This work investigates the capability of machine learning models, trained with space and ground GNSS observations, to produce fields of refractivity and IWV that describe the spatiotemporal morphology of atmospheric rivers (ARs) and quantify moisture associated with them to a degree sufficient for atmospheric studies. The reconstructed fields can be used to monitor ARs. It studies what LEO radio occultation (RO) constellation is appropriate to quantify the structure, location and timing of ARs.

This work investigates whether machine learning (ML) can offer an alternative to existing methods to map radio occultation (RO) products, allowing the extraction of information not visible in direct observations. ML can further improve the results of Bayesian interpolation, a state-of-the-art method to map RO observations. The results display improvements in horizontal and temporal domains, at heights ranging from the planetary boundary layer up to the lower stratosphere, and for all seasons.

We developed a new efficient algorithm leveraging orbital dynamics to collocate radio occultation soundings with microwave radiance soundings. This new algorithm is 99 % accurate and is much faster than traditional collocation-finding approaches. Speeding up collocation finding is useful for calibrating and validating microwave radiometers and for data assimilation into numerical weather prediction models. Our algorithm can also be used to predict collocation yield for new satellite missions.