Retrievals of smoke particle properties using AirMSPI measurements acquired during FIREX-AQ field campaign

Abstract

An optimization-based approach is developed to retrieve smoke particle properties from the observations acquired by the Airborne Multiangle SpectroPolarimetric Imager (AirMSPI) during the Fire Influence on Regional to Global Environments Experiment-Air Quality (FIREX-AQ) field campaign. This framework employs a three-step optimization strategy to mitigate the discontinuities between independently retrieved image patches. Moreover, it integrates a deep-neural-network-accelerated radiative transfer model to enhance retrieval efficiency. The retrieved aerosol products are compared with the AERONET reference data and the retrievals from the enhanced MODIS Airborne Simulator (eMAS). Under clear-sky conditions, retrieved aerosol optical depth (AOD) agrees with AERONET data within approximately ±0.05. For the smoke plumes with moderate to high AOD, the spatial patterns retrieved from AirMSPI are consistent with those from eMAS retrieval. The best agreement is achieved by excluding the four most oblique viewing angles (i.e., ±65° and ±57°). Sensitivity tests indicate that using three to five AirMSPI angles within ±45.8° of nadir balances the mitigation of plane-parallel atmosphere modeling error while preserving the information richness of multi-angle observations of a three-dimensional plume. Moreover, single-view-angle retrievals, even when initialized with multi-angle retrieval results, still exhibit discrepancies relative to multi-angle retrieval. This is likely due to the coupled effects of the temporal evolution of smoke properties and the spatial inhomogeneity of smoke particles within the plume. The latter causes the multi-angle views to capture different portions of the plume and challenges the plane-parallel atmosphere model used in the retrieval