|Working Group I: The Scientific Basis|
|Other reports in this collection|
14.2.2 Predictability in a Chaotic System
The climate system is particularly challenging since it is known that components in the system are inherently chaotic; there are feedbacks that could potentially switch sign, and there are central processes that affect the system in a complicated, non-linear manner. These complex, chaotic, non-linear dynamics are an inherent aspect of the climate system. As the IPCC WGI Second Assessment Report (IPCC, 1996) (hereafter SAR) has previously noted, “future unexpected, large and rapid climate system changes (as have occurred in the past) are, by their nature, difficult to predict. This implies that future climate changes may also involve surprises’. In particular, these arise from the non-linear, chaotic nature of the climate system Progress can be made by investigating non-linear processes and sub-components of the climatic system.” These thoughts are expanded upon in this report: “Reducing uncertainty in climate projections also requires a better understanding of these non-linear processes which give rise to thresholds that are present in the climate system. Observations, palaeoclimatic data, and models suggest that such thresholds exist and that transitions have occurred in the past Comprehensive climate models in conjunction with sustained observational systems, both in situ and remote, are the only tool to decide whether the evolving climate system is approaching such thresholds. Our knowledge about the processes, and feedback mechanisms determining them, must be significantly improved in order to extract early signs of such changes from model simulations and observations.” (See Chapter 7, Section 7.7).
Integrations of models over long time-spans are prone to error as small discrepancies from reality compound. Models, by definition, are reduced descriptions of reality and hence incomplete and with error. Missing pieces and small errors can pose difficulties when models of sub-systems such as the ocean and the atmosphere are coupled. As noted in Chapter 8, Section 8.4.2, at the time of the SAR most coupled models had difficulty in reproducing a stable climate with current atmospheric concentrations of greenhouse gases, and therefore non-physical “flux adjustment terms” were added. In the past few years significant progress has been achieved, but difficulties posed by the problem of flux adjustment, while reduced, remain problematic and continued investigations are needed to reach the objective of avoiding dependence on flux adjustment (see Chapter 8, Section 8.4.2; see also Section 188.8.131.52).
Another important (and related) challenge is the initialisation of the models
so that the entire system is in balance, i.e., in statistical equilibrium with
respect to the fluxes of heat, water, and momentum between the various components
of the system. The problem of determining appropriate initial conditions in
which fluxes are dynamically and thermodynamically balanced throughout a coupled
stiff system, such as the ocean-atmosphere system, is particularly difficult
because of the wide range of adjustment times ranging from days to thousands
of years. This can lead to a “climate drift”, making interpretation
of transient climate calculations difficult (see Chapter
8, Section 8.4.1).
The initialisation of coupled models is important because it produces the climate
base state or “starting point” for climate change experiments. Climate
model initialisation continues to be an area of active research and refinement
of techniques (see Chapter 8, Section 8.4). Most groups
use long integrations of the sub-component models to provide a dynamically and
thermodynamically balanced initial state for the coupled model integration.
However, there are at least as many different methods used to initialise coupled
models as there are modelling groups. See Stouffer and Dixon (1998) for a more
complete discussion of the various issues and methods used to initialise coupled
Since the SAR, improvements in developing better initialisation techniques
for coupled models have been realised. For instance, starting with observed
oceanic conditions has yielded improved simulations with reduced climate drift
(Gordon et al., 1999). Earlier attempts with this technique usually resulted
in relatively large trends in the surface variables (Meehl and Washington, 1995;
Washington and Meehl, 1996). Successfully starting long coupled integrations
from observations is important for a number of reasons: it simplifies the initialisation
procedure, saves time and effort, and reduces the overhead for starting new
coupled model integrations.
Such progress is important, but again further work is needed. We simply do not fully understand the causes of climate drift in coupled models (see Chapter 8, Section 8.4.2).
Other reports in this collection