Use of Kalman Filtering to Improve CYNSS Air-Sea Interaction Applications

The Cyclone Global Navigation Satellite System (CYGNSS) provides much improved all-weather sampling of ocean surface wind speeds throughout the global tropics and subtropics. The orbital configuration of the eight-member constellation also significantly enhances the ability to resolve sub-daily variability. Ocean surface winds are critical to many atmosphere-ocean interactions including the exchanges of momentum, heat and moisture. While CYGNSS is capable of retrieving winds in all-weather conditions, uncertainties are dependent upon the underlying characteristics of the retrieval scene (e.g. fetch, wave conditions) and observables (e.g. range corrected gain) and sampling can be impacted during high roll-angle orbital maneuvers. Thus, variable data quality and availability can significantly impact the utility of CYGNSS observations for process studies.

Kalman filtering (KF) is a well-known framework that can be used to incorporate measurement uncertainty information together with a state-transition model to perform sequential optimal estimation. We have developed a KF-based approach to locally estimate the time-varying underlying ocean surface wind state that leverages CYGNSS observations and dynamical tendencies from a forecast model to result in reduced error and gap-free surface wind estimates. These estimates allow for assessing the impacts of CYGNSS winds on surface air-sea interaction especially in those cases where sampling variability would otherwise preclude investigation. This work will focus specifically on the challenges of evaluating the tropical surface latent and sensible heat fluxes using CYGNSS observations. The development of the KF state-estimation, including estimation of measurement and process uncertainties, will be discussed. Results will highlight the impact of improved CYGNSS observations on surface heat flux estimates especially as related to tropical precipitating conditions where current passive microwave imager observations are typically unreliable or unavailable.