Not exact matches
No, that is not correct, both papers seek to determine whether the observational data are consistent with the
models, however Douglass et al use a statistical test that actually answers a different question, namely «is there a statistically significant difference between the mean
trend of the
ensemble and the observed
trend».
This is mainly a statistical treatment of how to test the reliability of
trends in short data - sets but Ben also has graphs of mid-troposphere from an
ensemble of results from one of the AR
models.
A comparison of observed sea ice decline with the
model ensemble spread can tell us only how likely an observed
trend is relative to that
ensemble.
The «
models used» (otherwise known as the CMIP5
ensemble) were * not * tuned for consistency for the period of interest (the 1950 - 2010
trend is what was highlighted in the IPCC reports, about 0.8 ºC warming) and the evidence is obvious from the fact that the
trends in the individual
model simulations over this period go from 0.35 to 1.29 ºC!
On my web site I have plotted 30 - year
trends for the three major temperature data series and for a 23 -
model ensemble.
Connolley and Bracegirdle (2007) show that expected
trends in a much larger sample of
models are very varied (though the
ensemble mean warms at about the rate seen in the Steig et al paper).
Also, about 2/3 of the individual
ensemble - members (46 out of 68) from all the
model runs have linear
trends that indicate at least a nominal weakening — this is significantly different from what one would be expected from a Binomial distribution with a 50 % probability.
1) Regarding the 1970s shift, Ray mentions that: «It's not evident why the smooth
trend in 20th century climate forcing should give rise to such an abrupt shift, and indeed the individual members of the
model ensemble do not show a clearly analogous shift.»
It's not evident why the smooth
trend in 20th century climate forcing should give rise to such an abrupt shift, and indeed the individual members of the
model ensemble do not show a clearly analogous shift.
In any specific
model, the range of short term
trends in the
ensemble is quite closely related to their simulation of ENSO - like behaviour.
We can derive the underlying
trend related to external forcings from the GCMs — for each
model, the underlying
trend can be derived from the
ensemble mean (averaging over the different phases of ENSO in each simulation), and looking at the spread in the
ensemble mean
trend across
models gives information about the uncertainties in the
model response (the «structural» uncertainty) and also about the forcing uncertainty — since
models will (in practice) have slightly different realisations of the (uncertain) net forcing (principally related to aerosols).
(The red line shows the histogram of the rank of the observed
trend at each grid point, within the CMIP5
ensemble spread - ideally, it would be flat, and the slope up to the left means that there are relatively more obs in the low end of the
model range than at the top end.)
Indeed,
model ensembles demonstrate that the
trends they yield for anthropogenic influences are quite independent of initial conditions, and the level of independence grows rather than diminishes as the timescale is lengthened from one or two decades to a longer timescale within a centennial interval.
[16] Lorenz P, Jacob D (2010) Validation of temperature
trends in the
ENSEMBLES regional climate
model runs driven by ERA40.
Van Haren et al (2012) also nicely illustrate the dependence of regional skill on lateral boundary conditions: simulations of (historic) precipitation
trends for Europe failed to match the observed
trends when lateral boundary conditions were provided from an
ensemble of CMIP3 global climate
model simulations, while a much better correspondence with observations was obtained when reanalyses were used as boundary condition.
Over a twenty year period, you would be on stronger ground in arguing that a negative
trend would be outside the 95 % confidence limits of the expected
trend (the one
model run in the above
ensemble suggests that would only happen ~ 2 % of the time).
In this way, we can obtain the expected range of projected climate
trends using the interannual statistics of the observed NAO record in combination with the
model's radiatively - forced response (given by the
ensemble - mean of the 40 simulations).
Then we add / subtract this scaled interannual regression map to / from the anthropogenically - forced component of the
trend over the next 30 years, the latter estimated from the
ensemble - mean of the CESM - LE (Fig. 8) or the
ensemble - mean of the 38 CMIP5
models (Fig. 9).
