OMR trends in structure was also detected in each extent type except for small extent. Similarly, a declining pattern of OMR trends in extent was detected in each structure type except for edge. These results provided clear evidence that the magnitude of warming is closely related to urbanization in terms of urban land cover and spatial range of EECP.

c. Possible bias of station conversion to station designation and the analysis of temperature trends

By using the land cover data of the 1980s and 2003, we obtained the classification results for structure type in the two periods. Temperature trends of stations with different structure types in the 1980s and 2003 are displayed in Table 3. The results from both the 1980s and 2003 schemes were generally consistent. Both observational temperature and OMR trends exhibited declining patterns across each structure type; however, the results of the 1980s scheme presented larger overall warming than that of the 2003 scheme. Observation temperature trends with the 1980s scheme for core, exposed, and edge type were 0.464°, 0.398°, and 0.340°C decade−1, respectively, which yielded larger trends than that of the 2003 scheme by 0.059°, 0.020°, and 0.003°C decade−1, respectively. If we consider the differences of observation temperature trends between core and edge types as traditional urban-minus-rural differences, which is a method of estimating the urbanization impact, the 2003 classification scheme yielded a difference of 0.068°C decade−1 while the 1980s scheme showed a rather larger value of 0.124°C decade−1. This large discrepancy between the two schemes suggests the importance of station classification in the estimation of urbanization influence. In fact, the results from the 1980s scheme are questionable because it is known that some of the originally defined exposed and edge stations were converted into core stations prior to 2003 owing to urban expansion. That is, the 1980s scheme was not able to accurately represent the actual station conditions, which led to an overestimation. This example illustrates that station conversion may cause additional bias to station classification and estimation of the urbanization impact in the context of rapid urbanization.

Table 3.

Temperature trends of stations with 1980s and 2003 schemes. (°C decade−1). The observation/observation minus reanalysis (Obs/OMR) trends among various structure types with both 1980s and 2003 schemes are not equal (sig < 0.01), as determined by ANOVA test. Multiple comparison test (LSD) indicates that Obs/OMR trends between core and edge, exposed and edge with the 1980s scheme are significantly different at the 0.01 level; Obs/OMR trends between core and edge with the 2003 scheme are significantly different at the 0.01 level.

Temperature trends of stations with 1980s and 2003 schemes. (°C decade−1). The observation/observation minus reanalysis (Obs/OMR) trends among various structure types with both 1980s and 2003 schemes are not equal (sig < 0.01), as determined by ANOVA test. Multiple comparison test (LSD) indicates that Obs/OMR trends between core and edge, exposed and edge with the 1980s scheme are significantly different at the 0.01 level; Obs/OMR trends between core and edge with the 2003 scheme are significantly different at the 0.01 level.
Temperature trends of stations with 1980s and 2003 schemes. (°C decade−1). The observation/observation minus reanalysis (Obs/OMR) trends among various structure types with both 1980s and 2003 schemes are not equal (sig < 0.01), as determined by ANOVA test. Multiple comparison test (LSD) indicates that Obs/OMR trends between core and edge, exposed and edge with the 1980s scheme are significantly different at the 0.01 level; Obs/OMR trends between core and edge with the 2003 scheme are significantly different at the 0.01 level.

Although these results are for an extreme case, they indicate that ongoing urbanization may cause further complexity in station designation because the typology is dynamic and varies through time. This issue should be considered in assessing the urbanization influence on temperature trends if rapid urbanization has occurred in the study area during a lengthy study period, because the designation of urban and rural stations can largely influence the results of analysis (Gallo et al. 1999).

Because different classification schemes reflect different natures of urbanization, they may yield different results, which leads to various interpretations of the urban influence. For example, Gallo et al. (1999) determined that three different designation schemes led to substantially different results although the differences in the three results were not statistically significant. Moreover, Peterson and Owen (2005) reported that metadata, or data that used to classify stations, are important for UHI research because the detection of UHI signal depends on the metadata applied. The interpretation inconsistencies of the urbanization influence on temperature among relevant studies may have originated from applications of different classification schemes. Therefore, it is necessary to select appropriate urban proxies that address the main features of urbanization and to establish a mechanism-based classification scheme.

