Research ArticleCLIMATOLOGY

Causes of irregularities in trends of global mean surface temperature since the late 19th century

See allHide authors and affiliations

Science Advances  06 Jun 2018:
Vol. 4, no. 6, eaao5297
DOI: 10.1126/sciadv.aao5297
  • Fig. 1 Observed, reconstructed, and CMIP5 simulated GST for the whole period 1891–2015.

    (A) Monthly GST anomalies (January 1891 to September 2017) from the three analyses that make up the WMO data set and the WMO average. Shown are the two warming and three slowdown periods. Estimated 95% confidence limits on the WMO data set are shown in brown. (B) Reconstructions of monthly GST 1891–2015 from each of the various forcing factors. This uses the average of 81,000 cross-validated regression equations calculated from data covering this whole period. (C) (a) Average cross-validated reconstruction (red) of observed monthly WMO GST (black) for the whole period 1891–2015. (b) As in (a) but using the CMIP5 40 model average (blue) of monthly GST.

  • Fig. 2 Reconstructions for slowdown period 1, 1896–1910, where time series include the overlapping longer period 1891–1913 and the observed series 1890–1914.

    (A) Average reconstruction of GST for 1891–1913. Shown are reconstructions (red), 95% confidence range of GST for the 54 regression equations (light blue), and average CMIP5 simulation of GST (blue) and WMO GST (black). Linear trends are shown from 1896 to 1910 using the REML trend method. (B) Reconstruction of WMO GST for 1891–1913. Shown for each panel are averaged observed GST (black) and each of the 54 regression equation estimates of GST (blue) for the different forcing factors and their combinations and for the three component WMO global temperature data sets. (C) Linear component of total temperature change for slowdown period 1: 1896–1910, for WMO average GST (observations), the reconstruction, its difference from WMO GST, and component forcings. Red lines show ±1σ uncertainties in the temperature changes. Stars denote significance at the 1% level or better. The REML trend method is used.

  • Fig. 3 Reconstructions for warming period 1, 1911–1940.

    (A) Average reconstruction of GST for warming period 1: 1911–1940. Linear trends are shown from 1911 to 1940 using the REML trend method. Otherwise as for Fig. 2A. (B) Reconstruction of WMO GST for warming period 1: 1911–1940. Otherwise as for Fig. 2B. (C) Linear component of total temperature change for warming period 1: 1911–1940. Stars denote significance at the 1% level or better. Otherwise as for Fig. 2C.

  • Fig. 4 Reconstructions for slowdown period 2, 1941–1975, where the time series include the lead-in period 1939–1940.

    (A) Average reconstruction of GST for slowdown period 2: 1941–1975. Thick trend lines are for 1941–1975, and thin lines are for 1951–1975. Otherwise as for Fig. 2A. (B) Reconstruction of WMO GST for slowdown period 2: 1941–1975. Otherwise as for Fig. 2B. (C) Linear component of total temperature change for slowdown period 2: 1951–1975. Stars denote significance at the 1% level or better. Otherwise as for Fig. 2C.

  • Fig. 5 Reconstructions for warming period 2, 1976–1997.

    (A) Average reconstruction of GST for warming period 2: 1976–1997. The linear trends are for 1976–1997. Otherwise as for Fig. 2A. (B) Reconstruction of WMO GST for warming period 2: 1976–1997. Stars denote significance at the 1% level or better. Otherwise as for Fig. 2B. (C) Linear component of total temperature change over warming period 2: 1976–1997. Stars denote significance at the 1% level or better. Otherwise as for Fig. 2C.

  • Fig. 6 Reconstructions for slowdown period 3, 1997–2015 and its three main sub-periods, where the observed time series include the lead-in period 1995–1996.

    (A) Average reconstruction of GST for slowdown period 3: 1997–2015. The linear trends are for 1998–2013 (thin lines) and 2001–2010 (thick lines). Otherwise as for Fig. 2A. (B) Reconstruction of WMO GST for slowdown period 3: 1997–2015. Otherwise as for Fig. 2B. (C) Linear components of total temperature change over slowdown period 3. (a) 1998–2013; (b) 2001–2010; (c) 2001–2013. Stars denote significance at the 1% level or better. Otherwise as for Fig. 2C. (D) Summary of the contributions of significant forcing (at <1% level) factors to GST trends (°C per decade) during (a) warming periods and (b) slowdown periods. The period 2001–2013 is used to represent slowdown period 3 where the appreciable but not significant ENSO-induced trend is shown.

