Temperature change plays a crucial role in global change sciences. In the past several decades, comprehensive findings have been achieved on temperature change in China for the past 100 years. Several time series have been created to illustrate the averaged surface air temperature for the country. The correlations of these series range from 0.73 to 0.97. It is also achieved in better data quality, wider spatial data coverage, improved homogeneity of time series, and enhanced reliability of findings. The results show an annual mean temperature increase by 0.78±0.27°C per 100 years in China for the period 1906–2005. After prolonging the period till 2007, it is found that 2007 is rated as the warmest year in the past 100 years. Although all the series, except one, reflect temperature changes in the eastern part of China before the 1930s, they represent the general temperature change in most parts of the country after the 1930s.
past 100 years ; air temperature series ; warming rate estimate ; global warming
In the past decades, global climate warming has become a major concern attracting broad attentions both at home and abroad. In this context, studying the basic facts of climate change, a fundamental part of the issue, is of theoretical and practical importance. Numerous studies have been made, both domestically and internationally, in the area, which creates a meaningful basis and background for improving people’s cognition of climate change. Chinese scientists have made studies on the effects of temperature change due to climate change [ Lin and Yu, 1990 ; Tang, 1996 ; Ren et al., 2005 ], but most of them dealt with effects occurring in the past 50 years. There are few studies on temperature change in the past 100 years, especially for the whole of China due to limited data availability. Apparently, it is extremely important to establish a nationwide temperature series, so as to study the trends of climate change in China. In the past, scientists analysed temperature series for China, using different datasets, stations, and methodologies. Most trends in temperatures are increasing. Two warming periods were determined, the 1920s–1940s and the 1980s respectively. However, the analyses of different datasets result in various estimates of trends and changes in temperature. Therefore, it is necessary to compare and assess these series systematically to improve the comprehensive temperature series for China.
Global change science, a relatively new discipline, emerged in the 1980s [ Ding and Sun, 2006 ], though the study of hemispheric and global temperature change started in the 1950s. In the 1960s, Mitchell  established a relatively reliable global and hemispheric mean temperature series for the past 100 years. From the 1970s to the 1990s, the hemispheric and global temperature series were developed using monthly mean temperature anomalies [ Borzenkova et al., 1976 ; Vinnikov et al., 1990 ]. In the 1980s, the global and hemispheric mean temperature series were established, which were one of the most comprehensive works at that time in the world [ Jones et al., 1982 ; Jones, 1988 ]. Hansen et al.  also created a global mean temperature series in 1987. All these series suggest a warming trend in the past 100 years, and estimate that global mean temperatures have risen by 0.5–0.6°C since 1880, though the time series show different patterns.
In recent years, both land and sea surface temperature data showed substantial improvements in quality, thanks to repeated corrections and supplementation efforts. Jones et al. [1994 ; 2003 ] corrected and updated land surface temperature data of CRU, and enlarged the data coverage in the Southern Hemisphere. Hansen et al.  and Peterson and Vose  also analysed global mean temperature series. The IPCC Fourth Assessment Report [ IPCC , 2007 ] published in 2007 presented new estimates of global warming trends for the past 100 years — since 1906, the global mean temperature increased by about 0.74°C. The temperature series and new assessment results have not only confirmed the ascending trend of global mean temperature, but also enhanced the reliability of estimates on warming.
In the last decades, an increasing number of Chinese studies on climate and associated climate change were accomplished. For example, Ye et al.  published the Contemporary Climate Studies in the early 1990s, a systematic review of findings derived from climate and climate change studies both in China and worldwide. Climate Change and Associated Impact Studies, written by Ding  , summarized the findings achieved by Chinese scientists on GHG (greenhouse gas) emissions, observed facts of climate change, climate modeling and prediction, and impacts of climate change in China. In recent years, a continuous flow of climate change studies have been published in China, exhibiting new findings [ Wang, 2001 ; Qin et al., 2005 ] based on previous studies.
Climate change in China over the past century is of major concern to climatologists. Trends in air temperature are more and more investigated. Zhu  summarized temperature change curves across China for the past 5,000 years, and analyzed temperature variations since the 20th century using instrumental records collected in Beijing, Tianjin, Shanghai, and Hong Kong. As early as in the 1960s, Tu  wrote a paper discussing climate warming in China. Wang et al.  analyzed climate warming trends across China in the 20th century. Many domestic scholars [ Zhang and Li , 1982 ; Tu , 1984 ; Wang, 1990 ; Wang et al., 1998 ; Tang and Lin, 1992 ; Ding and Dai, 1994 ; Lin et al., 1995 ] have studied temperature changes across China for the last 100 years in the global and Northern Hemispheric context since the 1980s. For example, Qian et al.  examined the footprints of climate change in seven regions of China and associated differences using the data produced by Wang et al.  . In 2005, Tang and Ren  reanalyzed temperature change in China over the past 100 years. After that, Wen et al.  compared air temperature and precipitation change in China using high resolution grid datasets of CRU. In addition, some scientists [ Ren and Zhou, 1994 ; Ding et al., 1998 ] investigated temperature changes on regional or weather station level.