d. Impact of land cover change associated with urbanization on temperature trends

Because the urbanization process has converted some stations into core types while others remained unchanged (Table 1), it offers the opportunity of examining temperature trends of stations with and without such conversion. As shown in Fig. 4, the largest warming, 0.221°C decade−1, occurred in unchanged core stations, which is clear evidence of the significant urbanization impact on temperature trends. In contrast, unchanged exposed and edge stations exhibited less pronounced warming at 0.135° and 0.099°C decade−1, respectively. This largest trend in unchanged core stations is highly likely related to internal changes because the energy consumption of unchanged core stations is significantly higher at 18 441 tce than that of other station types, which varied from 8832 to 1604 tce. Moreover, unchanged core stations are more likely to be located in large and well established cities, whereas converted stations tend to be located in smaller cities that are prone to land cover change. This pattern is supported by weighed EECP ratios, which are 1.236 and 0.778 for unchanged core and converted stations, respectively.

Fig. 4.

Observation (OBS) and OMR temperature trends for converted and unchanged stations. Core, exposed, and edge are unchanged stations; Ex to C are stations converted from exposed to core; Ed to C are stations converted from edge to core. Vertical lines of each bar denote the standard error. As determined by ANOVA test, the mean OBS/OMR trends among station groups are not equal (sig < 0.01). Multiple comparison test (LSD) indicates that OBS/OMR trends between unchanged core and all other groups are significantly different at the 0.05 level except expose to core; OBS/OMR trends between exposed to core and unchanged edge are significantly different at the 0.01 level.

Fig. 4.

Observation (OBS) and OMR temperature trends for converted and unchanged stations. Core, exposed, and edge are unchanged stations; Ex to C are stations converted from exposed to core; Ed to C are stations converted from edge to core. Vertical lines of each bar denote the standard error. As determined by ANOVA test, the mean OBS/OMR trends among station groups are not equal (sig < 0.01). Multiple comparison test (LSD) indicates that OBS/OMR trends between unchanged core and all other groups are significantly different at the 0.05 level except expose to core; OBS/OMR trends between exposed to core and unchanged edge are significantly different at the 0.01 level.

It is interesting to note that the conversions generally result in larger warming trends for all converted stations than those for unchanged stations. For example, the OMR trends for stations converted from exposed to core were higher at 0.167°C decade−1 than those for unchanged exposed stations at 0.135°C decade−1. Similar results were observed for stations converted from edge to core at 0.134°C decade−1 and unchanged edge stations at 0.099°C decade−1. Because the converted stations comprise approximately 85.6% of all core stations at 125 of 146, both the unchanged and converted core stations contributed to the observed strong warming for core type, which resulted from the aspects of urbanization including energy consumption and land cover change. It has been reported that urbanization-induced land use change near stations can exert significant influence on temperature (Gallo et al. 1996; Gallo et al. 1999). Nearby stations would be easily affected by this effect because temperature records are sensitive to changes in surrounding areas. The higher warming rates found in converted stations are consistent with previous findings (Lin et al. 2007) such that urban expansion leads to increases in temperature.

e. Influence of human activity on temperature trends

In addition to the land cover change associated with urbanization, human activity also significantly affects temperature trends. We used the mean energy consumption of the 3-km × 3-km window together with a weighed EECP ratio for representing the intensity and spatial extent of human activity to examine their impacts on temperature trends.

For mean energy consumption, as presented in Fig. 5, a statistically significant (sig < 0.01) positive relationship was apparent between energy consumption and OMR trends across all stations, which indicates that higher intensity of human activity leads to greater warming trends. A similar statistically significant (sig < 0.01) positive relationship between the EECP ratio and OMR trends is given in Fig. 6. These results may suggest that the spatial extent of human activity also has an obvious influence on temperature trends, which indicates that a larger spatial extent of human activity tends to cause greater warming.

Fig. 5.

Overall relationship between energy consumption and OMR trends. Energy consumption (independent variable) is the mean value of the 3-km × 3-km window, which represents the intensity of human activity. The slope of the regression is significant at the 0.01 level.

Fig. 5.

Overall relationship between energy consumption and OMR trends. Energy consumption (independent variable) is the mean value of the 3-km × 3-km window, which represents the intensity of human activity. The slope of the regression is significant at the 0.01 level.

