Data assimilation using sequential Monte Carlo methods in hydrological applications

Hydrologic Data Assimilation concerns the application of state estimation methods to hydrologic models. The hydrologic models used in this dissertation correspond to a lumped conceptual model (rainfall-runoff model) and a distributed physically-based model (land surface-atmosphere transfer scheme). Hydrologic systems are highly nonlinear with complex dynamics. Therefore, nonlinear/non-Gaussian estimation techniques should be used in the inference of the states and/or parameters of the hydrologic models. In this sense, sequential Monte Carlo methods (a.k.a. particle filters) have captured the interest of the scientific community and nowadays are widely utilized in complex state estimation problems. This dissertation introduces two data assimilation methods which are based on Kalman and particle filtering theory: the ensemble Gaussian particle filter and the standard particle filter with parameter resampling. The first hydrologic variable to be assimilated is the discharge of water at the outlet of the Zwalm catchment which is located in East-Flanders. The second variable of interest is the volumetric soil moisture content in the study area located in the Grand Duchy of Luxembourg. These variables are assimilated separately in experiments with different setups and different study areas. Overall, the results indicate an improvement in the estimation of the model output flows when the proposed methods are applied to specific hydrologic data assimilation problems