The above-mentioned studies have come up with the following main conclusions. In the past 100 years, China experienced a change in mean surface temperature which is quite similar to the observed changes in the Northern Hemisphere and worldwide. China has a more noticeable warming period in the 1920s–1940s than the world and Northern Hemisphere experienced. In the past 100 years, China’s mean surface temperature shows an increase of 0.50–0.80°C. Until 2006, 1998 is the warmest year in the Chinese meteorological records [ Wang and Gong, 2000 ].
A reliable temperature series is the basis for valuable results and future studies. Table 1 summarizes major mean temperature series established by Chinese climatologists since the 1980s. Zhang et al.  established a 5-year running mean temperature rating series (ZL series) from 1910 to 1979. The monthly mean temperature data was collected at 137 stations. Time series were determined for 7 defined regions. Missing temperature data were interpolated for the years before 1950 when data were scarce or absent. For data that could not be interpolated, linear interpolated, linear regression analyses based on data from other regions were applied. For regions with an insufficient number of stations, the temperature rating of typical stations was used as the temperature rating for the region. After that, Wang  converted temperature rating into temperature anomaly, and created another series (W series), using interpolated mean temperature data collected in Harbin, Beijing, Shanghai, and Guangzhou for the years before 1910.
|Series||Data||Stations||The method of region average||Starting year||Issued|
|ZL||Temperature ratings||137||7 regions, averaged regional temperature ratings, then national temperature ratings||1910||1982|
|W||Mean temperature of Harbin, Beijing, Shanghai, and Guangzhou for 1880–1990, temperature ratings since 1911||137||Averaged temperature ratings of 7 regions, then converting them into temperature anomaly for averaged national temperature series||1880||1990|
|TL||Monthly mean temperature||716||Arithmetic mean||1921||1992|
|LYT||Monthly mean temperature||711||10 regions, averaged regional anomaly series, then averaged national series||1873||1995|
|WYG||Mean temperature, temperature ratings, ice core, tree rings, and historical records for early stage||50||10 regions, 5 stations from each, averaged regional series, then averaged national series through weighted area||1880||1998|
|TR||Mean temperature derived from max and min temperature||616||5° × 5° weighted area for averaged national series||1905||2005|
|TD||Corrected and interpolated data and data collected from stations having an even distribution, mean temperature derived from max and min temperature||291||5° × 5° weighted area for averaged national series||1873||2006|
In 1992, Tang and Lin  established a monthly mean temperature series (TL series) for the period 1921–1990, based on arithmetically averaged monthly temperature data of 716 stations. Lin et al.  calculated the correlation coefficient between typical station (with long time) and other station by using the data of 711 selected weather stations across the country. The country was divided into 10 regions in line with the distribution of stations at a given significant level (> 0.01), in an attempt to derive a national series (LYT series) from regional averages. The temperature series include monthly and yearly mean temperatures for regional and national averages, starting from 1873. In 1998, Wang et al.  established a nationwide mean temperature series (WYG series) for the period of 1880–1996, through weighted averages of annual temperature series of 10 individual regions calculated by weighted area coefficients. They employed mean temperature and temperature anomaly data derived from temperature ratings, and employed ice core, tree ring, and ancient literatures to interpolate regional data for areas lacking instrumental records. The series has achieved a complete spatial coverage, with a full series for each individual region since 1880. This series is very relevant in understanding the spatial distribution of temperature changes in China.
Air temperature data collected at weather stations before 1950 are not consistent. They lack a unified observation time, complete spatial coverage, and continuous data (gaps). As a result, the temperature series for this period are inhomogeneous, and corrections are difficult. This impedes the improvement of national temperature series. To address this issue, Tang and Ren  used the averaged maximum and minimum temperatures before and after 1950 to calculate the monthly mean temperatures, and determined the temperature anomalies for a 5° × 5° grid, using observational data collected at 616 stations. Following this, they established a national temperature series (TR series) which covers the period 1905–2001, through the weighted areas method, which solves the abovementioned inhomogeneity problem to some extent. Although the mean temperatures derived from the maximum and minimum temperatures are somewhat different from the definite observed mean temperatures, whose difference can be ignored in studying temperature change, especially for the regional average [ Tang and Ding, 2007 ]. The relative analysis shows that the approach has achieved a noticeably enhanced homogeneity for the national temperature series. Tang  screened the temperature data collected before 1950 and corrected the false data to obtain a high homogeneity. The data provided by 630 stations were applied in an effort to take full advantage of the observational data which was available for the period and enhanced the continuity of the series. After that, a new national temperature series (TD series) was created based on data derived from relatively even distributed 291 stations. In addition, Wen et al.  created a temperature anomaly series for the period of 1901–2003, using data from a high resolution grid dataset released by CRU (CRU-TS2.1, 0.5° × 0.5°) for whole China. The series had a correlation coefficient of 0.84 with the WYG series. The datasets for under-represented areas in the western part of China offer a complete coverage, where the long distances between the stations automatically compromise the accuracy of the interpolation method.