  • Fig. 7 Global maps of observed and reconstructed temperature anomalies for some key results.

    (A) (a) Observed maps of temperature anomalies (°C) in 1907–1910. (b) Reconstructed anomalies. (c) Observed minus reconstructed anomalies. Significant differences at the 1% level in each 5° × 5° box for observed minus reconstructed anomalies (two-sided t test) are shown by stars. Dark red indicates warmest, and dark blue denotes coldest. (B) (a) Observed maps of temperature anomalies in 1944–1945. (b) Reconstructed anomalies. (c) Observed minus reconstructed anomalies. Otherwise as for (A). (C) Reconstructed linear component of worldwide temperature change due to TSI forcing, 2001–2010. Calculated changes for 5° × 5° boxes are ordinary least-squares trends multiplied by the period length. Dark red indicates warmest, and dark blue denotes coldest.

  • Table 1 Correlation skill of average reconstructed GST and CMIP5 average GST with WMO GST.
    PeriodLow pass (r)Significance (%)PeriodHigh pass (r)Significance (%)
    A. Slowdown period 1, 1896–1910
    Reconstruction, ≥2 months0.660.12–10 months0.32<0.01
    Reconstruction, ≥2 seasons0.720.22–10 seasons0.322.0
    Reconstruction, ≥2 years0.782.02–10 years0.800.05
    CMIP5, ≥2 months−0.342–10 months0.14
    CMIP5, ≥2 seasons−0.472–10 seasons0.341.0
    CMIP5, ≥2 years−0.752–10 years−0.47
    B. Warming period 1, 1911–1940
    Reconstruction, ≥2 months0.67<0.012–10 months0.39<0.01
    Reconstruction, ≥2 seasons0.710.022–10 seasons0.39<0.01
    Reconstruction, ≥2 years0.780.52–10 years0.590.1
    CMIP5, ≥2 months0.380.52–10 months0.160.5
    CMIP5, ≥2 seasons0.432–10 seasons0.19
    CMIP5, ≥2 years0.522–10 years−0.52
    C. Slowdown period 2, 1951–1975
    Reconstruction, ≥2 months0.64<0.012–10 months0.135.0
    Reconstruction, ≥2 seasons0.72<0.012–10 seasons0.360.05
    Reconstruction, ≥2 years0.80<0.012–10 years0.81<0.01
    CMIP5, ≥2 months0.242–10 months−0.02
    CMIP5, ≥2 seasons0.272–10 seasons0.03
    CMIP5, ≥2 years0.312–10 years0.28
    D. Warming period 2, 1976–1997
    Reconstruction, ≥2 months0.80<0.012–10 months0.28<0.01
    Reconstruction, ≥2 seasons0.86<0.012–10 seasons0.51<0.01
    Reconstruction, ≥2 years0.940.012–10 years0.87<0.01
    CMIP5, ≥2 months0.580.52–10 months0.161.0
    CMIP5, ≥2 seasons0.650.52–10 seasons0.20
    CMIP5, ≥2 years0.760.12–10 years0.505.0
    E. Slowdown period 3, 1998–2013
    Reconstruction, ≥2 months0.60<0.012–10 months0.250.05
    Reconstruction, ≥2 seasons0.690.052–10 seasons0.370.5
    Reconstruction, ≥2 years0.840.12–10 years0.800.02
    CMIP5, ≥2 months0.345.02–10 months0.09
    CMIP5, ≥2 seasons0.412–10 seasons0.20
    CMIP5, ≥2 years0.542–10 years0.26

