by Judith Curry
Santer et al. have a new paper in press entitled “Separating Signal and Noise in Atmospheric Temperature Changes: The Importance of Time Scale.”
Separating Signal and Noise in Atmospheric Temperature Changes: The Importance of Timescale, J. Geophys. Res., doi:10.1029/2011JD016263, in press.
B.D. Santer, Carl A. Mears, C. Doutriaux, Peter Martin Caldwell, Peter J. Gleckler, Tom M.L. Wigley, Susan Solomon, Nathan Gillett, Detelina P. Ivanova, Thomas R Karl, John R. Lanzante, Gerald A. Meehl, Peter A. Stott, Karl E Taylor, Peter Thorne, Michael F Wehner, Frank J. Wentz.
Abstract. We compare global-scale changes in satellite estimates of the temperature of the lower troposphere (TLT) with model simulations of forced and unforced TLT changes. While previous work has focused on a single period of record, we select analysis timescales ranging from 10 to 32 years, and then compare all possible observed TLT trends on each timescale with corresponding multi-model distributions of forced and unforced trends. We use observed estimates of the signal component of TLT changes and model estimates of climate noise to calculate timescale-dependent signal-to-noise ratios (S/N). These ratios are small (less than 1) on the 10-year timescale, increasing to more than 3.9 for 32-year trends. This large change in S/N is primarily due to a decrease in the amplitude of internally generated variability with increasing trend length. Because of the pronounced effect of interannual noise on decadal trends, a multi-model ensemble of anthropogenically-forced simulations displays many 10-year periods with little warming. A single decade of observational TLT data is therefore inadequate for identifying a slowly evolving anthropogenic warming signal. Our results show that temperature records of at least 17 years in length are required for identifying human effects on global-mean tropospheric temperature.
Key Points
- Models run with human forcing can produce 10-year periods with little warming
- S/N ratios for tropospheric temp. are ~1 for 10-yr trends, ~4 for 32-yr trends
- Trends >17 yrs are required for identifying human effects on tropospheric temp.
The paper is not available online, but Pielke Sr. has posted some excerpts on his blog, some of which are reproduced below:
From the conclusions:
We relied on control runs from the CMIP-3 multi-model archive for our estimates of climate noise. Estimates of externally forced climate-change signals were obtained from three different sets of satellite-based observations and from CMIP-3 simulations of 20th and 21st-century climate change. In contrast to almost all previous work, we compared modeled and observed TLT changes on multiple timescales (using maximally overlapping trends) rather than over a single period of record. For timescales less than the record length, this strategy reduces the impact of climate noise on estimates of the signal component of observed (and simulated) temperature trends. The fact that our pf′ values (even for 30-year TLT trends) are sensitive to the addition of a single year of observational data indicates the dangers of ignoring the effects of interannual variability on signal estimates, as was done, for example, in Douglass et al. [2007].
Because of the large effect of year-to-year variability on decadal trends, roughly 10% of the 10-year TLT trends in the 20CEN/A1B runs are less than zero (Figure 4A). This result shows that anthropogenically forced models can replicate the recent muted warming of the surface [Easterling et al., 2009; Knight et al., 2009] and the lower troposphere. Claims that minimal warming over a single decade undermine findings of a slowly-evolving externally-forced warming signal [e.g., as in Investor’s Business Daily, 2008; Happer, 2010] are simply incorrect.
On all timescales examined here, the TLT trends in the observational satellite datasets are not statistically unusual relative to model-based distributions of externally forced TLT trends. While this consistency is encouraging, it should be qualified by noting that: 1) The multi-model average TLT trend is always larger than the average observed TLT trend; 2) As the trend fitting period increases, values of pf decline, indicating that average observed trends are increasingly more unusual with respect to the multi-model distribution of forced trends. Possible explanations for these results include the neglect of negative forcings in many of the CMIP-3 simulations of forced climate change ), omission of recent temporal changes in solar and volcanic forcing [Wigley, 2010; Kaufmann et al., 2011; Vernier et al., 2011; Solomon et al., 2011], forcing discontinuities at the ‘splice points’ between CMIP-3 simulations of 20th and 21st century climate change [Arblaster et al., 2011], model response errors, residual observational errors [Mears et al., 2011b], and an unusual manifestation of natural internal variability in the observations (see Figure 7A).
JC comment: Note that the CMIP3 simulations were the basis for the “very likely” in the attribution statement, which was the subject of the uncertainty monster post.
Although we considered three different observational estimates of TLT changes (and one observational estimate of SST changes), our analysis does not comprehensively explore the impact of data uncertainties on model evaluation.
In summary, because of the effects of natural internal climate variability, we do not expect each year to be inexorably warmer than the preceding year, or each decade to be warmer than the last decade, even in the presence of strong anthropogenic forcing of the climate system. The clear message from our signal-to-noise analysis is that multi-decadal records are required for identifying human effects on tropospheric temperature. Minimal warming over a single decade does not disprove the existence of a slowly-evolving anthropogenic warming signal.
Pielke Sr only has minor criticisms of the paper. Looks like the editor doesn’t need to resign over this one :)
How well do CMIP3 models do in simulating natural internal variability?
Lets take a look at Fig. 9.7 in the IPCC AR4 WGI Report:
Figure 9.7. Comparison of variability as a function of time scale of annual global mean temperatures (°C2 yr–1) from the observed record (Hadley Centre/Climatic Research Unit gridded surface temperature data set (HadCRUT3), Brohan et al., 2006) and from AOGCM simulations including both anthropogenic and natural forcings. All power spectra are estimated using a Tukey-Hanning filter of width 97 years. The model spectra displayed are the averages of the individual spectra estimated from individual ensemble members. The same 58 simulations and 14 models are used as in Figure 9.5a. All models simulate variability on decadal time scales and longer that is consistent with observations at the 10% significance level. Further details of the method of calculating the spectra are given in the Supplementary Material, Appendix 9.C.
The point I want to make (and I made this point point in the Uncertainty Monster paper) is globally, the modeled spectral density of the variability, when compared with observations, is too high for periods of ~ 8-17 years, and too low for periods of 40-70 years. Because this is a log-log scale, it is somewhat difficult to eyeball this, and one might argue that this is all within the range of uncertainty of the climate models and the climate models and observations agree ‘ok.’
In the context of Santer et al.’s conclusions regarding “Trends >17 yrs are required for identifying human effects on tropospheric temp.”, they might be in store for a surprise if the cool phase of PDO (nominally of 60 year period) persists for 30 years. The model’s failure to capture the observed level of power in the periods 8-17 years and 40-70 years has biased Santer et al.’s conclusion towards the low end of the spectrum.
JC conclusion. Santer et al. have laid down the gauntlet with this paper in terms of providing a method for falsifying climate model simulations for the purpose of attribution of 20th and early 21st century temperature variations. It would be interesting to see the same study conducted for the CMIP5 simulations, but I suspect that there might not be much of a change.
So, what are your bets for the duration of the current period of “minimal warming” ?