Evaluation of the 15-year ROM SAF monthly mean GPS radio occultation climate data record

Abstract

We here present results from an evaluation of the Radio Occultation Meteorology Satellite Application Facility (ROM SAF) gridded monthly mean climate data record (CDR v1.0), based on Global Positioning System (GPS) radio occultation (RO) data from the CHAMP (CHAllenging Minisatellite Payload), GRACE (Gravity Recovery and Climate Experiment), COSMIC (Constellation Observing System for Meteorology, Ionosphere, and Climate), and Metop satellite missions. Systematic differences between RO missions, as well as differences of RO data relative to ERA-Interim reanalysis data, are quantified. The methods used to generate gridded monthly mean data are described, and the correction of monthly mean RO climatologies for sampling errors, which is essential for combining data from RO missions with different sampling characteristics, is evaluated. We find good overall agreement between the ROM SAF gridded monthly mean CDR and the ERA-Interim reanalysis, particularly in the 8-30km height interval. Here, the differences largely reflect time-varying biases in ERA-Interim, suggesting that the RO data record has a better long-term stability than ERA-Interim. Above 30-40km altitude, the differences are larger, particularly for the pre-COSMIC era. In the 8-30km altitude region, the observational data record exhibits a high degree of internal consistency between the RO satellite missions, allowing us to combine data into multi-mission records. For global mean bending angle, the consistency is better than 0.04%, for refractivity it is better than 0.05%, and for global mean dry temperature the consistency is better than 0.15K in this height interval. At altitudes up to 40km, these numbers increase to 0.08%, 0.11%, and 0.50K, respectively. The numbers can be up to a factor of 2 larger for certain latitude bands compared to global means. Below about 8km, the RO mission differences are larger, reducing the possibilities to generate multi-mission data records. We also find that the residual sampling errors are about one-third of the original and that they include a component most likely related to diurnal or semi-diurnal cycles.