Master internship Metocean multi-source data fusion non parametric filtering – Key words : Environmental statistics, metocean, spatio-temporal models, analogues. – Staff Valérie Monbet,
[email protected] Pierre Ailliot,
[email protected] – Place : Brest or Rennes. – Date : 2015
Context and objectives The amount of both observational and model-simulated data within the environmental, climate and ocean sciences has grown at an accelerating rate since the early 1980s. Observational (e.g. satellite, in-situ...) data are generally accurate but still subject to measurements errors and available with a complicated spatio-temporal sampling. Increasing computer power and understandings of physical processes have permitted to advance in models accuracy and resolution but purely model driven solutions may still not be accurate enough. Operational issues such as the forecast of natural hazards urge the development of models and algorithms which can combine efficiently the available sources of data. Data assimilation methods have been developed to incorporate observations into models (Bertino et al., 2003). The idea consists in improving the background solution of the model tacking into account observations information. One of the main backward of such methods, when high dimension physical system are studied, is that models require too large computation time. One alternative to reduce computation times is to use non-parametric statistical emulators of the system dynamic instead of the standard numerical solutions. For some meteorological parameters, the existing datasets (in situ data, satellite data, numerical model outputs) may now provide enough information in order to be able to imitate the physical dynamic using nearest neighbor methods (Yiou, 2014).
1
During this internship, we propose to treat the first steps of the project. On one hand, we will have to explore the available metocean dataset (https://meteobs-bzh.univ-rennes1.fr/). This database contains numerical model outputs and in situ observations. For example the figure shows a forecasted wind field (model output, blue) and observed wind at different locations (airports, sensor on boats, buoys, red). One would like to know more about the quality of the forecasts and one will build appropriate indices. On the other hand, one will start to build emulators of the dynamic of the observed variables (wind, temperature, etc) (Ailliot et al., 2014). One alternative way of work could be to start to build non parametric filtering algorithms (Tandéo et al., 2014). A PhD thesis will be proposer after the internship, if we get a grant. http://perso.univ-rennes1.fr/valerie.monbet/doc/FiltrageNpar_TheseIRMAR.pdf
References Ailliot P., Allard D., Monbet V., Naveau P. (2014) Stochastic weather generators : an overview of weather type model. In revision Bertino L, Evensen G, Wackernagel H. (2003). Sequential data assimilation techniques in oceanography. International Statistical Review, 71(2) :223-241. Tandeo P., Ailliot P. , Fablet R. , Ruiz J., Rousseau F. and Chapron B. (2014) The Analog Ensemble Kalman Filter and Smoother, Climate informatics, Boulder, Colorado, US. Yiou, P. (2014). AnaWEGE : a weather generator based on analogues of atmospheric circulation. Geoscientific Model Development, 7(2), 531-543.
2
