Bias Adjustment and Analysis of Chinese Daily Historical Radiosonde Temperature Data

+ Author Affiliations + Find other works by these authors
  • Corresponding author: Zijiang ZHOU, zzj@cma.gov.cn
  • Funds:

    Supported by the National Innovation Project for Meteorological Science and Technology (CMAGGTD003-5), China Meteorological Administration Special Public Welfare Research Fund (GYHY201506002), and National Key Research and Development Program of China (2017YFC1501801)

  • doi: 10.1007/s13351-021-9162-x

PDF

  • The discontinuities in historical Chinese radiosonde datasets are attributed to artificial errors. In order to reflect more realistically basic conditions of the atmosphere over China and provide more reasonable radiosonde data as input to climate change analysis and to atmospheric reanalysis data assimilation systems, this paper proposes a scheme to identify breakpoints and adjust biases in daily radiosonde observations. The ongoing ECMWF ReAnalysis-Interim (ERA-Interim) 12-h forecasts are used as reference series in the scheme, complemented by the ECMWF Twentieth Century Reanalysis (ERA-20C). A series of breakpoint identification schemes are developed and combined with metadata to detect breakpoints. The Quantile-Matching (QM) method is applied to test and adjust daily radiosonde data on 12 mandatory pressure levels collected at 80 sounding stations during 1979–2013. The adjusted temperatures on mandatory levels are interpolated to significant levels for temperature adjustment on these levels. The adjustment scheme not only solves the data discontinuity problem caused by changes in observational instruments and bias correction methods, but also solves the discontinuity problem in the 1200 minus 0000 UTC temperature time series on mandatory levels at individual sounding stations. Before the adjustment, obvious discontinuities can be found in the deviation field between the raw radiosonde data and ERA-Interim reanalysis with relatively large deviations before 2001. The deviation discontinuity is mainly attributed to the nationwide upgrade of the radiosonde system in China around 2001. After the adjustment, the time series of deviations becomes more continuous. In addition, compared with the adjusted temperature data on mandatory levels over 80 radiosonde stations in China contained in the Radiosonde Observation Correction Using Reanalyses (RAOBCORE) 1.5, the dataset adjusted by the method proposed in the present study exhibits higher quality than RAOBCORE 1.5, while discontinuities still exist in the time series of temperature at 0000, 1200, and 1200 minus 0000 UTC in RAOBCORE 1.5.
  • 加载中
  • Fig. 1.  Distribution of the 80 Chinese radiosonde stations used in this study.

    Fig. 2.  Monthly mean 30-hPa time series of radiosonde temperature anomalies at Nenjiang Station (50557) at (a) 0000 and (b) 1200 UTC, in which the red lines are regression fitting lines. (c) The 1200 minus 0000 UTC time series at 30 hPa before and after the adjustment.

    Fig. 3.  Annual mean 30-hPa temperature time series at (a) 0000 and (b) 1200 UTC at Yichun Station (50774) before (blue) and after (red) the adjustment. (c) Monthly mean difference of temperature between 1200 and 0000 UTC (1200 minus 0000) before (blue) and after (red) the adjustment and that from RAOBCORE 1.5 (green).

    Fig. 4.  As in Fig. 3, but for Xisha Station (59981) at 250 hPa.

    Fig. 5.  As in Fig. 3, but for Kunming Station (56778) at 200 hPa.

    Fig. 6.  All the final breakpoints at 80 stations and the number of stations with instrument changes or with recording and archiving method changes.

    Fig. 7.  The mean temperature adjustment from 1979 to 2013 on mandatory levels in ChinaADJ and in RAOBCORE 1.5 at the same 80 radiosonde stations.

    Fig. 8.  Monthly mean temperature deviations from ERA-Interim for the raw radiosonde data (red), ChinaADJ (blue), and RAOBCOR 1.5 (black) averaged over 80 radiosonde stations on six mandatory levels: (a) 30, (b) 70, (c) 100, (d) 250, (e) 500, and (f) 850 hPa.

    Fig. 9.  Averages of deviations from the ERA-Interim reanalysis at 80 stations for the raw data (red), ChinaADJ (blue), and data extracted from RAOBCORE 1.5 (green) on 12 mandatory levels.

    Fig. 10.  Time series of the 1200 minus 0000 UTC annual mean temperature averaged over 80 radiosonde stations in China before (red) and after (blue) the adjustment and that for the RAOBCORE (green) on six mandatory levels: (a) 30, (b) 70, (c) 100, (d) 200, (e) 400, and (f) 850 hPa.

    Fig. 11.  Temperature trends on 12 mandatory levels for the radiosonde data before (Raw) and after (ChinaADJ) the adjustment and for RAOBCORE 1.5.

    Fig. 12.  Trends of the (a, b) 50-, (c, d) 100-, (e, f) 300-, (g, h) 500-, and (i, j) 850-hPa temperature over 80 sounding stations during 1979–2013 (a, c, e, g, i) before and (b, d, f, h, j) after the adjustment.

    Fig. 13.  Time series of monthly mean temperature deviations from ERA-Interim on the significant levels within the layer of 150–100 hPa before (black) and after (red) the adjustment.

    Table 1.  Breakpoint numbers detected on eight mandatory levels

    Level (hPa)Number of breakpoints
    30353
    50352
    70322
    100323
    150345
    200353
    250333
    300331
    Download: Download as CSV
  • [1]

    Chen, Z., and S. Yang, 2014: Homogenization and analysis of China radiosonde temperature data from 1979 to 2012. Acta Meteor. Sinica, 72, 794–804. doi: 10.11676/qxxb2014.046. (in Chinese)
    [2]

    Dai, A. G., J. H. Wang, P. W. Thorne, et al., 2011: A new approach to homogenize daily radiosonde humidity data. J. Climate, 24, 965–991. doi: 10.1175/2010JCLI3816.1.
    [3]

    Della-Marta, P. M., and H. Wanner, 2006: A method of homogenizing the extremes and mean of daily temperature measurements. J. Climate, 19, 4179–4197. doi: 10.1175/JCLI3855.1.
    [4]

