Comparison of different machine learning approaches for tropospheric profiling based on COSMIC-2 data

Abstract

Abstract Precise and reliable information on the tropospheric temperature and water vapor profiles play a key role in weather and climate studies. Among the sensors supporting the observations of the troposphere, one can distinguish the Global Navigation Satellite System Radio Occultation (RO) technique, which provides accurate and high-quality meteorological profiles. However, external knowledge about temperature is essential to estimate other physical atmospheric parameters. To overcome this constraint, I trained and evaluated four different machine learning models comprising Artificial Neural Network (ANN) and Random Forest regression algorithms, where no auxiliary meteorological data is needed. To develop the models, I employed 150,000 globally distributed (45°S–45°N) RO profiles between October 2019 and December 2020. Input vectors consisted of bending angle or refractivity profiles from the Formosa Satellite-7/Constellation Observing System for Meteorology, Ionosphere, and Climate-2 mission together with the month, hour, and latitude of the RO event. While temperature, pressure, and water vapor profiles derived from the modern ERA5 reanalysis and interpolated to the RO location served as the models’ targets. Evaluation on the testing data set revealed a good agreement between all model outputs and ERA5 targets, where slightly better statistics were noted for ANN and refractivity inputs. Vertically averaged root mean square error (RMSE) did not exceed 1.7 K for the temperature and reached around 1.4 hPa and 0.45 hPa for the total and water vapor pressures. Additional validation with 477 co-located radiosonde observations and the operational one-dimensional variational product showed slightly larger discrepancies with the mean RMSE of around 1.9 K, 1.9 hPa, and 0.5 hPa for the temperature, pressure, and water vapor, respectively. Graphical Abstract