Which filter for data assimilation in water quality models? Focus on oxygen reaeration and heterotrophic bacteria activity

Abstract

With the development of sophisticated water quality models and the advances in computational power, data assimilation (DA) techniques, especially ensemble-based methods (the ensemble Kalman filter and particle filter), are attracting considerable attention in water quality modeling for improving the estimation of state variables and parameters in water quality models. The ensemble Kalman filter (EnKF) has become the most popular DA method while the particle filter (PF), which does not rely on Gaussian or quasi-linearity assumptions, is seldom applied in water quality modeling. Here, we present a comparison between the PF and EnKF for the update of model parameters related to river metabolism. The two filters are implemented in ProSe-PA, a hydro-biogeochemical software, and their performance is assessed on two synthetic case studies. The results indicate that PF and EnKF can estimate dissolved oxygen concentrations and the posterior probability distribution function of the associated parameters, either precisely for both filters in the case of a slightly nonlinear system (reaeration at the air–water interface) or more precisely for the PF in the case of a strongly nonlinear system (organic matter degradation) dominated by heterotrophic bacterial activities. Since the PF is more accurate, its usage is recommended for water quality modeling and guidelines are provided for its set-up.