Advanced Search
Article Contents

Homogenization of the Daily Land Surface Temperature over the Mainland of China from 1960 through 2017

Fund Project:

This work was supported by the National Science Fund for Distinguished Young Scholars (Grant No. 41925021) and the National Natural Science Foundation of China (Grant No. 41875106)


doi: 10.1007/s00376-021-1038-6

  • Land surface temperature (LST) is one of the most important factors in the land-atmosphere interaction process. Raw measured LSTs may contain biases due to instrument replacement, changes in recording procedures, and other non-climatic factors. This study attempts to reduce the above biases in raw daily measurements and achieves a homogenized daily LST dataset over China using 2360 stations from 1960 through 2017. The high-quality land surface air temperature (LSAT) dataset is used to correct the LST warming biases especially evident during cold months in regions north of 40ºN due to the replacement of observation instruments around 2004. Subsequently, the Multiple Analysis of Series for Homogenization (MASH) method is adopted to detect and then adjust the daily observed LST records. In total, 3.68 × 103 effective breakpoints in 1.65 × 106 monthly records (about 20%) are detected. A large number of these effective breakpoints are located over large parts of the Sichuan Basin and southern China. After the MASH procedure, LSTs at more than 80% of the breakpoints are adjusted within +/– 0.5ºC, and of the remaining breakpoints, only 10% are adjusted over 1.5ºC. Compared to the raw LST dataset over the whole domain, the homogenization significantly reduces the mean LST magnitude and its interannual variability as well as its linear trend at most stations. Finally, we perform preliminary analysis upon the homogenized LST and find that the annual mean LST averaged across China shows a significant warming trend [0.22ºC (10 yr)–1]. The homogenized LST dataset can be further adapted for a variety of applications (e.g., model evaluation and extreme event characterization).
    摘要: 地表土壤温度是陆-气相互作用过程中的重要因素之一。观测的地表土壤温度数据易受台站迁移、仪器更换、记录方式变更以及观测站周边环境改变等非气候因素的影响,从而产生不同程度的偏差。这种由非气候因素造成的变化统称为时间序列的非均一性。依据这样的非均一化序列得出的研究结果,极有可能会对序列的长期线性趋势以及极端气候事件的评估产生偏差,弱化真实气候变化的影响,甚至得到与真实变化相反的结论。因此,本文的研究工作主要是将中国2360个气象观测站点1960~2017年的逐日地表土壤温度数据进行均一化处理,尽可能地减少地表土壤温度数据的非均一性偏差。我们首先利用质量较高的观测气温数据修正了地表土壤温度自2005年起北纬40o以北地区冷月份出现的暖偏差现象。随后,应用多元均一化方法(MASH)对逐日地表土壤温度数据的非均一性进行检测和校订。总的来说,在1.65 × 106个月均记录中,共检测出3.68 × 103个有效断点记录(约占总数的20%),有效断点记录较多的站点主要分布在四川盆地地区以及中国南方的大部分地区。通过MASH方法调整之后,地表土壤温度序列有超过80%的偏差调整分布在+/– 0.5ºC之间,只有10%的偏差调整大小超过1.5ºC。我们将均一化后与原始的观测地表土壤温度数据进行比较,发现均一化过程显著降低了大部分站点地表土壤温度的平均量级、年际变率以及线性趋势。最后,我们还对均一化的地表土壤温度进行了初步分析,结果表明近年来中国年均地表土壤温度呈显著的增温趋势[0.22ºC (10 yr)–1]。本文的均一化逐日地表土壤温度的观测数据集今后可进一步为各科学研究应用提供可靠的数据支撑(例如:模型评估和极端事件表征等)。
  • 加载中
  • Figure 1.  (a) Location of the meteorological stations (dots) and the division of the nine subregions (black curves). Black dots (a total of 197 dots) denote the specific stations described in Section 3.1. (b) The number of daily stations with satisfactory data in each year. The top (bottom) of the whisker plot represents the 90th (10th) percentile of the station number, the top (bottom) of the box represents the 75th (25th) percentile, and the middle line represents the 50th percentile. The straight line indicates the final 2360 stations used in the current study (Additional information for the stations is shown in Table 1.)

    Figure 2.  Time series of the raw daily LST averaged across 197 stations (the black dots in Fig. 1a) in December for 1960−2017. The red solid line is a reference line to separate the years before and after 2005; the two blue dotted lines are the mean LSTs derived from the raw daily dataset during 1960−2005 and 2005−17. The green solid curve represents the daily LST time series after LSAT adjustments (section 3.1), and the red dotted line is the mean LST during 2005−17. The corresponding mean LST values averaged for different periods are also indicated.

    Figure 3.  Case station (ID 50862, 46.98ºN, 128.05ºE) for the raw and calculated LST, and the homogenized results (units: ºC). (a) Raw daily LST plotted against the calculated daily LST in each month during 1960−2004, where the black solid line is the reference (x = y) line, (b) raw (blue) and MASH homogenized (orange) daily LST, and (c) raw and calculated (green line) LST during 2003−07. The calculated daily LST is the LST adjusted by the LSAT (described in section 3.1).

    Figure 4.  Annual LST series at station ID 57710 (27.85ºN, 106.37ºE, a candidate station) in the YG subregion (units: ºC). The specific information of nine reference stations is ID 57714 (27.36ºN, 106.20ºE), ID 57717 (27.68ºN, 106.92ºE), ID 57713 (27.68ºN, 106.92ºE), ID 57606 (28.33ºN, 106.67ºE), ID 57720 (27.95ºN, 107.17ºE), ID 57614 (28.55ºN, 105.98ºE), ID 57803 (27.05ºN, 106.00ºE), ID 57718 (27.03ºN, 106.72ºE), and ID 57719 (27.07ºN, 106.97ºE), respectively. (a) Distribution of stations, including the candidate station (red dot), the nine nearest reference stations (green dots), and other stations (black dots) in the YG subregion, (b) the raw annual LST anomalies of the candidate station (red curve) and the nine reference stations (black curves), and (c) annual LST from the raw (red curve) and homogenized time series at the candidate station.

    Figure 5.  The number of effective breakpoints for the monthly LST records at the effective stations during 1960−2017; (a) the spatial distribution at all stations and (b) the accumulation in each month and each year across all stations.

    Figure 6.  Frequency distribution for the effective breakpoints of LST in the nine subregions and for the whole of the Chinese mainland. Biases at these breakpoints are adjusted with MASH.

    Figure 7.  Percentage of records with adjusted LST biases between –0.5ºC and 0.5ºC in different months over the nine subregions and the whole of the Chinese mainland.

    Figure 8.  Spatial distribution of the differences of the monthly means [(a), (b)] and standard deviations [(c), (d)] between the homogenized and raw LSTs in January [(a), (c)] and July [(b), (d)] (units: ºC). The filled dot and the “+” symbol indicate the mean LST differences passing or not passing the significance test (p = 0.05), respectively. The histogram plot in the lower-left corner shows the frequency distribution of the significant differences.

    Figure 9.  Distribution of linear trends of the monthly raw and homogenized LST in January [(a), (b)] and July [(c), (d)]. The filled dot (“+”) indicates stations with trends passing (not passing) the significance test (p = 0.05).

    Figure 10.  (a) Distribution of the multiyear mean homogenized LST and (b) its time series and linear trend averaged over all stations in China (LST/LSAT with the solid line/dotted line) for 1960−2017 (units: ºC).

