one year into the future. In addition, it provides a great number of applications with relevant climate information and helps decision-making
in agriculture, energy, health and hydrology, amongst other sectors of society. These predictions are influenced by the interactions
between ocean, sea ice, atmosphere and land.
Therefore, a suitable way to represent the Earth climate system at seasonal time-scales, is to use a dynamical coupled
ocean-sea-ice-atmosphere-land model.
Due to the complexity of the climate and the model processes involved, as well as the impact of small-scale processes that cannot
be explicitly represented, and the role of the large-scale processes of interest, it is not possible to describe the state of the climate
system with complete accuracy.
A way to improve climate predictions and to reduce the associated uncertainties is to use an ensemble of climate model predictions,
weighted according to their past performances.
Prediction ensembles improve significantly forecasts. The construction of these ensembles is normally
highly sensitive to initial conditions uncertainties and therefore not always optimal, leading to innaccuracies in the climate prediction.
In this talk, we propose a novel method for a dynamical ensemble selection, which samples in an optimum way the space of initial
condition uncertainties, and allows to trim large simulation ensembles as the experiments progresses.
This research will include robust machine-learning techniques to optimise the diagnostics and evaluation metric designed to
achieve the best informative outcome at the end of the climate experiment.
The objectives of this ambitious study will be achieved thanks to the usage of the computing resources provided by
the supercomputers of the Barcelona Supercomputing Center (BSC, Spain).
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner