Climate models are used extensively for understanding climate change and for developing policies to address it. Quantifying the uncertainty in the ocean component is important to understanding the importance of the oceans to the Earth's climate. Understanding the non-linearity of the ocean model's parameter space as exhibited by the resulting circulation will lead to improved representation of the simulated ocean and its deficiencies.
Intellectual Merit: Complex climate models make use of parameters to represent aspects of various physical processes. There is uncertainty in the outputs from the lack of knowledge of the parameters used to represent the model physics. The best method of propagating the uncertainty through the complex and non-linear models is to use large ensembles of runs. This feasibility study will demonstrate a method using a relatively small ensemble of simulations coupled with an emulator to examine the effect of uncertainty in parameter specification on ocean circulation in a model. The parameter space of the ocean component of a climate model will be explored, focusing on understanding the uncertainties associated with a set of ocean metrics (e.g. transports, strength of the meridional ocean circulation, sea level, regional and global SST). First, a 100-year experiment using the ocean model of Community Climate System Model parameter space, will be conducted for a full ensemble of 100 (fewer if for a reduced set of parameters) at coarse resolution (3°). Then, the model output will be used to develop a set of ocean based metrics to examine the model's uncertainty of its parameter space. Finally, these metrics will be used to create and explore an ensemble of 10 runs at a higher resolution (1°) by mapping differences between two parameter spaces to make inferences about uncertainties at 1°. The proposed method is known as Statistical Analysis of Computer Code Output or Design and Analysis of Computer Experiments, rather than the classical method of Monte Carlo ensembles. These methods allow the use of much smaller ensembles, i.e., order 100 rather than order of a 1000 runs. A minimum number of simulations are required to span the parameter space in a designed experiment. An emulator, or surrogate, is a fast statistical approximation to the original climate model. As such, the emulator can be used, instead of the complex model, to efficiently span the parameter space to estimate the probability distribution and to address questions relating to parameter setting and metric uncertainty.
Broader Impacts: The funding of this research will have broader impacts. The investigator is a member of an underrepresented group within both the Physical Oceanography and the Statistical and Mathematical communities. As such, her participation in meetings of both communities will increase their diversity. The multi-disciplinary aspect of the proposal will allow for the cross fertilization of ideas. Further, this proposal would produce an unprecedented dataset that would be available to researchers and projects beyond the scope of this proposal for analyses of various ocean-climate and statistical investigations. Understanding the uncertainty in a climate model and the estimates of what the future climate might look like is critical to the public's acceptance of climate predictions. As with weather forecasts, the public and policy makers are better able to understand the risk associated with an outcome (in this case, climate change), as well as trust the information, when the uncertainty is known. This research will give the public and policy makers the information and to external researchers, the methods for determining uncertainties of similarly sized problems.