    Table 1.  Number of stations in each subregion

    Full nameAbbreviationNumber of stations
    Huanghe-Huaihe-Haihe PlainHHH414
    Loess PlateauLP202
    Middle-lower Yangtze PlainYZ488
    Northeast China PlainNE183
    Northern arid and semiarid regionNA328
    Qinghai-Tibet PlateauTB90
    Sichuan Basin and surrounding regionsSC200
    South ChinaSE164
    Yunnan-Guizhou PlateauYG291
    Mainland of ChinaChina2360
    DownLoad: CSV
  • Aguilar, E., I. Auer, M. Brunet, T. C. Peterson, and J. Wieringa, 2003: Guidelines on climate metadata and homogenization. WMO/TD No. 1186.
    Cao, L. J., and Z. W. Yan, 2012: Progress in research on homogenization of climate data. Advances in Climate Change Research, 3, 59−67, https://doi.org/10.3724/SP.J.1248.2012.00059.
    Cao, L. J., P. Zhao, Z. W. Yan, P. Jones, Y. N. Zhu, Y. Yu, and G. L. Tang, 2013: Instrumental temperature series in eastern and central China back to the nineteenth century. J. Geophys. Res., 118, 8197−8207, https://doi.org/10.1002/jgrd.50615.
    Cao, L. J., Y. N. Zhu, G. L. Tang, F. Yuan, and Z. W. Yan, 2016: Climatic warming in China according to a homogenized data set from 2419 stations. International Journal of Climatology, 36, 4384−4392, https://doi.org/10.1002/joc.4639.
    China Meteorological Administration, 2003: Specifications for Surface Meteorological Observation. Meteorology Press, Beijing, 85−89. (in Chinese)
    Cohen, J., and D. Rind, 1991: The effect of snow cover on the climate. J. Climate, 4, 689−706, https://doi.org/10.1175/1520-0442(1991)004<0689:TEOSCO>2.0.CO;2.
    Du, J. Z., K. C. Wang, B. S. Cui, and S. J. Jiang, 2020: Correction of inhomogeneities in observed land surface temperatures over China. J. Climate, 33, 8885−8902, https://doi.org/10.1175/jcli-d-19-0521.1.
    Groisman, P. Y., T. R. Karl, and R. W. Knight, 1994: Observed impact of snow cover on the heat balance and the rise of continental spring temperatures. Science, 263, 198−200, https://doi.org/10.1126/science.263.5144.198.
    Guijarro, J. A., J. A. López, E. Aguilar, P. Domonkos, V. K. C. Venema, J. Sigró, and M. Brunet, 2017: Comparison of homogenization packages applied to monthly series of temperature and precipitation: The multitest project. Proc. 9th Seminar for Homogenization and Quality Control in Climatological Databases and Fourth Conference on Spatial Interpolation Techniques in Climatology and Meteorology, WCDMP-No. 85, World Meteorological Organization, Budapest, Hungary, 46−62.
    Hu, Q., and S. Feng, 2003: A daily soil temperature dataset and soil temperature climatology of the contiguous United States. J. Appl. Meteor., 42, 1139−1156, https://doi.org/10.1175/1520-0450(2003)042<1139:adstda>2.0.co;2.
    Li, Q. X., 2016: Climate data homogeneity studies in China: Progresses and prospects. Advances in Meteorological Science and Technology, 6, 67−74, https://doi.org/10.3969/j.issn.2095-1973.2016.03.009.
    Li, Z., and Z. W. Yan, 2010: Application of multiple analysis of series for homogenization to Beijing daily temperature series (1960−2006). Adv. Atmos. Sci., 27, 777−787, https://doi.org/10.1007/s00376-009-9052-0.
    Li, Z., Z. W. Yan, L. J. Cao, and P. D. Jones, 2018: Further-adjusted long-term temperature series in China based on MASH. Adv. Atmos. Sci., 35, 909−917, https://doi.org/10.1007/s00376-018-7280-x.
    Li, Z., Z. W. Yan, Y. N. Zhu, N. Freychet, and S. Tett, 2020: Homogenized daily relative humidity series in China during 1960−2017. Adv. Atmos. Sci., 37, 318−327, https://doi.org/10.1007/s00376-020-9180-0.
    Liu, X. N., Z. H. Ren, and Y. Wang, 2008: Differences between automatic-observed and manual-observed surface temperature. Journal of Applied Meteorological Science, 19, 554−563, https://doi.org/10.3969/j.issn.1001-7313.2008.05.006. (in Chinese with English abstract
    McMichael, B. L., and J. J. Burke, 1998: Soil temperature and root growth. HortScience, 33, 947−951, https://doi.org/10.21273/HORTSCI.33.6.947.
    Peterson, T. C., and Coauthors, 1998: Homogeneity adjustments of in situ atmospheric climate data: A review. International Journal of Climatology, 18, 1493−1517, https://doi.org/10.1002/(sici)1097-0088(19981115)18:13<1493::aid-joc329>3.0.co;2-t.
    Peterson, T. C., and Coauthors, 2002: Recent changes in climate extremes in the Caribbean region. J. Geophys. Res., 107, 4601, https://doi.org/10.1029/2002jd002251.
    Qian, B. D., E. G. Gregorich, S. Gameda, D. W. Hopkins, and X. L. Wang, 2011: Observed soil temperature trends associated with climate change in Canada. J. Geophys. Res., 116, D02106, https://doi.org/10.1029/2010jd015012.
    Ren, Z. H., G. A. Wang, and F. L. Zou, 2013: Difference between soil temperatures obtained through automatic observation and manual observation and analysis of its causes. Acta Pedologica Sinica, 50, 657−663, https://doi.org/10.11766/trxb201210060395. (in Chinese with English abstract
    Shi, P. J., S. Sun, M. Wang, L. Ning, J. A. Wang, Y. Y. Jin, X. T. Gu, and W. X. Yin, 2014: Climate change regionalization in China (1961−2010). Science China Earth Sciences, 57, 2676−2689, https://doi.org/10.1007/s11430-014-4889-1.
    Szentimrey, T., 1999: Multiple analysis of series for homogenization (MASH). Proc. of the 2nd Seminar for Homogenization of Surface Climatological Data, Geneva: WMO, 27−46.
    Szentimrey, T., 2013: Theoretical questions of daily data homogenization. Idojaras, 117, 113−122.
    Szentimrey, T., 2014: Manual of Homogenization Software MASHv3.03. Hungarian Meteorological Service, 69 pp.
    Tao, S. Y., C. B. Fu, Z. M. Zeng, and Q. Y. Zhang, 1991: Two long-term instrumental climatic data bases of the People's Republic of China. Oak Ridge National Laboratory ORNL/CDIAC-47, Oak Ridge, https://doi.org/10.3334/CDIAC/cli.ndp039.
    Vancutsem, C., P. Ceccato, T. Dinku, and S. J. Connor, 2010: Evaluation of MODIS land surface temperature data to estimate air temperature in different ecosystems over Africa. Remote Sensing of Environment, 114, 449−465, https://doi.org/10.1016/j.rse.2009.10.002.
    Wang, X. L., and Y. Feng, 2013: RHtestsV4 user manual. Climate Research Division, Atmospheric Science and Technology Directorate Science and Technology Branch, Environment Canada, Toronto, Ontario, Canada, 29 pp. [Available online from http://etccdi.pacificclimate.org/software.shtml]
    Wang, X. L., H. F. Chen, Y. H. Wu, Y. Feng, and Q. Pu, 2010: New techniques for the detection and adjustment of shifts in daily precipitation data series. J. Appl. Meteor. Climatol., 49, 2416−2436, https://doi.org/10.1175/2010jamc2376.1.
    Wu, L. Y., and J. Y. Zhang, 2014: Strong subsurface soil temperature feedbacks on summer climate variability over the arid/semi-arid regions of East Asia. Atmospheric Science Letters, 15, 307−313, https://doi.org/10.1002/asl2.504.
