Deep Learning-Based GNSS-R Global Vegetation Water Content: Dataset, Estimation, and Uncertainty

Vegetation water content (VWC) is a crucial parameter for understanding vegetation dynamics and hydrological cycle on Earth. With rapid climate changes in recent years, monitoring VWC with high spatiotemporal coverage on a global scale is of paramount importance. Yet, traditional in situ measurements are constrained in remote and densely vegetated regions. Additionally, existing spaceborne remote sensing methods face challenges due to poor cloud penetration capabilities, soil moisture interference, and inadequate temporal resolution. Spaceborne global navigation satellite system reflectometry (GNSS-R) has demonstrated promising potential to overcome these limitations in vegetation monitoring. In this study, we propose a scheme for deep learning-based GNSS-R VWC assessment, leveraging a rapidly growing amount of GNSS-R data with an unprecedented sampling rate. We introduce a triplet dataset, which consists of measurements from the cyclone GNSS (CYGNSS), global land data assimilation system (GLDAS), and soil moisture active passive (SMAP), spanning over three years. Validation is performed using several benchmark models with the proposed dataset. Furthermore, the models' predictive uncertainty is quantified with Monte Carlo (MC) dropout technique to provide a trustworthy representation of estimations. Experimental evaluation of the models demonstrates good consistency between the estimated VWC and ground truth, with a minimum root mean square deviation (RMSD) of 1.0988 kg/m<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathbf {^{2}}$</tex-math></inline-formula> and a bias of 0.002kg/m<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\mathbf {^{2}}$</tex-math></inline-formula> over a twelve-month test period. Moreover, a daily global VWC estimation is achieved through the proposed pipeline, filling the gaps of current products and enabling rapid measurements with enhanced temporal availability. We will make the proposed dataset publicly available.