A Bayesian technique for refining the uncertainty in global energy model forecasts

Abstract

Global energy models have a large degree of uncertainty associated with them. This consists of uncertainty in the model structure as well as uncertainty in the exogenous input parameters. This paper combines Monte Carlo methods with Bayesian updating techniques to provide a method for refining the uncertainty in the Edmonds-Reilly global energy model. The Bayesian updating technique uses likelihood-based windows constructed from actual observations of the output variables to filter out the model simulations that do not conform with the observed output. The windows are based on outputs of energy consumption and carbon emissions. Two alternative model structures are examined: one with uncorrelated input parameters and the other with correlated input parameters. Statistical properties are calculated to measure the effects of windowing on the output distributions. The partial rank correlations between the inputs and outputs and between the inputs are also determined. The prior distributions and correlation structure of the inputs are then revised through the updating process. An application of the windowing process illustrates the effects of capping carbon emissions on the input structure.