    Easterling, D. R., and T. C. Peterson, 1995: A new method for detecting undocumented discontinuities in climatological time series. Int. J. Climatol., 15, 369–377. doi: 10.1002/joc.3370150403.
    [5]

    Free, M., and D. J. Seidel, 2005: Causes of differing temperature trends in radiosonde upper air data sets. J. Geophys. Res. Atmos., 110, D07101. doi: 10.1029/2004JD005481.
    [6]

    Gaffen, D. J., 1994: Temporal inhomogeneities in radiosonde temperature records. J. Geophys. Res. Atmos., 99, 3667–3676. doi: 10.1029/93JD03179.
    [7]

    Guo, Y. J., and Y. H. Ding, 2009: Long-term free-atmosphere temperature trends in China derived from homogenized in situ radiosonde temperature series. J. Climate, 22, 1037–1051. doi: 10.1175/2008JCLI2480.1.
    [8]

    Guo, Y. J., and Y. H. Ding, 2010: Impacts of reference time series on the homogenization of radiosonde temperature. Adv. Atmos. Sci., 28, 1011–1022. doi: 10.1007/s00376-010-9211-3.
    [9]

    Guo, Y. J., Q. X. Li, and Y. H. Ding, 2009: The effect of artificial bias on free air temperature trend derived from historical radiosonde data in China. Chinese J. Atmos. Sci., 33, 1309–1318. doi: 10.3878/j.issn.1006-9895.2009.06.16. (in Chinese)
    [10]

    Haimberger, L., 2007: Homogenization of radiosonde temperature time series using innovation statistics. J. Climate, 20, 1377–1403. doi: 10.1175/JCLI4050.1.
    [11]

    Haimberger, L., C. Tavolato, and S. Sperka, 2012: Homogenization of the global radiosonde temperature dataset through combined comparison with reanalysis background series and neighboring stations. J. Climate, 25, 8108–8131. doi: 10.1175/JCLI-D-11-00668.1.
    [12]

    Houghton, J. T., Y. H. Ding, D. J. Griggs, et al., 2001: Climate Change 2011: The Scientific Basis. Cambridge University Press, Cambridge, 881 pp.
    [13]

    Jones, P. D., and A. Moberg, 2003: Hemispheric and large-scale surface air temperature variations: An extensive revision and an update to 2001. J. Climate, 16, 206–223. doi: 10.1175/1520-0442(2003)016<0206:HALSSA>2.0.CO;2.
    [14]

    Lanzante, J. R., S. A. Klein, and D. J. Seidel, 2003: Temporal homogenization of monthly radiosonde temperature data. Part I: Methodology. J. Climate, 16, 224–240. doi: 10.1175/1520-0442(2003)016<0224:THOMRT>2.0.CO;2.
    [15]

    Luers, J. K., and R. E. Eskridge, 1998: Use of radiosonde temperature data in climate studies. J. Climate, 11, 1002–1019. doi: 10.1175/1520-0442(1998)011<1002:UORTDI>2.0.CO;2.
    [16]

    Ramella Pralungo, L., L. Haimberger, A. Stickler, et al., 2014: A global radiosonde and tracked balloon archive on 16 pressure levels (GRASP) back to 1905 - Part 1: Merging and interpolation to 00:00 and 12:00 GMT. Earth Syst. Sci. Data, 6, 185–200. doi: 10.5194/essd-6-185-2014.
    [17]

    Sherwood, S. C., J. R. Lanzante, and C. L. Meyer, 2005: Radiosonde daytime biases and late-20th century warming. Science, 309, 1556–1559. doi: 10.1126/science.1115640.
    [18]

    Thorne, P. W., P. D. Jones, T. J. Osborn, et al., 2002: Assessing the robustness of zonal mean climate change detection. Geophys. Res. Lett., 29, 1920. doi: 10.1029/2002GL015717.
    [19]

    Trewin, B. C., and A. C. F. Trevitt, 1996: The development of composite temperature records. Int. J. Climatol., 16, 1227–1242. doi: 10.1002/(SICI)1097-0088(199611)16:11<1227::AID-JOC82>3.0.CO;2-P.
    [20]

    Vincent, L. A., X. L. Wang, E. J. Milewska, et al., 2012: A second generation of homogenized Canadian monthly surface air temperature for climate trend analysis. J. Geophys. Res. Atmos., 117, D18110. doi: 10.1029/2012JD017859.
    [21]

    Wang, X. L., 2008: Penalized maximal F test for detecting undocumented mean shift without trend change. J. Atmos. Oceanic Technol., 25, 368–384. doi: 10.1175/2007JTECHA982.1.
    [22]

    Wang, X. L., and Y. Feng, 2010: RHtestsV3 User Manual. Climate Research Division, Atmospheric Science and Technology Directorate, Science and Technology Branch, Environment Canada, Toronto, Ontario, Canada, 26 pp.
    [23]

    Wang, X. L., Q. H. Wen, and Y. H. Wu, 2007: Penalized maximal t test for detecting undocumented mean change in climate data series. J. Appl. Meteor. Climatol., 46, 916–931. doi: 10.1175/JAM2504.1.
    [24]

    Wang, X. L., H. F. Chen, Y. H. Wu, et al., 2010: New techniques for the detection and adjustment of shifts in daily precipitation data series. J. Appl. Meteor. Climatol., 49, 2416–2436. doi: 10.1175/2010JAMC2376.1.
    [25]

    Wang, X. L., Y. Feng, and L. A. Vincent, 2014: Observed changes in one-in-20 year extremes of Canadian surface air temperatures. Atmos.–Ocean, 52, 222–231. doi: 10.1080/07055900.2013.818526.
    [26]

    Xu, W. H., Q. X. Li, X. L. Wang, et al., 2013: Homogenization of Chinese daily surface air temperatures and analysis of trends in the extreme temperature indices. J. Geophys. Res. Atmos., 118, 9708–9720. doi: 10.1002/jgrd.50791.
    [27]

    Zhai, P. M., 1997: Some gross errors and biases in China’s historical radiosonde data. Acta Meteor. Sinica, 55, 563–572. doi: 10.11676/qxxb1997.055. (in Chinese)
    [28]