Basically speaking, the above-mentioned temperature series were mainly built on temperature ratings, mean temperatures of regular observations, and average of maximum and minimum temperatures. Some series were supplemented with proxy data, including historical records, ice core assessment, and tree ring analysis. Proxy data were mainly employed to qualitatively improve the interpolations for the western regions in early years. Mainly four regional averaging methods were applied — arithmetic mean, weighted average based on regional averages, weighted area average based on regional area averages, and grid area based on weighted average. Noticeable improvements lead to more reliable national temperature series for the past 100 years.
This section focuses on the CRU, WYG, LYT, TR, and TD temperature series which cover a long period and have been frequently used in climate change studies. The W and TL series are left out, as the W series shows explicit similarities to the WYG series, and the TL series to the LYT series. The ZL series is also not more closely discussed because of its estimation through 5-year running average ratings. All the five first mentioned series have been extended to 2007 based on the latest data available.
Figure 1 depicts the anomalies of the five Chinese series. All curves show similar values after 1951, though the data and methods employed are not exactly the same. It is also very apparent that the curves differ widely before 1950. For example, in the warm period of the 1920s–1930s, TR, CRU and TD show negative values compared with LYT and WYG. In addition, in the cold period of the early 20th century, TR, TD, and CRU curves are lower than LYT and WYG. This probably associates with the data and methods applied for calculating.
Annual mean temperature anomaly series (relative to 1971–2000) in China
Table 2 presents correlation coefficients for the five Chinese series and the global (GL) / Northern Hemispheric (NH) series from 1906 to 2005. For the five Chinese series only, the correlation coefficients range from 0.73 to 0.97, with the highest correlation found between TD and TR. The high correlations imply that all the series have relatively close variations. And we can also see that Chinese series gradually close to global series. The trends in Chinese series rather agree with those in the global and Northern Hemispheric series.
It is important to have a good estimate of the magnitude of temperature trends in studying climate change. Table 3 summarizes the increase rates of temperature derived from different series. For the convenience of comparison with IPCC Fourth Assessment Report [ IPCC , 2007 ], the temperature series from 1906 to 2005 are taken. As a result, increase rates range from 0.34 to 1.20°C per 100 years. CRU shows the strongest increase by 1.20°C per 100 years, followed by TR and TD. LYT shows the least increase by 0.34°C per 100 years. The differences are mainly caused by the differences from the early 20th century to 1950. In the early period, most parts of western China were not covered by weather stations. In addition, the data collected in the same period are noticeably poor in homogeneity. Taking into account the impacts of the aforesaid factors and differences in the series themselves, the determination of increase for the period before 1950 is very vague. To understand the trend in China for the past 100 years, we use the averages of the five temperature series. Consequently, we have a temperature increase rate of 0.78°C per 100 years. After deducting the estimation errors, the warming rate varies at 0.78±0.27°C per 100 years at the 95% confidence level. The result increases by 0.25°C, compared with previous estimates. The increase rate differs in different periods. For example, the period of 1908–2007 shows a slightly higher increase.
For warm years, the Chinese and NH/GL temperature series show different characteristics. Table 4 summarizes the five warmest years for the period 1906–2007. Of the Chinese series, WYG, TR, and TD experienced the warmest year in 2007, followed by 1998. For CRU, the warmest year was 1998, while for LYT it was 1946. Taking into account the fact that 3 of the 5 temperature series have used more original data compared with CRU, and that CRU has included the data collected from weather stations in adjacent countries and regions, it can be concluded that 2007 is the warmest year. However, it is not pronounced that 2007 is the warmest year in the GL and NH series.
The discrepancies might be explained by three major aspects. First, the data employed are not exactly uniform. The WYG series was estimated from temperature ratings at 50 stations, with ice core, tree rings, and historical records being the proxy data for the early period. LYT series use regular observed data collected at 711 stations. TR, TD, and CRU series work entirely or mostly on the averages of the maximum and minimum temperatures. Second, the selected stations and the number are different, and so are the spatial distribution and coverage. Third, the methodology applied in calculating the regional averages are also slightly different.