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/4/6/eaao5297/DC1

    section S1. Key forcings used expressed in absolute units

    section S2. Separating the two AMO time series and ENSO from the IPO

    section S3. The REML linear temperature change method

    section S4. CMIP5 model experiments used

    section S5. Structure of the regression equations and cross-correlation of predictors

    section S6. Comparison of annual reconstruction and CMIP5 root mean square errors with estimated WMO GST annual uncertainties

    section S7. Reconstruction and CMIP5 correlation statistics for 1941–1975

    section S8. Substituting ERA Interim GST for WMO GST for 1997–2015

    section S9. Response times of GST to TSI forcing

    section S10. Further details of the WMO data sets

    section S11. Structure of multiple regression residuals and cross-correlation of predictors

    fig. S1. Volcanic, solar, and GA forcings in original units without smoothing, expressed as anomalies from their 1961–1990 averages.

    fig. S2. Additional details of monthly time series of the predictors.

    fig. S3. Maximum likelihood plot of the serial correlation of the residuals and their variance for the reconstructed linear component of WMO temperature change over 2001–2013.

    fig. S4. Identifying the boundaries between consecutive slowdown and warming periods using trend analysis.

    fig. S5. Reconstructed and CMIP5 root mean square errors for annual data compared to 1σ uncertainties in annual WMO GST values including additional uncertainties due to the spread of the three component data sets.

    fig. S6. Comparing the reconstructions using ERA Interim and WMO GST during slowdown period 3.

    fig. S7. As in the Fig. 6C panels for 1998–2013, 2001–2010, and 2001–2013 in the main text but for ERA Interim.

    fig. S8. Original (unfiltered) and e-folded monthly TSI time series January 1981 to December 2015 reflecting the three sets of delayed e-folding response times used in the paper.

    table S1. CMIP5 models and historical experiments used in this paper.

    table S2. Combinations of predictors used in the 18 main regression equations.

    table S3. Correlation skill of average reconstruction GST and CMIP5 average GST with WMO GST in slowdown period 2, 1941–1975.

    table S4. Cross-correlation of monthly predictors 1891–2015.

    table S5. Cross-correlation of monthly predictors 1951–2015.

    References (5759)

  • Supplementary Materials

    This PDF file includes:

    • section S1. Key forcings used expressed in absolute units
    • section S2. Separating the two AMO time series and ENSO from the IPO
    • section S3. The REML linear temperature change method
    • section S4. CMIP5 model experiments used
    • section S5. Structure of the regression equations and cross-correlation of predictors
    • section S6. Comparison of annual reconstruction and CMIP5 root mean square errors with estimated WMO GST annual uncertainties
    • section S7. Reconstruction and CMIP5 correlation statistics for 1941–1975
    • section S8. Substituting ERA Interim GST for WMO GST for 1997–2015
    • section S9. Response times of GST to TSI forcing
    • section S10. Further details of the WMO data sets
    • section S11. Structure of multiple regression residuals and cross-correlation of predictors
    • fig. S1. Volcanic, solar, and GA forcings in original units without smoothing, expressed as anomalies from their 1961–1990 averages.
    • fig. S2. Additional details of monthly time series of the predictors.
    • fig. S3. Maximum likelihood plot of the serial correlation of the residuals and their variance for the reconstructed linear component of WMO temperature change over 2001–2013.
    • fig. S4. Identifying the boundaries between consecutive slowdown and warming periods using trend analysis.
    • fig. S5. Reconstructed and CMIP5 root mean square errors for annual data compared to one-sigma uncertainties in annual WMO GST values including additional uncertainties due to the spread of the three component data sets.
    • fig. S6. Comparing the reconstructions using ERA Interim and WMO GST during slowdown period 3.
    • fig. S7. As in the Fig. 6C panels for 1998–2013, 2001–2010, and 2001–2013 in the main text but for ERA Interim.
    • fig. S8. Original (unfiltered) and e-folded monthly TSI time series January 1981 to December 2015 reflecting the three sets of delayed e-folding response times used in the paper.
    • table S1. CMIP5 models and historical experiments used in this paper.
    • table S2. Combinations of predictors used in the 18 main regression equations.
    • table S3. Correlation skill of average reconstruction GST and CMIP5 average GST with WMO GST in slowdown period 2, 1941–1975.
    • table S4. Cross-correlation of monthly predictors 1891–2015.
    • table S5. Cross-correlation of monthly predictors 1951–2015.
    • References (57–59)

    Download PDF

    Files in this Data Supplement:

Navigate This Article