Fig. 6.

The overall relationship between effective energy consumption pixel (EECP) ratio and OMR trends. The weighed EECP ratio is the weighted percentage of EECP of the three circular buffer zones with radii of 10, 20, and 30 km, which represent the spatial extent of human activity. The slope of the regression is significant at the 0.01 level.

Fig. 6.

The overall relationship between effective energy consumption pixel (EECP) ratio and OMR trends. The weighed EECP ratio is the weighted percentage of EECP of the three circular buffer zones with radii of 10, 20, and 30 km, which represent the spatial extent of human activity. The slope of the regression is significant at the 0.01 level.

The strength of the relationship between human activity and OMR trends varied among various station types. Table 4 presents the regression results across each structure type. The signs of the slope were consistently positive across all structure types but at different significant levels. The positive relationship between energy consumption and OMR trends was strong and statistically significant in core stations, particularly the unchanged core type, and became weaker in exposed- and edge-type stations, indicating lower significance. Such results were expected because those core stations registered a strong human impact because of the high intensity of human activity; thus, a stronger warming trend was anticipated. Further, the results of exposed and edge stations suggest that a lower intensity of human activity in nonurbanized stations leads to weaker influence on temperature trends. In contrast, the EECP ratio had an overall better relationship with OMR trends than that of energy consumption, which is at high significance and was observed in all structure types except edge stations.

Table 4.

Slopes of linear regression between human activity and OMR trends. Figures in parentheses indicate R2.

Slopes of linear regression between human activity and OMR trends. Figures in parentheses indicate R2.
Slopes of linear regression between human activity and OMR trends. Figures in parentheses indicate R2.

These results reveal that the magnitude of warming is closely related to human activity, both in intensity and spatial extent, particularly for core stations that experience strong urbanization impacts. However, for exposed and edge stations that were affected by slight urbanization impact, the relationship between human activity and OMR trends was rather poor, which may indicate that human activity-related energy consumption is not the primary impact factor for temperature trends in these stations.

Considering the significant role of energy consumption in local temperature trends, we conducted an analysis presented in Table 5 similar to that in Table 2 but with OMR trends aggregated in three energy consumption groups of high, middle, and low at 4000 and 700 tce (see the  appendix) to reanalyze the influence of energy consumption and to verify the impact of land cover change. For the high energy consumption group (>4000 tce), OMR trends were 0.181°C decade−1, which were significant larger (sig < 0.01) than the 0.113° and 0.111°C decade−1 for middle and low energy consumption groups, respectively. For the structure types, the declining pattern of OMR trends from core to edge was also evident in each energy consumption group although impaired to some degree because of the limited number of station, which suggests the dependence of warming on structure type.

Table 5.

OMR trends with respect to energy consumption groups (°C decade−1). Figures in parentheses represent the station numbers of each category. The mean OMR trends among energy consumption groups are not equal (sig < 0.01), as determined by ANOVA test. Multiple comparison test (LSD) indicates that the mean OMR trends between high and low, high and middle are significantly different at the 0.01 level.

OMR trends with respect to energy consumption groups (°C decade−1). Figures in parentheses represent the station numbers of each category. The mean OMR trends among energy consumption groups are not equal (sig < 0.01), as determined by ANOVA test. Multiple comparison test (LSD) indicates that the mean OMR trends between high and low, high and middle are significantly different at the 0.01 level.
OMR trends with respect to energy consumption groups (°C decade−1). Figures in parentheses represent the station numbers of each category. The mean OMR trends among energy consumption groups are not equal (sig < 0.01), as determined by ANOVA test. Multiple comparison test (LSD) indicates that the mean OMR trends between high and low, high and middle are significantly different at the 0.01 level.