For a national series, representation is an important issue. Abovementioned analysis shows that the quantity of data and spatial distribution of weather stations employed in establishing the series make a difference in spatial representation. In the following section the analysis and assessment of the representation of these temperature series are presented.
As Wang et al.  concluded, the entire country was divided into 10 regions, including the western provinces and Taiwan. Five typical stations of each region (50 in total) were selected. A fine representation is given, thanks to a relatively even spatial coverage, though the series had fewer stations employed, compared with other series. Its high representativeness was proven by a correlation coefficient of 0.99 between the WYG series and the mean temperature series derived from a 1° x1° grid dataset after 1951. It is noted that the uncertainty of proxy data and data scarcity in the western region before 1950 has, to some extent, compromised the representation in earlier period.
The LYT series used 330 climate stations after 1951, with a relatively even distribution and fine spatial representation. Before 1950, data from 546 stations were used, which has the largest number of stations employed among the temperature series. The country was divided into 10 regions, with the western region starting to have typical stations from the 1930s. Consequently, the series had a better reliability after the 1930s. Before the 1930s, LYT mainly reflected the change in the eastern part of the country. The series have some problems with homogeneity, as it uses original mean temperature data.
Before 1950, TR and TD selected data from 231 and 291 stations respectively. After 1951, TR and TD used data collected at 616 and 291 stations respectively, which revealed a good representation. TD series and the averaged series of 971 stations have a correlation coefficient of 0.99. Additionally in TD, interpolations employed the data from 630 stations were executed, which improve its representation. Both TR and TD are impacted by limited data on grid coverage for different periods, such as 4–6 grids available for the period prior to the 20th century, 7–24 for 1901–1930, 25–38 for 1931–1950, and exceeding 43 after 1951 respectively. The correlation analysis shows that the TD series is able to reflect 2/3 of temperature change in China in the 1920s. After then, the temperature change of whole China could be basically depicted by the TD series with great improvement of representation. Both TR and LYT show similar characteristics. CRU series has its original data mainly from published statistics, containing less domestic information, compared with other series. In this context, its representation deserves further study. Overall speaking, WYG series presents a complete coverage, while others mainly reflect temperature change in the eastern part of China before the 1920s or 1930s, though they are able to reflect the general picture of climate change in most parts of the country in the latter period.
It is evidently difficult to study temperature changes across China for the past 100 years, as the lack of data and other factors. Two major factors cause the difficulties. Firstly, the incomplete coverage of data has led to an insufficient representation of early time series, and to an inhomogeneity between early and late time periods (before and after 1950). Secondly, unregulated observation standards in the first half of the 20th century also compromised the homogeneity of mean temperature records. Fortunately, efforts with remarkable outcomes have been made to address the problems: (1) strict quality control has been imposed to verify original data using diverse methods, allowing more and better historical observational data to be utilized; (2) a nationwide coverage using ice core, tree rings, and historical records has been achieved; and (3) mean temperatures derived from averaged highest and lowest temperatures have been used to address the inhomogeneity caused by different observing time. Based on these outcomes, a number of long-term temperature series have been established.
The comprehensive analysis of existing temperature series and their associated representation came up with the following conclusions: (1) for the period 1906–2005, China has experienced an average temperature increase by 0.78±0.27°C; (2) 2007 is the warmest year in China since 1906, followed by 1998; and (3) all temperature series, except the WYG series with complete coverage, mainly reflect changes in the eastern part of China before the 1920s or 1930s, but they are able to reflect the general climate change in most parts of the country since then.
Unfortunately, the lack of historical data will prevail and must be considered in the future. In the context of data coverage, proxy data is limited and can not reach the accuracy of instrumental observations. Another possible alternative is to make interpolations using the data collected from adjacent countries and regions, but its utility and rationality needs to be verified. Additionally, previous studies ignored the impacts of site-relocation and instrument-upgrade, which need to be verified and corrected in future studies. Environmental changes, especially urbanization and land use change, cause higher uncertainty in the observations. Recent studies [ Zhou and Ren, 2005 ; Zhang and Ren, 2005 ; Tang et al., 2008 ] show that the fast development of the Chinese economy (e.g., urbanization), produces noticeable effects on observed meteorological data series. The issue is so complicated to be dealt with in the existing temperature series. Urbanization is of a positive effect on temperature increase [ Zhou and Ren , 2005 ], so the real temperature increase occurred in China may be smaller than the range given in this paper.
This research is supported by the National Science & Technology Pillar Program during the Eleventh Five-Year Plan Period of China (2007BAC03A01) and the Climatic Change Project of China Meteorological Administration (CCCSF2008-10).