Empirical analysis and modeling of errors of atmospheric profiles from GPS radio occultation

摘要

The utilization of radio occultation (RO) data in atmospheric studiesrequires precise knowledge of error characteristics. We present results ofan empirical error analysis of GPS RO bending angle, refractivity, drypressure, dry geopotential height, and dry temperature. We find very goodagreement between data characteristics of different missions (CHAMP,GRACE-A, and Formosat-3/COSMIC (F3C)).In the global mean, observational errors (standard deviation from "true"profiles at mean tangent point location) agree within 0.3% in bendingangle, 0.1% in refractivity, and 0.2 K in dry temperature at allaltitude levels between 4 km and 35 km.Above 35 km the increase of the CHAMP raw bending angle observationalerror is more pronounced than that of GRACE-A and F3C leading to alarger observational error of about 1% at 42 km. Above ≈20 km,the observational errors show a strong seasonal dependence at highlatitudes. Larger errors occur in hemispheric wintertime and are associatedmainly with background data used in the retrieval process particularlyunder conditions when ionospheric residual is large. The comparison betweenUCAR and WEGC results (both data centers have independent inversionprocessing chains) reveals different magnitudes of observational errors inatmospheric parameters, which are attributable to different backgroundfields used. Based on the empirical error estimates, we provide a simpleanalytical error model for GPS RO atmospheric parameters for the altituderange of 4 km to 35 km and up to 50 km for UCAR raw bending angle andrefractivity. In the model, which accounts for vertical, latitudinal, andseasonal variations, a constant error is adopted around the tropopauseregion amounting to 0.8% for bending angle, 0.35% for refractivity,0.15% for dry pressure, 10 m for dry geopotential height, and 0.7 Kfor dry temperature. Below this region the observational error increasesfollowing an inverse height power-law and above it increasesexponentially.For bending angle and refractivity we also include formulations for errorcorrelations in order to enable modeling of full error covariance matricesfor these primary data assimilation variables.The observational error model is the same for UCAR and WEGC data but due tosomewhat different error characteristics below about 10 km and above about20 km some parameters have to be adjusted. Overall, the observationalerror model is easily applicable and adjustable to individual errorcharacteristics.