An Improved Particle Filter Algorithm Based on Ensemble Kalman Filter and Markov Chain Monte Carlo Method

Abstract

Data assimilation (DA) has developed into an important method in Earth science research due to its capability of combining model dynamics and observations. Among various DA methods, the particle filter (PF) is free from the constraints of linear models and Gaussian error distributions. Thus, it is now receiving increasing attention in DA. However, the particle degeneracy still remains a major problem in practical application of PF. In this paper, an improved PF is proposed based on ensemble Kalman filter (EnKF) and the Markov Chain Monte Carlo (MCMC) method. It uses an EnKF analysis to define the proposal density of PF instead of the prior density, thus reducing the risk of particle degeneracy. Furthermore, when particle degeneracy happens, resampling is performed follow by an MCMC move step to increase the diversity of particles, thus reducing the potential of particle impoverishment and improving the accuracy of the filter. Finally, the improved PF is tested by assimilating brightness temperatures from the Advanced Microwave Scanning Radiometer (AMSR-E) into the variance infiltration capacity (VIC) model to estimate soil moisture in the NaQu network region at the Tibetan Plateau. The experiment results show that the improved PF can provide more accurate assimilation results and also need fewer particles to get reliable estimations than the EnKF and the standard PF, thus demonstrating the effectiveness and practicality of the improved PF.