Models That Predict the Pause

There's a very interesting paper out today in Nature Climate Change:

"Well-estimated global surface warming in climate projections selected for ENSO phase," James S. Risbey, Stephan Lewandowsky, Clothilde Langlais, Didier P. Monselesan, Terence J. O’Kane & Naomi Oreskes, Nature Climate Change (2014).

What they found is that those climate models that, by chance, reproduce ENSOs since 1950, do show a slowdown in surface warming for the 1998-2012 period.

Climate models make projections, not forecasts. They calculate climate over long time periods -- a few decades at least -- due to forcing variations, but aren't good at projecting internal variability like ENSOs and the PDO which take up or release heat (and which average out to zero over many decades). Slowdowns ("pauses") occur in the 15-year trends in the models, but forecasting one from, say, 1998 requires setting up a model with its ocean in that particular ENSO state.

Here, for example, is a histogram of all the model projections for two 15-year periods. Most models overpredicted the last 15 years, but overunderpredicted the previous 15 years (which I've called the anti-pause):

This group looked at all CMIP5 models, and culled all except for models whose internal variability was, by chance, close to the Nino3.4 surface temperature:

.To select this subset of models for any 15-year period, we calculate the 15-year trend in Niño3.4 index24 in observations and in CMIP5models and select only thosemodels with aNiño3.4 trend within a tolerance window of 0.01K/yr of the observed Niño3.4 trend. This approach ensures that we select only models with a phasing of ENSO regime and ocean heat uptake largely in line with observations. In this case we select the subset of models in phase with observations from a reduced set of 18 CMIP5 models where Niño3.4 data were available25 and for the period since 1950 when Niño3.4 indices are more reliable in observations.

Then here is their result:

So the in-phase models projections are very close to the bottom of the all-model spread. It's not perfect, as they note:

This method of phase aligning to select appropriate model trend estimates will not be perfect as the models contain errors in the forcing histories27 and errors in the simulation of ENSO (refs 25,28) and other processes. Further, ENSO is not the only process generating natural variability on these timescales and so the method used here can be only approximate. Nonetheless, the phaseselection method provides a fairer and more appropriate basis to compare model projection trends over decadal-scale periods than use of the entire multi-model envelope. When the phase of naturalvariability is taken into account, the model 15-year warming trends in CMIP5 projections well estimate the observed trends for all 15-year periods over the past half-century.