I took a look at the new model at climateprediction.net, the one you said includes cloud formation. What struck me about the following press release from climateprediction.net is that they threw out 18% of their model runs because they didn't like the results - either the drift was too high or it showed a cooling, not a warming. Like Michael Crichton says (he's an MD by training) you could not get a new drug approved this way. It is unclear to me what, if anything, they are trying to do to link their model results to empirical observation.
________________________
Method used to Calculate Histogram and Maps for G8 Press Release
The data for these figures was assembled from a subset of the climateprediction.net ensemble of general circulation models. The ensemble is based upon the Hadley Centre AM3 atmospheric model coupled to a thermodynamic slab ocean, while the physical parameters representing processes in the model are perturbed within the current range of uncertainty deemed appropriate by experts in the respective fields.
The experiment consists of three stages:
Stage 1 - A calibration stage with fixed ocean temperatures to determine what heat flux is necessary at the ocean surface to maintain a stable atmosphere with pre-industrial greenhouse forcing.
Stage 2 - A control stage to examine whether the perturbed atmosphere can remain stable when coupled to a thermodynamic ocean and pre-industrial greenhouse gases
Stage 3 - A double carbon dioxide run to determine the model response to a known greenhouse gas forcing
The published figures show the equilibrium response in surface temperature to current greenhouse forcing. The current forcing (the radiative power per unit area applied to the climate system, compared to pre-industrial times) was taken as the most probable total forcing as listed in the IPCC third assessment report as 2.25 Wm-2 global mean.
The equilibrium response to a double carbon dioxide forcing was inferred by assuming the temperature would follow a decaying exponential function, and then interpolating the trend obtained from the fifteen year Stage 3 simulation. This temperature response was then scaled by the ratio of current greenhouse forcing (compared to pre-industrial) to the double carbon dioxide forcing in the experiment (2.25/3.74).
Some runs were discounted from the analysis:
Those runs with large instabilities in the control simulation were ignored (drift greater than 2K)
Some runs with apparently negative temperature response were ignored because the cooling is due to a feedback between the tropical slab ocean and the overlying cloud cover. This feedback would not exist in a model with a dynamic ocean, or indeed in the real world where heat transport from elsewhere in the ocean would compensate for the effect.
Thus 10482 of an original 12638 simulations are used and each remaining member of the ensemble is allocated a scaled global mean temperature response for present day greenhouse gas forcing. The distribution of these responses is shown in the published histogram.
The figures representing global summer temperature response to present day forcing in a high and low sensitivity case are constructed with the aid of ERA-40 reanalysis data for the 1960's. The surface of the planet is divided up into a 96x73 grid, and the mean of June, July and August for the years 1960-1969 is taken from the ERA-40 data for each grid-cell. This map is assumed to represent approximately pre-industrial temperatures.
Two simulations from the climateprediction.net ensemble were then taken; one from the 5th percentile of global mean temperature response to greenhouse forcing, and one from the 95th percentile. An eight year mean taken between years 7 and 15 of both control and double carbon dioxide is recorded by the experiment for each grid-cell.
The temperature response in each grid-cell during the third stage is assumed to follow an exponential decay with the same time constant as that of the global mean temperature. Thus, a constant scaling factor converts the difference between the 8 year mean temperatures into an equilibrium temperature response in each grid-cell. This equilibrium response is again scaled by the ratio of present day greenhouse forcing (compared to pre-industrial) to the double CO2 forcing.
Hence, the three globes shown are constructed as follows:
'Pre-industrial' - a mean of June, July and August temperatures taken for the years 1960-1969 from the ERA-40 reanalysis dataset.
'Low sensitivity' - the above temperature map added to the equilibrium temperature response evaluated for each grid-cell, and scaled to most probable forcing due to present day greenhouse gas levels. The model shows a global mean temperature response of 1.34 degrees to this greenhouse forcing.
'High sensitivity' - as a above, but using a model with a global mean temperature response of 4.29 degrees
Ben Sanderson, Dave Frame, Myles Allen, 1/7/2005
climateprediction.net |
