A probabilistic approach for the retrieval of mineral dust properties from infrared spaceborne observations

Mineral dust has far reaching impact on atmospheric dynamics and the biosphere, altering both the hydrological and the carbon cycle, as well as on human health and the economy. Scattering and absorption effects of dust are enhanced in the terrestrial infrared due to the Si-O resonance bands and thus dust remote sensing with infrared sounders such as IASI (Infrared Atmospheric Sounding Interferometer) is well motivated. This work pertains to current updates on the Infrared Mineral Aerosol Retrieval Scheme (IMARS) on IASI hyperspectral data. IMARS algorithm estimates probabilistically the atmospheric state with respect to desert dust and ice clouds based on simulations of the observed signal for various dust and ice cloud properties.

A preprocessor compresses IASI radiance data in three pseudo-channels exploiting their high redundancy and accounting for the unequal distribution of their information content. From these, four distinct  brightness temperatures differences (BTD) are formed with respect to these channels, which reflect the spectral variation of dust or cloud or surface signal. By varying dust particle size distributions (s), mineralogical compositions (c), infrared optical depths (τ) and layer heights(h) we construct a simulation database constituting 6000 brightness temperature difference sets. The deviation of simulated and observed BTDs by means of a Gaussian metric yields a probability distribution function (PDF), with which the state vector as well as its probability and uncertainty are determined. The calculation of aerosol optical depth (AOD), dust layer temperature, dust effective radius, and other dust properties follows from correspondingly adding contributions (s, c, τ, h) weighted by this PDF. The ice cloud retrieval is realized in the same manner using cloud optical properties from a range of parameterizations found in literature. The distinction between dust and cloud is generally based on quality flagging in terms of the emission temperature relative to the approximated surface temperature and its expected range, and the (dust/cloud) probabilities and uncertainties with stricter criteria for the so-called dust belt. Finally, the IMARS pixel-wise product offers four levels of quality filtering in terms of probabilities, uncertainties, quality flags and information entropy.

Preliminary evaluation of IMARS AOD against AERONET coarse mode AOD obtained by the Spectral Deconvolution Algorithm, is done with data of mild level quality filtering restricted in the dustbelt and using Barnes objective analysis. Results show an overall moderate correlation (and small bias) and a stronger one for focused AERONET stations.