This indicates that internal variability will dominate over the forced response for NAO
trends over the next 30 years, regardless of whether the forced response is estimated from the
ensemble - mean of the CESM - LE or the CMIP5
models.
If you wish to test whether the
model ensemble trends are significantly different from reality.
The very high significance levels of
model - observation discrepancies in LT and MT
trends that were obtained in some studies (e.g., Douglass et al., 2008; McKitrick et al., 2010) thus arose to a substantial degree from using the standard error of the
model ensemble mean as a measure of uncertainty, instead of the standard deviation or some other appropriate measure of
ensemble spread.
Nevertheless, almost all
model ensemble members show a warming
trend in both LT and MT larger than observational estimates (McKitrick et al., 2010; Po - Chedley and Fu, 2012; Santer et al., 2012).
The mean of the
model ensemble is what we think we would get if we had thousands of replicate Earths and averaged their
trends over the same period.
The
model ensemble trend in the period 1910 to 1945 is still only 1/2 of that.
And that this is reflected in individual
model runs but as the timing of events such as El Nino / La Nina, volcanic eruptions etc. is unpredictable when projections are made based on
ensemble runs then they will tend to average out and the projection will show a fairly steady
trend.
So, the proper comparison is not the observed
trend with the mean, it is with the range of trajectories produced by the
model ensemble — that prediction interval is what the scatter of squiggles in fig. 2 shows.
However, I also know that the
ensemble trend is a very murky number, coming as it does from a range of
models with different parametization factors etc..
Using data from the KNMI data explorer website the
model ensemble mean of the CMIP3 + runs for the 20th century, the
trend is 0.05 C / decade.
The
model ensemble trend (for CMIP3 anyway) is 1/3 the warming rate of the observations.
The discrepancy can be seen with box - and - whiskers of the
ensemble, but it pervades all
models... (Re Fig. 5) In addition, even IPCC's seemingly broad concession somewhat understates the problem, as all (not «most») CMIP5 RCP8.5 runs and
models run too hot, as shown in the following boxplot: IPCC's entire discussion of 15 - year
trends is completely worthless....
If there are such departures (in the sign of
trends) at the global scale in PCC 2007 Figure SPM.4, how can climate
model ensembles capture the bigger local falling
trends?
an analysis of the full suite of CMIP5 historical simulations (augmented for the period 2006 - 2012 by RCP4.5 simulations, Section 9.3.2) reveals that 111 out of 114 realisations show a GMST
trend over 1998 - 2012 that is higher than the entire HadCRUT4
trend ensemble... During the 15 - year period beginning in 1998, the
ensemble of HadCRUT4 GMST
trends lies below almost all
model - simulated
trends whereas during the 15 - year period ending in 1998, it lies above 93 out of 114
modelled trends.
The very high significance levels of
model — observation discrepancies in LT and MT
trends that were obtained in some studies (e.g., Douglass et al., 2008; McKitrick et al., 2010) thus arose to a substantial degree from using the standard error of the
model ensemble mean as a measure of uncertainty, instead of the
ensemble standard deviation or some other appropriate measure for uncertainty arising from internal climate variability... Nevertheless, almost all
model ensemble members show a warming
trend in both LT and MT larger than observational estimates (McKitrick et al., 2010; Po - Chedley and Fu, 2012; Santer et al., 2013).
Now, if a warming
trend of some kind resumes before AR6, the IPCC and the climate science community will consider themselves off the hook, even if the warming
trend predicted by AR4
models, shifted to the right in time, remains below the lower boundary of the AR4
model ensemble.