    Zhai, P. M., and R. E. Eskridge, 1996: Analyses of inhomogeneities in radiosonde temperature and humidity time series. J. Climate, 9, 884–894. doi: 10.1175/1520-0442(1996)009<0884:AOIIRT>2.0.CO;2.
  • 2. Zijiang ZHOU and Zhe CHEN.pdf

  • 加载中
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

Bias Adjustment and Analysis of Chinese Daily Historical Radiosonde Temperature Data

    Corresponding author: Zijiang ZHOU, zzj@cma.gov.cn
  • 1. National Meteorological Information Center, China Meteorological Administration, Beijing 100081, China
  • 2. National Center for Atmospheric Research, Boulder, Colorado 80301, USA
  • 3. Beijing Meteorological Information Center, Beijing Meteorological Service, Beijing 100089, China
Funds: Supported by the National Innovation Project for Meteorological Science and Technology (CMAGGTD003-5), China Meteorological Administration Special Public Welfare Research Fund (GYHY201506002), and National Key Research and Development Program of China (2017YFC1501801)

Abstract: The discontinuities in historical Chinese radiosonde datasets are attributed to artificial errors. In order to reflect more realistically basic conditions of the atmosphere over China and provide more reasonable radiosonde data as input to climate change analysis and to atmospheric reanalysis data assimilation systems, this paper proposes a scheme to identify breakpoints and adjust biases in daily radiosonde observations. The ongoing ECMWF ReAnalysis-Interim (ERA-Interim) 12-h forecasts are used as reference series in the scheme, complemented by the ECMWF Twentieth Century Reanalysis (ERA-20C). A series of breakpoint identification schemes are developed and combined with metadata to detect breakpoints. The Quantile-Matching (QM) method is applied to test and adjust daily radiosonde data on 12 mandatory pressure levels collected at 80 sounding stations during 1979–2013. The adjusted temperatures on mandatory levels are interpolated to significant levels for temperature adjustment on these levels. The adjustment scheme not only solves the data discontinuity problem caused by changes in observational instruments and bias correction methods, but also solves the discontinuity problem in the 1200 minus 0000 UTC temperature time series on mandatory levels at individual sounding stations. Before the adjustment, obvious discontinuities can be found in the deviation field between the raw radiosonde data and ERA-Interim reanalysis with relatively large deviations before 2001. The deviation discontinuity is mainly attributed to the nationwide upgrade of the radiosonde system in China around 2001. After the adjustment, the time series of deviations becomes more continuous. In addition, compared with the adjusted temperature data on mandatory levels over 80 radiosonde stations in China contained in the Radiosonde Observation Correction Using Reanalyses (RAOBCORE) 1.5, the dataset adjusted by the method proposed in the present study exhibits higher quality than RAOBCORE 1.5, while discontinuities still exist in the time series of temperature at 0000, 1200, and 1200 minus 0000 UTC in RAOBCORE 1.5.

    • Radiosonde data are an important component of global atmospheric observations. Compared to satellite data, radiosonde data cover a longer period and demonstrate great advantages in vertical resolution of observations. Therefore, radiosonde data are a unique and important data source that can be used to describe the atmospheric condition on upper levels, and play a crucial role in the data assimilation and real-time weather forecasting. However, during decades of radiosonde observations, changes in measurement instruments and correction methods have resulted in data discontinuities and deviations, which have great impacts on data assimilation systems. The radiosonde data with large deviations would be excluded during the data assimilation process to ensure the spatial consistency. Note that radiosonde stations are relatively few, and the elimination of certain radiosonde data will further reduce the utilization of radiosonde data. Meanwhile, not eliminating radiosonde data with artificial biases will inevitably affect the reanalysis products. Therefore, it is extremely important to effectively adjust radiosonde data.

      In the early stages of studies for the climate change, Thorne et al. (2002), Lanzante et al. (2003), Free and Seidel (2005), and others adjusted radiosonde data based on statistical analysis and metadata information. However, most of these studies focused on the adjustment of monthly mean radiosonde data, while data assimilation systems require daily adjustment of real-time daily sounding data. For this reason, Haimberger (2007) and Haimberger et al. (2012) produced the Radiosonde Observation Correction Using Reanalyses (RAOBCORE) 1.5, which is an adjusted daily temperature dataset from the global radiosonde network. RAOBCORE has been used as input to data assimilation systems for reanalysis products such as the ECMWF ReAnalysis-Interim (ERA-Interim), the Modern-Era Retrospective Analysis for Research and Applications (MERRA), the Japanese Reanalysis (JRA), and so on.

      In 2015, the National Meteorological Information Center (NMIC) of China Meteorological Administration began a project to produce the first generation of global atmospheric reanalysis product. Assimilation of high-quality radiosonde data in China is one characteristic of this reanalysis product. At present, there are more than 120 radiosonde stations in China. Due to differences in observation time, methods, instruments, and bias correction algorithms over the past decades, radiosonde data are inhomogeneous both spatially and temporarily, affecting the results of regional climate change studies in China. Therefore, it is necessary to make a reasonable adjustment to the daily radiosonde data in China.

      The first radiosonde station in China was built in 1951, and most radiosonde stations have been built gradually since 1957. Radiosonde instruments and technical specifications related to observations and reports changed considerably in the 1950s; radiosonde instruments changed frequently from the 1950s to 1960s; in 2000 and 2001, a radiosonde system upgrade was carried out for all stations in China; since 2002, L-band radar and electronic sonde have been gradually deployed at all radiosonde stations in China. These artificial factors could result in discontinuities in the radiosonde data over China. A number of studies have been conducted on the homogenization of radiosonde temperature data in China. Zhai et al. (1996) and Zhai (1997) analyzed the artificial bias in the Chinese historical radiosonde temperature dataset by using the Easterling–Peterson method (Easterling and Peterson, 1995), and argued that there might exist relatively large errors in the time series of radiosonde temperatures in the 1960s and 1970s. Guo et al. (2009) implemented a two-phase regression method to test the inhomogeneity of radiosonde data in China. They found that the inhomogeneity of the time series of radiosonde data in China can be attributed largely to changes in observational instruments and radiation adjustment methods. By using the ERA-Interim reanalysis product as their reference series, Chen and Yang (2014) checked and adjusted radiosonde temperature data collected at 125 radiosonde stations in China. They found that the inhomogeneity in the monthly mean radiosonde temperature during the period of 1979–2012 arises largely from the upgrade of the sounding system (including changes in the radiation correction method) in the period of 2000–2001 and instrument changes after 2002.

