In two posts about a year ago and more recently in my book, we compared the satellite-based sea surface temperature anomalies and CMIP3-based climate model simulations on an ocean-basin basis. Refer to Satellite-Era Sea Surface Temperature Anomalies Versus IPCC Hindcasts/Projections Part 1 and Part 2. The models performed—I’m looking for an appropriate word—pathetically. The easiest way to portray how poorly the models matched the global satellite-based sea surface temperature data is with Figure 9from Part 1 of those posts. On a zonal-mean (latitudinal) basis, that graph compares the linear trends of the simulations of sea surface temperature (CMIP3 multi-model mean) to the observed (Reynolds OI.v2 SST anomaly data) trends. The modelers appear not to understand how or why the sea surface temperatures have warmed over the past 30 years. Refer to the discussion of that illustration under the heading of SST ANOMALY TREND COMPARISONS ON ZONAL MEAN BASES.
And when one considers that land surface temperatures are in part a product of sea surface temperatures, one wonders how climate scientists/modelers could ever attempt to simulate land surface temperatures when the oceans are modeled so poorly.
The model-simulated trends of land surface temperature are shown in Figure 1 on a zonal-mean basis. While the trends of the model-mean of land surface temperature simulations don’t match the observed rates of warming from pole to pole, they are much better than the model depiction of sea surface temperature anomaly trends. The area with the greatest divergence is the Arctic—yet we hear so much about how the models are able to simulate polar amplification.
The graphs for this post present the multi-model ensemble mean of the CMIP5 models that have been prepared for the upcoming 5th Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC). The use of the model mean was discussed at length in the post Part 2 – Do Observations and Climate Models Confirm Or Contradict The Hypothesis of Anthropogenic Global Warming?, under the heading of CLARIFICATION ON THE USE OF THE MODEL MEAN. And we’re using scenario RCP 8.5 because it has the most models and ensemble members and its hindcast/projection is typical of most of the other scenarios during this period. Refer to Preview of CMIP5/IPCC AR5 Global Surface Temperature Simulations and the HadCRUT4 Dataset. (CMIP3 data [20C3M / SRES A1B] used in the IPCC’s AR4 also appears as a reference in one graph and in Table 1.) The observations dataset is the GISS Land-Ocean Temperature Index data. All three datasets are available through the KNMI Climate Explorer. There I used the ocean masking option to acquire only land surface temperature data for the climate model and GISS datasets. The ocean masking also eliminates any differences that could result from GISS extending land surface temperature data out over the oceans in areas of sea ice. I used the GISS data because it is infilled and close to being spatially complete over the continental land masses during this period.
Figure 2 is a similar zonal means graph, but in it, the model mean of the CMIP3 models has been added. The CMIP5 models appear to have a better grasp of the Arctic warming trends, but the increases in the CMIP5 models have also caused them to diverge more from the observations at other latitudes.
REGIONAL MODEL-DATA COMPARISONS
We’ll use the same regions that were presented by the IPCC in Figure 9.12 of AR4, but we’ll present the land surface temperature anomalies and linear trends based on the monthly data for the last (almost) 30 years, January 1982 to October 2011. Again, the KNMI Climate explorer ocean mask was used on the GISS LOTI and the CMIP5 multi-model ensemble mean data, leaving only the observed and simulated, respectively, land surface temperature anomaly data. The coordinates and designators I used are clarified on page 9 of the Supplementary Materials for Chapter 9 of AR4. The IPCC based their breakdown of the continental land masses on the Giorgi and Francisco (2000) papers Evaluating uncertainties in the prediction of regional climate, and Uncertainties in regional climate change prediction: a regional analysis of ensemble simulations with the HadCM2 coupled AOGCM. Both papers are paywalled, but a map of the regions is available online and is presented as Figure 3. Note: I’ve also included Antarctic data, which was not presented by the IPCC.
Table 1 shows the regions and their observed and modeled linear trends for the period of January 1982 to October 2011. (If you’re wondering why the data ends last October, the GISS data at the KNMI Climate Explorer ended in October 2011 at the time I was downloading it, a few weeks ago.) Included are the trends of the CMIP3 (20C3M / SRES A1B) and CMIP5 (RCP 8.5) models. I’ve also included the errors presented by the models as a percentage of the observed trends. [Percent error = ((Model Trend – Obs. Trend)/Obs. Trend)*100]. Click on Table 1 for a full-sized version.
Table 1 – Click to Enlarge
If you were to examine the modeled trends and the observed trends, the models performed well in many of the regions, and there are others where they performed very poorly. And if you were to compare the trends of the newer CMIP5 models and the older CMIP3 models, there were improvements in some regions, but in others, the differences between the models and observations increased. Globally, the model trend worsened with the newer models; this was caused by the error in the trend of the Southern Hemisphere data increasing from 51% to 68%. The models aren’t simulating the Southern Hemisphere land surface temperatures too well. But referring to Figure 4, the trend of the modeled global land surface temperatures is reasonably good in a comparison to the observed trend.
A NOTE ABOUT A FEW OF THE REGIONS
There are a few regions where climate shifts are very evident in recent years. For example, let’s compare the climate model simulations and observations of land surface temperature anomalies for the Western North America (WNA) region, but we’ll end the data in 2006. See Figure 5. The modeled trend is basically a perfect match for the observations during this period. Can’t get much better than that.
Now let’s add the next five years or so of data, Figure 6. As illustrated, Western North America land surface temperatures dropped about 0.7 deg C in 2007 and fundamentally stayed there. There is nothing to indicate that temperatures there are making an effort to climb back up to where they were. That downward shift in temperature dropped the linear trend of the observed surface temperatures from 0.396 deg C/decade to 0.132 deg C/decade, or a decline of about two-thirds. That’s a chunk. Yet the simulated surface temperature of the climate models continues to climb higher.
Eastern Asia and Tibet regions are also showing recent (2009) downward shifts in temperature, but it’s a little soon to tell if the temperature anomalies will remain at those levels long enough to greatly impact the observed trends.
REGIONAL MODEL-DATA COMPARISON GRAPHS
There are 22 regions illustrated in the IPCC’s Figure 9.12. I’ve also created graphs of global and Antarctic land surface temperature anomalies. That’s a lot of graphs. Some visitors would prefer to look at the data smoothed with a 13-month running-average filter to reduce the seasonal- and weather-related variations. Minimizing the month-to-month variations also allows the temperature scale on the graph to be reduced, which allows the differences in trends to be seen easier. And there are visitors who would prefer to see the raw monthly data with all of its variability to put things in a different perspective. With the raw data, the extent of the month-to-month variations in temperature anomalies stands out. Take the Western North America data for example again, Figure 7. In its raw form, the monthly data shows a change of more than 8 deg C from November 1985 to January 1986. That’s a massive swing in a few months time, especially when we consider the warming trend projected by the models is 0.37 deg C/decade. (Keep in mind, we’re looking at anomalies, not absolute temperatures.)
Since the two presentations have their advantages, I will present the graphs in raw and smoothed forms. So not to overwhelm the post with a bunch of graphs, I’ve uploaded them and provided links to them in table form, allowing you to select which regions you wish to view and in which format. See Table 2. Use Table 1 and the map in Figure 3 as references. Happy viewing.
|TABLE 2||GRAPH TYPE|
|WNA||Western N. America||Click||Click|
|CNA||Central N. America||Click||Click|
|ENA||Eastern N. America||Click||Click|
|SSA||Southern S. America||Click||Click|
In the comparisons of the GISS land surface temperature anomalies and the multi-model mean of the CMIP5 coupled climate models on a regional basis, the models performed well at simulating the trend of the last 30 years in only certain parts of the globe. In general, the models performed well regionally in Europe and in most of Africa and most of Asia. On the other hand, in general, the models performed poorly at simulating the regional temperature trends in the Americas, Australia, Southern Africa and Southeast Asia. The models matched the warming trend of Antarctica extremely well. Refer to Figure 8.
That raises the question: why do the climate models simulate land surface temperature trends well for the last 30 years in some parts of the globe but not others?
As illustrated and discussed here and here about a year ago and more recently in my book, the CMIP3 climate models do a very poor job of simulating ocean-basin sea surface temperatures since 1982. Very poor. We showed this with time-series and zonal-mean graphs. Remember that the global oceans represent about 70% of the surface area of the globe. In this post we showed that the models poorly simulated the trends of land surface temperature anomalies in the Americas, Australia, South Africa, and Southeast Asia , which covers about another—what?—10%. That leaves only 20% of the surface area of the globe where the climate models do a respectable job of simulating the surface temperatures trends for the past 30 years. Since 80% of the surface area of the globe is modeled poorly, it might lead one to believe the good performance of the models in the other 20% is simply happenstance—an accident, a fluke, a stroke of luck.
And what does the IPCC have to say about the capabilities of models to simulate temperatures on a regional level? Keep in mind that the following quote pertains to the CMIP3 models that appeared only in Figure 2 and Table 1 of this post. The fifth paragraph of the IPCC’s AR4, in Chapter 9, Understanding and Attributing Climate Change, under the heading of “9.4.5 Summary”, reads:
An important development since the TAR has been the detection of an anthropogenic signal in surface temperature changes since 1950 over continental and sub-continental scale land areas. The ability of models to simulate many aspects of the temperature evolution at these scales (Figure 9.12) and the detection of significant anthropogenic effects on each of six continents provides stronger evidence of human influence on the global climate than was available to the TAR. Difficulties remain in attributing temperature changes at smaller than continental scales and over time scales of less than 50 years. Attribution at these scales has, with limited exceptions, not yet been established. Temperature changes associated with some modes of variability, which could be wholly or partly naturally caused, are poorly simulated by models in some regions and seasons and could be confounded with the expected temperature response to external forcings. Averaging over smaller regions reduces the natural variability less than averaging over large regions, making it more difficult to distinguish changes expected from external forcing. In addition, the small-scale details of external forcing and the response simulated by models are less credible than large-scale features. Overall, uncertainties in observed and model-simulated climate variability and change at smaller spatial scales make it difficult at present to estimate the contribution of anthropogenic forcing to temperature changes at scales smaller than continental and on time scales shorter than 50 years.
The IPCC appears a bit optimistic with the first two sentences in the above quote. Some readers might think the first two sentences are contradicted by the rest of that paragraph.
MY FIRST BOOK
As illustrated and discussed in If the IPCC was Selling Manmade Global Warming as a Product, Would the FTC Stop their deceptive Ads?, the IPCC’s climate models cannot simulate the rates at which surface temperatures warmed and cooled since 1901 on a global basis, so their failings on a regional basis as discussed in this post come as no surprise.
Additionally, the IPCC claims that only the rise in anthropogenic greenhouse gases can explain the warming over the past 30 years. Satellite-based sea surface temperature disagrees with the IPCC’s claims. Most, if not all, of the rise in global sea surface temperature is shown to be the result of a natural process called the El Niño-Southern Oscillation, or ENSO. This is discussed in detail in my first book, If the IPCC was Selling Manmade Global Warming as a Product, Would the FTC Stop their deceptive Ads?, which is available in pdf and Kindle editions. A copy of the introduction, table of contents, and closing can be found here.
The modeled and observed surface temperature data presented in this post are available through the KNMI Climate Explorer: