This post presents observed 17-year (204-month) Sea Surface Temperature (SST) anomaly trends versus those of the climate models from the IPCC AR4, and presents it as the casual observer would look at it in terms of how well the models portray sea surface temperatures during the last 30 years, which is the satellite era of Sea Surface Temperature data.
This is a post about perceptions, about what the casual observer would perceive if he or she were to investigate the data.
The first post here to use 17-year trends, 17-Year And 30-Year Trends In Sea Surface Temperature Anomalies: The Differences Between Observed And IPCC AR4 Climate Models, received a good amount of attention. Roger Pielke Sr prepared a post about it, and Anthony Watts cross posted it at WattsUpWithThat. The “17-year and 30-year trend post” showed how poorly the IPCC AR4 climate models hindcast and projected the most recent 17-year and 30-year trends in Global and Hemispheric Sea Surface Temperature anomalies. But the graphs also presented the multidecadal variations in trends for the model mean and observation-based data since 1900. This generated a post from Tamino, over at his blog Open Mind. Maybe Tamino was not aware that the vast majority of the readers here understand that a model mean would average out any multidecadal variations in the individual model runs, if any existed. And maybe Tamino was not aware that the readers here also understand that very few of the models included in the IPCC AR4 model mean actually have multidecadal variations, because we have presented and discussed this, and the reasons for it, in earlier posts and comments. Yet another example of this was presented in the reply post to Tamino. Regardless of whether Tamino was aware, it still appears his post was solely intended to distract his readers from the conclusions of the “17-year and 30-year trend post,” which were summarized in Table 1.
This post presents time-series graphs that compare Global and Hemispheric satellite-era Sea Surface Temperature anomalies to the model mean of the hindcasts and projections of the AR4 climate models. Time-series graphs and linear trends are easy to understand. They are used and discussed quite frequently here. Like the “17-year and 30-year trend post”, 204-month (17-year) linear trends are used. The choice of 204-months (17-years) is based on the Santer et al (2011) paper, Separating Signal and Noise in Atmospheric Temperature Change: The Importance of Timescale. In the abstract, Santer et al (2011) conclude with:
“Our results show that temperature records of at least 17 years in length are required for identifying human effects on global-mean tropospheric temperature.”
Since Sea Surface Temperature anomalies are not as noisy as Lower Troposphere Temperature anomalies, we’ll assume that 17 years would also be an acceptable timescale to present sea surface temperature trends on global and hemispheric bases.
The satellite era of Sea Surface Temperature observations is only 30 years old, so we’ll use a relatively simple presentation. We’ll look at the linear trends of the models and observations on Global, Northern Hemisphere, and Southern Hemisphere bases for the first 17 years of the data and the last 17 year of data. This will show how well (or poorly) the linear trends of the IPCC AR4 model hindcasts and projections matched the trends of the observation-based data during these two periods. The results may surprise some of you.
1. The dataset used for observations is Reynolds OI.v2 Sea Surface Temperature data.
2. This post does not provide statistical analyses like you might find at Lucia’s The Blackboard.Only the data and the linear trends are presented. Anyone is welcome to carry the analyses further.
3. Each of the of the graphs presents all of the observed and model Sea Surface data, from 1982 to present, even though the trends for 204-months are being illustrated. The unused data is shown as dashed lines.
4. The two periods presented in the post share a common factor: the 1997/98 El Niño. This El Niño event was described as the El Niño of the century due to its strength. It was massive on all accounts, and is plainly visible in the graphs of both periods. The 1997/98 El Niño makes its presence known in the data toward the end of the first 204-month period, which runs from January 1982 to December 1998, and toward the beginning of the second period, which runs from November 1994 to October 2011. This will undoubtedly be raised in some of your comments.
Figure 1 compares the trends Global Sea Surface Temperature anomaly observations for the first 17-year period to the trends of the IPCC AR4 Model Mean. Not too surprisingly, the two trends are relatively close. The trend of the observed Global Sea Surface Temperature anomalies is approximately 0.123 deg C per decade, while the hindcast and projection of the models present a trend of 0.115 deg C per decade. The trend of the observations is only 7% higher than the model trend. Not bad. The model trend actually appears conservative during this period.
Unfortunately for the models, during the last 17-year period, Figure 2, the rise in observed global sea surface temperature anomalies (0.04 deg C per decade) is only 26% of the rise projected by the models (0.155 deg per decade). That’s not too good when we consider the rise in Sea Surface Temperature is supposed to be forced by Anthropogenic Greenhouse Gases and that the model mean is supposed to represent the forced component of the all of the models, without the noise caused by the internal variability inherent in the individual model ensemble members.
Figure 3 compares the observed trend versus the model mean trend for the Sea Surface Temperature anomalies of the Northern Hemisphere for the initial 17-year period. The observed trend from 1982 through 1988 (0.198 deg C per decade) is about 55% higher than that hindcast and projected by the models (0.128 deg C per decade). But during the last 17 years, Figure 4, the observed trend (0.061 deg C per decade) is 64% lower than the trend of the model mean (0.171 deg C per decade). In other words, the trend of the model mean was too low during the first 17 years of this 30-year Sea Surface Temperature dataset and way too high during the last 17 years.
The problems for the climate models persist, as one would expect, in the Southern Hemisphere. For the first 17-year period, Figure 5, the observed trend in Southern Hemisphere Sea Surface Temperature anomalies (0.065 deg C per decade) is 62% of the trend hindcast and projected by the model mean (0.105 deg C per decade). And it gets even worse. The trend of the observed rise in Sea Surface Temperature anomalies during the last 17 years of the Southern Hemisphere (0.025 deg C per decade) is only 17% of the trend that was hindcast and projected by the models. In other words, for the oceans of the Southern Hemisphere, which represent about 40% of the surface area of the globe, the linear trend of the Sea Surface Temperature anomalies that was hindcast and projected by the models is more than 5 times higher than what was observed. More than 5 times. Let’s try it one more time; the average of all of the climate models used in the IPCC AR4 have projected Sea Surface Temperature trends that are more than 5 times higher than what has been experienced in the Southern Hemisphere, or 40% of the surface area of the globe, for the last 17 years.
So what impression is the casual observer left with if he or she were to investigate how well climate models can hindcast and project sea surface temperatures over 17-year periods, a time span that is appears to be acceptable to the who’s-who of climate scientists that helped prepare the Santer et al (2011) paper? Not a very good impression. They can see that the observed Sea Surface Temperature trends and those projected by the climate models only appear to come close to matching one another on a global basis, but that the match is only good for the first 17-year period of the satellite-era Sea Surface Temperature data. They can see that the models do not come close to matching observations in either hemisphere during the first or last 17-year periods.
The casual observer may investigate further and discover explanations for the poor behavior of the models, such as the models don’t do a good job of reproducing natural variability, and the models aren’t initialized to reproduce the multidecadal variations observed in the surface temperature record, and the like. And do you know what those explanations sound like to the casual observer? Excuses. And do you know what they sound like to a not-so-casual observer, like me? Excuses.
Both the Reynolds OI.v2 Sea Surface Temperature data and the IPCC AR4 Hindcast/Projection (TOS) data used in this post are available through the KNMI Climate Explorer. The HADISST data is found at the Monthly observations webpage, and the model mean data is found at the Monthly CMIP3+ scenario runswebpage.