A Super-Ensemble (SE) methodology is proposed to improve the short-term ocean predictions of Sea Surface Temperature (SST) in the Mediterranean Sea.
The adopted methodology consists in a multi-linear regression applied to a Multi-Physics Multi-Model Super-Ensemble dataset (MMSE), a collection of different operational forecasting systems together with other ad-hoc simulations, created by modifying selected numerical model parameterizations.
Our results show that SE estimate can improve the representation of the ocean state with respect to the best contributing member and the performances are dependent on the choice of the unbiased estimator and the length of the training period. We carried out sensitivity studies in order to find weakness and strengths of the method. The quality of the originating dataset has the main impact to the MMSE Root Mean Square Error (RMSE). The MMSE estimate changes appreciably by changing the number and the kind of ensemble members. Other formulations of the regression algorithm, including an extension to the usage of horizontal Empirical Orthogonal Functions as a filter for the signal during the training and test period, have been proposed in order to fix regression algorithm problems.
Lecturer
Jenny Pistoia
CMCC, Bologna
Bologna, Italy, Viale Aldo Moro, 42 - Viale Aldo Moro, 42, Bologna, Italy -
20 Mar 2014
Organized by
- CMCC - Euro-Mediterranean Center on Climate Change