A General Framework for Global Retrievals of Trace Gases From IASI: Application to Methanol, Formic Acid, and PAN

Retrieving concentrations of minor atmospheric trace gases from satellite observations is challenging due to their weak spectral signature. Here we present a new version of the ANNI (Artificial Neural Network for Infrared Atmospheric Sounding Interferometer, IASI) retrieval framework, which relies on a hyperspectral range index (HRI) for the quantification of the gas spectral signature and on an artificial feedforward neural network to convert the HRI into a gas total column. We detail the different steps of the retrieval method, especially where they differ from previous work, and apply the retrieval to three important volatile organic compounds: methanol (CH3OH), formic acid (HCOOH), and peroxyacetyl nitrate (PAN). The comparison of the retrieved columns with those from an optimal estimation inversion retrieval shows an overall excellent agreement: differences occur mainly when the sensitivity to the target gas is low and are consistent with the conceptual differences between the two approaches. We present retrieval examples over selected regions, comparison with previously developed products, and the global seasonal distributions including the first global distributions of PAN on a daily basis. The ANNI retrieval has been carried out on the whole time series of IASI observations (2007-2018), so that currently over 10years of twice-daily global CH3OH, HCOOH, and PAN total column distributions have been produced. This unique data set opens avenues for tackling important questions related to sources, transport, and transformation of volatile organic compounds in the global atmosphere. Plain Language Summary Volatile organic compounds are atmospheric trace gases of various origins (e.g., biogenic, anthropogenic, and biomass burning). For many of them, their sources, removal processes, and impacts on atmosphere and climate are not well understood. One major reason for this is the paucity of measurements of their concentration. In this study, we describe the innovative ANNI (Artificial Neural Network for Infrared Atmospheric Sounding Interferometer, IASI) method that allows us to calculate here the abundance of three important volatile organic compounds (methanol, formic acid, and peroxyacetyl nitrate) from infrared measurements of the Earth's surface and atmosphere performed by the IASI satellite instrument. The method consists of two steps. First, the spectral signature, associated with the trace gas of interest, is extracted from the observed spectrum. The second step relies on an artificial neural network to transform the observed signature into a gas abundance. This technique provides a high level of sensitivity and accuracy compared with conventional physical methods. Furthermore, its computational efficiency makes it possible to readily process all the measurements (over 1,200,000/day) recorded since the launch of IASI in 2006. The result is a unique day-to-day global picture of the three gases, which represents a benchmark for tackling their uncertainties.