The temperature forecasting track record of the IPCC

Guest essay by Euan Mearns of Energy Matters

In geology we use computer models to simulate complex processes. A good example would be 4D simulation of fluid flow in oil and gas reservoirs. These reservoir models are likely every bit as complex as computer simulations of Earth’s atmosphere. An important part of the modelling process is to compare model realisations with what actually comes to pass after oil or gas production has begun. It is called history matching. At the outset, the models are always wrong but as more data is gathered they are updated and refined to the point that they have skill in hind casting what just happened and forecasting what the future holds. This informs the commercial decision making process.

The IPCC (Intergovernmental Panel on Climate Change) has now published 5 major reports, the First Assessment Report (FAR) in 1990. This provides an opportunity to examine what has been forecast with what has come to pass. Examining past reports is quite enlightening since it reveals what the IPCC has learned in the last 24 years.

I conclude that nothing has been learned other than how to obfuscate, mislead and deceive.

Figure 1 Temperature forecasts from the FAR (1990). Is this the best forecast the IPCC has ever made? It is clearly stated in the caption that each model uses the same emissions scenario. Hence the differences between Low, Best and High estimates are down to different physical assumptions such as climate sensitivity to CO2. Holding the key variable constant (CO2 emissions trajectory) allows the reader to see how different scientific judgements play out. This is the correct way to do this. All models are initiated in 1850 and by the year 2000 already display significant divergence. This is what should happen. So how does this compare to what came to pass and with subsequent IPCC practice?

I am aware that many others will have carried out this exercise before and in a much more sophisticated way than I do here. The best example I am aware of was done by Roy Spencer [1] who produced this splendid chart that also drew some criticism.

Figure 2 Comparison of multiple IPCC models with reality compiled by Roy Spencer. The fact that reality tracks along the low boundary of the models has been made many times by IPCC sceptics. The only scientists that this reality appears to have escaped are those attached to the IPCC.

My approach is much more simple and crude. I have simply cut and pasted IPCC graphics into XL charts where I compare the IPCC forecasts with the HadCRUT4 temperature reconstructions. As we shall see, the IPCC have an extraordinary lax approach to temperature datums and in each example a different adjustment has to be made to HadCRUT4 to make it comparable with the IPCC framework.

Figure 3 Comparison of the FAR (1990) temperature forecasts with HadCRUT4. HadCRUT4 data was downloaded from WoodForTrees [2] and annual averages calculated.

Figure 3 shows how the temperature forecasts from the FAR (1990) [3] compare with reality. It should be quite clear that the best model is the Low Model. I cannot easily find the parameters used to define the Low, Best and High models but the report states that a range of climate sensitivities from 1.5 to 4.5˚C are used. It should be abundantly clear that the Low model is the one that lies closest to the reality of HadCRUT4. The High model is already running about 1.2˚C too warm in 2013.

Figure 4 The TAR (2001) introduced the hockey stick. The observed temperature record is spliced onto the proxy record and the model record is spliced onto the observed record and no opportunity to examine the veracity of the models is offered. But 13 years have since past and we can see how reality compares with the models in that very short time period.

I could not find a summary of the Second Assessment Report (SAR) from 1994 and so jump to the TAR (third assessment report) from 2001 [4]. This was the year (I believe) that the hockey stick was born (Figure 4). In the imaginary world of the IPCC, Northern Hemisphere temperatures were constant from 1000 to 1900 AD with not the faintest trace of Medieval Warm Period or Little Ice Age where real people either prospered or died by the million. The actual temperature record is spliced onto the proxy record and the model world is spliced onto that to create a picture of future temperature catastrophe. So how does this compare with reality?

Figure 5 From 1850 to 2001 the IPCC background image is plotting observations (not model output) that agree with the HadCRUT4 observations. Well done IPCC! The detail of what has happened since 2001 is shown in Figure 6. To have any value or meaning all of the models should have been initiated in 1850. We would then see that the majority are running far too hot by 2001.

Figure 5 shows how HadCRUT4 compares with the model world. The fit from 1850 to 2001 is excellent.  That is because the background image is simply plotting observations in this period. I have nevertheless had to subtract 0.6˚C from HadCRUT4 to get it to match the observations while a decade earlier I had to add 0.5˚C. The 250 year x-axis scale makes it difficult to see how models initiated in 2001 now compare with 13 years of observations since. Figure 6 shows a blow up of the detail.

