Deep Generative Regression Models For Soil Moisture Retrieval From GNSS-R Observations

Soil moisture content (SMC) is a crucial component in both the water and carbon cycles, significantly influencing weather and climate. Over recent years, numerous remote sensing platforms have been established to track soil moisture. Leading missions, such as NASA’s SMAP and the ESA’s SMOS, offer global-scale coarse-resolution soil moisture monitoring using active and passive microwave imaging at a moderate revisit frequency on the scale of days. Regrettably, current products’ spatial and temporal resolution is often insufficient for many applications. Global navigation satellite system reflectometry (GNSS-R) is a prime example of a signal-of-opportunity (SoOP) which has been shown to be highly sensitive variations of SMC. An important source of such signals is the NASA Cyclone GNSS (CYGNSS) mission which consists of eight low-orbit observatories, each utilizing a zenith antenna to receive direct global positioning system (GPS) signals, as well as two nadir-looking antennas to receive reflected GPS signals, that are used to derive a delay-Doppler map (DDM). In order to extract SMC from DDM signals, the traditional approach involves inverting parameterized forward models which account for aspects like dielectric properties of the soil incidence angle among others. Despite their grounding in physical processes, these models typically involve simplified assumptions and demonstrate increased sensitivity to the values of different parameters. To address this challenge, a new line of research tries to address this challenge by utilizing machine learning models and treating the problem as that of supervised regression e.g.¹.