Identification Method - Climate Change: Identification and Projections - page 67

6
Identification Method
6.1. The current state of affairs
Not all GCM parameters came from fundamental physics, and some need
to be adjusted, more or less empirically, like those relating to the genesis of
clouds and their feedback effects, or the biochemical processes of
vegetation, or micro-turbulence, or many others. Centers of modeling give
little information on the way in which they adjusted the parameters of their
model. Obtaining globally stable and realistic temperature levels in
simulation is in itself not a trivial matter. Large-scale identification of these
models is out of reach 1 . However, identification limited to a small number of
critical parameters, known as closure parameters, can be envisaged
[JAR 10]. This can be done by using techniques of identification, for
instance by nonlinear filtering (UKF: Unscented Kalman Filter) .
Nevertheless, we are still a long way from a black box type identification,
and the result is also highly dependent on the quality of the rest of the model.
An even more restricted approach consists of detecting and attributing to a
given input the responsibility in a given element of observed climatic
behavior: typically, to detect the “fingerprint” of various human activities in
warming.
Complete black box type identification can only be achieved with models
of reduced complexity, like those presented in Chapter 4, and it is reasonable
1 According to the author of box 9.1 of the AR5, “the need for model tuning may increase
model uncertainty. There have been recent efforts to develop systematic parameter
optimization methods, but owing to model complexity they cannot yet be applied to fully
coupled climate models”.