Evaluation of the ROMEX datasets-Bias and uncertainty estimates

The Radio Occultation Modeling Experiment (ROMEX) is a loosely organized international program to test the impact of radio occultation (RO) observations on numerical weather prediction (NWP) forecasts. An unprecedented number of RO observations (30,000-40,000 profiles per day) has been collected for September-November 2022 and are available free of charge for testing in NWP models. This unique, high-resolution RO dataset also affords the opportunity for scientific studies on atmospheric phenomena such as tropical cyclones, atmospheric rivers, upper troposphere/lower stratosphere structure, gravity waves, and the planetary boundary layer. The ROMEX dataset contains RO data from 13 public-sector and commercial missions, obtained from different satellites and instruments in various orbits and processed by different centers. The ROMEX dataset’s heterogeneity and widespread use requires that it be carefully and thoroughly evaluated. This evaluation will be useful for data providers, processors, and users. An advantage of the ROMEX dataset is that a large number of independent scientists and evaluators are considering the same datasets, allowing for easier and more robust comparisons of processing methods and results.

We have estimated the bias and uncertainty (random error statistics) of the 13 ROMEX missions, with a focus on the three largest ROMEX datasets-Spire, Yunyao, and COSMIC-2, which together comprise approximately 78% of the ROMEX data. To eliminate small differences associated with different processing centers, we use the UCAR-processed data in our evaluations. We use multiple datasets for evaluation, including ERA5 and JRA-3Q reanalyses, short-range GFS and ECMWF forecasts, and radiosondes. We used the three-cornered hat method to estimate the uncertainties (random error statistics) and estimated the biases in a number of different ways, including RO-to-RO, RO-to-model, and RO-to-radiosonde comparisons. Our results indicate that the RO data from all missions have similar biases and uncertainties between about 8 and 30 km MSL (350 hPa to 5 hPa). They show that RO biases are extremely small in this “golden zone,” likely under 0.2%, confirming the common assumption that RO observations are unbiased. Below 8 km and above 30 km there are differences that can be important in NWP assimilation and research studies. The overall conclusion is that all the ROMEX datasets are high quality and should be useful for assimilation in NWP models, climate monitoring, and other scientific studies and operations.