      The above two adjustments to the operational observation system led to a systematic decrease in the monthly mean radiosonde temperature data. Note that the above homogenization work has been carried out only for the monthly mean radiosonde temperature data, so it cannot meet the requirements for data assimilation that needs daily radiosonde data as input. As mentioned above, RAOBCORE is a good daily radiosonde correction dataset. We obtained RAOBCORE 1.5 from https://doi.pangaea.de/10.1594/PANGAEA.823609 (Ramella Pralungo et al., 2014). An evaluation of RAOBCORE 1.5 shows that there are only 80 Chinese radiosonde stations, which have the relatively complete data in RAOBCORE 1.5. The daily radiosonde temperature data at the other 40 Chinese radiosonde stations have not yet been adjusted. In addition, the evaluation of RAOBCORE 1.5 shows that the radiosonde data over China after 2008 has not been adjusted yet in this version, and some biases still exist (as shown in the following parts of this paper). Furthermore, the early homogenization work was conducted only on mandatory levels, whereas there is actually much more radiosonde data on significant levels than on mandatory levels. Therefore, another issue that needs to be solved in the development of a reanalysis product is how to provide the adjusted radiosonde temperature data on significant levels.

      To address the issues mentioned above, this study proposes a test and adjustment scheme to solve the discontinuity problem in the daily radiosonde temperature data in China. The scheme is applied to adjust radiosonde temperature on 12 mandatory levels and significant levels at 80 sounding stations. The adjusted data are compared with data at the same stations contained in RAOBCORE 1.5. This paper is organized as follows. Section 2 introduces the data used in the present study, the reference series, and the test and adjustment scheme. Evaluation of the adjusted data on mandatory levels and significant levels is performed in Sections 3 and 4, respectively. Conclusions and discussion are presented in Section 5.

    2.   Data and methods
    • The raw daily radiosonde data used in the present study are extracted from the quality-controlled “China Upper-Level Dataset on Mandatory Levels (V2.1)”. Twice-daily temperatures (0000 and 1200 UTC) collected at 80 radiosonde stations (Fig. 1) in China during the period of 1979–2013 are adjusted and compared with the data contained in RAOBCORE 1.5. The mandatory levels selected for the present study are 30, 50, 70, 100, 150, 200, 250, 300, 400, 500, 700, and 850 hPa. The observations are conducted at 0000 and 1200 UTC. Data on significant levels include all the radiosonde data at individual stations.

      Figure 1.  Distribution of the 80 Chinese radiosonde stations used in this study.

    • Metadata information comes from the latest historical document of China’s radiosonde stations compiled by the NMIC, which includes information about instrument changes, radiosonde data archival and calculation methods, and relocation information about sounding stations. After 1979, techniques for the upper-air detection and observation were greatly improved. China–US and China–Finland collaborations played a critical role in the automation of the upper-air detection and continuous update of observational instruments in China. The sounding systems of China have been compared with those of the US and Finland several times. In particular, a comparison of the sounding systems of China and Finland, conducted in Zhengzhou, China, greatly promoted the upper-air detection technology in China. In 1989, the World Meteorological Organization (WMO) Internatio-nal Radiosonde Comparison Phase III was conducted in Dzhambul, USSR. An obvious disadvantage was found in the sounding and wind measurement systems in China. The electronic L-band wind measurement radar and GPS navigation system for wind measurement were developed in China against this background. Electronic sonde and L-band wind measurement radar were successfully developed in 2001, and began to be deployed for operational service in 2002. The new instruments were updated at 13 stations in 2002, 10 stations in 2003, 12 stations in 2004, 29 stations in 2005, 4 stations in 2006, 11 stations in 2008, and 27 stations in 2009. By 2010, the change to new instruments at all 120 stations had been completed.

      In addition, comprehensive corrections were conducted for all radiosonde stations in 2001 and 2002, and the system residual deviation was processed in the real-time detection mode at individual stations.

      Overall, the operational sounding system in China has experienced multiple periods of changes in both observational instruments and technical specifications. The application of metadata information will further increase the identification accuracy of breakpoints.

    • The 12-h forecast fields of the ERA-Interim reanalysis product are selected to be the reference series, which are calculated from the corrected data of 80 radiosonde stations in China contained in RAOBCORE 1.5 and their deviations from ERA-Interim forecasts in the Global Radiosonde and tracked-balloon Archive on Sixteen Pressure levels (GRASP; Ramella Pralungo et al., 2014). The ERA-Interim 12-h forecasts are selected to be the reference series based on the considerations that, on the one hand, the discontinuities and spatial distribution of the reanalysis data are relatively reasonable; on the other hand, the forecast fields are more independent of sounding observations compared with reanalysis fields, which increases the independence of the reference series. The ECMWF Twentieth Century Reanalysis (ERA-20C) product is used as the complementary reference series, which is advantageous in data independence and continuity because only the surface pressure is assimilated into ERA-20C. The complementary series is used only for identification of breakpoints and is not used for the calculation of adjusted variables. The distance inverse proportional relationship is applied to interpolate data from the two reference series to the selected sounding sites.