    Xu, W. H., C. H. Sun, J. Q. Zuo, Z. G. Ma, W. J. Li, and S. Yang, 2019: Homogenization of monthly ground surface temperature in China during 1961−2016 and performances of GLDAS reanalysis products. J. Climate, 32, 1121−1135, https://doi.org/10.1175/jcli-d-18-0275.1.
    Xue, Y. K., and Coauthors, 2018: Spring land surface and subsurface temperature anomalies and subsequent downstream late spring-summer droughts/floods in North America and East Asia. J. Geophys. Res., 123, 5001−5019, https://doi.org/10.1029/2017jd028246.
    Xue, Y. K., R. Vasic, Z. Janjic, Y. M. Liu, and P. C. Chu, 2012: The impact of spring subsurface soil temperature anomaly in the western U.S. on North American summer precipitation: A case study using regional climate model downscaling. J. Geophys. Res., 117, D11103, https://doi.org/10.1029/2012jd017692.
    Yan, Z. W., Z. Li, Q. X. Li, and P. Jones, 2010: Effects of site change and urbanisation in the Beijing temperature series 1977−2006. International Journal of Climatology, 30, 1226−1234, https://doi.org/10.1002/joc.1971.
    Zeng, X. B., Z. Wang, and A. H. Wang, 2012: Surface skin temperature and the interplay between sensible and ground heat fluxes over arid regions. Journal of Hydrometeorology, 13, 1359−1370, https://doi.org/10.1175/jhm-d-11-0117.1.
    Zhou, C. L., and K. C. Wang, 2016: Land surface temperature over global deserts: Means, variability, and trends. J. Geophys. Res., 121, 1 4344−1 4357, https://doi.org/10.1002/2016jd025410.
    Zhou, C. L., K. C. Wang, and Q. Ma, 2017: Evaluation of eight current reanalyses in simulating land surface temperature from 1979 to 2003 in China. J. Climate, 30, 7379−7398, https://doi.org/10.1175/jcli-d-16-0903.1.
    Zhou, L. T., and R. H. Huang, 2006: Characteristics of interdecadal variability of the difference between surface temperature and surface air temperature in spring in arid and semi-arid region of Northwest China and its impact on summer precipitation in North China. Climatic and Environmental Research, 11, 1−13, https://doi.org/10.3878/j.issn.1006-9585.2006.01.01. (in Chinese with English abstract
    Zhou, L. T., and L. M. Wen, 2016: Characteristics of temporal and spatial variations in land-air temperature difference in China and its association with summer rainfall. Climatic and Environmental Research, 21, 621−632, https://doi.org/10.3878/j.issn.1006-9585.2016.15196. (in Chinese with English abstract
  • [1] LI Zhen, YAN Zhongwei, 2010: Application of Multiple Analysis of Series for Homogenization to Beijing Daily Temperature Series (1960--2006), ADVANCES IN ATMOSPHERIC SCIENCES, 27, 777-787.  doi: 10.1007/s00376-009-9052-0
    [2] GUO Yanjun, DING Yihui, 2011: Impacts of Reference Time Series on the Homogenization of Radiosonde Temperature, ADVANCES IN ATMOSPHERIC SCIENCES, 28, 1011-1022.  doi: 10.1007/s00376-010-9211-3
    [3] ZHANG Guo, ZHOU Guangsheng, CHEN Fei, WANG Yu, , 2014: Analysis of the Variability of Canopy Resistance over a Desert Steppe Site in Inner Mongolia, China, ADVANCES IN ATMOSPHERIC SCIENCES, 31, 681-692.  doi: 10.1007/s00376-013-3071-6
    [4] Zhen LI, Zhongwei YAN, Lijuan CAO, Phil D. JONES, 2018: Further-Adjusted Long-Term Temperature Series in China Based on MASH, ADVANCES IN ATMOSPHERIC SCIENCES, 35, 909-917.  doi: 10.1007/s00376-018-7280-x
    [5] Huqiang ZHANG, LI Yaohui, GAO Xuejie, 2009: Potential Impacts of Land-Use on Climate Variability and Extremes, ADVANCES IN ATMOSPHERIC SCIENCES, 26, 840-854.  doi: 10.1007/s00376-009-8047-1
    [6] Zhen LI, Zhongwei YAN, Yani ZHU, Nicolas FREYCHET, Simon TETT, 2020: Homogenized Daily Relative Humidity Series in China during 1960−2017, ADVANCES IN ATMOSPHERIC SCIENCES, 37, 318-327.  doi: 10.1007/s00376-020-9180-0
    [7] FU Weiwei, 2012: Altimetric Data Assimilation by EnOI and 3DVAR in a Tropical Pacific Model: Impact on the Simulation of Variability, ADVANCES IN ATMOSPHERIC SCIENCES, 29, 823-837.  doi: 10.1007/s00376-011-1022-7
    [8] BIAN Lingen, LIN Xiang, 2012: Interdecadal Change in the Antarctic Circumpolar Wave during 1951--2010, ADVANCES IN ATMOSPHERIC SCIENCES, 29, 464-470.  doi: 10.1007/s00376-011-1143-z
    [9] LI Zhen, YAN Zhongwei, TU Kai, WU Hongyi, 2015: Changes of Precipitation and Extremes and the Possible Effect of Urbanization in the Beijing Metropolitan Region during 1960-2012 Based on Homogenized Observations, ADVANCES IN ATMOSPHERIC SCIENCES, 32, 1173-1185.  doi: 10.1007/s00376-015-4257-x
    [10] Bo SUN, 2018: Asymmetric Variations in the Tropical Ascending Branches of Hadley Circulations and the Associated Mechanisms and Effects, ADVANCES IN ATMOSPHERIC SCIENCES, 35, 317-333.  doi: 10.1007/s00376-017-7089-z
    [11] LI Zhen, YAN Zhongwei, TU Kai, LIU Weidong, WANG Yingchun, 2011: Changes in Wind Speed and Extremes in Beijing during 1960--2008 Based on Homogenized Observations, ADVANCES IN ATMOSPHERIC SCIENCES, 28, 408-420.  doi: 10.1007/s00376-010-0018-z
    [12] WANG Xinmin, ZHAI Panmao, WANG Cuicui, 2009: Variations in Extratropical Cyclone Activity in Northern East Asia, ADVANCES IN ATMOSPHERIC SCIENCES, 26, 471-479.  doi: 10.1007/s00376-009-0471-8
    [13] A.M.Selvam, M.Radhamani, 1994: Signatures of a Universal Spectrum for Nonlinear Variability in Daily Columnar Total Ozone Content, ADVANCES IN ATMOSPHERIC SCIENCES, 11, 335-342.  doi: 10.1007/BF02658153
    [14] Zhang Qiang, Cao Xiaoyan, Wei Guoan, Huang Ronghui, 2002: Observation and Study of Land Surface Parameters over Gobi in Typical Arid Region, ADVANCES IN ATMOSPHERIC SCIENCES, 19, 121-135.  doi: 10.1007/s00376-002-0039-3
    [15] Xinping XU, Fei LI, Shengping HE, Huijun WANG, 2018: Subseasonal Reversal of East Asian Surface Temperature Variability in Winter 2014/15, ADVANCES IN ATMOSPHERIC SCIENCES, 35, 737-752.  doi: 10.1007/s00376-017-7059-5
    [16] WEN Jun, WEI Zhigang, LU Shihua, CHEN Shiqiang, AO Yinhuan, LIANG Ling, 2007: Autumn Daily Characteristics of Land Surface Heat and Water Exchange over the Loess Plateau Mesa in China, ADVANCES IN ATMOSPHERIC SCIENCES, 24, 301-310.  doi: 10.1007/s00376-007-0301-9
    [17] Haoxin ZHANG, Naiming YUAN, Zhuguo MA, Yu HUANG, 2021: Understanding the Soil Temperature Variability at Different Depths: Effects of Surface Air Temperature, Snow Cover, and the Soil Memory, ADVANCES IN ATMOSPHERIC SCIENCES, 38, 493-503.  doi: 10.1007/s00376-020-0074-y
    [18] WANG Shaowu, ZHU Jinhong, CAI Jingning, 2004: Interdecadal Variability of Temperature and Precipitation in China since 1880, ADVANCES IN ATMOSPHERIC SCIENCES, 21, 307-313.  doi: 10.1007/BF02915560
    [19] Li Wei, Yu Rucong, Zhang Xuehong, 2001: Impacts of Sea Surface Temperature in the Tropical Pacific on Interannual Variability of Madden-Julian Oscillation in Precipitation, ADVANCES IN ATMOSPHERIC SCIENCES, 18, 429-444.  doi: 10.1007/BF02919322
    [20] Jiangyu MAO, Ming WANG, 2018: The 30-60-day Intraseasonal Variability of Sea Surface Temperature in the South China Sea during May-September, ADVANCES IN ATMOSPHERIC SCIENCES, 35, 550-566.  doi: 10.1007/s00376-017-7127-x

