Bridging Particle-Resolved and Global Scales: Online Emulation of Aerosol Mixing State in E3SMv3 and Its Impacts on CCN and Cloud Droplet Number

Le résumé fourni par la source

Aerosol mixing state, the distribution of chemical species among individual particles, influences hygroscopicity, cloud condensation nuclei (CCN) activity, and droplet activation. Most global models assume complete internal mixing within each aerosol mode. We present the online integration of a machine-learning mixing state emulator into the Energy Exascale Earth System Model version 3 (E3SMv3). The XGBoost-based emulator, trained on particle-resolved PartMC-MOSAIC simulations, predicts two entropy-based indices at each time step, quantifying the mixing of black carbon (BC) with all other accumulation-size species and between hydrophobic and hydrophilic species. Each index is mapped to a two-subpopulation representation of the accumulation mode, with distinct hygroscopicities, for the activation calculation. In AMIP-style diagnostic and interactive simulations, the emulator diagnoses substantially more external mixing than the default assumption, reducing the surface-mean BC index from ~0.90 to ~0.60 and the hydrophobic index from ~0.95 to ~0.41. The BC mixing state has negligible CCN impact (<0.1%), whereas the hydrophobic–hydrophilic mixing state reduces global-mean CCN by ~5.5%, exceeding 10–20% over dust-influenced regions. A semi-quantitative comparison shows the emulator moving the BC coating ratio toward single-particle observations at all 22 sites, with error reductions of 11–88%. Against reanalysis-derived CCN, the emulator reduces the bias where the internal-mixing assumption most overestimates aerosol hygroscopicity and improves the spatial correlation. The interactive simulations shift regional in-cloud droplet number distributions toward MODIS retrievals, most coherently over the Southern Ocean and marine stratocumulus regions. Mixing state emulators trained on particle-resolved simulations thus offer a transferable framework for improving aerosol representation in global models.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.