Observation, analysis, and modeling of deep radio occultation signals: Effects of tropospheric ducts and interfering signals

GPS radio occultation (RO) signals are sometimes observed very deep in the Earth's shadow. To investigate these phenomena, one of the FORMOSAT-3/COSMIC RO receivers was set to track RO signals deep below the limb, down to a height of straight line −350 km on 5–6 October 2010. Analysis of the spectrograms revealed the existence of two types of signals below −200 km, RO signals induced by tropospheric propagation and interfering signals not transmitted by the occulted GPS. The RO signals induced by tropospheric propagation arrive from impact heights corresponding to inversion layers. Wave optics modeling of RO signals showed that deep signals exist when the refractivity gradient exceeds critical (super-refraction). The existence of such signals is a diffractional phenomenon, which offers a new quality control parameter to identify occultations that may be affected by super-refraction. This is important for RO data assimilation in weather models in the moist lower troposphere because assimilation of RO data affected by super-refraction is an ill-conditioned problem. Detection of the tropospheric ducts also may be useful for evaluation of radio wave propagation conditions. For infinitely horizontally extended ducts, the deep signals are extended in duration, have amplitudes of about 0.1%, and exist for only elevated ducts. For ducts of limited horizontal extension, the deep signals are shorter in duration, have amplitude of about 1%, and may exist for both elevated and surface ducts. The interfering signals were found in about half of occultations. Based on frequency modeling, in most cases, the interfering signal was identified with non-occulted GPS. The disturbance of retrieved bending angle induced by an interfering signal from a non-occulted GPS in a region of strong defocusing and significant spectral spread of RO signal was modeled and determined to be quite large, up to 10% of bending angle. However, the probability of occurrence of such interference (not estimated), generally, must be low. Removal of the interfering signal by high-pass filtering was tested, but it introduced a disturbance of bending angle of about the same magnitude as the interfering signal itself.