Assessing Near‐Surface Soil Moisture Assimilation Impacts on Modeled Root‐Zone Moisture for an Australian Agricultural Landscape

Abstract

Soil moisture content is an important component of the hydrologic cycle, particularly over vegetation rootzone depths where its variation is linked to the relative fractions of evaporative and sensible heat flux (LE and H) feedbacks to the lower atmosphere, surface runoff, and groundwater recharge [Brutsaert, 2005]. Quantifying these processes across catchments using land surface models (LSMs), therefore, depends on soil moisture state prediction. Improved characterization of root-zone soil moisture quantities has the potential to contribute toward better predictions for a range of hydrological processes — information that will ultimately benefit agricultural and land-use management decisions (e.g., better irrigation scheduling), numerical weather prediction (NWP; e.g., through improved LE and H feedbacks), and emergency management (e.g., improved flood prediction). While an imperfect model structure means that improving certain model variables will not necessarily lead to improvements in predictions of all other model variables [Drusch, 2007], improved root-zone soil moisture can translate to improvement in predictions of other water-balancerelated quantities [Pipunic et al., 2013]. Therefore, the ability to routinely improve root-zone moisture prediction is an important aim, and the impact on other hydrologic variables of interest may contribute to a better understanding of model structural inaccuracies. Inherent LSM uncertainty, resulting from errors in input data (meteorological forcing and parameter information on soil and vegetation properties) and model structural inaccuracies, is the impetus for data assimilation techniques such as the ensemble Kalman filter [EnKF: Evensen, 1994], where observed information is used to sequentially update/correct LSM states through time, based on both modeled and observed error statistics. For routine constraint of root-zone soil moisture prediction across catchments, assimilating relevant remotely sensed data is ideal given their broad spatial coverage at regular repeat intervals. Brightness temperature observations from passive microwave remote sensors have proven particularly suitable for deriving spatial estimates of soil moisture [Kerr et al., 2010; Njoku et al., 2003]. However, these estimates have major limitations, including coarse spatial resolution (>10 km) and shallow sensing depth, which varies depending on a sensor’s spectral frequency and the near-surface moisture conditions but is typically within the top few centimeters of soil at most. Therefore, the impact from assimilating such data products must be able to adequately translate to the model’s deeper layers in order to improve root-zone estimates. A number of studies using synthetic data or in situ field data have shown near-surface moisture assimilation can improve root-zone predictions [e.g., Pipunic et al., 2013; Kumar et al., 2009; Pipunic et al., 2008; Walker et al., 2001; Entekhabi et al., 1994], with some modest improvements to deeper soil moisture from assimilating remotely sensed near-surface moisture shown by Reichle et al. [2007]. More recent work demonstrating the value of assimilating remotely sensed near-surface moisture for 18