(a) Global radiative forcing components used in our study, (b) decomposition of the four AER components including indirect aerosol effects, and (c) spatial decomposition of the effective and noneffective AER. Scaled coastal HadOST (blue) and coastal BE anomalies (red) in comparison with (d) 60°N–60°S HadOST (black) and (e) as in (d), but with coastal ERSSTv4, coastal GISTEMP, and 60°N–60°S ERSSTv4.
TWR to estimate the TCR adjustment factors for WMGHGs, AER, and VOL for (left) NHem/SHem and (right) Land/Ocean: (a),(b) Allforcing warming contributions in CMIP5, HadCM3, HadOST, and the response model, (c),(d) WMGHG only, (e),(f) AER only, and (g),(h) VOL only contributions in HadCM3 and the response model. Modeled VOL (negative) temperature response is shifted by +10 and +25 years merely for better readability. The timeline of volcanic eruptions (scaled radiative forcing) is shown in black in (g) and (h). All data are low-pass filtered to remove interannual variability. The boxes at the bottom show the inferred (diagnosed) warming ratios (TWRD) for WMGHG and AER using the product of the ratios of observed (red) and modeled (orange) allforcing TWRs in (a) and (b), multiplied by the modeled TWRs (orange) for WMGHG in (c) and (d) and AER in (e) and (f). The estimated warming ratios (TWRE) refer to the simulated response model TWR using TWRD. Both values are given in light purple. Only the 30-yr period of strongest differential warming is used for the central TWR estimates. VOL TWR is only a function of the fast response.
Summary panel for all the necessary response model parameters, including their justification. (top left) Global, hemispheric, and Land/Ocean TCR scaling factors for WMGHGs, AER, and VOL based on the findings shown in Fig. 1. (bottom left) Forcing response time estimates and sensitivities used in this analysis are provided, including their source. Color codes are used for better readability. The pink labels in the lower box refer to the original AER-TWRD. In gray are the associated coupling factors. Surface temperature trends in (a) HadOST, (b) CMIP5, and (c) HadCM3 for 1978–2017.
(a) Fractional variance (square of the model error) for impulse response model uncertainty (green), total radiative forcing uncertainty (blue), and internal variability uncertainty (gray). The 1σ (32nd–68th percentiles) range is shown. We note that internal variability is no response model uncertainty in a strict sense as it is added post hoc (i.e., onto the calculated temperature). The peaks in the response model uncertainty coincide with volcanic eruptions (e.g., Tambora in 1816). (b) The internal variability from selected CMIP5 piControl runs is contrasted with the unforced residuals from the GMST datasets used in this study. Observed and modeled time series are low-pass filtered with a 30-yr smoothing radius. The standard error is provided in parentheses.
Illustration of the ENSO influence on our results. In the upper graph in each panel, the observations are plotted against the response after adding MEI variability to the time series. The lower graph shows the raw impulse response model results against the ENSO-corrected suite of observational data. (a) Land (brown; including sea ice grid points), (b) NHem (red), (c) (green), (d) Ocean (purple; excluding sea ice grid points), and (e) SHem (blue). Observations from the HadOST composite (pale gray), Cru4CW (yellow), and BE (black) are shown. Explained variances R2 are given for non-ENSO corrected, model-adjusted (MEI), and observation-adjusted (MEI) (Foster and Rahmstorf 2011) low-pass-filtered correlations. The WWII correction factors are applied to both instrumental temperature time series in each panel (except Land). TCR values associated with alternative response model results are provided on the right of each panel (1.2–2.0 K).
Evolution of the response model from forcing and response times as applied in Haustein et al. (2017) (H17), with AER as used in CMIP5 (old AER) and the current version using CEDS AER (new AER). Note that the WWII bias correction is only applied in case of new AER in order to illustrate the impact (no change in Land only). The results are shown for (a) Land, (b) NHem, (c) Global, (d) Ocean, and (e) SHem. The two dashed lines in the lower graph of each panel indicate the variability of the result as a function of the ECS value applied in the response model. The default value of 3.0 K corresponds with our central estimate.
Unforced residual observed variability. Impulse response model (IRM) minus HadOST for (a) NHem, (b) Global, (c) SHem, (d) Land, and (f) Ocean. HadOST Global as in (b) is compared to CruCW4 and BE Global in (e). A 30-yr lowess smooth is added in each plot. The revised AMV index is shown in (a). The MEI is added in (b). Note that the rhs y-axis labels for AMV in (a) and MEI in (b) are different.
Spatial map of correlation coefficients R over time between 1850 and 2016. Positive correlations are shown in red and negative correlations in black. Annual means are used. (a) Time series of the global response model vs HadOST composite. (b) As in (a), but with MEI noise added to the global response model time series. (c) Time series of the NHem response model vs HadOST. (d) The improved AMV index (van Oldenborgh et al. 2009) vs HadOST. The AMV/NAVI region is highlighted with a red box. (e) As in (c), but with a 5-yr running means applied to both NHem and HadOST. (f) Combination of (c) and (e) where both regressors are detrended and low-pass filtered with a 5-yr running mean. (g) As in (d), but with both AMV and HadOST being detrended. (h) As in (e), but with a 10-yr running mean. (j) As in (f), but with a 10-yr running mean. (k) As in (d), but with both AMV and HadOST being detrended and low-pass filtered with a 20-yr running mean. (m) As in (e), but with a 20-yr running mean. (n) As in (f), but with a 20-yr running mean. The SHem area is shown in semitransparent colors to highlight the NHem region of interest.
Decadal GMST anomalies for the Twentieth Century Reanalysis, all observational data used in this study including the new Hybrid SST dataset (Cowtan et al. 2018), the CMIP5 subset, and the NorESM1-M global circulation model. Decades from (top) 1850–59 to (bottom) 2010–17 are shown in each row. All anomalies are given relative to the 1901–2000 baseline period. The 1940–49 decade that is affected by the WWII warm bias is highlighted by the red box.