Uncertainty propagation within an integrated model of climate change

This paper demonstrates a methodology whereby stochastic dynamical systems are used to investigate a climate model's inherent capacity to propagate un certainty over time. The usefulness of the methodology stems from its abili ty to identify the variables that account for most of the model's uncertain ty. We accomplish this by reformulating a deterministic dynamical system ca pturing the structure of an integrated climate model into a stochastic dyna mical system. Then, via the use of computational techniques of stochastic d ifferential equations accurate uncertainty estimates of the model's variabl es are determined. The uncertainty is measured in terms of properties of pr obability distributions of the state variables. The starting characteristic s of the uncertainty of the initial state and the random fluctuations are d erived from estimates given in the literature. Two aspects of uncertainty a re investigated: (1) the dependence on environmental scenario - which is de termined by technological development and actions towards environmental pro tection; and (2) the dependence on the magnitude of the initial state measu rement error determined by the progress of climate change and the total mag nitude of the system's random fluctuations as well as by our understanding of the climate system. Uncertainty of most of the system's variables is fou nd to be nearly independent of the environmental scenario for the time peri od under consideration (1990-2100). Even conservative uncertainty estimates result in scenario overlap of several decades during which the consequence s of any actions affecting the environment could be very difficult to ident ify with a sufficient degree of confidence. This fact may have fundamental consequences on the level of social acceptance of any restrictive measures against accelerating global warming. In general, the stochastic fluctuation s contribute more to the uncertainty than the initial state measurements. T he variables coupling all major climate elements, such as CO2 concentration of ocean surface temperature change are among the most sensitive variables to any kind of uncertainties. (C) 1998 Elsevier Science B.V. All rights re served.