Near-Real-Time Aerosol Retrievals Via Neural Networks: Application to OMPS Limb Profiler Measurements

During major volcanic and wildfire activity, rapid processing of satellite measurements is critical to coordinate public and scientific responses. However, processing satellite measurements in near-real-time (NRT) can be challenging for computationally expensive retrieval approaches, as this may require reducing the number of wavelengths considered in NRT processing or the introduction of additional approximations to reduce runtime at the cost of accuracy. Alternatively, given a sufficient data set of traditional retrievals, a surrogate modeling approach via neural networks (NN) can offer NRT processing capabilities without the simplifications required for traditional methods. We demonstrate our methodology using retrievals of aerosol extinction profiles at six wavelengths from measurements by the Ozone Mapping and Profiler Suite (OMPS) Limb Profiler instrument on board the Suomi-NPP satellite. We compare our NN approach to the existing OMPS LP aerosol data product for recent years, including the Australian wildfire events in 2019-2020 and the Hunga Tonga-Hunga Ha'apai eruption in 2022, and discuss limitations of our approach.