CMIP6 and AR6, a preview – Watts Up With That?

By Andy May

The new IPCC report with the abbreviation “AR6” is to appear between April 2021 (physical-scientific basis) and June 2022 (synthesis report). I bought some very strong hip waders to prepare for the events. For those who don’t already know, sturdy hip boots are required when wading in the sewer. I also looked at the output of the CMIP6 model published so far in the KNMI Climate Explorer. I thought I was going to share some of what I found.

Figure 1. The 13-person CMIP6 model ensemble from the KNMI Climate Explorer. The global temperature anomaly from the 1981-2010 average is plotted for all model runs. All 13 model runs are drawn from the ensemble. Note that two runs were recorded for three models.

There are currently two model ensembles on the site, one contains 68 model runs and the other contains 13. Figure 1 shows all 13 runs for the smaller ensemble. The Canadian Center for Climate Modeling ensemble (CanESM5 P1 and P2) consists of two runs. There are also two for NCAR (CESM2-1 and CESM2-2) and two for the Jamstec model (MIROC). Figure 1 shows a total of ten models. The models use historical forcing before 2014 and projected forcing after 2014.

All curves are global mean temperature anomalies from the 1981-2010 mean temperature. Note that the dispersion of the results in the 19th century is more than one degree C. The current spread of the results isn’t much narrower, and the spread around 2100 is over two degrees. All of these model runs use the emissions scenario ssp245 which, as far as I can tell, is the CMIP6 version of RCP 4.5. So it is the middle scenario.

Two of the runs are pretty wild, the light blue model which is very low in 1960 and very high in 2100 is the Met Office Hadley Center model run UKESM1.0-LL (2018). The Canadian model run “CanESM5 P 1” follows the same path, but is hidden by UKESM1.0. The other runs are combined in a one-degree scrum.

In Figure 2 we show three of the model runs. Both Canadian model runs are shown in blue and orange, and one of the NCAR models is shown in gray. The black is the 13-model run “Ensemble Mean” as calculated by the KNMI Climate Explorer. I chose an ensemble on the website. I didn’t make this ensemble myself. The ensemble mean was also provided by KNMI Climate Explorer, I did not calculate it.

Figure 2. Three individual model runs are compared with the mean of the 13-person ensemble. The vertical axis is the anomaly from the 1981-2010 average in degrees C.

Historical forces are used before 2014 and projected values ​​afterwards. The blue and orange curves are from two runs of a single Canadian model. The two runs differ by over 0.2 ° C in 2010 and 2011, and over 0.5 ° C in some months. There are several periods in which the model runs are clearly out of phase for several years. Examples are 2001-2003 and 2014 to 2017. The period from 2015 to 2019 is a mess.

In Figure 3, the ensemble mean shown in Figure 2 is compared with three weather reanalysis datasets that were also downloaded from the KNMI Climate Explorer. The weather reanalysis records are shown in the faint lines.

Figure 3. Three weather reanalysis data sets are compared with the mean value of the model ensemble from Figure 2.

Re-weather analysis is performed after the weather data has been recorded, but using a weather computer model. The reanalysis has many thousands of observations that it can take into account, so the output is generally reasonably reliable, at least in my opinion. Note that all three weather reanalysis datasets, NOAA, NCEP, and ERA5 (European Center for Medium-Range Weather Forecasts) are in phase and tracking each other. Over a period of up to three years, the mean of the ensemble model is hopelessly out of phase with re-analysis. This occurs when the model had historical data (before 2014) and after.


I am unimpressed by the CMIP6 models. The total warming since 1900 is less than one degree, but the variation in the model results in Figure 1 is never less than one degree. It’s often more than that, especially in the 1960s. The models obviously do not reproduce the natural climate cycles or vibrations such as AMO, PDO and ENSO. As can be seen in Figure 2, they are often completely out of phase for years, even if only two runs of the same model. I used the Canadian model as an example, but the two NCAR model runs (CESM2) are no better. In fact, they are worse in 2010-2011 and 2015-2019 as you can see in Figure 4.

Figure 4. Comparison of the two runs of the NCAR CESM2 model with the ensemble mean and a Canadian model run.

The AR5 report was an expensive iteration of AR4. Both gave up all hope of finding solid evidence of human impact on the climate and tried to use climate models to show that humans somehow control the climate with our greenhouse gas emissions. They tried to use solid evidence in SAR, the second report, and TAR, the third report, but it was shown to be wrong on both attempts. You can read about their SAR humiliation and TAR hockey stick humiliation here.

Forty years of work and billions of dollars spent and still no evidence that humans control the climate. Models are all they have, and these actions do not instill confidence in the models. As we discussed in our previous post, averaging models don’t improve results. Ensembles are no better than their member models and can be worse. When the individual runs are out of phase, you destroy data and the natural cycles by averaging them. See the Wyatt, Curry, and Javier evidence here and here. If you want to convince this observer, you have to do a lot better. And a word to taxpayers, why are we paying these huge sums of money just to do the same thing over and over again? Conclusion: will AR6 differ from AR4 or AR5? Are these documents worth paper and ink?

Like this:

To like Loading…

Comments are closed.