Unscented weighted ensemble Kalman filter for soil moisture assimilation

A new data assimilation technique, unscented weighted ensemble Kalman filter (UWEnKF) was developed based on the scaled unscented transformation and ensemble Kalman filter (EnKF). In UWEnKF, the individual members selected are unequally weighted and symmetric about the expectation. To investigate the performance of UWEnKF, nine assimilation experiments with different ensemble sizes (161, 1601, 16001) and different assimilation frequencies (every 6 h, every 12 h, every 24 h) were designed to assimilate soil surface (5 cm) moisture data observed at station HY in the upper reaches of the Yellow River, in the northeastern of Tibetan plateau, China into the Richards equation. The results showed that the performance of the filter was greatly affected by random noise, and the filter was sensitive to ensemble size and assimilation frequency. Increasing the ensemble size reduced the effects of random noise on filter performance in several independent assimilation runs (i.e., it decreased the differences between the results of the several independent assimilation runs). Reducing the assimilation frequency also reduced the effects of random noise on filter performance. UWEnKF gave more accurate soil moisture model results than EnKF for all ensemble sizes and assimilation frequencies at all soil depths. Additionally, EnKF may have different performances according to different initial conditions, but not for UWEnKF. Precipitation and soil properties uncertainties had some impact on filter performance. Thus, UWEnKF is a better choice than EnKF, while it is more computationally demanding, for improving soil moisture predictions by assimilating data from many sources, such as satellite-observed soil moisture data, at a low assimilation frequency (e.g., every 24 h).