Developments in statistics applied to hydrometeorology: imputation of streamflow data and semiparametric precipitation modeling

Keywords :

• dynamic regression models
• statistical modeling of precipitation amounts
• extended generalized Pareto distribution
• Bernstein polynomials
• nonparametric kernel estimator

In the first part of this PhD thesis we propose an approach for streamflow imputation based on dynamic regression models, more specifically, a multiple linear regression with ARIMA residual modeling. We apply this method for reconstructing the data of eight stations situated in the Durance watershed in the south-east of France. The results showed that, without making use of additional variables, we manage to accurately reconstruct missing blocks of various lengths, ranging up to 20 years.
The second part of this work addresses the statistical modeling of precipitation amounts. We develop two semiparametric models based on a new class of distributions, the extended generalized Pareto (EGPD). We compare the performance of these methods with the one obtained by applying EGPD, on both simulated samples and two precipitation data sets from south-east of France. The results show a reduced estimation error compared to EGPD, this effect being even more obvious as the sample size increases.

President:

Mr Stéphane Girard (INRIA Alpes)

Directors:

• Mme Clémentine Prieur (Professeur - Université Grenoble Alpes )
• Mme Anne Catherine Favre (Professeur - Université Grenoble Alpes )

Raporteurs:

• Mme Véronique Maume-Deschamps (Professeur - Université Claude Bernard Lyon 1 )
• Mme Valérie Mombet (Professeur - Université de Rennes 1 )

Examinators:

• Mr Philippe Naveau (Directeur de recherche - LSCE CNRS )
• Mr Benjamin Renard (Chargé de recherche - IRSTEA )