Table 6 presents temperature trends with respect to station conversions. The results are similar to those presented in Fig. 4 but are disaggregated into three energy consumption groups. Although the results of the several cross groups were significantly affected by small sample size, the general conclusions are the same. In high energy consumption group, the largest warming at 0.266°C decade−1 was observed in unchanged core stations. The strongest warming can be attributed to a high energy consumption of 25014 tce because the energy consumption for other stations types was significantly lower and varied from 11 913 to 5757 tce. As expected, it was reaffirmed that converted stations exhibit larger OMR trends than unchanged stations. The same patterns were observed in the middle energy consumption group except for unchanged core stations. Noting that the strong warming for unchanged core stations disappeared in middle energy consumption group, which can be attributed to a decrease of energy consumption to 2010 tce; however, this result should be interpreted with caution because of the small sample size. Meanwhile, the two converted stations exhibited higher warming than their unchanged counterparts. It also provides us clear evidence that converted stations generated more warming when the influence of energy consumption is minimized. These results confirm the important role of energy consumption and also demonstrate that warming associated with conversion is robust, which reflects the impact of land cover change.

Table 6.

OMR trends for unchanged and converted stations with respect to various energy consumption groups (°C decade−1). Figures in parentheses represent the station numbers of each category. Multiple comparison (LSD) test shows that OMR trends of unchanged core are significantly different with the other types of the high energy consumption group at the 0.05 significant level; edge to core are excluded because of small sample size. OMR trends between exposed to core and unchanged edge are significantly different at the 0.1 significant level.

OMR trends for unchanged and converted stations with respect to various energy consumption groups (°C decade−1). Figures in parentheses represent the station numbers of each category. Multiple comparison (LSD) test shows that OMR trends of unchanged core are significantly different with the other types of the high energy consumption group at the 0.05 significant level; edge to core are excluded because of small sample size. OMR trends between exposed to core and unchanged edge are significantly different at the 0.1 significant level.
OMR trends for unchanged and converted stations with respect to various energy consumption groups (°C decade−1). Figures in parentheses represent the station numbers of each category. Multiple comparison (LSD) test shows that OMR trends of unchanged core are significantly different with the other types of the high energy consumption group at the 0.05 significant level; edge to core are excluded because of small sample size. OMR trends between exposed to core and unchanged edge are significantly different at the 0.1 significant level.

4. Discussion

Our results indicate that the urbanization impact on temperature trends varies among stations of various structure and extent types as well as among energy consumption groups, which reveals that the inner complexities of warming related to urbanization could differ in various regions according to factors such as rate of development or urban structure (Fujibe 2009). We determined that unchanged core stations exhibited the largest warming while converted core stations presented relatively fewer large warming trends; however, when the influence of energy consumption is minimized, the strong warming in unchanged core stations disappeared. Some studies reported similar features that the warming is not significant in urban areas relative to rural stations, such as those in some European cities (Jones et al. 2008), because such cities have likely completed the urbanization process. Urban infrastructure in a well-established city no longer increases its ability to retard outgoing longwave radiation, and the anthropogenic heat inputs are stable (Parker 2010). In contrast, for a country experiencing rapid urbanization and economic growth, such as China, the opposite is observed. Several studies have indicated that the urbanization impact on temperature trends in China is clear and strong (Zhou et al. 2004; Ren et al. 2008; Yang et al. 2011). Because these unchanged core stations are located in urban areas throughout the entire study period with no significant land cover changes occurring nearby, the urbanization impact on temperature may rely more on human activity than on land cover change. This theory is illustrated by the positive relationship in Table 4, which indicates strong correlations among human activity and OMR trends. Moreover, the critical role of human activity in the urbanization impact has been also confirmed by other studies, which indicate that urban areas could experience further warming because of the increase in human activity when urbanization has been completed or is in a slow process. For example, although most domains of Tokyo had already been converted to urban surfaces with unchanged or declining populations by the 1960s, the urban heat islands became amplified. The intensified UHI was accompanied by an increasing number of vehicles and tall buildings (Fujibe 2009). A case study in Vienna showed that a city with a constant population can resolve increasing trends in excess urban temperature (Böhm 1998). From the results of our study and other global research, we can conclude that internal changes such as increases in energy consumption owing to economic development and increasing living standards can significantly influence temperature trends.

For converted core stations, land cover change in addition to the influence of human activity contributes the higher warming rates, as has been described in many previous studies (Hale et al. 2008; Hale et al. 2006; He et al. 2007). The OMR trends are larger in stations converted from exposed to core and edge to core than those in unchanged exposed and edge stations. Although these differences in trends are not statistically significant (Fig. 4), they illustrate some sharp contrasts between stations with and without such conversion.