Due to internal climate variability, in any given 15 - year period the observed GMST
trend sometimes lies near one end of a
model ensemble an effect that is pronounced in Box 9.2, Figure 1a, b since GMST was influenced by a very strong El Niño event in 1998
Firstly, that the statistical distribution of the observed 15 - year global temperature
trends since 1880 isn't distinguishable from the distribution of 15 - year global temperature
trends derived from an
ensemble of
model simulations, and, secondly, whether simulated global temperature
trends over 15 years since 1950 lie in the same tail of the statistical distribution as the observed 15 - year temperature
trends or whether the simulated and observed
trends lie in opposite tails of the distribution largely depends on whether the simulated and observed ENSO variablity over the 15 - year periods are in phase or out of phase by chance.
Between 801 and 1800 ce, the surface cooling
trend is qualitatively consistent with an independent synthesis of terrestrial temperature reconstructions, and with a sea surface temperature composite derived from an
ensemble of climate
model simulations using best estimates of past external radiative forcings.
Arguing against the
model vs real world comparison «Here Judith is (I think) referring to the mismatch between the
ensemble mean (red) and the observations (black) in that period... However, the observations are well within the spread of the
models and so could easily be within the range of the forced
trend + simulated internal variability.»
Just for one prominent example, when NOAA and / or NASA cite 2014 as being the hottest year on record, in a context of stating their official positions concerning climate change — as were 1998, 2005, and 2010 similarly cited — then for purposes of verifying the AR5
model ensemble, what they are really saying is that the
trend of peak hottest years is what matters most to them as climate scientists, not the central
trend of observed temperatures.
The
trend in peak hottest years starting in 1998 and continuing on through 2005, 2010, and now 2014 is roughly 0.1 C per decade, as is illustrated in the graphic shown below, which is an adaption of the Ed Hawkins graphic referenced by David Apell several weeks ago in a comment he posted in response to the «Spinning the «warmest year»» article... As shown in the above graphic, if a
trend of peak hottest years starts in 1998 and is then extrapolated at a rate of +0.1 C per decade out to the year 2035, the extrapolated
trend just skirts the lower boundary of the
model ensemble range interval described by IPPC AR5 RCP (all 5 - 95 % range).
But if researchers run the same
model, or an
ensemble of
models, multiple times and the results diverge from each other or from the observed
trends, he cautions, «planners should handle them with kid gloves.
The temperature
trends during the past decades as observed and in the (
ensemble mean)
model response (Fig. 4) are roughly consistent with each other, which indicates that much of the land warming is a response to the warming of the oceans.
Internal variability is small enough such that an MMH - type mean
trend plus internal variability
ensemble model fails.
C / decade and the simulated
ensemble mean over the
models, calculated from the grid boxes of the
models where observations exist (which is flawed in my opinion, since excluding of mostly the high latitudes from the
model data may emphasize a warm bias in lower latitudes in the
models making them appear warmer than they are, but a possible cold bias of the global observations data set is not excluded in this way) had a
trend of 0.3 deg.
Thick «X» marks indicate the observational linear
trends, while thin marks represent the
trends in historical
ensembles from the five CMIP5
models having more than 10 members.
Same as Fig. 3 but for rank of observation for SAT
trend (1986 — 1990) among the climate
model ensembles.
In this study, we primarily investigate the reliability of the climatology (long - term mean of
model simulation) of large - scale features of climate
model ensembles, but we also consider the
trend for surface air temperature where transient simulations are available (that is, for the coupled ocean — atmosphere
models).
They estimate that a negative
trend over a 20 year period would be outside the 95 % CI predictions of current IPCC
model ensembles.
The SAT
trend can be calculated only for the
model ensembles with historical simulations.
The
model outputs are generally presented as an average of an
ensemble of individual runs (and even
ensembles of individual runs from multiple
models), in order to remove this variability from the overall picture, because among grownups it is understood that 1) the long term
trends are what we're interested and 2) the coarseness of our measurements of initial conditions combined with a finite
modeled grid size means that
models can not predict precisely when and how temps will vary around a
trend in the real world (they can, however, by being run many times, give us a good idea of the * magnitude * of that variance, including how many years of flat or declining temperatures we might expect to see pop up from time to time).