Assessing Performances of Multivariate Data Assimilation Algorithms with SMOS, SMAP, and GRACE Observations for Improved Soil Moisture and Groundwater Analyses

Natthachet Tangdamrongsub,Jianzhi Dong,Peter Shellito

doi:10.3390/w14040621

Abstract

Multivariate data assimilation (DA) of satellite soil moisture (SM) and terrestrial water storage (TWS) observations has recently been used to improve SM and groundwater storage (GWS) simulations. Previous studies employed the ensemble Kalman approach in multivariate DA schemes, which assumes that model and observation errors have a Gaussian distribution. Despite the success of the Kalman approaches, SM and GWS estimates can be suboptimal when the Gaussian assumption is violated. Other DA approaches, such as particle smoother (PS), ensemble Gaussian particle smoother (EnGPS), and evolutionary smoother (EvS), do not rely on the Gaussian assumption and may be better suited to non-Gaussian error systems. The objective of this paper is to evaluate the performance of these four DA approaches (EnKS, PS, EnGPS, and EvS) in multivariate DA systems by assimilating satellite data from the Soil Moisture and Ocean Salinity (SMOS), Soil Moisture Active Passive (SMAP), and Gravity Recovery And Climate Experiment (GRACE) missions into the Community Atmosphere and Biosphere Land Exchange (CABLE) land surface model. The analyses are carried out in Australia’s Goulburn River catchment, where in situ SM and groundwater data are available to comprehensively validate the DA performance. Results show that all four DA approaches have outstanding performances and improve correlation coefficients of SM and GWS estimates by ~20% and 100%, respectively. The EvS outperforms the others, but its benefit is relatively marginal compared to Gaussian approaches (e.g., EnKS). This is due to the fact that SM and TWS error distributions in this study are close to Gaussian: a suitable condition for, e.g., EnKS, EnGPS. The robust performance of EvS appears to be the optimal approach for jointly assimilating multi-source hydrological observations to improve regional hydrological analyses.

Highlights

The accurate estimation of terrestrial water storage (TWS) is critical for regional water resource analyses
This study aims to assess the performance of four different Data assimilation (DA) algorithms on multivariate DA using satellite soil moisture (SM) and TWS data
This study assessed the performance of four different multivariate DA algorithms used to assimilate Soil Moisture and Ocean Salinity (SMOS), Soil Moisture Active Passive (SMAP), and Gravity Recovery And Climate Experiment (GRACE) data into the Community Atmosphere and Biosphere Land Exchange (CABLE) model to improve SM and groundwater storage (GWS) simulations in the Goulburn River catchment in Australia

Summary

Introduction

The accurate estimation of terrestrial water storage (TWS) is critical for regional water resource analyses. TWS can be calculated using a land surface model (LSM), which incorporates complex land–surface processes into simulations of a wide range of hydrologic variables [1]. Data assimilation (DA) is an important technique for correcting model state variables with satellite data, resulting in more robust hydrologic variables [3]. Univariate DA usually improves only one state variable associated with the assimilated observation, while the effect on other variables can be insignificant or negative. Tian et al [6] reported a reduction in the accuracy of groundwater storage (GWS) estimates after assimilating satellite SM observations. Multivariate DA, as compared to univariate DA, uses a suite of satellite observations to simultaneously constrain multiple model states, resulting in a better overall representation of hydrologic systems [7,8]

Objectives

Methods

Findings

Conclusion