Get Citation+

Export:  

Share Article

Manuscript History

Manuscript received: 21 January 2021
Manuscript revised: 25 May 2021
Manuscript accepted: 28 May 2021
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Homogenization of the Daily Land Surface Temperature over the Mainland of China from 1960 through 2017

    Corresponding author: Aihui WANG, Wangaihui@mail.iap.ac.cn
  • 1. Nansen-Zhu International Research Center, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China.
  • 2. University of Chinese Academy of Sciences, Beijing 100029, China

Abstract: Land surface temperature (LST) is one of the most important factors in the land-atmosphere interaction process. Raw measured LSTs may contain biases due to instrument replacement, changes in recording procedures, and other non-climatic factors. This study attempts to reduce the above biases in raw daily measurements and achieves a homogenized daily LST dataset over China using 2360 stations from 1960 through 2017. The high-quality land surface air temperature (LSAT) dataset is used to correct the LST warming biases especially evident during cold months in regions north of 40ºN due to the replacement of observation instruments around 2004. Subsequently, the Multiple Analysis of Series for Homogenization (MASH) method is adopted to detect and then adjust the daily observed LST records. In total, 3.68 × 103 effective breakpoints in 1.65 × 106 monthly records (about 20%) are detected. A large number of these effective breakpoints are located over large parts of the Sichuan Basin and southern China. After the MASH procedure, LSTs at more than 80% of the breakpoints are adjusted within +/– 0.5ºC, and of the remaining breakpoints, only 10% are adjusted over 1.5ºC. Compared to the raw LST dataset over the whole domain, the homogenization significantly reduces the mean LST magnitude and its interannual variability as well as its linear trend at most stations. Finally, we perform preliminary analysis upon the homogenized LST and find that the annual mean LST averaged across China shows a significant warming trend [0.22ºC (10 yr)–1]. The homogenized LST dataset can be further adapted for a variety of applications (e.g., model evaluation and extreme event characterization).

摘要: 地表土壤温度是陆-气相互作用过程中的重要因素之一。观测的地表土壤温度数据易受台站迁移、仪器更换、记录方式变更以及观测站周边环境改变等非气候因素的影响,从而产生不同程度的偏差。这种由非气候因素造成的变化统称为时间序列的非均一性。依据这样的非均一化序列得出的研究结果,极有可能会对序列的长期线性趋势以及极端气候事件的评估产生偏差,弱化真实气候变化的影响,甚至得到与真实变化相反的结论。因此,本文的研究工作主要是将中国2360个气象观测站点1960~2017年的逐日地表土壤温度数据进行均一化处理,尽可能地减少地表土壤温度数据的非均一性偏差。我们首先利用质量较高的观测气温数据修正了地表土壤温度自2005年起北纬40o以北地区冷月份出现的暖偏差现象。随后,应用多元均一化方法(MASH)对逐日地表土壤温度数据的非均一性进行检测和校订。总的来说,在1.65 × 106个月均记录中,共检测出3.68 × 103个有效断点记录(约占总数的20%),有效断点记录较多的站点主要分布在四川盆地地区以及中国南方的大部分地区。通过MASH方法调整之后,地表土壤温度序列有超过80%的偏差调整分布在+/– 0.5ºC之间,只有10%的偏差调整大小超过1.5ºC。我们将均一化后与原始的观测地表土壤温度数据进行比较,发现均一化过程显著降低了大部分站点地表土壤温度的平均量级、年际变率以及线性趋势。最后,我们还对均一化的地表土壤温度进行了初步分析,结果表明近年来中国年均地表土壤温度呈显著的增温趋势[0.22ºC (10 yr)–1]。本文的均一化逐日地表土壤温度的观测数据集今后可进一步为各科学研究应用提供可靠的数据支撑(例如:模型评估和极端事件表征等)。

    • Land surface temperature (LST) is a fundamental parameter in the climate field that is indicative of the thermal radiance from the ground surface. It directly influences the sensible and latent heat fluxes from the land surface to the planetary boundary layer (Zeng et al., 2012; Zhou and Wen, 2016). Changes in LST and soil temperature can also be used as indicators of climate change (McMichael and Burke, 1998; Qian et al., 2011). Xue et al. (2018) found that soil temperature has a relationship with subsequent downstream climate variables and can be used as a predictor of extreme hydrological events. Furthermore, several researchers have considered the difference between LST and land surface air temperature (LSAT) as a teleconnection factor to predict changes in precipitation in downstream regions (Zhou and Huang, 2006; Xue et al., 2012). Therefore, through its local and teleconnection effects, LST plays a significant role in modulating climate and climatic variability.

      Long-term instrumental LST data have been collected and stored for many decades in China. As one of the conventional observations, LST is liable to be influenced by changes in non-climatic factors (e.g., station migration, equipment replacement, and changes in the environment around the station location) (Li, 2016). Among these factors, equipment replacement is always conducted at multiple stations simultaneously [e.g., see Fig. 1 of Xu et al. (2019)]. Ren et al. (2013) compared the soil temperature observed from both manual and automatic techniques at the same stations during the same period in China and found that significant differences existed. Furthermore, the sites of approximately 80% of observation stations have been relocated at least once since 1950 due to the growth and expansion of cities (Cao et al., 2013). If we keep such inhomogeneities and directly analyze the raw data, which may inaccurately describe actual climate variations, then we will potentially reach incorrect conclusions, particularly in the representation of long-term trends and extremes (Peterson et al., 2002; Xue et al., 2012). Therefore, it is necessary to homogenize the raw observed LST before it can be used as the “ground truth” in various applications.

