Can Natural Variability Save Climate Models?

Climate scientists Patrick Michaels and Chip Knappenberger have a blockbuster post on the Cato Institute blog. They claim to have uncovered a “clear example of IPCC ideology trumping fact.”

As is widely known, global mean surface temperature (GMST) has not increased over the past 13-plus years, contributing to a growing divergence between global warming predictions and observations.

Figure source: John Christy and Robert McKnider

While acknowledging there are “differences” between modeled and observed temperatures for “periods as short as 10 to 15 years,” the IPCC’s 2013 Fifth Assessment Report (AR5) claims models and observations “agree” over the 62-year period from 1951 to 2012 (Summary for Policymakers, p. 15). Moreover, the IPCC has “very high confidence” the models’ long-term GMST trends are “consistent with observations” (Chapter 9, p. 769). The chart below illustrates “model response error” during two 15-year periods and the longer 62-year period.

In each panel, red hatching shows observed temperatures as compiled by the UK Hadley Center; the gray bars show GMST trend distribution from 114 climate models. IPCC AR5, Chapter 9, Box 9.2.

Panel (c) appears to depict a close match between simulations and observations. But when Michaels and Knappenberger unpack the information incorporated in the graphic, they find that 90 out of 108 models hind-cast more warming than actually occurred.

Okay, that makes IPCC’s “very high confidence” seem misplaced, but why is Michaels and Knappenberger’s column a blockbuster? Because of what they show next. 

The standard rejoinder to skeptics is: ‘Just wait, once the pause is over, global warming will resume, and then your criticism that IPCC models are tuned too hot will collapse.’

Since the IPCC claims models and observations align over the long term, Michaels and Knappenberger investigate how well models would match observations over the 80-year period from 1951 to 2030 under three scenarios of how global surface temperature might behave between now and 2030.

In Scenario 1, the “pause” continues and the warming rate in 2014-2030 remains what it was in 2001-2013 (essentially zero). In Scenario 2, warming resumes at the long-term 1951-2012 rate (0.107ºC/decade). In Scenario 3, warming resumes at the 1977-1998 rate (0.17ºC/decade) — the rate from the start of recent global warming until the pause.

Here’s the stunner. Even in the warmest scenario, fewer than 5% of model simulations of the long-term, 80-year trend agree with observations by 2020 and fewer than 2.5% agree by 2030.

“In other words,” Michaels and Knappenberger comment, “statisticians would say the models disagree with the observations with ‘very high confidence'” — exactly the reverse of the IPCC’s assessment.

Here’s how the two Cato researchers explain the chart above:

Under Scenario 1, the observed trend (beginning  in 1951) would fall below the 5th percentile of the distribution of model simulations in the year 2018 and beneath the 2.5th percentile in 2023. Under Scenario 2, the years to reach the 5th and 2.5th percentiles are 2019 and 2026, respectively. And under Scenario 3, the observed trend would fall beneath the 5th percentile of model simulated trends in the year 2020 and beneath the 2.5th percentile in 2030.

The credibility of today’s state-of-the-art climate models may not survive the next 5-15 years, Michaels and Knappenberger suggest:

Within the next 5 to 15 years, the long-term observed trend (beginning in 1951) will more than likely fall so far below model simulations as to be statistically recognized as not belonging to the modeled population of outcomes. This disagreement between observed trends and model trends would be complete.  

I couldn’t help wondering, though, how well the models would perform in a more aggressive warming scenario.

In panel (b) of the IPCC chart above, the observed warming rate for 1984-1998 is 0.26ºC/decade — considerably faster than the ensemble model rate for that period (0.16ºC/decade). According to IPCC AR5, the mismatch between models and observations during both 1984-1998 and 1998-2012 may be due to “internal decadal climate variability, which sometimes enhances and sometimes counteracts the long-term externally forced trend” (Chapter 9, p. 769).*

So I asked Mr. Knappenberger to test the models’ agreement with long-term observations using a new ‘third’ scenario in which internal variability once again “enhances” the “externally forced trend” and global warming resumes at the 1984-1998 rate of 0.265ºC/decade. He kindly obliged.

This graphic shows the three scenarios along with the ensemble model mean during 1951-2030. The model mean continues to overshoot all scenarios including the new scenario 3 in which warming resumes at 0.265ºC/decade.

The next graphic shows the percentage of model simulations that match each of the three scenarios.

While the under the new scenario 3, the models never reach outright statistical failure (<2.5%), they nevertheless perform very poorly, matching observations less than 5% of the time.  As Mr. Knappenberger explained to me:

Basically, the model average never anticipated such a long “pause” and as a result, the observations have a devil of a time catching back up to the model projections.

– – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – –

* Funny, when the IPCC’s 2001 Third Assessment Report (TAR) was proclaiming the 1990s the warmest decade and 1998 the warmest year of the millennium, I don’t recall the IPCC giving any credit to internal decadal climate variability. If I’m not mistaken, the IPCC did not begin ascribing any portion of recent warming to natural variability until they needed natural variability to explain the pause.