The tenth IMSC, Beijing, China, 2007 - International Meetings on ...

Empirical probabilistic seasonal forecasts for Canada
Speaker: V. Kharin
V. Kharin
Canadian Centre for Climate Modelling & Analysis
Slava.Kharin@ec.gc.ca
A method for constructing empirical deterministic and probabilistic seasonal forecasts
in Canada is proposed. <strong>The</strong> performance of the empirical forecasts is evaluated and compared
to the performance of a multi-model ensemble of seasonal hindcasts produced with four global
numerical models. Special attention in developing and validating empirical forecasts is paid to
avoiding artificial skill enhancement. <strong>The</strong> 20th century empirical seasonal hindcasts of near
surface temperature in Canada are best skillful in colder seasons for time leads up to several
months. Little skill is found in transient seasons. One-month lead empirical seasonal
forecasts are competitive to the corresponding numerical model seasonal forecasts. Seasonal
precipitation hindcasts in Canada are not very skillful.
Regime predictability in an atmospheric model or analogue prediction revisited
Speaker: Frank Kwasniok
Frank Kwasniok
University of Exeter, United Kingdom
Predictability properties of atmospheric flow are diagnosed and analysed empirically using
nonlinear signal processing techniques. <strong>The</strong> notion of predictability is relaxed from accurate
prediction of individual trajectories to a coarse-grained view in which only probabilities of
visiting certain regions of state space or regimes are forecast. <strong>The</strong> regimes and a probabilistic
model of the transitions between them are determined simultaneously by fitting a hidden
Markov model to a data set of the system. <strong>The</strong>n regime predictability is examined as a function
of initial condition using the Markov model; states with particularly high or low regime
predictability are identified. States with high regime predictability are characterized by a large
projection of their posterior distribution of regime membership onto the slowliest decaying
eigenmode of the Markov model. Predictability information is refined by using an analogue
model of the regime posterior. <strong>The</strong> method bears the potential of improving prediction by
combining dynamical models of the system with the present statistical model. <strong>The</strong> ideas
outlined above are applied to the classical Lorenz system as well as an atmospheric model
with realistic variability and teleconnection patterns.
79