      Figure 1.  (a) Location of the meteorological stations (dots) and the division of the nine subregions (black curves). Black dots (a total of 197 dots) denote the specific stations described in Section 3.1. (b) The number of daily stations with satisfactory data in each year. The top (bottom) of the whisker plot represents the 90th (10th) percentile of the station number, the top (bottom) of the box represents the 75th (25th) percentile, and the middle line represents the 50th percentile. The straight line indicates the final 2360 stations used in the current study (Additional information for the stations is shown in Table 1.)

      To homogenize the LST dataset, several methods have been developed and applied in different countries (e.g., Hu and Feng, 2003; Zhou et al., 2017; Xu et al., 2019). Hu and Feng (2003) applied a quality-control method to the in situ soil temperature in the United States and then reproduced a high-quality soil temperature dataset at multiple soil layers. Zhou et al. (2017) applied the RHtest software package to homogenize a monthly observed LST dataset from approximately 2200 stations from 1979 to 2003 in China and then used the homogenized LST to assess eight reanalysis products. Xu et al. (2019) incorporated additional ancillary and metadata into a raw LST dataset and then constructed a homogenized monthly LST dataset from 686 meteorological stations in China. Previous studies have mainly focused on homogenizing LST datasets on monthly time scales. The daily LST can describe temporal variations at synoptic scales and is a needed parameter in weather forecasting as well as in modeling applications (Xue et al., 2012; Wu and Zhang, 2014). For instance, the daily LST from atmospheric reanalysis products has been widely used in climate research, but most of these projects underestimate the LST in China (e.g., Zhou and Wang, 2016; Zhou et al., 2017; Xu et al., 2019). Therefore, a publicly accessible homogenized and high-quality, in situ, daily LST dataset in China is urgently needed.

      Since the 1980s, various methods and techniques have been developed and used to eliminate data inhomogeneities (Peterson et al., 1998; Aguilar et al., 2003; Cao and Yan, 2012). However, there is no universal homogenization method for all climatic elements across different spatial and temporal scales. Each homogenizing method has its advantages and limitations, and the choice and applicability of one method versus another are often dictated by a few influencing factors, such as the station density, availability of metadata, and the type of variable (Peterson et al., 1998). Globally, two methods are popularly used to homogenize in situ observations. One is the RHtest method (Wang et al., 2010; Wang and Feng, 2013) and the other is the Multiple Analysis of Series for Homogenization (MASH, Szentimrey, 1999, 2013, 2014). The RHtest method applies a two-phase regression model to calculate linear trends of a target time series, detect their statistically significant breakpoints, and finally, adjust them using the quantile-matching adjustment method. The MASH method uses a point-to-point comparison to detect breakpoints through an intercomparison of station observations within the same climatic area. Then, the method homogenizes the breakpoints relied on by its neighbors (a detailed description is provided in section 2.2). Because the MASH method is not overly restricted by metadata, it is widely used to homogenize meteorological variables. Li and Yan (2010) applied the MASH method to the daily air temperature of Beijing with additional reference information (with or without metadata) and found that the MASH method could successfully detect the majority of inhomogeneities in the raw dataset. For a long-term historical station observation dataset, it is usually difficult to preserve and obtain metadata (Tao et al., 1991), and this is the case for the in situ measured LST dataset in China. Nevertheless, the MASH method allows users to perform data homogenization in the case of no metadata. As a result, we select the MASH method in the present work.

      In this study, we aim to develop a high-quality LST dataset that can be used in future applications for long-term climatic and soil-related researches. Using the MASH method, we examine and identify the inhomogeneities in the daily in situ collected LST from 2360 stations in the mainland of China. The manuscript is organized as follows. Section 2 describes the dataset and methods used in this study. Section 3 presents the detailed homogenization procedure. Section 4 demonstrates and discusses the changes in LST due to homogenization. The conclusions are presented in section 5.

    2.   Data and Methods
    • The raw daily LST dataset from meteorological stations in China, spanning the period 1960−2017, was obtained from the National Meteorological Information Center of the China Meteorological Administration (NMIC/CMA). Fundamental quality control (e.g., screening for unreasonable extreme values) was performed on this dataset before it was released to the public (http://data.cma.cn/data). In China, a large number of meteorological stations were established in the early 1950s, and the station density and records were maintained with relatively high continuity and stability after 1960 (Shi et al., 2014; Cao et al., 2016). Currently, more than 2400 national meteorological stations exist in the mainland of China. Land surface temperature is one of the factors regularly measured and collected every day at the meteorological stations. Before 2000, LST was recorded and stored manually four times a day (at 0200, 0800, 1400, and 2000 Beijing time, Beijing time=UTC+8) with a surface and bent stem earth mercurial thermometer. After 2000, the observation system was updated gradually to employ a platinum resistance temperature sensor, which is an automatic instrument and is more effective than mercurial thermometers. After the update, the observation frequency became hourly (twenty-four times a day, Ren et al., 2013). The renewal period is concentrated during 2000−07 across the mainland of China, and the observation systems north of 40ºN were updated in 2004 [see Fig. 1 of Xu et al. (2019)]. The arithmetic average of the observations collected at multiple times within a day is regarded as the daily LST in the data collection. Liu et al. (2008) compared differences between automatic and manual observed LSTs in daily, monthly, and annual scales during the parallel observation period and found that the substantial differences mainly appear in snow-covered regions, e.g., northern Heilongjiang, northern Inner Mongolia, and most of Xinjiang provinces. In Southern China, the observed differences between the two instruments are very small.

      In the present work, daily LST measurements from up to 2360 meteorological stations are used. The majority of stations are located in relatively low elevation regions, which are usually densely populated (Fig. 1a). For example, there is a much higher density of stations in the Yellow River and Yangtze River basins, Southeast China, and the Central China Plain compared to other regions. The station distribution is relatively sparse in high elevations and sparsely populated areas, such as Northwest China and the Qinghai-Tibet Plateau regions. In addition, the number of effective stations (i.e., stations with measurements) was 1476 in 1960, after which it increased generally over time (stable during 1980−2010). In 2017 the number of effective stations increased to 2628 (Fig. 1b).

    • The MASH method was developed by Szentimrey (1999, 2014) and is based on the hypothesis test method to detect possible breakpoints at a given significance level. Through comparisons of measurements at different stations within the same climate region, the MASH method does not require prior assumptions of a homogeneous time series. This method has been widely applied to detect and adjust the inhomogeneity of raw meteorological observations at ground stations (e.g., Li et al., 2018, 2020). Guijarro et al. (2017) tested nine commonly used homogenization methods (including the RHtest and MASH methods) and compared their homogenized performances in terms of the root mean square errors and trends of the air temperature time series. They found that the results obtained by MASH method were comparatively reliable (Guijarro et al., 2017).

      To homogenize daily records, the MASH method must first homogenize the corresponding monthly data. Therefore, the daily LST at each station is first aggregated to monthly values, and then, the breakpoints of the monthly LST are detected and adjusted. Finally, the MASH method is again applied to homogenize the daily LST with the incorporation of the homogenized monthly values. Detailed information of the MASH method, including the mathematics it uses and its technique, is provided in its online manual (https://www.met.hu/en/omsz/rendezvenyek/homogenization_and_interpolation/software/). The latest version of MASH, v3.03, is used in this study. According to the description in the MASH manual, two different models can be selected and used. One is the additive model, which requires that the targeted dataset has a normal distribution (e.g., temperature). The other is the multiplicative model and can be used for a quasi-lognormally distributed data series (e.g., precipitation). Moreover, the Monte Carlo method is used in MASH to detect the inhomogeneity of time series with its threshold of precision (0.1, 0.05, or 0.01). In general, the smaller the Monte Carlo’s threshold is, the more stringent the detection of inhomogeneity will be (Szentimrey, 2014). Using different thresholds might reduce the number of breakpoints, but the frequency distribution patterns of breakpoints are similar (figures not shown). In this study, we choose the additive model and define a Monte Carlo threshold of 0.05 as the significance level.

