Using neural networks for near-real-time aerosol retrievals from OMPS Limb Profiler measurements

Abstract

Among aerosol characterization methods, limb scattering measurements provide both near-global coverage and information about how aerosol is vertically distributed through the atmosphere. Near-real-time retrievals of aerosols produced by volcanic eruptions are particularly important for aviation safety, but the radiative transfer modeling of scattering processes performed by traditional retrieval methods can be too computationally expensive for near-real-time applications without simplifying assumptions, depending on the instrument and available computational resources. Here we present a near-real-time approach based on neural networks (NNs) for aerosol retrievals from the Ozone Mapping and Profiler Suite’s Limb Profiler (OMPS LP) instrument aboard the Suomi National Polar-orbiting Partnership satellite. We find it is at least 60 times faster than the current operational code and on average achieves agreement within 20 % at most altitudes and latitudes with sensitivity and non-negligible aerosol abundances. We also apply our trained NNs to measurements of the recent Shiveluch and Ruang eruptions from NOAA-21’s OMPS LP and find results consistent with the operational retrieval algorithm, indicating our methodology generalizes to future iterations of the same instrument without retraining the NNs.