All of these results help to explain mechanisms by which various aspects of urbanization influence temperature trends. For a region undergoing rapid urban expansion, the impact of urbanization on temperature relies on both the physical changes in land surface and human activity. However, in a well-established city, because most of the land has been converted to urban surfaces, the function of urbanization on temperature is mainly determined by changes in the intensity and spatial extent of human activity such as economic development, growth in population and energy consumption, and increasing numbers of vehicles and large buildings. In most cases, however, these factors are combined.

Although urbanization can affect temperature trends, it does not necessarily determine local temperature change. We observed large variations in temperature trends within each station type; such trends at urban stations were not always larger than those at rural stations. Many stations received a strong human impact but showed slight positive or even negative trends. Large variations in temperature trends may relate to local and regional factors such as local human activity and other land use effects such as vegetation and water, which can mitigate the warming caused by urbanization (He et al. 2007). These results indicate the internal complexity of temperature change that could be affected by multiple natural and anthropogenic factors.

An additional uncertainty that may have influenced our results is the frequently mentioned inhomogeneities in climate data, which is an inevitable problem in many parts of world. It is widely accepted that inhomogeneities can arise from many factors including station relocation and changes in observing practices, instruments, and the environment (Liu and Li 2003) and can introduce large biases in the climate analysis. Although several scholars reported that inhomogeneities exist in Chinese temperature time series (Li et al. 2004a,b; Li and Yan 2009), no homogeneity adjustment was made to observation temperatures in this work partly because the metadata of stations are currently unavailable in the data sharing system. The absence of such data creates adjustment challenges because the actual cause for each suspicious value cannot be determined, and it is unknown whether the value is a climatic signal of rapid change or is caused by nonclimatic factors. The primary reason is that homogeneity adjustment may sacrifice the information of the urbanization impact on temperature to a degree. Land use change, urbanization, and other changes that occurred near the stations are important sources of inhomogeneity (Wu et al. 2007); however, these factors are of primary concern in this paper. The urbanization effect and inhomogeneities induced by other factors are usually intertwined and cannot be completed separated, isolation of the urban warming effect from other inhomogeneities remains a complicated task (Kukla et al.1986; Peterson 2003). In this case, the urbanization signal would be inevitably undetermined after homogeneity adjustment under current situation. Further, the choice of adjustment methods and reference stations can have large impact on the estimated temperature trends at the same stations (Li and Yan 2009), and the magnitude of this uncertainty is comparable to the urbanization impact and would confound the analysis.

Despite these factors, we believe that inhomogeneity has only minor influence on our results because we used averaged trends from different station groups including many stations. Li and Yan (2009) reported that homogeneity adjustments have only a minor influence on the large-scale mean climate trends because local inhomogeneities compensate for each other when the averages are calculated. Moreover, other studies using adjusted homogenous data also detected significant warming trends related to urbanization in China (Zhou et al. 2004; Ren et al. 2008). It appears that basic conclusions of urbanization impacts on temperature trends from previous studies in China are insensitive to homogeneity adjustment. New homogeneous climate data in the future will help resolve this issue, such as the upcoming second version of China Homogenized Historical Temperature Dataset (CHHT) ( Q. X. Li 2013, personal communication), which will significantly improve the quality of climate records in China and could be potentially useful in climate analysis. Although we cannot fully attribute the observed strong warming signal to the urbanization impact, analysis based on a large number of stations still provides positive evidence to strengthen our theory.

5. Conclusions

In this study, we investigated the impact of urbanization on temperature trends from the aspects of physical changes in land surface and internal changes related to human activity. Results based on our new classification scheme confirmed the preliminary recognition that both aspects play vital roles in and together contribute to the observed warming at local and regional scales, particularly at urbanized stations. This fact suggests that land cover change not only represents changes in typology but also internal changes such as a significant increase in human activity. However, the latter means of change usually is poorly reflected in the current land classification system. Therefore, there is an urgent need for development of better measures of urbanization that address its main features in addition to a more suitable classification scheme in urban climate research such as a mechanism-based scheme. It has become increasing difficult to find stations that are completely free of the urbanization impact; thus, the combined use of multiple data sources for station designation and the introduction of an alternative method such as OMR should be considered.

