by Roger Andrews
Some time ago I posted a graph showing how the IPCC’s 21st century temperature projections for the “worst case” RCP85 emissions scenario could be replicated almost exactly using the IPCC’s CO2 radiative forcing estimates for the scenario, a climate sensitivity of 2.2C and nothing else. (Note that I don’t claim this as an original discovery. Others, I think Clive Best is one, have presented similar graphs in the past):
Author note, early am UK time August 28th : Dave Rutledge has pointed out that Figure 1 gives a climate sensitivity of around 3.0C when calculated using the present-day CO2 concentration of ~400 ppm and a 2100 concentration of 936 ppm, the IPCC’s official estimate for RCP85. This is higher than my best-fit estimate of 2.2C because I used 289 ppm CO2 in 1750 as the baseline and the IPCC’s official 2100 estimate of 1,233 ppm CO2 equivalent, which includes other greenhouse gases such as methane and nitrous oxide expressed as CO2.
Although there’s actually nothing very surprising about this result because CO2 is effectively the only input the IPCC’s climate models receive in the 21st century. Only about half of the ~4 degrees C of warming they project under the RCP85 scenario, however, is contributed by the direct radiative impacts of CO2. The other half comes from positive feedbacks generated by the interaction of CO2 with other model variables, so if the models are to predict the right amount of warming it’s important that they correctly quantify these feedbacks.
And do they?
Well, we really haven’t checked. We have enough model output and observational data to say whether they do or not, but nobody seems to have looked at them. So here we will rectify this omission.
There are three basic feedback mechanisms that operate, at least in theory, to enhance the warming impacts of CO2 (IPCC AR4 Section 8.6 discusses them in more detail):
Water vapor, itself a greenhouse gas, will cause a strong positive temperature feedback if temperatures increase and relative humidity stays the same and will cause an appreciable positive feedback even if it decreases slightly.
Clouds: Clouds block sunlight, causing negative feedbacks, and also absorb outgoing longwave radiation, causing positive feedbacks. Whether the net result is positive or negative depends on a number of poorly-understood factors.
Ice and snow: Less ice and snow causes positive feedbacks by lowering the Earth’s albedo and by allowing increased absorption of solar radiation.
Output data for three model variables that simulate humidity, clouds and ice/snow, either directly or indirectly, are available from the CMIP5 (Coupled Model Intercomparison Project Phase 5) model results posted on KNMI Climate Explorer. The three variables are:
Near-surface relative humidity
Precipitation (which since precipitation comes from clouds should be a proxy for clouds – there is no model variable that quantifies clouds per se)
Sea ice area (no snow cover model variable is available)
And the acid test of whether these model variables provide meaningful results is of course whether they adequately hindcast observations. Let’s see how well they do:
Figure 2 compares the CMIP5 near-surface relative humidity multi-model means with observed relative humidity at 1000mb (data from ESRL). Observations show short-term fluctuations of up to a percent and an overall decrease in relative humidity since 1948. The models show a flat line with a few minor wiggles:
Figure 3 compares the CMIP5 global sea ice area means with global (Arctic + Antarctic) sea ice extent data since 1978 obtained from NASA. The model hindcasts diverge significantly from observations in the mid-1990s and after 2007. (The model sea ice area fractions have been converted into ice extent in sq km so they can be compared directly with observations. The conversion will not be exact but should be close enough for comparison purposes.)
Figure 4 compares observed precipitation over land (data from the CRU TS3.21 data set at KNMI) with CMIP5 model precipitation hindcasts over land since 1900. The models show a decrease in rainfall between 1962 and 1992 that isn’t reflected in the observations and also fail to hindcast the observed increase in precipitation after 1940:
It’s possible that some of these mismatches may reflect observational inaccuracies, but as shown in Figure 5 we run into further difficulties when we compare model output for the three variables with CO2 radiative forcings between 1900 and 2100 (estimated from observed CO2 concentrations before 2010 and from the RCP85 CO2-equivalent concentrations after 2010 using w/sq m = 5.35 ln(CO2(2)/CO2(1)). Since there is no way of reconciling the units in these cases the scales are adjusted for an “eyeball” best fit:
Relative humidity shows an overall negative correlation with CO2 forcings in the 20th century but a strong positive correlation in the 21st, which seems implausible (note that the scale is inverted so that CO2 and relative humidity move in the same sense in the 21st century). Model precipitation tracks CO2 forcings after 2000 but not before, which also does not inspire confidence in the model precipitation predictions. Sea ice area shows a consistent inverse relationship with CO2 between 1900 and 2100 (note the inverted scale again) but the model hindcasts can’t be checked before 1978 because there are no global sea ice cover data to compare them against.
What do we conclude from these results? Well, they’re not as bad as they could be – at least the models give absolute values for relative humidity, sea ice cover and precipitation that are close to observed values. But do they hindcast observed trends well enough to allow us to place faith in the projections? Look at it this way. If these were temperature hindcasts, would we accept them as proof that the IPCC’s warming projections for the 21st century were realistic? I don’t think we would.
So here we have another problem with the IPCC’s climate models. Feedbacks roughly double predicted warming in the 21st century, but the model variables that most closely define the amplitude and sense of these feedbacks don’t adequately simulate what is happening in the atmosphere.
But if they don’t simulate what is happening in the atmosphere, what do they simulate? That isn’t hard to answer. They simulate CO2. Figure 6 compares 21st century CMIP5 output for ten RCP85 model output variables with 21st century CO2 radiative forcing inputs, and every one of them increases or decreases in lockstep or near-lockstep with CO2 – even soil moisture content. (Note that scales are again sometimes inverted and that the sea level pressure data are from the CMIP3 a1b emissions scenario used in the IPCC AR4 – the CMIP5 models don’t give usable sea level pressure values.)
Results like these make it difficult to escape the conclusion that the IPCC’s climate models are basically nothing more than CO2 recyclers. We give them CO2 in. They give us CO2 out.