Figure 6 The single vertical grid line is the year 2000. The blue line is HadCRUT4 (reality) moving sideways while all of the models are moving up.

The detailed excerpt illustrates the nature of the problem in evaluating IPCC models. While real world temperatures have moved sideways since about 1997 and all the model trends are clearly going up, there is really not enough time to evaluate the models properly. To be scientifically valid the models should have been run from 1850, as before (Figure 1), but they have not. Had they been, by 2001 they would have been widely divergent (as 1990) and it would be easy to pick the winners. But they are brought together conveniently by initiating the models at around the year 2000. Scientifically this is bad practice.

Figure 7 IPCC future temperature scenarios from AR4 published in 2007. It seems that the IPCC has taken on board the need to initiate models in the past and in this case the initiation date stays at 2000 offering the same 14 years to compare models with what came to pass.

For the Fourth Assessment Report (AR4) [5] we move on to 2007 and the summary shown in Figure 7. By this stage I’m unsure what the B1 to A1F1 scenarios mean. The caption to this Figure in the reports says this:

Figure SPM.5. Solid lines are multi-model global averages of surface warming (relative to 1980–1999) for the scenarios A2, A1B and B1, shown as continuations of the 20th century simulations. Shading denotes the ±1 standard deviation range of individual model annual averages. The orange line is for the experiment where concentrations were held constant at year 2000 values. The grey bars at right indicate the best estimate (solid line within each bar) and the likely range assessed for the six SRES marker scenarios. The assessment of the best estimate and likely ranges in the grey bars includes the AOGCMs in the left part of the figure, as well as results from a hierarchy of independent models and observational constraints. {Figures 10.4 and 10.29}

Implicit in this caption is the assertion that the pre-year 2000 black line is a simulation produced by the post-2000 models (my bold). The orange line denotes constant CO2 and the fact that this is a virtual flat line shows that the IPCC at that time believed that variance in CO2 was the only process capable of producing temperature change on Earth. I don’t know if the B1 to A1F1 scenarios all use the same or different CO2 increase trajectories. What I do know for sure is that it is physically impossible for models that incorporate a range of physical input variables, initiated in the year 1900, to be closely aligned and to converge on the year 2000 as shown here. It is a physical impossibility as demonstrated by the IPCC models published in 1990 (Figure 1).

So how do the 2007 simulations stack up against reality?

Figure 7 Comparison of AR4 models with reality. Since 2000, reality is tracking along the lower bound of the models as observed by Roy Spencer and many others. If anything, reality is aligned with the zero anthropogenic forcing model shown in orange.

Last time out I had to subtract 0.6˚C to align reality with the IPCC models. Now I have to add 0.6˚C to HadCRUT4 to achieve alignment. And the luxury of tracking history from 1850 has now been curtailed to 1900. The pre-2000 simulations align pretty well with observed temperatures from 1940 even though we already know that it is impossible for the pre-2000 simulations to have been produced by a large number of different computer models programmed to do different things – how can this be? Post 2000, reality seems to be aligned best with the orange no CO2 rise /  no anthropogenic forcing model.

From 1900 to 1950 the alleged simulations do not in fact reproduce reality at all well (Figure 8). The actual temperature record rises at a steeper gradient than the model record. And reality has much greater variability due to natural processes that the IPCC by and large ignore.

Figure 8 From 1900 to 1950 the alleged AR4 simulations actually do a very poor job of simulating reality, HadCRUT4 in blue.

Figure 9 The IPCC view from AR5 (2014). The inconvenient mismatch 1900 to 1950 observed in AR4 is dealt with by simply chopping the chart to 1950. The flat blue line is essentially equivalent to the flat orange line shown in AR4.

The fifth assessment report (AR5) was published this year and the IPCC current view on future temperatures is shown in Figure 9 [6]. The inconvenient mismatch of alleged model data with reality in the period 1900 to 1950 is dealt with by chopping that time interval off the chart. A very simple simulation picture is presented. Future temperature trajectories are shown for a range of Representative Concentration Pathways (RCP). This is the completely wrong approach since the IPCC is no longer modelling climate but different human, societal and political choices, that result in different CO2 trajectories. Skepitcalscience provides these descriptions [7]:

RCP2.6 was developed by the IMAGE modeling team of the PBL Netherlands Environmental Assessment Agency. The emission pathway is representative of scenarios in the literature that lead to very low greenhouse gas concentration levels. It  is a “peak-and-decline”  scenario; its radiative forcing level first reaches a value of around 3.1 W/m2  by mid-century, and returns to 2.6 W/m2  by 2100. In order to reach such radiative forcing levels, greenhouse gas emissions (and indirectly emissions of air pollutants) are reduced substantially, over time (Van Vuuren et al. 2007a). (Characteristics quoted from van Vuuren et.al. 2011)

AND

RCP 8.5 was developed using the MESSAGE model and  the IIASA Integrated Assessment Framework by  the International  Institute  for  Applied  Systems  Analysis  (IIASA),  Austria.  This  RCP  is characterized by increasing greenhouse gas emissions over time, representative of scenarios in the literature that lead to high greenhouse gas concentration levels (Riahi et al. 2007).

This is Mickey Mouse science speak. In essence they show that 32 models programmed with a low future emissions scenario have lower temperature trajectories than 39 models programmed with high future emissions trajectories.

The models are initiated in 2005 (the better practice of using a year 2000 datum as employed in AR4 is ditched) and from 1950 to 2005 it is alleged that 42 models provide a reasonable version of reality (see below). We do not know which, if any, of the 71 post-2005 models are included in the pre-2005 group. We do know that pre-2005, each of the models should be using actual CO2 et al concentrations and since they are all closely aligned we must assume they all use similar climate sensitivities. What the reader really wants to see is how varying climate sensitivity influences different models using fixed CO2 trajectories and this is clearly not done. The modelling work shown in Figure 9 is effectively worthless. Nevertheless, let us see how it compares with reality.

Figure 10 Comparison of reality with the AR5 model scenarios.

With models initiated in 2005 we have only 8 years to compare models with reality. This time I have to subtract 0.3˚C from HadCRUT4 to get alignment with the models. Pre-2005 the models allegedly reproduce reality from 1950. Pre-1950 we are denied a view of how the models worked then. Post-2005 it is clear that reality is tracking along the lower limit of the two uncertainty envelopes that are plotted. This is an observation made by many others [e.g 1].

Concluding comments

• To achieve alignment of the HadCRUT4 reality with the IPCC models the following temperature corrections need to be applied: 1990 +0.5; 2001 -0.6; 2007 +0.6; 2014 -0.3. I cannot think of any good reason to continuously change the temperature datum other than to create a barrier to auditing the model results.
• Comparing models with reality is severely hampered by the poor practice adopted by the IPCC in data presentation. Back in 1990 it was done the correct way. That is all models were initiated in 1850 and used the same CO2 emissions trajectories. The variations in model output are consequently controlled by physical parameters like climate sensitivity and with the 164 years that have past since 1850 it is straight forward to select the models that provide the best match with reality. In 1990, it was quite clear that it was the “Low Model” that was best almost certainly pointing to a low climate sensitivity.
• There is no good scientific reason for the IPCC not adopting today the correct approach adopted in 1990 other than to obscure the fact that the sensitivity of the climate to CO2 is likely much less than 1.5˚C based on my and others’ assertion that a component of the Twentieth Century warming is natural.
• Back in 1990, the IPCC view on climate sensitivity was a range from 1.5 to 4.5˚C. In 2014 the IPCC view on climate sensitivity is a range from 1.5 to 4.5˚C. 24 years have past and billions of dollars spent and absolutely nothing has been learned! The wool has been pulled over the eyes of policy makers, governments and the public to the extent of total brain washing. Trillions of dollars have been misallocated on energy infrastructure that will ultimately lead to widespread misery among millions.
• In the UK, if a commercial research organisation were found cooking research results in order to make money with no regard for public safety they would find the authorities knocking at their door.

References

[1] Roy Spencer: 95% of Climate Models Agree: The Observations Must be Wrong

[2] Wood For Trees

[3] IPCC: First Assessment Report – FAR

[4] IPCC: Third Assessment Report – TAR

[5] IPCC: Fourth Assessment Report – AR4

[6] IPCC: Fifth Assessment Report – AR5

[7] Skepticalscience: The Beginner’s Guide to Representative Concentration Pathways