China has experienced rapid economic growth and urbanization during the study period. Nearly half of the stations in our study area have been converted to core stations, which are urbanized stations according to our classification scheme. Strong warming occurred in these converted stations owning to significant land cover change nearby. Moreover, energy consumption also explains a large portion of warming in the unchanged core stations. Our results suggest that the station conversion problem must be considered for analysis including fast urbanization. Otherwise, additional uncertainty can be introduced and may confound the interpretation of urban influence because the results are largely influenced by the manner in which stations are classified.

In the estimation of the urbanization impact, the dichotomy of urban and nonurban stations has an inherent limitation because the urbanization impact on temperature trends is usually manifested in the form of continuous influence and declines with the urban–rural gradient. Although studies on urbanization influenced by discrete station groups are helpful in identifying the urban influence, they are not effective for quantifying the influence because the various benchmarks present challenges in comparing the estimations. Therefore, it is preferable to consider the urbanization impact in the form of continuous impact, the magnitude of which is determined by the function of various aspects of urbanization such as land cover change, human activity, and other related factors.

Acknowledgments

This work was supported by the National Natural Science Foundation of China No. 41130534, Oceanography Science Foundation for the Youth Scholars of State Oceanic Administration (2013256) and National Natural Science Foundation of China No. 40635028. We thank three anonymous reviewers for their insightful comments, which greatly improved the quality of the manuscript. We also thank Yao Feng for proofreading the manuscript.

APPENDIX

Application of the Threshold Technique

The criterion of effective energy consumption pixel (EECP) was determined by a threshold technique developed by De Laat and Maurellis (2006). In this method, a particular value of energy consumption is first chosen as a threshold. Temperature trends of stations with energy consumption above and below the threshold are then averaged at each threshold. This threshold technique was conducted for the full range of energy consumption, and the results are plotted in Fig. A1. Temperature trends above and below the threshold generally decreased with a decrease in the energy consumption threshold. Two critical points of the energy consumption threshold occur at approximately 4000 and 700 tce. When the energy consumption threshold is higher than 4000 tce, temperature trends above and below the threshold decrease rapidly with energy consumption thresholds. This declining pattern becomes weaker when the energy consumption threshold is lower than 4000 tce. Temperature trends tend to rise in fluctuation with a decrease in energy consumption thresholds when the energy consumption threshold is lower than 700 tce. This suggests that the human impact on temperature trends becomes less significant at this point. Thus, we defined an energy consumption of more than 700 tce as the EECP, by which method the human impact on temperature trends is apparent and detectable. In addition, these values of energy consumption, 4000 and 700 tce, were used to determine the three energy consumption groups in section 3e.

Fig. A1.

Threshold results for energy consumption and OMR trends. Temperature trends against the energy consumption threshold denote the mean temperature trends for stations above or below the corresponding energy consumption threshold value. The two vertical lines denote two critical points at 4000 and 700 tce, respectively.

Fig. A1.

Threshold results for energy consumption and OMR trends. Temperature trends against the energy consumption threshold denote the mean temperature trends for stations above or below the corresponding energy consumption threshold value. The two vertical lines denote two critical points at 4000 and 700 tce, respectively.

Similarly, we used the threshold method to determine the weighed EECP ratio of the three buffer zones of each station. The threshold results for the weighted EECP ratio and OMR trends are plotted in Fig. A2. Therefore, similar critical points can be found at EECP ratios of 0.6 and 0.17, which were used to classify extent types into large, medium, and small.

Fig. A2.

Threshold results for EECP ratio and OMR trends. Vertical lines denote two critical points at 0.6 and 0.17, respectively.

Fig. A2.

Threshold results for EECP ratio and OMR trends. Vertical lines denote two critical points at 0.6 and 0.17, respectively.

