A unified data-driven method to derive hydrologic dynamics from global SMAP surface soil moisture and GPM precipitation data

Authors: MAO, Y. - University Of Washington; Crow, Wade; NIJSSEN, B. - University Of Washington

Submitted to: Water Resources Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/15/2019
Publication Date: 1/26/2020
Citation: Mao, Y., Crow, W.T., Nijssen, B. 2020. A unified data-driven method to derive hydrologic dynamics from global SMAP surface soil moisture and GPM precipitation data. Water Resources Research. 56:E2019WR024949. https://doi.org/10.1029/2019WR024949.
DOI: https://doi.org/10.1029/2019WR024949

Interpretive Summary: Land surface models attempt to accurately estimate the flow of water into and out of the vertical layer of soil containing roots (i.e., the vegetation root zone). As such, they are useful for a broad range of agricultural applications including: in-season crop yield prediction, agricultural drought monitoring and irrigation scheduling. They also form the basis for attempts to improve weather and climate forecasting by accurately accounting for the flux of water and energy between the land surface and the lower atmosphere. However, to make useful predictions, these models require the accurate specification of multiple parameters and typically require extensive calibration. In many cases, such calibration is not feasible due to a lack of available ground-based observations for comparison. This paper explores the use of two new types of hydrologic observations (i.e., large-scale estimates of surface soil moisture and rainfall acquired from satellite remote sensing) as a source of information to improve the performance of an existing land surface model. Specifically, using a simple regression approach, we identify structural problems with the land surface model without reliance on any ground-based observations. This approach will eventually be used by operational drought and crop-system modelers to improve their ability to forecast the effects of water stress on agricultural productivity.

Technical Abstract: The new satellite-observed Soil Moisture Active Passive (SMAP) and the Global Precipitation Measurement (GPM) datasets contain rich information about land surface hydrologic processes. In this study, a unified regression method is proposed and applied to these global datasets to quantify factors governing surface soil moisture (SSM) dynamics. Two simple forms of regressors are implemented: 1) the linear regressors of SSM itself and a precipitation input, and 2) the two linear regressors with an additional interaction term. Results of the coefficients fitted on the 3-year global SMAP and GPM data show that the unified regression method can reproduce or mimic the SSM characteristics found by several recent studies, including the SSM exponential decay rate, the fraction of precipitation retained in surface soil layer, and the effective depth of hydrologic storage. Additionally, including the interaction regressor provides a novel way to derive the sensitivity of infiltration/runoff partition process to antecedent SSM level without the need for streamflow observation data. Comparing these SMAP/GPM regression with comparable results from a model-based global SSM dataset provides new insight into the suitability of model structure and parameterization. In particular, relative to the satellite data, the physically-based model retains moisture longer in the top layer, under-predicts the sensitivity of runoff infiltration partitioning to top-layer soil moisture, and exhibits less spatial variation in SSM dynamics. This study demonstrates that data-driven methods are capable of recovering useful process-level insight from SMAP SSM retrievals and informing process representation in hydrologic models.