      In the original MASH procedure, the monthly/annual value is set as missing if any missing day/month exists within that month/year. Such a strict criterion will lead to the loss of many useful records because many stations only have few days missing in a specific month. Furthermore, the loss of useful data is not the intention of data homogenization. Here, we use a lenient threshold condition for the missing value judgment to retain as many useful observation stations as possible. At each station, the monthly LST is set as a missing value when more than nine days of daily LST is missing in a given month. We also investigated the impact of different criteria (i.e., 8, 9, 10 days missing) for the count of missing months and found that there is little difference when varying criteria for a 28, 29, 30, and 31 day month. Meanwhile, the annual LST is set as a missing value when more than three continuous monthly values are missing. Finally, we remove stations with available annual values totaling less than 30 years of the full 58 years. Eventually, 2360 stations remained and were homogenized by the MASH method (shown as the straight solid line in Fig. 1b). According to the principle of MASH as well as the spatial variability of LST, we perform MASH in each individual climate region. Based on the natural conditions for agricultural production and also in consideration of climate characteristics, we divide the mainland of China into nine subregions (Fig. 1a): the Huanghe-Huaihe-Haihe Plain (HHH), Loess Plateau (LP), Middle-lower Yangtze Plain (YZ), Northeast China Plain (NE), Northern arid and semiarid region (NA), Qinghai-Tibet Plateau (TB), Sichuan Basin and surrounding regions (SC), South China (SE), and the Yunnan-Guizhou Plateau (YG). The shapefile of each subregion is available from the Institute of Geographical Sciences and Resources, Chinese Academy of Sciences (http://www.resdc.cn/data.aspx?DATAID=275), and the number of effective stations in each subregion is shown in Table 1.

      Full nameAbbreviationNumber of stations
      Huanghe-Huaihe-Haihe PlainHHH414
      Loess PlateauLP202
      Middle-lower Yangtze PlainYZ488
      Northeast China PlainNE183
      Northern arid and semiarid regionNA328
      Qinghai-Tibet PlateauTB90
      Sichuan Basin and surrounding regionsSC200
      South ChinaSE164
      Yunnan-Guizhou PlateauYG291
      Mainland of ChinaChina2360

      Table 1.  Number of stations in each subregion

      In addition, other statistical methods, including linear regression and standard deviation, are used to comparatively analyze the raw and homogenized LST dataset. A two-tailed Student’s test is used to test the significance of these statistics.

    3.   Inhomogeneity of the raw climate time series
    • As mentioned in the introduction section, automatic instruments for LST measurements began to replace manual ones in 2004 in northern China (mainly north of 40ºN, including Xinjiang Province, NE, and NA subregions, shown as black dots in Fig. 1a), which resulted in LSTs increasing abruptly in cold months since 2005. To understand this result, we select one cold month (December) as an example to analyze the time series of LST over northern China. Figure 2 shows the daily LST time series averaged across 197 stations in northern China in December from 1960 to 2017. There is a distinct jump in 2005, after which the magnitude of LST increases remarkably. The mean value is –8.1ºC for 2005−17, which is 8ºC higher than that for 1960−2005 (–16.1ºC). This phenomenon is prevalent at all 197 stations. Xu et al. (2019) found a similar problem in the same regions (the NE and NA subregions) when they compared the differences of LST measurements from manual and automatic instruments during parallel observation periods. The warm shift phenomenon is principally caused by the change of the observation system from a manual to an automatic one around 2004. In winter, when the ground surface is covered by snow, the manual instrument measures the temperature at the snow surface, whereas the automatic instrument sensor measures the temperature at the soil surface under snow. Because snow is a strong insulator of heat, it can absorb both longwave radiation from the ground surface transmitted upward and shortwave radiation from the atmosphere transmitted downward (Cohen and Rind, 1991; Groisman et al., 1994). Therefore, the measured LST from the automatic instrument would change much more slowly under snow and cannot represent real LST change features (Liu et al., 2008; Ren et al., 2013). When the manual sensor was buried in snow, it would be taken out to measure the snow surface temperature instead of the soil surface temperature (China Meteorological Administration 2003).

      Figure 2.  Time series of the raw daily LST averaged across 197 stations (the black dots in Fig. 1a) in December for 1960−2017. The red solid line is a reference line to separate the years before and after 2005; the two blue dotted lines are the mean LSTs derived from the raw daily dataset during 1960−2005 and 2005−17. The green solid curve represents the daily LST time series after LSAT adjustments (section 3.1), and the red dotted line is the mean LST during 2005−17. The corresponding mean LST values averaged for different periods are also indicated.

      More specifically, we take a station (ID 50862, 46.98ºN, 128.05ºE) at the NE subregion as an illustrative example. Figure 3 shows the abrupt warming shift in cold months (from October to the following April) since 2005. A similar phenomenon appears at all 197 stations (black dots in Fig. 1a) in the NE and NA subregions. In the MASH procedure, possible breakpoints are determined by comparing the values at the candidate station with those at the nine nearest reference stations. If those surrounding reference stations show similarly abrupt changes or shifts as the candidate station, then the MASH method cannot detect them (Li, 2016). Due to how MASH works, it will fail to deal with the specific problematic phenomenon mentioned above (compare the blue and orange lines in Fig. 3b). Thus, before applying the MASH method, we perform a preprocessing procedure for the LSTs from these 197 stations.

      Figure 3.  Case station (ID 50862, 46.98ºN, 128.05ºE) for the raw and calculated LST, and the homogenized results (units: ºC). (a) Raw daily LST plotted against the calculated daily LST in each month during 1960−2004, where the black solid line is the reference (x = y) line, (b) raw (blue) and MASH homogenized (orange) daily LST, and (c) raw and calculated (green line) LST during 2003−07. The calculated daily LST is the LST adjusted by the LSAT (described in section 3.1).

      There is a very close relationship between the LSAT and LST, and their differences determine surface heat fluxes (Zeng et al., 2012). The observed LSATs have undergone a strict quality control process and do not show such remarkable warming shifts. Therefore, the LSAT is used as a reference to correct the LST warming bias at each station. Simplistically and practically, it is assumed that the characteristics of the changes in LSATs are consistent with those in LSTs at the same station. This assumption is reasonable because they have a robust relationship and show similar variabilities (Vancutsem et al., 2010). The following procedure is used to adjust the LST via the LSAT. First, we construct a regression equation between the LST (dependent variable) and LSAT (independent variable) for each cold month (from October to the following April) at each of the 197 stations from 1960 to 2004. Second, the regression coefficient and the raw daily LSAT in the current month are used to compute the corresponding daily LST from 2005 to 2017. The computed LSTs are regarded as the “calculated LSTs” at the 197 stations. To verify the reasonableness and applicability of the above hypothetical relationship, we compare the raw and calculated LSTs in Fig. 3a. All the dots spread around the reference y-x line, and the monthly absolute mean differences between the calculated and raw LSTs are generally less than 1ºC. In cold months, the absolute differences are less than 0.5ºC, except in April (October), when there is a relatively large error in the mean differences of approximately –2ºC (2ºC). However, the standard deviation of the raw LST during 1960−2005 in April (October) is 3ºC (6ºC) and is higher than the mean difference. Therefore, it still can be concluded that the calculated LSTs are consistent with the raw LSTs during the whole period. Similar results are also shown at other stations. After the above process, the distinct warm shift of the LST in cold months is removed (Fig. 3c). The calculated daily LST across all 197 stations averaged in December from 2005 to 2017 (green line in Fig. 2) also shows consistency with the raw daily LST from 1960 to 2005, and the mean value for 2005−17 (–15.9ºC) is close to that for 1961−2005 (–16.1ºC). The above-calculated LSTs at 197 stations for 2005−17 are added to the raw dataset and replace the values at the same time and station.