    • The RHtests V5 (Wang and Feng, 2010) software package is implemented to verify the breakpoints. This package includes PMFred (the Penized Maximal F test; Wang, 2008) and PMTred (the Penized Maximal t test; Wang et al., 2007) methods for breakpoint testing. The PMFred method does not need the reference series as input, but PMTred does need. In our study, the PMTred method is implemented to statistically test breakpoints in daily observations. This method has been widely applied to multiple datasets of daily observations (Dai et al., 2011; Xu et al., 2013). There should not be an obvious discontinuity in the series of deviations between the twice-daily radiosonde observations at 0000 and 1200 UTC. Therefore, in addition to testing the 0000- and 1200-UTC series, an extra test is conducted for the 1200 minus 0000 UTC series. Similarly, the testing of breakpoints is also conducted for the monthly mean radiosonde temperature by using the PMTred method and statistical averaging of the ERA-Interim 12-h forecasts. In addition, considering possible discontinuities in the reference series, the PMFred method is applied to test breakpoints in the reference series.

      The scheme for the identification of breakpoints is described below. Breakpoints in the reference series are tested first. If a discontinuity is found at a specific time in the reference series as well as in the raw data, this breakpoint is regarded as being introduced by the implementation of the reference series. This point will not be retained in the final determination of breakpoints. Secondly, NMIC has produced a detailed metadata dataset of Chinese radiosonde stations by verification of radiosonde monthly reports. All the Chinese radiosonde stations are organized to check and collect their metadata information, including changes in historical observation service and historical documents. In this situation, the judgment on breakpoints is based mainly on a combination of statistical tests and breakpoint metadata verification. A breakpoint is retained only when the test shows a confidence level above 95% and the metadata also support the identification. Furthermore, if there is missing information in the metadata, the breakpoint can still be identified without support of the metadata, given that the test of monthly mean temperature time series at 0000, 1200, or 1200 minus 0000 UTC is significant. Finally, the two reference series (ERA-Interim and ERA-20C) are used to verify the breakpoints. When a breakpoint appears in both of the series tests and exceeds the 95% confidence level, this point is retained even if there is no metadata to support the identification.

    • In most previous studies of the monthly mean radiosonde data adjustment, the mean value adjustment method was implemented to adjust raw radiosonde observations (Guo et al., 2009; Chen and Yang, 2014). This method can adjust the mean climatic state and climatic change trend of a physical variable (Houghton, 2001; Jones and Moberg, 2003). However, for daily observations, the adjustment of mean deviation caused by artificial errors cannot meet the need for studies of extremes. For this reason, various adjustment methods have been deve-loped specifically for daily observations to match the data before and after the breakpoint (Trewin and Trevitt, 1996; Della-Marta and Wanner, 2006; Xu et al., 2013). These methods can reduce the impacts of non-climatic factor errors on the data, including its variability and distribution. The present study applies the quantile-matching (QM) method (Wang et al., 2010) to adjust the time series of daily radiosonde temperature after the breakpoints have been identified. Following the approach of Vincent et al. (2012) and Wang et al. (2014), the 10-yr observations before and after the breakpoints minus the reference series are used for adjustment.

    • The adjusted temperature data on mandatory levels can be obtained by using the method described above. However, the input fields for an assimilation system include not only data on mandatory levels, but also data on significant levels. In order to maintain the consistency of adjustments in the vertical direction, it is necessary to adjust data on significant levels as well. The adjustment on significant levels is realized by use of the temperature adjustments on mandatory levels. For temperature $ {T}_{{x}_{0}} $ on a specific significant level of any radiosonde profile, temperature adjustment values ($ {\Delta T}_{\rm{a}} $ and $ {\Delta T}_{\rm{b}} $) on mandatory levels above and below the significant level are used to compute the adjustment value ($ \Delta T $) for the significant level. The algorithm is written as below:

      $$ \Delta T={\Delta T}_{\rm{a}}+({\Delta T}_{\rm{b}}-{\Delta T}_{\rm{a}})\times \frac{\log\left(P/{P}_{\rm{a}}\right)}{\log\left({P}_{\rm{b}}/{P}_{\rm{a}}\right)}, $$
      $$ {T}_{x}={T}_{{x}_{0}}+\Delta T, $$

      where $ {T}_{x} $ is the adjusted temperature on the significant level, P is the pressure of this level, and $ {P}_{\rm{a}} $ and $ {P}_{\rm{b}} $ are the pressures of mandatory levels above and below the significant level.

    3.   Analysis of the adjusted data on standard pressure levels
    • Figure 2 displays the monthly mean time series of 0000- and 1200-UTC radiosonde temperature anomalies at Nenjiang Station (50557) at 30 hPa and the 1200 minus 0000 UTC series before and after the adjustment. The red lines are regression fitting lines. Based on the method described in Section 2, five significant breakpoints that occurred on 1 December 1988, 29 December 1993, 31 December 1999, 28 January 2001, and 31 July 2005 have been identified. Note that the radiation correction method was modified at this station on 31 December 1999, the sounding system was upgraded on 28 January 2001, the “59-701” radiosonde was replaced by an L-band radar-radiosonde on 31 July 2005, and metadata information supports the breakpoints occurring at the above three time points. Before the adjustment (blue line in Fig. 2c), obvious discontinuities can be found in the time series of 1200 minus 0000 UTC, especially in July 2005; after the adjustment (red line in Fig. 2c), the data become more continuous.

      Figure 2.  Monthly mean 30-hPa time series of radiosonde temperature anomalies at Nenjiang Station (50557) at (a) 0000 and (b) 1200 UTC, in which the red lines are regression fitting lines. (c) The 1200 minus 0000 UTC time series at 30 hPa before and after the adjustment.

      Figures 3a and b present the annual mean temperature time series of Yichun Station (50774) at 0000 and 1200 UTC at 30 hPa before and after the adjustment. Before the adjustment, significant discontinuities can be found on 1 June 1984, 1 December 1997, 1 January 2001, and 1 August 2005; after the adjustment, the continuity of the dataset has obviously been increased, which is also reflected in the 1200 minus 0000 UTC time series shown in Fig. 3c. Temperature differences at 1200 UTC before and after the adjustment are smaller than those at 0000 UTC, suggesting that biases in the time series of nighttime temperature (1200 UTC) are relatively small. Figure 3c also displays the 1200 minus 0000 UTC time series (in green) from RAOBCORE 1.5, which clearly shows that the discontinuity still exists around 1989 and 2009 even after the adjustment.