REFERENCES

REFERENCES
Böhm
,
R.
,
1998
:
Urban bias in temperature time series—A case study for the city of Vienna, Austria
.
Climatic Change
,
38
,
113
128
.
Cai
,
M.
, and
E.
Kalnay
,
2004
:
Climate (communication arising): Impact of land use change on climate
.
Nature
,
427
,
214
214
.
De Laat
,
A. T. J.
, and
A. N.
Maurellis
,
2006
:
Evidence for influence of anthropogenic surface processes on lower tropospheric and surface temperature trends
.
Int. J. Climatol.
,
26
,
897
913
.
Englehart
,
P. J.
, and
A. V.
Douglas
,
2003
:
Urbanization and seasonal temperature trends: Observational evidence from a data-sparse part of North America
.
Int. J. Climatol.
,
23
,
1253
1263
.
Fan
,
H. L.
, and
D. J.
Sailor
,
2005
:
Modeling the impacts of anthropogenic heating on the urban climate of Philadelphia: A comparison of implementations in two PBL schemes
.
Atmos. Environ.
,
39
,
73
84
.
Fujibe
,
F.
,
2009
:
Detection of urban warming in recent temperature trends in Japan
.
Int. J. Climatol.
,
29
,
1811
1822
.
Gallo
,
K. P.
,
D. R.
Easterling
, and
T. C.
Peterson
,
1996
:
The influence of land use land cover on climatological values of the diurnal temperature range
.
J. Climate
,
9
,
2941
2944
.
Gallo
,
K. P.
,
T. W.
Owen
,
D. R.
Easterling
, and
P. F.
Jamason
,
1999
:
Temperature trends of the U.S. historical climatology network based on satellite-designated land use/land cover
.
J. Climate
,
12
,
1344
1348
.
Grimmond
,
S.
,
2007
:
Urbanization and global environmental change: Local effects of urban warming
.
Geogr. J.
,
173
,
83
88
.
Hale
,
R. C.
,
K. P.
Gallo
,
T. W.
Owen
, and
T. R.
Loveland
,
2006
:
Land use/land cover change effects on temperature trends at U.S. Climate Normals stations
.
Geophys. Res. Lett.
, 33, L17033, doi:10.1029/2006GL026358.
Hale
,
R. C.
,
K. P.
Gallo
, and
T. R.
Loveland
,
2008
:
Influences of specific land use/land cover conversions on climatological normals of near-surface temperature
.
J. Geophys. Res.
, 113, D14113, doi:10.1029/2007JD009548.
He
,
J. F.
,
J. Y.
Liu
,
D. F.
Zhuang
,
W.
Zhang
, and
M. L.
Liu
,
2007
:
Assessing the effect of land use/land cover change on the change of urban heat island intensity
.
Theor. Appl. Climatol.
,
90
,
217
226
.
Ichinose
,
T.
,
K.
Shimodozono
, and
K.
Hanaki
,
1999
:
Impact of anthropogenic heat on urban climate in Tokyo
.
Atmos. Environ.
,
33
,
3897
3909
.
Jones
,
P. D.
,
P. Y.
Groisman
,
M.
Coughlan
,
N.
Plummer
,
W. C.
Wang
, and
T. R.
Karl
,
1990
:
Assessment of urbanization effects in time-series of surface air-temperature over land
.
Nature
,
347
,
169
172
.
Jones
,
P. D.
,
D. H.
Lister
, and
Q.
Li
,
2008
:
Urbanization effects in large-scale temperature records, with an emphasis on China
.
J. Geophys. Res.
,
113
,
D16122
, doi:10.1029/2008JD009916.
Kalnay
,
E.
, and
M.
Cai
,
2003
:
Impact of urbanization and land use change on climate
.
Nature
,
423
,
528
531
.
Karl
,
T. R.
,
H. F.
Diaz
, and
G.
Kukla
,
1988
:
Urbanization: Its detection and effect in the United States climate record
.
J. Climate
,
1
,
1099
1123
.
Kukla
,
G.
,
J.
Gavin
, and
T. R.
Karl
,
1986
:
Urban warming
.
J. Climate Appl. Meteor.
,
25
,
1265
1270
.
Li
,
Q.
,
X.
Liu
,
H.
Zhang
,
T. C.
Petersen
, and
D. R.
Easterling
,
2004a
:
Detecting and adjusting temporal inhomogeneity in Chinese mean surface air temperature data
