Saturday, March 1, 2014

New paper falsifies climate model predictions at 95% confidence level, "very low" chance they could have predicted the "pause"

A new paper published in Nature Climate Change deals a near-fatal blow to climate models, finding the chances were "very low" that the models could have predicted the 'pause' in global warming over the past 20 years. The paper is lead authored by John Fyfe, a co-chair of the IPCC, who also published a recent paper in Nature finding that there has been no statistically-significant global warming for the past 20 years.

The paper falsifies climate model simulations at a 95% confidence level, stating, "the observed trends over this period lie outside the 5–95% range of simulated trends, or in other words, they are inconsistent with the simulated combination of internal variability and response to natural and anthropogenic forcings."

Recent observed and simulated warming


Nature Climate Change
 
4,
 
150–151
 
 
doi:10.1038/nclimate2111
Published online
 
Figure 1 shows observed3 (red) and simulated (black) trends over the past 20 years (1993–2012) in global mean surface temperature plotted against corresponding trends in eastern tropical Pacific sea surface temperature. As pointed out by Fyfe and colleagues1, the observed rate of global warming over this period is less than that simulated in all but two of 117 CMIP5 simulations. Figure 1shows an even more pronounced discrepancy over the eastern tropical Pacific, with the observed cooling trend being substantially more negative than that in any of the 117 CMIP5 simulations. The observations in Fig. 1 lie on the straight line that best fits the simulated global and eastern tropical Pacific temperature trends over the period from 1993 to 2012 — indicating that the observed global mean trend could be inferred from the observed tropical Pacific trend and the relationship between these two variables in the models.
Figure 1: Trends in global mean surface temperature and eastern tropical Pacific sea surface temperature for 1993–2012.
Trends in global mean surface temperature and eastern tropical Pacific sea surface temperature for 1993-2012.
Observed trends (red) are averages over 100 reconstructions of the HadCRUT4 dataset3. Simulated trends (black) are from 117 realizations of the climate from 37 CMIP5 models and their 5–95% ranges are shown with the black ellipse. The straight line is the best fit to the simulated global mean and eastern tropical Pacific trends, with a correlation of 0.63. As in Kosaka and Xie2 the eastern tropical Pacific is defined as the region east of the dateline and between 20° S and 20° N, and as in Fyfe et al.1 the simulations are sampled only where corresponding observations exist.
Because observations are sparse in polar regions, the calculated global mean trends could be less than actual trends given indications of rapid warming in the Arctic over the satellite record4. In our analysis, trends in both models and observations are computed only where adequate observations are available in situ, making this a robust like-for-like comparison of models and observations.Figure 2 shows observed (a) and model-average (b) trend maps over the past 20 years (1993–2012) computed for locations where adequate observations are available in situ. Over this period most of the observed regions exhibited warming, but much of Siberia, the eastern Pacific Ocean and the Southern Ocean cooled5. The regions of cooling over Siberia and the eastern Pacific Ocean are not seen in the simulated trends, although some Southern Ocean cooling is suggested on average. Figure 2b shows that for about 21% of grid cells with sufficient observational coverage the observed trends over this period lie outside the 5–95% range of simulated trends, or in other words, they are inconsistent with the simulated combination of internal variability and response to natural and anthropogenic forcings.
Figure 2: Trends in global surface temperature for 1993–2012.
Trends in global surface temperature for 1993-2012.
a, Observed trends. b, Average simulated trends from 117 simulations of the climate by 37 CMIP5 models. As in Fyfe et al.1 the simulations are sampled only where corresponding observations exist. Trends are computed only at grid points with at least 50% temporal coverage. The rectangles encompass the eastern tropical Pacific region2. In b the stippling indicates where the observed trends are outside the 5–95% range of the simulated trends.
Kosaka and Xie2 concluded that the current hiatus is part of internal climate variability tied to La Niña-like decadal cooling, but we point out that internal climate variability alone does not readily explain the difference between simulated and observed trends over this period, given that none of the 117 CMIP5 simulations captured the current eastern tropical Pacific cooling trend. Although on average the models show realistic 20-year trend variance in this region based on the limited observational record (Supplementary Fig. 1), and do not generally underestimate interannual variability associated with the El Niño Southern Oscillation5 (Supplementary Fig. 2), CMIP5 simulations of internal variability in the tropical Pacific do exhibit pronounced systematic errors5 and it remains possible that the models underestimate the probability of large internally generated cooling trends in this region. We further note that the models simulate externally forced warming in this region since about 1970 (Supplementary Fig. 1), which is likely to be associated in part with simulated weakening of the Walker circulation567, whereas observed sea surface temperatures cooled and the Walker circulation strengthened over the past 20 years25.
In conclusion, we agree with Kosaka and Xie2 that accounting for cooling in the eastern tropical Pacific could, in principle, reconcile recent observed and simulated global warming. However, based on the CMIP5 ensemble of climate simulations, the probability of simulating the recently observed eastern tropical Pacific cooling with a freely running climate model under the CMIP5 radiative forcing protocol is very low, and hence so too is the probability of simulating the observed global temperature change over the past 20 years.

1 comment:

  1. http://stevengoddard.wordpress.com/2014/04/29/shock-news-97-of-climate-models-overpredict-warming/

    ReplyDelete