We can’t predict the climate on a local, regional, or continental scale

This is part of a series that Tony Cox and I are doing that drills down to the most important points and papers, with proper references, as a definitive resource.The models are wrong: not just “unverified”, not just “uncertain”, but proven to have failed. — Jo

Joint Post: Tony Cox and Jo Nova

Across different regions, and different time-spans over the last century, the models fail.

Koutsoyiannis and Anagnostopolous  et al show those models can’t model the recent century, and because the models fail to predict regional and smaller scale effects it’s impossible that they could predict longer and global values.^[i]

On 30 year time frames, the original observations are nothing like the models projections on a local scale. (Click to enlarge).

The models should retrospectively match the actual temperature over the past 100 years. This test of retrospectivity is called hindcasting. If a model has valid assumptions about the climatic effect of variables such as greenhouse gases, particularly CO₂, then the model should be able to match past known data.

“…all the models were “irrelevant with reality” at the 30 year climate scale…”

When tested, the global climate models failed to hindcast the climate at 55 locations around the world. The models were even worse at reproducing what happened across 70 sites spread throughout continental USA.  Their outputs show the models are unable to model the known climate.

70 sites were used around the US. Models couldn’t model this size region accurately. (Click to enlarge)

…

They compared 55 stations around the world with model outputs. Top, temperature. Bottom, precipitation. Outputs did not correlate well with observations on a point scale. (Click to enlarge).

Models were also tested across different time-spans, including the 30 year scale. In 2008 Koutsoyiannis found that while the models had some success with the monthly data all the models were “irrelevant with reality” at the 30 year climate scale.^[ii]

At a monthly timescale the models can reproduce the seasonal cold-warm or wet-dry periods for all 55 stations studied. However, they fail on a yearly scale. For example at Durban, South Africa, not a single model output showed the 1.5°C  fall in temperature from 1920 -1960 (see the top graphs), instead every one predicted a constant increase. At all stations examined, there was not a single model run that reproduced the time series of all the variables examined.

Koutsoyannis answers the critics

An earlier version of the tests published as Koutsoyiannis’s 2008 was criticized by Schmidt in the Real Climate blog.^[iii] Kousouyannis addressed all those points and expanded the study form 8 to 55 points in 2011 and got results which were even more conclusive.

Schmidt’s criticism was 4-fold; that Koutsoyiannis uses a regional comparison, few models, real temperatures not anomalies and too short a time period. Koutsoyiannis et al point out that the period from 1990-2008 was the period in which IPCC modeling had occurred; the IPCC had argued that regional effects from global warming would occur (so their regional predictions ought to be tested); model ensembles were used by Koutsoyiannis; and since the full 100 year temperature and rainfall data was used in intra-annual and 30 year periods by Koutsoyiannis, anomalies were irrelevant.

The global warming models fail at almost every test; they could not predict the future in 1990; they cannot predict the present and they could not replicate or match the past.