Upscaling soil moisture from point scale to field scale: Toward a general model

Abstract

Field-scale soil moisture measurements are valuable but rarely available because the resolution of most satellite soil moisture products is too coarse, while most in situ sensors provide only point-scale data. Previous upscaling approaches for such data are mostly site-specific, and none are suitable to upscale data from the thousands of stations in existing monitoring networks. To help fill this gap, this research aims to develop a more broadly applicable upscaling approach using data from the Marena, Oklahoma, In Situ Sensor Testbed and a cosmic-ray neutron rover. Rover survey data were used to measure average soil moisture for the ∼64-ha field on 12 dates in 2019–2020. Statistical modeling was used to identify the soil, terrain, and vegetation properties influencing the relationships between the field-scale rover data and point-scale in situ data from six monitoring sites. Site-specific linear upscaling models estimated the field average soil moisture with root mean squared error (RMSE) values ranging from 0.007 to 0.017 cm3 cm−3, but such models are not transferrable between sites. To create a more general model, Least Absolute Shrinkage and Selection Operator regression was used with a leave-one-out cross-validation approach to identify the key predictors for upscaling. The resulting parsimonious model required only the point-scale observations and sand content data and achieved RMSE values ranging from 0.006 to 0.031 cm3 cm−3 for the six monitoring sites. The texture-based model demonstrated reasonable accuracy and is a promising step toward a general model that could be broadly applied for upscaling point-scale in situ monitoring stations.