Soil Moisture Estimation From Spire GNSS-Reflectometry Reflectivity: Semi-Empirical and Linear Regression Methods

This study develops semi-empirical and linear regression algorithms to estimate near-surface soil moisture (SM) using reflectivity observations of Global Navigation Satellite System reflection (GNSS-R) signals collected by Spire Global commercial satellites. The semi-empirical method employs a forward model of GNSS-R reflectivity considering soil dielectric properties, vegetation attenuation, and surface roughness effects. As direct roughness measurements are unavailable globally, we obtain effective roughness attenuation maps using Cyclone GNSS (CYGNSS) reflectivity and SMAP SM and vegetation optical depth (VOD) data. Systematic biases between Spire and CYGNSS observations are also characterized and corrected. Daily gridded corrected Spire reflectivity observations at a 36-km resolution are used to invert soil dielectric permittivity and then SM. Linear regression parameters are also determined by correlating corrected Spire reflectivity measurements to SMAP SM data for each grid. Both algorithms are applied to six months of Spire data from late Jan to Jul 2024 and evaluated over Australia. The linear regression approach achieves better overall agreement with SMAP (correlation coefficient: 0.85, root mean squared difference (RMSD): 0.05 cm³/cm³) compared to the semi-empirical method (correlation: 0.64, RMSD: 0.07 cm³/cm³). The semi-emprical method demonstrates relatively better accuracy over sparsely vegetated regions with RMSD smaller than 0.06 cm³/cm³. Both methods show degraded performance over dense vegetation. Improved characterization of vegetation and surface roughness effects is essential to enhance retrieval accuracy. This work contributes to positioning GNSS-R as a complementary technology for independent SM observations.