      Figure 3.  Annual mean 30-hPa temperature time series at (a) 0000 and (b) 1200 UTC at Yichun Station (50774) before (blue) and after (red) the adjustment. (c) Monthly mean difference of temperature between 1200 and 0000 UTC (1200 minus 0000) before (blue) and after (red) the adjustment and that from RAOBCORE 1.5 (green).

      Figure 4 displays the annual mean temperature time series of Xisha Station at 0000 and 1200 UTC at 250 hPa, which is the southernmost station in China. Opposite to the situation at stations in the north, temperature differences before and after the adjustment at 1200 UTC are obviously larger than those at 0000 UTC. Statistical analysis indicates that the multi-year average temperature at 1200 UTC is higher than that at 0000 UTC, which reflects the fact that this station is located in a low-latitude region, where the solar elevation angle is still large at 1200 UTC, and therefore the incident solar radiation remains strong at this time. This is the main reason why temperature differences before and after the correction at 1200 UTC are larger than those at 0000 UTC. Before the adjustment, significant discontinuity exists in the dataset due to the upgrade of the sounding system in January 2001 and the instrument change to the L-band radar-radiosonde in June 2005. After the adjustment, the data continuity has improved. Before the adjustment, the annual mean temperature time series at 0000 and 1200 UTC both exhibit a decreasing trend; after the adjustment, however, a slightly increasing trend is shown in both the temperature series, which is also reflected in the 1200 minus 0000 UTC temperature time series after the adjustment (Fig. 4c). In addition, before the adjustment, the Root Mean Square Error (RMSE) of the 1200 minus 0000 UTC time series after June 2005 is relatively small; after the adjustment, not only has the continuity of the entire time series increased significantly, but the RMSE of the time series before June 2005 has also been distinctly reduced. The 1200 minus 0000 UTC temperature time series at Xisha Station based on RAOBCORE 1.5 is also displayed in Fig. 4c (in green), showing an obvious discontinuity around 2005.

      Figure 4.  As in Fig. 3, but for Xisha Station (59981) at 250 hPa.

      The annual mean temperature time series of Kunming Station, which is located on the Yungui Plateau, at 0000 and 1200 UTC at 200 hPa is displayed in Fig. 5. Similar to the situation at stations in the south, temperature differences before and after the adjustment at 1200 UTC are obviously larger than those at 0000 UTC, which is attributed to the fact that the elevation of Kunming Station is high and the multi-year average temperature at 1200 UTC is higher than that at 0000 UTC. At 1200 UTC, the solar elevation angle is still large, and the solar radiation is also large at this time. Discontinuities exist in these temperature time series before the adjustment, which is mainly because of the upgrade of the sounding system in January 2001 and the change to the L-band radar-radiosonde. In January 2001, temperature abruptly decreased significantly at the two observational time points. After the adjustment, it is found that the continuity of the data has increased greatly, which is also reflected in 1200 minus 0000 UTC annual mean temperature time series before and after the adjustment (Fig. 5c). Before the adjustment, the RMSE of the 1200 minus 0000 UTC temperature time series after June 2005 is relatively small; after the adjustment, not only has the continuity of the entire time series increased significantly, but the RMSE of the time series before June 2005 has also been distinctly reduced. The time series of 1200 minus 0000 UTC temperature at Kunming Station based on RAOBCORE 1.5 is also displayed in Fig. 5c (in green), which still shows an obvious discontinuity around 2005. Compared to our study, the decrease in RMSE in the RAOBCORE 1.5 is less significant.

      Figure 5.  As in Fig. 3, but for Kunming Station (56778) at 200 hPa.

    • Breakpoints on mandatory levels in the time series of radiosonde temperature collected at 80 radiosonde stations are identified by using the method described in Section 2.2. The results are listed in Table 1. The statistical results indicate that the number of breakpoints is smallest in the lower troposphere at 850 and 700 hPa, and the number at 850 hPa is only 55% of that at 30 hPa. In the middle and upper troposphere and lower stratosphere, the number of breakpoints increases significantly compared to that in the lower troposphere. This is mainly because instrument errors gradually accumulate from lower to high levels, and the upper levels are more affected by solar radiation.

      Level (hPa)Number of breakpoints
      30353
      50352
      70322
      100323
      150345
      200353
      250333
      300331

      Table 1.  Breakpoint numbers detected on eight mandatory levels

      Since the 1960s, the “59-701” radiosonde had been adopted in China. Most of the sounding stations in China began to use the “59-701” radiosonde in the 1970s, whereas a few stations began to use the “59-701” radiosonde as late as in the 1980s and 1990s. Starting from 2002, the “59-701” radiosonde began to be replaced by the L-band radar-radiosonde at stations in China. The times of replacement were different at various stations, and the replacement was not completed at all stations until 2011. In addition, various statistical methods like the manual analysis, automatic analysis, and a combination of manual and automatic analyses were adopted during different periods to digitize radiosonde temperature observations on mandatory levels. The approaches for the selection of mandatory levels and the correction parameters and methods involved in various statistical analyses are different, and they may lead to discontinuities in the final output of radiosonde temperature. Figure 6 shows all the final breakpoints at 80 stations and the number of stations experiencing instrument changes or method changes for recording and archiving observations. Breakpoints are concentrated in 2000 and 2001, which is attributed to the nationwide upgrade of the sounding system in these two years. The upgrade included modifications of major correction parameters and algorithms, which resulted in obvious discontinuities in the Chinese radiosonde temperature dataset then. In addition, the “59-701” radiosonde was replaced by L-band radar-radiosonde at all sounding stations after 2002, which also contributed to the data discontinuity. More breakpoints also appeared in the 1980s, some of which corresponded to statistical method changes and instrument replacements. However, some of the breakpoints that appeared in the 1980s did not correspond to any metadata that could support the discontinuity. This is possibly related to some missing information in the metadata. Overall, the detected breakpoints correspond well to metadata information, and the years of peak breakpoints are consistent with those of instrument changes, sounding system upgrade, or method changes for data recording and archiving. This result indicates that the adjusted dataset in our study is supported by the metadata and is highly reliable.

      Figure 6.  All the final breakpoints at 80 stations and the number of stations with instrument changes or with recording and archiving method changes.