    • After the above preliminary adjustments, we apply the MASH method at all 2360 stations in China. The first step of MASH is to detect the temporal breakpoints of the LST time series at each station. The breakpoint (also called the change-point) is where the time series displays significant differences before and after that point. The time series shows a distinct leap at the breakpoint. To help understand this process, we take one station (ID 57710 at 27.85ºN, 106.37ºE) in the YG subregion as an illustrative example (Fig. 4). The LST at the candidate station (the red dot in Fig. 4a) is first compared with the values from the nine closest stations (the green dots in Fig. 4a). Compared with the other nine reference time series (Fig. 4b), a breakpoint appears in 1970 in the candidate time series, of which the mean LST during 1960−70 is 21.1ºC, 3ºC higher than that during 1970−2017 (18.1ºC). The variability of the homogenized time series (the blue line in Fig. 4c) is akin to the nine neighbor reference time series. The time series has a warming trend [0.09ºC (10 yr)–1], showing the opposite tendency compared to the raw time series [–0.26ºC (10 yr)–1].

      Figure 4.  Annual LST series at station ID 57710 (27.85ºN, 106.37ºE, a candidate station) in the YG subregion (units: ºC). The specific information of nine reference stations is ID 57714 (27.36ºN, 106.20ºE), ID 57717 (27.68ºN, 106.92ºE), ID 57713 (27.68ºN, 106.92ºE), ID 57606 (28.33ºN, 106.67ºE), ID 57720 (27.95ºN, 107.17ºE), ID 57614 (28.55ºN, 105.98ºE), ID 57803 (27.05ºN, 106.00ºE), ID 57718 (27.03ºN, 106.72ºE), and ID 57719 (27.07ºN, 106.97ºE), respectively. (a) Distribution of stations, including the candidate station (red dot), the nine nearest reference stations (green dots), and other stations (black dots) in the YG subregion, (b) the raw annual LST anomalies of the candidate station (red curve) and the nine reference stations (black curves), and (c) annual LST from the raw (red curve) and homogenized time series at the candidate station.

      Repeating the above procedure for all stations in each of the nine subregions, we identify the breakpoints at all 2360 stations from the monthly LST time series for 1960−2017. To better illustrate the results, we subjectively divided MASH results into two parts. The first part occurs when the absolute adjustment value is greater than 0.5ºC, it is treated as the effective adjustment and its corresponding breakpoint as the effective breakpoint. The other part occurs when the absolute adjusted value is less than 0.5ºC, it is regarded as the minor adjustment, in which inhomogeneity is relatively small and the MASH procedure has little effect on the time series. There are a total of 3.68 × 103 effective breakpoints counted in Fig. 5. The plot clearly shows that the monthly LSTs at the majority of stations contain 5−25 effective breakpoints (Fig. 5). Of all 2360 stations, 35 stations contain over 40 effective breakpoints, of which most are located in the SC and SE subregions (the red dots in Fig. 5a), while 21 stations scattered across China contain no effective breakpoints (the dark blue dots in Fig. 5a). The effective breakpoints are relatively concentrated in the SE, HHH, and SC subregions, where there are at least 20 effective breakpoints identified in most stations. Having aggregated the numbers of effective breakpoints at all stations (Fig. 5b), nearly 500 effective breakpoints occur each year on average. However, the number of effective breakpoints shows distinct interannual variations and evidently increases over time, in particular since 2000. In 2003, when the replacement of the LST measurement instruments occurred at many stations throughout China, more than 1200 effective breakpoints are identified. At this time, the number of effective breakpoints reaches its maximum value. For the monthly LST time series, the number of effective breakpoints in cold months (January to April, and November to December) is higher than in warm months (May to September) in every year. Previous documents and reports have also mentioned that the LST measurement instruments used for in situ observation systems have been gradually changed from manual to automatic ones since 2000 (Liu et al., 2008; Xu et al., 2019). Furthermore, it is well known that urbanization has developed rapidly during recent decades in China. The fact that some stations suffered from changes in the surrounding environment may also induce abrupt changes in the LST records, which is another reason for the increase in the number of breakpoints (Yan et al., 2010).

      Figure 5.  The number of effective breakpoints for the monthly LST records at the effective stations during 1960−2017; (a) the spatial distribution at all stations and (b) the accumulation in each month and each year across all stations.

    • To explore the LST adjustments identified in the MASH procedure, we count the number of effective breakpoints at all stations in each subregion and throughout the mainland of China. There are approximately 1.6 × 106 monthly records, of which 3.28 × 105 records (approximately 20% of the total records) show effective adjustments in MASH. In each subregion (Fig. 6), the majority of effective adjustments vary between 0.5ºC and 1.5ºC and between –0.5ºC and –1.5ºC (2.95 × 105 or 90% of all effective adjustments), within which two peaks are found around 0.5ºC to 1.0ºC and –0.5ºC to –1.0ºC. Except for the TB subregion, stations in other subregions show that the number of effective breakpoints with positive adjustments is greater than the number with negative ones. Moreover, minor adjustments are also counted in Fig. 7. We find that the minor adjustments account for 80% of the total adjustments in most subregions and in all of the mainland of China, which is consistent with the above results (effective adjustments records account for approximately 20% of the total records). Obviously, the percentage of minor adjustments in the SE subregion between July and October is low (less than 60%, red line with an inverted triangle in Fig. 7). This indicates that the percentage of effective adjustments in SE subregion is relatively large (over 40%). Except for the SE subregion, the YZ (NE) subregion in the cold season shows the highest (lowest) percentage of minor adjustments, above 90% (below 80%). Approximately 80% of the total records in the HHH subregion have experienced minor adjustments. Considering the large variability of the daily data, it is difficult to directly homogenize the daily time series. Therefore, MASH identifies the inhomogeneities from the monthly time series and applies the result to estimate the inhomogeneities of daily time series through smooth interpolation. We also plotted the frequency distribution of daily adjusted records (figures not shown) and it indeed exhibits similar characteristics with the monthly results.

      Figure 6.  Frequency distribution for the effective breakpoints of LST in the nine subregions and for the whole of the Chinese mainland. Biases at these breakpoints are adjusted with MASH.

      Figure 7.  Percentage of records with adjusted LST biases between –0.5ºC and 0.5ºC in different months over the nine subregions and the whole of the Chinese mainland.

    4.   Characteristics of raw and homogenized LSTs during 1960−2017
    • To comprehensively understand the effects of inhomogeneities on the LST, we first compute three statistical metrics (mean, standard deviation, and linear trend) from both the homogenized and raw LST at all stations, and we then compare their differences. To illustrate the results in different seasons, we use the statistical metrics in January and July of multiple years to represent the results in summer and winter, respectively (Fig. 8). It should be noted that the differences of the monthly means (or the standard deviations) are the homogenized minus raw datasets.

