Developing new methods for Satellite Remote Sensing of Soil Moisture

Abstract

The Optical Trapezoid Model (OPTRAM) has been widely used for high-resolution mapping of surface soil moisture (top 0-5 cm) using optical satellite data. Despite its strengths, it faces limitations in parameter calibration, particularly in defining the wet edge in areas with high moisture content, and in applying a single set of parameters across heterogeneous landscapes. Moreover, Shortwave transformed reflectance used as input in this model is affected by variables not related to soil moisture and all these limitations introduce uncertainties in the model outputs. This dissertation addresses OPTRAM’s limitations through three methodological advancements aimed at improving the accuracy, and consistency of OPTRAM-derived soil moisture estimates. First, a new variant of OPTRAM is introduced to enhance its ability to distinguish not only soil moisture but also surface water bodies such as lakes and rivers, as well as saturated pixels. Using Landsat-8 surface reflectance data across the Central Valley, California, the modified model demonstrated improved performance in differentiating land and water pixels and reduced sensitivity to user-defined parameters. Quantitative comparison showed that the new variant achieved higher consistency in soil moisture estimates, with R² values between parameter sets ranging from 0.67 to 0.76, compared to 0.47 to 0.52 for the original OPTRAM. Second, a landcover-specific calibration approach (OPTRAM-LS) is proposed, in which separate parameter sets are derived for different land cover types, as opposed to a uniform calibration for the entire study region (OPTRAM-CV). Leveraging Sentinel-2 surface reflectance and the Cropland Data Layer dataset, 20-m resolution soil moisture maps were generated for the Central Valley. Comparative validation against in situ observations and SMAP-Hydro Blocks (a 30-m soil moisture dataset) showed that OPTRAM-LS significantly outperformed OPTRAM-CV, reducing the average root mean square error from 0.09 to 0.05 m³m⁻³. Finally, a novel method called Microwave-Guided OPTRAM (MG-OPTRAM) has been developed to further improve soil moisture retrieval by fusing optical and microwave remote sensing data. MG-OPTRAM integrates Soil Moisture Active Passive (SMAP) observations at 9-km resolution into OPTRAM calibration, enabling us to extend the soil moisture estimations beyond SMAP’s operational period (2015–present) back to 2000, and downscale SMAP data to a finer resolution (500-m) using Moderate Resolution Imaging Spectroradiometer (MODIS) observations. Application in the Central Valley demonstrated that MG-OPTRAM reliably reconstructs SMAP-like soil moisture. Validation against in situ and radar-derived soil moisture data confirmed its robustness at coarse and high resolution, although some uncertainty persists at finer resolutions due to model simplicity and subpixel heterogeneity. Together, the three methodological advancements represent a connected and progressive refinement of the OPTRAM framework. By adopting the enhanced OPTRAM formulation, shifting from region-wide to land cover-specific calibration, and advancing to SMAP pixel-level calibration as well as filtering the soil moisture-related reflectance, MG-OPTRAM emerges as the most evolved version of the model. It integrates the strengths of the previous improvements and fulfills the dissertation’s goal of developing an accurate, high-spatial-resolution soil moisture dataset spanning a long historical period. Based on these findings, this dissertation recommends future research directions including: (1) incorporating environmental covariates such as precipitation and soil texture; (2) exploring Synthetic Aperture Radar (SAR)-based calibration potential; and (3) enhancing the model framework to better identify soil moisture-related reflectance characteristics. These improvements can further strengthen the performance and accuracy of soil moisture estimated by OPTRAM.