.
Adv. Atmos. Sci.
,
21
,
260
268
.
Li
,
Q.
,
H.
Zhang
,
X.
Liu
, and
J.
Huang
,
2004b
:
Urban heat island effect on annual mean temperature during the last 50 years in China
.
Theor. Appl. Climatol.
,
79
,
165
174
.
Li
,
Q.
, and
Coauthors
,
2010
:
Assessment of surface air warming in northeast China, with emphasis on the impacts of urbanization
.
Theor. Appl. Climatol.
,
99
,
469
478
.
Li
,
Y.
, and
X. Y.
Zhao
,
2012
:
An empirical study of the impact of human activity on long-term temperature change in China: A perspective from energy consumption
.
J. Geophys. Res.
,
117
,
D17117
, doi:10.1029/2012JD018132.
Li
,
Z.
, and
Z.-W.
Yan
,
2009
:
Homogenized daily mean/maximum/minimum temperature series for China from 1960–2008
.
Atmos. Oceanic Sci. Lett.
,
2
,
237
243
.
Lin
,
W. S.
, and
Coauthors
,
2007
:
A numerical study of the influence of urban expansion on monthly climate in dry autumn over the Pearl River Delta, China
.
Theor. Appl. Climatol.
,
89
,
63
72
.
Liu
,
X. N.
, and
Q. X.
Li
,
2003
:
Research of the inhomogeneity test of climatological data series in China
.
Acta Meteor. Sin.
,
17
,
492
502
.
Narumi
,
D.
,
A.
Kondo
, and
Y.
Shimoda
,
2009
:
Effects of anthropogenic heat release upon the urban climate in a Japanese megacity
.
Environ. Res.
,
109
,
421
431
.
Parker
,
D. E.
,
2010
:
Urban heat island effects on estimates of observed climate change
.
Wires Climate Change
,
1
,
123
133
.
Peng
,
S. S.
, and
Coauthors
,
2012
:
Surface urban heat island across 419 global big cities
.
Environ. Sci. Technol.
,
46
,
696
703
.
Peterson
,
T. C.
,
2003
:
Assessment of urban versus rural in situ surface temperatures in the contiguous United States: No difference found
.
J. Climate
,
16
,
2941
2959
.
Peterson
,
T. C.
, and
T. W.
Owen
,
2005
:
Urban heat island assessment: Metadata are important
.
J. Climate
,
18
,
2637
2646
.
Peterson
,
T. C.
,
K. P.
Gallo
,
J.
Lawrimore
,
T. W.
Owen
,
A.
Huang
, and
D. A.
McKittrick
,
1999
:
Global rural temperature trends
.
Geophys. Res. Lett.
,
26
,
329
332
.
Ren
,
G. Y.
,
Y. Q.
Zhou
,
Z. Y.
Chu
,
J. X.
Zhou
,
A. Y.
Zhang
,
J.
Guo
, and
X. F.
Liu
,
2008
:
Urbanization effects on observed surface air temperature trends in north China
.
J. Climate
,
21
,
1333
1348
.
Ren
,
Y. Y.
, and
G. Y.
Ren
,
2011
:
A remote-sensing method of selecting reference stations for evaluating urbanization effect on surface air temperature trends
.
J. Climate
,
24
,
3179
3189
.
Ren
,
Z. H.
,
A. Y.
Xiong
, and
F. L.
Zou
,
2007
:
The quality control of surface monthly climate data in China (in Chinese)
.
J. Appl. Meteor. Sci.
,
18
,
516
523
.
Wu
,
L. H.
,
L. L.
Kang
,
H. Y.
Chen
, and
X. M.
Ma
,
2007
: Effect of meteorological observation environment variability on homogeneity of temperature series in Zhejinag Province. Meteor. Sci. Technol.,35, 152–156.
Xiao
,
H. L.
, and
Q. H.
Weng
,
2007
:
The impact of land use and land cover changes on land surface temperature in a karst area of China
.
J. Environ. Manage.
,
85
,
245
257
.
Yang
,
X.
,
Y.
Hou
, and
B.
Chen
,
2011
:
Observed surface warming induced by urbanization in east China
.
J. Geophys. Res.
,
116
,
D14113
, 10.1029/2010JD015452.
Zhou
,
L. M.
, and
Coauthors
,
2004
:
Evidence for a significant urbanization effect on climate in China
.
Proc. Natl. Acad. Sci. USA
,
101
,
9540
9544
.