    • The temperature time series after the adjustment in our work (hereafter ChinaADJ) on mandatory levels is compared with that extracted from RAOBCORE 1.5 at the same 80 radiosonde stations (Fig. 7). It is found that temperatures in the two datasets are close in the troposphere, and an average difference of 0.067°C is found in the stratosphere. The average difference at 0000 UTC is 0.13°C at 70, 50, and 30 hPa, while the difference at these levels reaches 0.48°C at 1200 UTC. This result indicates that the difference in the adjusted temperature between the two datasets comes mainly from the difference at 1200 UTC. Overall, the adjustment values are negative at all standard pressure levels in both datasets, suggesting that the raw data have a warm bias.

      Figure 7.  The mean temperature adjustment from 1979 to 2013 on mandatory levels in ChinaADJ and in RAOBCORE 1.5 at the same 80 radiosonde stations.

    • The time series of monthly mean deviations from the ERA-Interim for raw radiosonde data, ChinaADJ, and RAOBCORE 1.5 on six mandatory levels averaged over 80 radiosonde stations are displayed in Fig. 8. It is clearly shown that the deviations from the ERA-Interim for the raw dataset without adjustment are significant on all mandatory levels before 2001, and the deviations have reduced somewhat since 2001. After the adjustment, the deviations in ChinaADJ are greatly reduced before 2001, and the time series of deviations during the period 1979–2013 become more continuous on all mandatory levels. By comparing ChinaADJ with RAOBCORE 1.5, it is found that the deviations from the ERA-Interim in the layer 250–70 hPa for ChinaADJ are obviously smaller than those for RAOBCORE 1.5, while the deviations at 30 hPa within the layer 850–500 hPa are similar for the two datasets.

      Figure 8.  Monthly mean temperature deviations from ERA-Interim for the raw radiosonde data (red), ChinaADJ (blue), and RAOBCOR 1.5 (black) averaged over 80 radiosonde stations on six mandatory levels: (a) 30, (b) 70, (c) 100, (d) 250, (e) 500, and (f) 850 hPa.

      Averages of deviations from the ERA-Interim reana-lysis on mandatory levels (Fig. 9) indicate that the deviations over 400–70 hPa for ChinaADJ are smaller than those for RAOBCORE 1.5. At 500 hPa, the deviation from the ERA-Interim for RAOBCORE 1.5 is close to zero, while that for ChinaADJ is 0.07°C. At 30 hPa, the deviation for ChinaADJ is around −0.2°C, larger than that for the raw dataset, which is close to zero. Overall, the deviations from the ERA-Interim for the corrected data are much smaller than those for the raw data in almost the entire layer and smaller than those for RAOBCORE 1.5 on most of the mandatory levels.

      Figure 9.  Averages of deviations from the ERA-Interim reanalysis at 80 stations for the raw data (red), ChinaADJ (blue), and data extracted from RAOBCORE 1.5 (green) on 12 mandatory levels.

    • The 1200 minus 0000 UTC annual mean temperature time series over 80 radiosonde stations in China before and after the adjustment and that for RAOBCORE 1.5 on six mandatory levels are presented in Fig. 10. Obvious discontinuities on almost all mandatory levels around 2001 (the discontinuity is not significant at 400 hPa around 2001) are observed. This is attributed to the nationwide upgrade of the sounding system consecutively conducted in July 2000 and January 2001, respectively. In addition, the “59-701” model radiosonde had been gradually replaced by the L-band radar-radiosonde since 2002 at all radiosonde stations in China. The replacement was completed at different times for individual stations. Unlike the consistent discontinuity around 2001, the discontinuity shown in the 1200 minus 0000 UTC temperature time series on individual mandatory levels after 2002 (Fig. 10) is not consistent because the instrument replacement at the sounding stations was completed at different times. However, discontinuity still exists during the period of 2004–2006 on various mandatory levels. After the adjustment, the homogeneity of the 1200 minus 0000 UTC temperature time series has been greatly improved on all mandatory levels, and the discontinuities that occurred around 2001 and after 2002 no longer exist in ChinaADJ. In contrast, the homogeneity of the 1200 minus 0000 UTC temperature time series at the 80 stations in RAOBCORE 1.5 is not as ideal as expected, and distinct discontinuities still exist especially after 2008.

      Figure 10.  Time series of the 1200 minus 0000 UTC annual mean temperature averaged over 80 radiosonde stations in China before (red) and after (blue) the adjustment and that for the RAOBCORE (green) on six mandatory levels: (a) 30, (b) 70, (c) 100, (d) 200, (e) 400, and (f) 850 hPa.

    • Figure 11 shows the temperature trends on mandatory levels before and after the adjustment. Before the adjustment, the temperature shows an increasing trend below 300 hPa and a decreasing trend above 300 hPa. After the adjustment, however, the trend is close to zero at 150 hPa. Above 150 hPa, the decreasing trend weakens; and below 200 hPa, temperature exhibits an increasing trend. The increasing trend intensifies within the layer of 700–200 hPa and the increasing trend weakens at 850 hPa. Compared with RAOBCORE 1.5, the temperature decreasing trend weakens and the increasing trend intensifies in both datasets, but the amplitudes on different mandatory levels are different.

      Figure 11.  Temperature trends on 12 mandatory levels for the radiosonde data before (Raw) and after (ChinaADJ) the adjustment and for RAOBCORE 1.5.

      Temperature trends at individual stations exhibit a significant decreasing trend during the period of 1979–2013 within layer of 30–100 hPa before the adjustment; after the adjustment, the decreasing trend distinctly weakens at all stations (Figs. 12a–d). At 300 hPa, the trends are not significant at most of the stations before the adjustment, and after the adjustment, more stations show an increasing trend (Figs. 12e, f). At 500 hPa, most of the stations show an increasing trend, and after the adjustment, the increasing trend becomes more significant at these stations (Figs. 12g, h). At 850 hPa, the increasing trend is weaker before the adjustment than that after the adjustment (Figs. 12i, j).

