Exploration opérationnelle de la prévision hydrologique d'ensemble avec assimilation automatique de données

Authors: Abaza, Mabrouk
Advisor: Anctil, François
Abstract: The aim of this research project is to identify an automatic data assimilation method that could be used in an operational flood forecasting context. The Ensemble Kalman Filter (EnKF) and the variational assimilation method were first compared on the GR4Jsnow model. Results showed that the EnKF provides the best performance for all forecast horizons when compared to various implementations of the variational method. The performance of the variational method varied from one watershed to another. The deterministic hydrological forecasts issued from the EnKF generate better performance than forecasts without assimilation, which is not always the case for the variational method. The naïve output assimilation is not recommended for forecasts beyond a 2-day horizon. The EnKF was next implemented on a semi-distributed hydrological model (Hydrotel) to improve the initial conditions of the hydrological ensemble predictions (HEP), for the snow-free flows on the au Saumon watershed. The verification of the quality of the HEP showed a considerable improvement in both performance and reliabilty in comparison to the results of model without assimilation. Manual assimilation provides performance similar to the EnKF, but with much less reliability. The 1000-member HEP obtained from the EnKF can be reduced up to 50 members without any loss of reliability or performance. Similar results were obtained for snow accumulation and melt periods, especially when updating the snow water equivalent in addition to the state variables previously identified for the snow-free flows, namely the soil water content and the land flow. Different Canadian meteorological ensemble prediction systems (EPS) were also compared to assess the impact of spatial resolution of these systems on the quality of the HEP – they were also compared to deterministic products. The results showed that EPS are superior to the deterministic products. Global and regional EPS share the same performance, in contrast the regional product that turned out more reliable and characterize the uncertainty of the forecast.
Document Type: Thèse de doctorat
Issue Date: 2014
Open Access Date: 20 April 2018
Permalink: http://hdl.handle.net/20.500.11794/25488
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
Description SizeFormat 
31246.pdfTexte8.13 MBAdobe PDFThumbnail
All documents in CorpusUL are protected by Copyright Act of Canada.