Assessing sequential data assimilation techniques for integrating GRACE data into a hydrological model

Abstract

The time-variable terrestrial water storage (TWS) products from the Gravity Recovery And Climate Experiment (GRACE) have been increasingly used in recent years to improve the simulation of hydrological models by applying data assimilation techniques. In this study, for the first time, we assess the performance of the most popular data assimilation sequential techniques for integrating GRACE TWS into the WorldWide Water Resources Assessment (W3RA) model. We implement and test stochastic and deterministic ensemble-based Kalman filters (EnKF), as well as Particle filters (PF) using two different resampling approaches of Multinomial Resampling and Systematic Resampling. These choices provide various opportunities for weighting observations and model simulations during the assimilation and also accounting for error distributions. Particularly, the deterministic EnKF is tested to avoid perturbing observations before assimilation (that is the case in an ordinary EnKF). Gaussian-based random updates in the EnKF approaches likely do not fully represent the statistical properties of the model simulations and TWS observations. Therefore, the fully non-Gaussian PF is also applied to estimate more realistic updates. Monthly GRACE TWS are assimilated into W3RA covering the entire Australia. To evaluate the filters performances and analyze their impact on model simulations, their estimates are validated by independent in-situ measurements. Our results indicate that all implemented filters improve the estimation of water storage simulations of W3RA. The best results are obtained using two versions of deterministic EnKF, i.e. the Square Root Analy

On the Use of Adaptive Ensemble Kalman Filtering to Mitigate Error Misspecifications in GRACE Data Assimilation

“…Due to the narrow range of storage dynamics in surface soil moisture, the improvement was less significant in this layer with the groundwater level, as the most significant contributor, receiving the greatest improvement. This corroborates the results from previous studies (e.g., Girotto et al, ; Khaki, Hoteit, et al, ; Khaki, Ait‐El‐Fquih, et al, ; Schumacher et al, ; Shokri, Walker, van Dijk & Pauwels ; Tian et al, ; Zaitchik et al, ). Conversely, the accuracy of the EnKF with initial α Q =2 (twice the reference error magnitude) slightly degraded in both soil layers and groundwater, comparing to OL.…”

“…It has been proven that the ensemble Kalman filter (EnKF) can improve the accuracy of hydrologic models by merging observations with model predictions. Such observations include soil moisture (e.g., Aubert et al, ; Crow & Ryu, ; Houser et al, ; Pauwels et al, ; Reichle et al, , ; Walker & Houser, ), snow water equivalent (e.g., Barrett, ; Slater & Clark, ; Sun et al, ), streamflow (e.g., Lee et al, ; Clark et al, ), groundwater levels (e.g., Hendricks Franssen et al, ), turbulent heat fluxes (e.g., Bateni & Entekhabi, ; Pipunic et al, ; Xu et al, ) , microwave radiances (e.g., Dechant & Moradkhani, ), and terrestrial water storage (TWS; e.g., Ellett et al, ; Forman & Reichle, ; Forman et al, ; Girotto et al, ; ; Houborg et al, ; Khaki, Ait‐El‐Fquih, et al, ; Khaki, Hoteit, et al,; Khaki, Schumacher, et al, ; Kumar et al, ; Li et al, ; Li & Rodell, ; Smith, ; Tian et al, ; van Dijk et al, ; Zaitchik et al, ). The idea behind the EnKF is to combine observations and model estimates of state variables considering their relative error covariances.…”

On the use of the GRACE normal equation of inter-satellite tracking data for estimation of soil moisture and groundwater in Australia

“…In addition, it is challenging to improve surface soil moisture varying rapidly in time, using a monthly mean GRACE observation. Tian et al (2017) utilized the satellite soil moisture observation from Soil Moisture and Ocean Salinity (SMOS, Kerr et al, 2001) in addition to GRACE data for their data assimilation and showed a clear improvement in the top soil moisture estimate. The GC approach with complementary observations at higher temporal resolution should be considered, particularly to enhance the surface soil moisture computation.…”

Performance of Different Ensemble Kalman Filter Structures to Assimilate GRACE Terrestrial Water Storage Estimates Into a High‐Resolution Hydrological Model: A Synthetic Study

“…The ensemble Kalman filter (EnKF) is a very useful data assimilation (DA) tool to reduce the model errors at each time step by updating the model state variables (Evensen, ). DA methods have been successfully applied in hydrologic modeling to assimilate observations from various sources into the model, including soil moisture (e.g., Aubert et al, ; Crow & Ryu, ; Houser et al, ; Pauwels et al, ; Reichle et al, , ), snow water equivalent (e.g., Barrett, ; Slater & Clark, ; Sun et al, ), streamflow (e.g., Clark et al, ; Lee et al, ), groundwater levels (e.g., Hendricks Franssen et al, ), microwave radiances (e.g., Dechant & Moradkhani, ), and TWS (e.g., van Dijk et al, ; Ellett et al, ; Forman & Reichle, ; Forman et al, ; Girotto et al, , ; Houborg et al, ; Khaki, Ait‐El‐Fquih, et al, ; Khaki, Hoteit, et al, ; Khaki, Schumacher, et al, ; Kumar et al, ; Li & Rodell, ; Li et al, ; Smith, ; Tian et al, ; Zaitchik et al, ). Therefore, assimilating the GRACE TWS retrievals into a hydrological model yields more reliable water storage estimates in which the drawbacks of both approaches are mitigated.…”