Evaluation and interpretation of modeling bias for biomass burning aerosols in AeroCom models

Biomass burning (BB) injects aerosols into the atmosphere and can thereby affect the earth climate and human health. Yet the modeling of BB aerosols exhibits significant bias. Here we present a comprehensive evaluation of AeroCom model simulations with satellite observations to understand such uncertainties. A total of 59 model runs using 17 models from three AeroCcom Phase III experiments (i.e., Biomass Burning emissions, CTRL2016, and CTRL2019 experiment) and 14 satellite products are involved. AOD (aerosol optical depth) at 550 nm wavelength during the fire season over three typical fire regions (Amazon, South Hemisphere Africa, and Boreal North America, or AMAZ, SHAF, and BONA) is the focus of our study, although we also consider AE and SSA from POLDER.

The 14 satellite products are shown to have quite substantial differences from AERONET observation. But we show that such differences have little impact on the model evaluation which is mainly affected by modeling bias. Through the comparison with POLDER observation, we found the modeled AOD are biased by -93% ~ 174% with most models showing significant underestimations even for the most recent modeling experiment (i.e., CTRL19). SHAF is among the three regions with the strongest underestimation in general. By scaling up the input emissions, such negative bias would be significantly reduced, which, however, has little impact on the day-to-day correlation between models and observations.

On top of the satellite-based model evaluation, we interpret the model diversity from the aspect of aerosol emissions, lifetime, and MEC (mass extinction coefficient), which are further linked with specific parameters in models. These three parameters cause similar levels of AOD diversity, which is quite different from the modeled aerosols during non-fire season when the contribution of lifetime is predominant. During the fire season, diversity caused by lifetime is strongly affected by local deposition; as a matter of fact, models tend to do quite poorly in simulating precipitation strength. Modeled MECs show significant correlations with aerosol wet-growth (which is known to be challenging to models) and AE (Angstrom Exponent) for some involved models. Comparisons with POLDER observed AE suggests some models tend to underestimate AE and thus MEC, which might be responsible for the overall AOD underestimation in certain models. Additionally, we show that model AOD biases correlate with satellite observed formaldehyde columns, suggesting SOA formation may be insufficiently captured by models.