Analysis of the Diurnal Cycle in RO Data Using Bayesian Interpolation
Presenter:
Stephen S. Leroy
Atmospheric and Environmental Research
Jet Propulsion Laboratory, Caltech; Danish Meteorological Institute
Poster
Bayesian interpolation fits irregularly distributed data using basis functions without over-fitting. It is ideally suited to developing level 3 climatologies of global data which are sparse, non-uniform, and semi-random in distribution such as GNSS radio occultation (RO) data. We had previously developed an implementation of Bayesian interpolation for mapping RO data on a sphere using spherical harmonics as basis functions. It succeeded in mapping RO data without introducing a bias due to singularities in the RO sampling density and without reference to an atmospheric model. It produces an accurate error analysis and an inference of the effective horizontal spatial resolution of the climatology.
We will present an extension of Bayesian interpolation into the time domain in order to map the diurnal cycle using RO data. The basis set consists of spherical harmonics in (horizontal) space and sinusoids in time-of-day. The time-of-day can optionally be UTC time or solar time: dependence on UTC time is useful for the provision of level 3 climatologies for comparison to climate models; dependence on solar time is useful for testing tidal models of the upper atmosphere for measuring the amplitude of the migrating tides. Using COSMIC RO data we have determined the optimal regularizer to be used in analyzing the diurnal cycle, the regularizer being the smoothness condition that prevents the over-fitting of data. We will show that estimates of the error covariance matrix produced by the Bayesian mapping of the diurnal cycle are consistent with a bootstrap analysis of the sampling error. Finally, we will discuss likely research applications for the mapping algorithm. First and foremost, the outcome will be a test of models of the diurnal cycle in the upper air, particularly in the upper stratosphere where information on the diurnal cycle is sparse. There are already indications that the ECMWF model itself does not adequately represent in the migrating semidiurnal cycle in the upper stratosphere adequately enough for the removal of sampling error. Second, we will discuss the consequences of under-sampling the diurnal cycle by future GNSS RO constellations, which has clear implications for monitoring the climate using RO.
Poster in PDF:
