Towards Locally Relevant Global Soil Moisture Monitoring Leveraging Remote Sensing and Modeling for Water Resources Applications

Abstract

Accurate and detailed soil moisture estimates can critically shape cross-sectoral water resources decision-making. From local to regional scales, monitoring of agricultural water demands, droughts, floods, landslides, and wildfires can benefit from high-resolution soil moisture information. However, soil moisture highly varies in space and time, and as a result, it is challenging to obtain detailed information at the stakeholder-relevant spatial scales. This dissertation leverages advances in satellite remote sensing, hyper-resolution land surface modeling, high-performance computing, and machine learning to bridge this data gap. Chapter 2 introduces a novel cluster-based Bayesian merging scheme that combines NASA's SMAP satellite observations and hyper-resolution land surface modeling for obtaining satellite-based surface soil moisture retrievals at an unprecedented 30-m spatial resolution. This approach's scalability and accuracy are demonstrated in Chapter 3 by introducing SMAP-HydroBlocks, the first satellite-based surface soil moisture dataset at a 30-m resolution over the United States (2015-2019). Using this dataset, Chapter 4 assesses the multi-scale properties of soil moisture spatial variability and the persistence of this variability across spatial scales. This analysis maps where detailed information is critical for solving water, energy, and carbon scale-dependent processes and how much variability is lost when data is only available at coarse spatial scales. Using machine learning, Chapter 5 demonstrates the value of high-resolution soil moisture for drought monitoring and crop yield prediction at farmer's field scales (250-m resolution). This dissertation provides a novel pathway towards global monitoring of water resources' dynamics at locally relevant spatial scales.