RESEARCH THEMES
Rainfall models
Rainfall models| stochastic models
​
During my PdH thesis at INRS-ETE (Québec, CANADA), I proposed different versions of the Neyman-Scott process, a point-process model widely used to represent the statistical properties of rainfall at small temporal scales. More precisely, I applied cubic copulas to represent the dependence between rain cell intensity and duration.
My current project at LTHE (Grenoble, FRANCE) focuses on rainfall extremes at daily and sub-daily scales. Using a weather pattern classification, the distribution of rainfall extremes is modelled with a mixture of heavy-tailed distribution. The main contribution of this work is the regionalization (RFA) of the most severe rainfall events.
Hydrological models
Calibration of hydrological models | Heteroscedasticity, autocorrelation and skewness of residual errors
​
At the university of Newcastle and at the University of Adelaide (AUSTRALIA), I worked on the calibration of CRR-type hydrological models (e.g. GR4J). Using a Bayesian framework, I proposed general guidelines regarding the treatment of residual errors, which usually are highly heteroscedastic, autocorrelated and skewed.
A major achievement of this project was the BATEA software developed by Dmitri Kavetski. I contributed to the alpha-version of this software which offers powerful algorithms for Bayesian inference and prediction, as well as an interface, scripting facilities, and connections to external user simulation models.
Mixture models
Mixture of skewed distributions
​
At IREQ (Hydro-Québec research institute, CANADA), I developped mixture of gamma and Gumbel distributions. Hidden-Markov models were used to represent succession of dry and wet periods when the mxiture models were applied to annual peak flows and volumes.
Copulas
Copula selection
​
A copula is a multivariate distribution which can be used to represent the dependence structure between several random variables. At INRS-ETE( Québec, CANADA), I compared different methods for the selection of the best copula. With David Huard and Anne-Catherine Favre, we proposed a new method using a Bayesian framework.