Assessing the impacts of using energy balance models to estimate probabililty distributions of equilibrium climate sensitivity

Thursday, 18 December 2014
Chris E Forest1, Ashley Warner2, Klaus Keller2 and Andrei P Sokolov3, (1)Earth and Environmental Systems Institute, The Pennsylvania State University, University Park, PA, United States, (2)Pennsylvania State University Main Campus, University Park, PA, United States, (3)Massachusetts Institute of Technology, Joint Program on the Science and Policy of Global Change, Cambridge, MA, United States
Probability distributions for equilibrium climate sensitivity (ECS) are estimated using climate models that have a range of complexities. Here, we investigate the ability to identify ECS using low complexity models using two statistical methods. By using a climate model with intermediate complexity model to provide pseudo-observations with known climate system properties, we test whether the parameters for a low complexity model can be identified to yield similar model responses across a wide range of ECS values. In this work, the Diffusive Ocean Energy balance CLIMate model (DOECLIM) was calibrated to match the output from 639 simulations of the MIT Integrated Global System Model (IGSM), where each IGSM simulation has a different set of values for three key climate system properties: equilibrium climate sensitivity, vertical diffusivity of temperature anomalies and net aerosol forcing. The energy balance model estimates the globally averaged climate state based on simplified model physics. The IGSM represents the zonal mean state of the atmosphere and ocean using physics used in higher complexity models and produces hindcasts and projections that include internal climate variability. The DOECLIM parameters were estimated for each IGSM run to find the parameter settings for the simpler model that would best match the global mean surface temperature and the global-mean ocean heat content as simulated in the more complex model. When successful, this allows for the simpler model to be used as an emulator over a range of model parameter settings. Two statistical calibration techniques were used and compared. The first was Differential Evolution, a genetic algorithm that produces a single set of parameters providing best fit values; the second was a Markov Chain Monte Carlo method, which produces a joint probability distribution for the DOECLIM parameters. We analyzed the statistical skill, including potential biases, that exist when calibrating the energy balance model to IGSM output. In particular, the estimated DOECLIM ECS values tended to be lower than their corresponding IGSM values, particularly for runs with low ocean diffusivity. We will discuss these results in the context of other estimates of probability distributions for ECS that use similar energy balance models.