Decadal Warning Systems - Reducing Disaster: Early Warning Systems for Climate Change - page 299

Fig. 16.2 Schematic illustrating the potential of initialization to narrow uncertainties in decadal
climate predictions. A schematic representation of prediction in terms of probability. The probabil-
ity distribution corresponding to a forced simulation is in red with the deeper shades indicating
higher probability. The probabilistic forecast is in blue . The sharply peaked forecast distribution
based on initial conditions broadens with time as the infl uence of the initial conditions fades until
the probability distribution of the initialized prediction approaches that of an uninitialized projec-
tion (Adapted from IPCC 2013 , Box 11.1, Fig. 3)
Standard climate model projections (IPCC 2007 ) do simulate natural internal
variability, but on average it will not be in phase with reality. In order to predict the
evolution of natural internal variability, it is necessary to start from the present
state of the climate system, just as it is necessary to start from conditions today to
predict tomorrow's weather. This is achieved by initializing climate models with
observations. Internal variability is not necessarily predictable, but any skill would
narrow the near-term uncertainties in climate change projections, as illustrated in
Fig. 16.2 . Initialization may also narrow uncertainties by correcting errors in
model responses to previous external forcing factors. Decadal climate prediction
is much less mature than seasonal forecasting, but there is currently a substantial
international effort to build and evaluate climate predictions for the coming years
to a decade or two (e.g. Meehl et al. 2013 ; Smith et al. 2012b ), as summarized in
this chapter.