      Figure 12.  Trends of the (a, b) 50-, (c, d) 100-, (e, f) 300-, (g, h) 500-, and (i, j) 850-hPa temperature over 80 sounding stations during 1979–2013 (a, c, e, g, i) before and (b, d, f, h, j) after the adjustment.

    4.   Adjustment of daily radiosonde temperature on significant levels
    • Significant levels refer to those where the vertical profiles of the temperature or dewpoint undergo a prominent change. Information on significant levels can reflect the structure of atmospheric stratification, including the inversion layer, isothermal layer, tropopause, cloud layer, and obvious low temperature layer. Unlike mandatory levels, significant levels do not have specified pressure values. In other words, significant levels are determined completely on the actual atmospheric condition at each observation time. Therefore, significant levels have no continuity in time. In addition, the homogenization methods cannot be applied to test and adjust data on significant levels. However, in the real-time observations, the number of significant levels can be more than triple the number of mandatory levels. For any reanalysis product, if the adjustment is performed only for data on mandatory levels but not for data on significant levels, data inconsistency between vertical levels is very likely to occur. For this reason, it is necessary to adjust information on significant levels.

      According to the method described in Section 2.3, the adjusted data on mandatory levels are interpolated to significant levels to adjust data on these levels. The time series of monthly mean deviations from the ERA-Interim on all significant levels within the layer of 150–100 hPa for ChinaADJ are displayed in Fig. 13. Large deviations from the ERA-Interim are seen on upper significant levels for the raw data before 2001. After 2001, the deviations on the same levels decrease. After the adjustment, the deviations from the ERA-Interim are greatly reduced for the data before 2001, and the time series of mean deviations on significant levels become more consistent during the period of 1979–2013.

      Figure 13.  Time series of monthly mean temperature deviations from ERA-Interim on the significant levels within the layer of 150–100 hPa before (black) and after (red) the adjustment.

    5.   Conclusions and discussion
    • In this study, the quality-controlled twice-daily radiosonde data on 12 mandatory levels collected at 80 sounding stations in China were tested and corrected. In total, 1884 temperature time series were processed. The results indicate that, for stations in southern China and Tibetan Plateau, the bias adjustments at 1200 UTC are obviously larger than those at 0000 UTC, and the opposite is true for stations in northern China. This is mainly because the solar elevation angle is still higher at 1200 UTC in the south and Tibetan Plateau, and the solar radiation is larger at 1200 UTC than that at 0000 UTC at these stations. In contrast, the solar radiation is larger at 0000 UTC than that at 1200 UTC at stations in northern China. The above results reflect the impact of solar radiation on the radiosonde temperature bias at different observational times. Considering the temperature deviations caused by solar radiation, many previous studies have pointed out that the radiosonde temperature time series in the nighttime (1200 UTC) can be a good reference series for the bias adjustment of radiosonde temperature in the daytime (0000 UTC) (Gaffen, 1994; Luers and Eskridge, 1998; Lanzante et al., 2003; Sherwood et al., 2005). However, the results of our study indicate that, due to differences in latitude and station elevation, using the nighttime series of radiosonde data is not an appropriate approach for all radiosonde stations in China. Guo and Ding (2010) tested the inhomogeneity of the monthly mean radiosonde temperature in China during the period of 1958–2005 by using the nighttime series of radiosonde data and reanalysis products as the reference series respectively. The results indicate that the breakpoints identified by the nighttime series (1200 UTC) as the reference series is only half that identified by using the reanalysis as the reference series.

      By comparing the raw radiosonde data with the ERA-Interim reanalysis dataset, the relatively large deviations are found on mandatory levels for the raw data before 2001, which is largely attributed to the evaluation system. Since the 1980s, WMO has entrusted the Met Office to evaluate systematic and random errors in radiosonde data worldwide based on the ECMWF 6-h forecast products. Large biases are found in China radiosonde data before 2000, and more than 50% of the radiosonde stations in China were often listed as unreliable by WMO (Guo and Ding, 2010). For this reason, the China Meteorological Administration (CMA) upgraded the entire sounding system in China in 2000 and 2001, including a comprehensive revision of several error correction meth-ods, such as the radiation error correction parameters (Guo and Ding, 2009). After the upgrade, deviations of Chinese radiosonde temperature data from the ECMWF reanalysis product have been greatly reduced. However, the upgrade of the sounding system in 2000 and 2001 cannot solve the problem of large deviations in the historical radiosonde data before 2001, which is also the reason why large deviations from the ERA-Interim still exist in the raw radiosonde data before 2001. After the adjustment in our study, the deviations mentioned above have been significantly reduced on all mandatory levels, and the time series of mean deviations have become more continuous.

      In addition, in our previous work (Chen and Yang, 2014), it was found that after test and adjustment are conducted separately for radiosonde temperature data at 0000 and 1200 UTC, discontinuities still exist at points of time corresponding to metadata in the 1200 minus 0000 UTC temperature time series. To address this problem, the test and breakpoint identification for the 1200 minus 0000 UTC temperature time series was introduced in our study. Results of the tests reveal obvious discontinuities in the 1200 minus 0000 UTC temperature time series for both the raw radiosonde data and RAOBCORE. The homogeneity of the 1200 minus 0000 UTC temperature time series is intensified for ChinaADJ and the RMSE is reduced.

      Overall, our work can effectively test and adjust the discontinuity problem induced by artificial errors in the daily radiosonde temperature data in China. For the temperature trends after the adjustment, it is found that the cooling trends in the upper troposphere and lower stratospheres are weakened, while the warming trends in the middle and lower troposphere are intensified. This result is consistent with the results of previous studies (Guo et al., 2009; Chen and Yang, 2014). Compared with the radiosonde temperature data at 80 stations in China extracted from RAOBCORE, it is found that no adjustment has been conducted for RAOBCORE since 2008, and obvious discontinuities are still found in the temperature time series at 0000 and 1200 UTC as well as 1200 minus 0000 UTC. Therefore, the data quality in ChinaADJ is better than that in RAOBCORE.

      Acknowledgments. The authors thank Dr. Leopold Haimberger for providing valuable suggestions to improve the manuscript.

Reference (28)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return