Impact of Increased ROMEX GNSS Radio Occultation Data on Forecast Performance in the Korean Integrated Model

Abstract. Global Navigation Satellite System Radio Occultation (GNSS RO) data can be assimilated to improve global numerical weather prediction (NWP). While simulation studies and observing system experiments have indicated benefits from increasing RO data volume, the extent to which large-volume RO assimilation improves forecasts in an operational NWP configuration has been less thoroughly quantified. As part of the international Radio Occultation Modeling Experiment (ROMEX), which assesses the forecast impact of GNSS RO observations, this study evaluates the effect of increased GNSS RO data volume on the forecast performance of the Korean Integrated Model (KIM), the operational global NWP model of the Korea Meteorological Administration (KMA). Four baseline experiments were conducted: NoRO (without GNSS RO observations), CTL (operational-level GNSS RO volume), EXP_20k (≈20 000 profiles d⁻¹), and EXP_35k (≈35 000 profiles d⁻¹). To investigate the sensitivity of the EXP_35k results to the assimilation configuration, additional experiments were conducted by adjusting the observation-error inflation factor and the refractivity coefficient k₁ (the dry-term coefficient in the refractivity equation). A comparison of NoRO and CTL results showed that the assimilation of GNSS RO observations in KIM generally improved temperature and geopotential height forecasts in the lower stratosphere and upper troposphere. With larger GNSS RO datasets, EXP_20k showed relatively consistent improvements across most regions and levels, whereas EXP_35k maintained the global-mean improvement in 100 hPa temperature forecasts but showed weaker improvement in the tropics at that level. It also produced a pronounced increase in tropical 500 hPa geopotential height root-mean-square error, clearest through the first three forecast days and weakening only gradually through day 5. These results indicate that increasing the RO data volume does not translate linearly into forecast improvement and that the response of the current KIM assimilation configuration depends on the region, pressure level, and forecast variable. Increasing the observation-error inflation factor mitigated part of the mid-level geopotential height degradation but also indicated that this change weakens the upper-level temperature improvement attributable to RO data. In the k₁ adjustment experiments, increasing k₁ in KIM produced more favorable results than did decreasing it for certain variables and levels but did not directly resolve the main degradation observed in EXP_35k. The analysis-increment diagnostics showed similar tropical mid-level increment statistics across the three experiments, suggesting that the degradation is more likely related to accumulated differences in the analysis field or the forecast error growth than to excessive per-cycle analysis adjustment. These results show that the forecast impact of large-volume GNSS RO data is sensitive to the observation-error configuration and observation-operator coefficient. They further highlight that maximizing the value of future high-density GNSS RO observing systems depends not only on increased observation availability but also on continued improvements in observation-error modeling, observation handling, and observation operators within data assimilation systems.