Interpreting effective hydrologic depth estimates derived from soil moisture remote sensing: A Bayesian non-linear modelling approach

Authors: HYUNGLOK, KIM - Oak Ridge Institute For Science And Education (ORISE); Crow, Wade

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/21/2023
Publication Date: 11/2/2023
Citation: Hyunglok, K., Crow, W.T. 2023. Interpreting effective hydrologic depth estimates derived from soil moisture remote sensing: A Bayesian non-linear modelling approach. Remote Sensing of Environment. 908. https://doi.org/10.1016/j.scitotenv.2023.168067.
DOI: https://doi.org/10.1016/j.scitotenv.2023.168067

Interpretive Summary: Over the past decade, large advances have been made in our ability to track soil moisture dynamics from space. One potential application of this tracking is the constraint of surface water fluxes (e.g., precipitation, evapotranspiration and runoff)-since the sum of these fluxes must be balanced by a corresponding temporal change in soil moisture levels. However, achieving this constraint requires a parameter describing the effective hydrologic depth of satellite-derived soil moisture estimates. Unfortunately, previous attempts to derive this depth parameter using remote sensing observations have resulted in widely varying estimates. This paper examines the theoretical challenges associated with estimating the depth parameter and argues that current remote sensing resources cannot resolve it due to limitations in their temporal frequency and accuracy – resulting in estimated depth parameter values that are highly uncertain in nature and prone to spurious effects. By better identifying these challenges, this manuscript will guide future research aimed at the development of water management tools for existing remote sensing resources.

Technical Abstract: The water balance equation (WBE) describes the relationship between net water inflows into a system and concurrent changes in storage during a time interval ('t). It is a key tool for a range of hydrologic and water resources analyses – including estimating water supply, understanding the terrestrial water cycle, and determining storage and flux trends associated with anthropogenic climate change. When applied to the land surface, the WBE links water fluxes like precipitation (P), evapotranspiration (ET), drainage (D) and surface runoff (R) to concurrent soil moisture changes within a discrete soil layer ('SM). While traditional hydrologic analysis has focused on the accuracy of WBE-estimated fluxes, there has been increased recent interest in fitting the WBE using remotely sensed P and SM data to infer water balance parameters – particularly the length parameter 'Z required to link net fluxes and to 'SM within a time intervalperiod 't. However, obtaining physically interpretable values for WBE parameters like 'Z is a potentially ill-posed problem due to a range of factors including: simplifying assumptions imbedded in a WBE implementation, missing hydrologic constraints (e.g., time-varying variables other than P and SM), and accuracy and revisit limitations in available remote sensing data. In addition, WBE parameters obtained from classical maximum likelihood estimation approaches can varydiffer significantly when different regulation practices are performed during the optimization processes. Here, using athe Bayesian non-linear modeling approach, we demonstrate how these factors can impede the identification of useful 'Z estimates. Specifically, we find that 'Z estimates obtained via the inversion of the WBE using remotely sensed datasets are likely to be spuriously biased high due to limitations in the temporal resolution and accuracy of satellite-based as 'SM estimates – as well as the neglect of other hydrological components in the WBE. Results suggests that estimates of 'Z obtained through the inversion of satellite-based P and other water fluxes are highly effective in nature and great care should be exercised when attempting to interpret them physically. Instead, they should be defined as effective parameters reflecting the characteristics of their fitted data sets.