      Figure 8.  Spatial distribution of the differences of the monthly means [(a), (b)] and standard deviations [(c), (d)] between the homogenized and raw LSTs in January [(a), (c)] and July [(b), (d)] (units: ºC). The filled dot and the “+” symbol indicate the mean LST differences passing or not passing the significance test (p = 0.05), respectively. The histogram plot in the lower-left corner shows the frequency distribution of the significant differences.

      In winter, 81 (279) stations display significantly positive (negative) differences (p = 0.05) between the homogenized and raw monthly mean LSTs (Fig. 8a). Among these stations, 163 (58% of 279) stations with significantly negative differences are located north of 40ºN, while some are in the SE subregion. In summer, stations with significant LST differences are relatively concentrated in the YG, SC, SE, and NA subregions (Fig. 8b), where 325 stations had significant LST differences, of which 117 (208) stations showed positive (negative) differences. In both winter and summer, the significant negative differences generally vary between –2ºC and –1ºC, while the significant positive differences are within 1ºC (histogram plots in Figs. 8a, b).

      We use the standard deviation to represent the interannual variability, and its difference between homogenized and raw data can be regarded as the changes of the interannual variability of LST. In winter, the interannual variability of the homogenized monthly LST is much smaller than that of the raw values (Fig. 8c). All 197 stations north of 40ºN (Fig. 1a) show negative differences. Stations with negative differences are also prevalent in the TB subregion. In summer (Fig. 8d), the interannual variability of the homogenized mean exhibits a remarkable difference from that of the raw mean at 102 stations, of which 100 stations display negative differences, which indicates that our homogenization process reduces the interannual variability remarkably at most stations for both winter or summer.

      To explore the changes in the long-term tendency of LST due to the MASH process, we also computed the linear trend for the homogenized and raw LSTs at all 2360 stations in winter and summer during 1960−2017. In winter (Figs. 9a, b), 144 stations (6% of all 2360 stations) have negative linear trends in the raw dataset, of which only the trends at three stations are significant (p =0.05). Most stations show positive trends (94% of all 2360 stations), of which the LST trend at 57% of stations is significant (p =0.05). Stations with significant positive LST trends are concentrated in the NA, NE, LP, HHH, and TB subregions, and stations with negative trends are mainly located in the YG and YZ subregions. After homogenization, the number of stations with negative trends decreases from 144 to 45, and the LST trends at all these stations are not significant. The percentage of stations with significant positive LST trends (57%) does not change, but the distribution differs greatly. Compared to the linear trend of raw LSTs, stations with significant positive trends are reduced north of 40ºN (the NE subregion and Xinjiang Province) but increase in the SE and TB subregions. In summer (Figs. 9c, d), stations with negative trends increase and occupy almost the entire southern part of the Yellow River Basin. There are in total 648 stations with negative trends and 12% of these are significant for the raw LSTs. This number increases to 659, while only 2% of them are significant for the homogenized LSTs, indicating that the homogenization process reduces the LST trends overall in the mainland of China.

      Figure 9.  Distribution of linear trends of the monthly raw and homogenized LST in January [(a), (b)] and July [(c), (d)]. The filled dot (“+”) indicates stations with trends passing (not passing) the significance test (p = 0.05).

      Finally, we perform a preliminary analysis of the homogenized LST dataset in terms of its temporal and spatial variations. Figure 10 displays the distribution of the annual mean homogenized LST and its time series averaged in China. It is worth mentioning that the number of missing measurements in 2016 is relatively large (also seen in Fig. 1b), in particular to the south of the Yellow River basin. Therefore, the mean value in 2016 is given as missing. Notably, the annual mean LST has below zero degrees Celsius at very few stations distributed in the NE subregion, while most stations show positive values. Moreover, for the entire period of 1960−2017, the annual mean LST shows a significant warming trend of approximately 0.22ºC (10 yr)–1, which is consistent with previous studies (Zhou et al., 2017; Xu et al., 2019). The trend of the annual mean LSAT shows a similar magnitude [0.21ºC (10 yr)–1], and the annual mean LST is generally 2ºC higher than the annual mean LSAT.

      Figure 10.  (a) Distribution of the multiyear mean homogenized LST and (b) its time series and linear trend averaged over all stations in China (LST/LSAT with the solid line/dotted line) for 1960−2017 (units: ºC).

    5.   Summary and discussion
    • Long-term station observed datasets are the fundamental basis of climate research. A homogenized long-term LST record will greatly help us to thoroughly understand land-atmosphere interaction processes. This study is devoted to developing a homogenized long-term LST dataset from 2360 stations in the Chinese mainland from 1960 to 2017 based on the NMIC/CMA in situ measured LST at meteorological stations. The main results are summarized as follows.

      Due to the replacement of instruments from manual to automatic ones around 2004 north of 40ºN, the time series of the LSTs at 197 stations display remarkable warming shifts during cold months. This warming bias fails to be homogenized by the MASH method. Therefore, the raw LST records at the above stations are first adjusted using the high-quality LSAT under the assumption of the same monthly variability of both the LST and LSAT at the same station.

      Then, the MASH method is applied to homogenize the LST dataset in China. During 1960−2017, 5−25 significant breakpoints are detected in the LST monthly time series at most stations. Stations in southern China and most parts of the Sichuan Basin exhibited more intensive breakpoints than those in other subregions. The number of breakpoints has increased since 2000. Notably, in 2004, breakpoints are detected at more than 1200 stations. For most breakpoints, the absolute adjusted biases due to MASH are between 0.5ºC and 1.5ºC, with peaks around 0.5ºC and –0.5ºC. Moreover, comparing the difference between the homogenized and raw LSTs, the MASH process generally reduces the magnitude, interannual variability, and linear trends of LST. The interannual variabilities and linear trends of the homogenized LSTs are also reduced at the majority of stations, especially in the regions north of 40ºN in winter.

      It should be noted that there are some limitations of the new homogenized LST dataset north of 40ºN. We preliminarily adjusted warm shifts of LSTs in cold seasons from 2005 onwards simply by using a linear relationship between the LST and LSAT, but many other factors can also directly affect changes in LST, including precipitation frequency, solar radiation, vegetation growth, surface roughness (Zeng et al., 2012; Zhou et al., 2017). In a recently published article, Du et al. (2020) corrected LSTs through LSATs while considering the influences of snow depth and solar radiation. They found that the LST-LSAT relationship is stable and near-linear under snow-free conditions, but that it is sensitive to both snow depth and solar radiation. Therefore, the above uncertainty should be considered when this dataset is applied in further research.

      In summary, we provide a 58-year (1960−2017) homogenized daily LST dataset with an abundant number of stations (2360) in China. The current paper only presents the homogenization process of the LST dataset and some preliminary analyses for LSTs, but our work extends beyond this. We have applied the MASH method to soil temperature at six soil depths (0 cm, 5 cm, 10 cm, 15 cm, 20 cm, and 40 cm). The results for soil temperatures in other soil depths are similar to those for LST, so we do not present them in this paper. The long-term homogenized LST and soil temperature in multiple soil depths contain a relatively dense number of stations and can have a variety of applications, such as for model evaluation and climate and soil-related research.

      Acknowledgements. This work was supported by the National Science Fund for Distinguished Young Scholars (Grant No. 41925021) and the National Natural Science Foundation of China (Grant No. 41875106). The new homogenized daily LST dataset in this research is available from the authors for various applications.

Reference

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return