In the past year or so, the blogosphere has been full of claims concerning short-term global temperature trends and their supposed falsification of the scientific consensus on anthropogenic global warming as set forth in the IPCC Fourth Assesment Report (AR4). In this post, I’ll present a comparison of the AR4 near-term projection to smoothed observed trends. This comparison shows the recent observed trend to be somewhat below the projection, but still well within a reasonable confidence interval.
The above mentioned contrarian claims include, of course, the oft-repeated canard that global warming has “stopped” (or slowed significantly) since 1998, or that IPCC projections are based on climate models that are “failing abjectly”, as the Cato Institute (Pat Michaels) put it.
The various claims have been rebutted effectively: see, for example, this recent post at Real Climate demolishing Monckton’s latest travesty, or the latest discussion of short term temperature “trends” and other curve-fitting follies at Open Mind.
In general, the contrarians tend to gloss over or underestimate the uncertainties in the short-term model projections and/or corrresponding observations. Short-term linear trends tend to fluctuate wildly about the more stable longer-term trend, sometimes well above the projected trend and sometimes well below. For example, the eight-year trend in the GISTemp global temerature record (from NASA) was relatively flat from 2001-2008, but stood at 0.35 deg/decade two years earlier.
Meanwhile, the longer term trends are relatively stable, and indeed are mostly higher in the 2000s. Indeed, in all the main global surface temperature data sets both the 20-year trend and trend from 1979 forward was higher at the end of 2008 than at the beginning of the decade.
But having waded through the back and forth volleys of spaghetti graphs of detailed monthly model ensembles and observations, I’m convinced that a straightforward comparison of smoothed projections and observations would bring a different perspective and, perhaps, some welcome clarity.
Such a comparison is found for the 2001 IPCC TAR (Third Assessment Report) projections in a 2007 Science “brief” by Rahmstorf et al. (PDF available here).
The following graph is an enhanced version of the key figure and can be found in Tamino’s post on this subject. The IPCC projections were based on climate model run ensembles for various emission scenarios from 1990 forward. The figure shows observations up to 2007 for NASA GISTemp (red) and HADCrut (blue) were mainly in the upper range of projections (dotted lines and shaded area).
In order to undertake a similar analysis comparing current observations to the latest IPCC projection, we first derive a smoothed temperature record to the end of 2008, using a 21-point lowess smooth (data used goes back to 1970 and is baselined to 1980-1999).
Both temperature records rise fairly smoothly over the last three decades, with GISS slightly above HADCrut for the 2000s.
As in TAR, the latest IPCC projections found in AR4 (Fourth Assessment Report) were based on climate model ensembles, run for a set of standard emission scenarios, only this time from 2000 forward. The projections were derived by extracting key graphs in Chapter 10 of the WG1 (Working Group 1) report. First we give the detailed projections based on multi-model means for the A1B and A2 scenarios, as extracted from Figure 10.5 from AR4 Chapter 10 (with the use of a bitmap extraction tool.
In the current decade, the two scenarios are quite close, with a slight increasing divergence as we approach 2030. Here’s a closer look at 2000-2010:
As one would expect, the observed annual values show considerable interannual variation. But even the model means show some variation, pointing up one of the advantages of the smoothing approach. The 2001-2008 trend for the A1B is 0.24 deg/decade, while the 2000-2010 linear trend is 0.21 deg/decade, an artifact of the random variation within the model ensemble. So the 2001-8 period for that scenario yields an exaggerated expected trend. (It just so happens that 2008 is the warmest year in the first decade of the A1B model mean annual series, and 2001 the coolest).
In the first decade of the projection period there is minimal divergenge (A1B is just over 0.2 deg/decade and A2 is just below 0.19 deg/decade). So we will use the illustrated 0.2 deg/decade as our smoothed trend.
Finally, we have used AR4 Figure 10.26 to impute confidence intervals for the smoothed trend. That figure gives a standard deviation of about 0.06 deg deg C in 2000, rising to 0.12 deg C (A2) or 0.13 deg C (A1B) by 2010 for the A1B scenario. (This zoomed PDF graph of the A1B smoothed projection to 2020 shows an example of the PDF markup used to estimate these values).
Using a z-score of 1.63 (corresponding to a 90% confidence interval) yields limit trends of 0.1 deg/decade up to 0.3 deg/decade. This interval may be slightly underestimated (since it does not take into account the small divergence between the various scenarios), but seems a reasonable approximation.
Putting all the elements together gives the figure at the top of the post:
So for 2000-2008, the IPCC smoothed projection was an average of 0.33 deg +/- 0.13 deg (90% confidence interval) above the 1980-99 baseline. Both NASA GISS (0.26 deg) and HADCrut (o.25 deg) were within that range, albeit in the lower part.
- We are only 9 years into the 31-year early 21st century projection period (2000-2030), so it’s far too early to draw any definite conclusions concerning the attribution of model-observation discrepancies.
- There continues to be a growing divergence between GISS and HADCrut data sets (undoubtedly due in large part to the inclusion of the Arctic region temperature estimates in GISS).
- TAR and AR4 both projected similar trends in the 2000-10 period. Yet observations were well above the TAR projection, while observations (extended only two more years to 2008) are below the AR4 trend. This is due in large part to the change in baseline methodology for matching model projections to the past.
Indeed, smoothing of the observations shows the import of this last point. As shown above, the smoothed observation trends are already about 0.05 deg C below the IPCC trend line at the beginning of the projection period (and never “catches up”). And as seen below, the projected trend sits well above the smoothed observations even back to 1990.
This suggests that a different baselining methodology might show a better match between the smoothed projection and observations. But that is a discussion for another time.
[Update: A few references on lowess smoothing are listed here:
Worked out example: