Site icon Watts Up With That?

Gavin’s Twitter Trick

David Middleton

Last week, Larry Kummer posted a very thoughtful article here on WUWT:

A climate science milestone: a successful 10-year forecast!

At first glance, this did look like “a successful 10-year forecast:

Figure 1. A successful 10-year forecast?

The observations track closer to the model ensemble mean (P50) than most other models and the 2016 El Niño spikes at least a little bit above P50.  Noticing that this was CMIP3 model, Larry and others asked if CMIP5 (the current Climate Model Intercomparison Project) yielded the same results, to which Dr. Schmidt replied:

Figure 2. A failed 10-year forecast.

The CMIP5 model looks a lot like the CMIP3 model… But the observations bounce between the bottom of the 95% band (P97.5) and just below P50… Then spike to P50 during the 2016 El Niño.  When asked about the “estimate of effect of misspecified forcings, Dr. Schmidt replied:

Basically, the model would look better if it was adjusted to match what actually happened.

The only major difference between the CMIP3 and CMIP5 model outputs was the lower boundary of the 95% band (P97.5), which lowered the mean (P50).

Figure 3. Improving accuracy by increasing imprecision.

CMIP-5 yielded a range of 0.4° to 1.0° C in 2016, with a P50 of about 0.7° C. CMIP-3 yielded a range of 0.2° to 1.0° C in 2016, with a P50 of about 0.6° C.

They essentially went from 0.7 +/-0.3 to 0.6 +/-04.

Progress shouldn’t consist of expanding the uncertainty… unless they are admitting that the uncertainty of the models has increased.

Larry then asked Dr. Schmidt about this:

Dr. Schmidt’s answer:

“Not sure”?  That instills confidence.  He seems to be saying that the CMIP5 model (the one that failed worse than CMIP3) may have had “more coherent forcing across the ensemble, more realistic ENSO variability, greater # of simulations.”

I’m not a Twitterer, but I do have a rarely used Twitter account and I just couldn’t resist joining in on the conversation.  Within the thread, there was a discussion of Hansen et al., 1988.  And Dr. Schmidt seemed to be defending that model as being successful because the  2016 El Niño spiked the observations to “business-as-usual.”

Figure 4.  Hansen et al., 1988 is still an epic fail.  The monster El Niño of 2016 is not “business-as-usual.”

I asked the following question:

No answer.  Dr. Schmidt is a very busy person and probably doesn’t have much time for Twitter and blogging.  So, I don’t really expect an answer.

In his post, Larry Kummer also mentioned a model by Zeke Hausfather posted on Carbon Brief…

Figure 5.  Another failed climate model.  The 2016 El Niño is not P50 weather.

El Niño events like 1998 and 2016 are not high probability events.  On the HadCRUT4 plot below, I have labeled several probability bands:

Standard Deviation Probability Band % Samples w/ Higher Values
+2σ P02.5 2.5%
+1σ P32 32.0%
Mean P50 50.0%
-1σ P68 68.0%
-2σ P97.5 97.5%

Yes… I am assuming that HadCRTU4 is reasonably accurate and not totally a product of the Adjustocene.

I removed the linear trend, calculated a mean (P50) and two standard deviations (1σ & 2σ).  Then I added the linear trend back in to get the following:

Figure 6.  HadCRUT4 (Wood for Trees) with probability bands.

The 1998 El Niño spiked to P02.5.  The 2016 El Niño spiked pretty close to P0.01.  A strong El Niño should spike from P50 toward P02.5.

All of the models fail in this regard.  Even the Mears-ized RSS satellite data exhibit the same relationship to the CMIP5 models as the surface data do.

–

The RSS comparison was initialized to 1979-1984.  The 1998 El Niño spiked above P02.5.  The 2016 El Niño only spiked to just above P50… Just like the Schmidt and Hausfather models.  The Schmidt model was initialized in 2000.

This flurry of claims that the models don’t “run hot” because the 2016  El Niño pushed the observations toward P50 is being driven by an inconvenient paper that was recently published in Nature Geoscience (discussed here, here and here).

21 September 2017 0:27

Factcheck: Climate models have not ‘exaggerated’ global warming

ZEKE HAUSFATHER

A new study published in the Nature Geosciences journal this week by largely UK-based climate scientists has led to claims in the media that climate models are “wrong” and have significantly overestimated the observed warming of the planet.

Here Carbon Brief shows why such claims are a misrepresentation of the paper’s main results.

[…]

Carbon Brief

All (95%) of the models run hot, including “Gavin’s Twitter Trick”.  From Hansen et al., 1988 to CMIP5 in 2017, the 2016 El Niño spikes toward the model ensemble mean (P50)… Despite the fact that it was an extremely low probability weather event (<2.5%).  This happens irrespective of then the models are initialized.  Whether the models were initialized in 1979 or 2000, the observed temperatures all bounce from the bottom of the 95% band (P97.5) toward the ensemble mean (P50) from about 2000-2014 and then spike to P50 or slightly above during the 2016 El Niño.

Exit mobile version