AEROCOM/AEROSAT: an intercomparison of AAOD & SSA in model and satellite data

<p>Absorbing aerosol has the capacity to warm the climate, but the amount of warming is highly uncertain. AAOD (Absorptive Aeorosl Optical Depth) is an optical measure of the abundance of this absorbing aerosol, comprising mineral dust, black and brown carbon and can be retrieved from satellite measurements providing an almost global view on absorbing aerosol.</p><p>In this study we evaluate AEROCOM models with satellite observations of AAOD and SSA (Single Scattering Albedo) and interpret the discrepancies. Over source regions, diversity in model AAOD is mostly due to emissions even though models employ different assumptions regarding the imaginary refractive index. On the one hand this suggests emissions to be a major error source, on the other hand it suggests that the AEROCOM ensemble as a whole may have a bias with regards to MAC (Mass Absorption Coefficient). We show that in the models AAOD scales almost linearly with emissions (either black carbon or dust) and this allows the use of observations as a constraint.<span>  </span>In contrast, model diversity in AOD is shown to depend in almost equal measure on emissions, lifetimes and MECs (Mass Extinction Coefficient). We also analyse mineral dust and black carbon lifetimes by considering the contrast in AAOD over source regions and over outflow regions, and again provide observations constraints.</p><p>While the older Phase II models generally underestimate AAOD, Phase III models tend to straddle the observations, with some models over-estimating and other models underestimating AAOD. Emissions seem to be the driving factor in this difference. The amount of diversity is larger in the Phase III than Phase II models.</p><p>This study was conducted using four satellite datasets of AAOD and SSA. These datasets were extensively evaluated with AERONET. Dearth of observations prevents global assesment of the satellite retrievals. However, we show that model evaluation is relatively independent of the chosen dataset, even though we identify significant biases between the datasets.</p>