Evaluation of CMIP6 model simulations of PM2.5 compositions in China

The radiative effects of various PM2.5 compositions play an important role in global climate change. The latest Coupled Model Interomparison Project Phase 6 (CMIP6) offers simulations of PM2.5 compositions from multiple models. Yet there remains little understanding of the model performance on PM2.5 compositions over China, due in part to lack of observational data. Here, we evaluate near-surface concentrations of PM2.5 and its main PM2.5 compositions (OA, BC, sulfate, nitrate and ammonium) simulated by fourteen CMIP6 models. For this purpose, we collect observational data between 2000 and 2014 for total PM2.5 from a satellite-based dataset and for PM2.5 compositions from the literature. Our study shows that the total PM2.5 concentrations, when calculated from outputted PM2.5 compositions based on a specified formulas from the literature (Turnock et al. 2020), of each model have a peak (between 28 and 96ppb) over eastern China, consistent with the satellite-based dataset (between 32 and 84ppb). However, the total PM2.5 concentrations outputted directly from six models (CESM2-WACCM, EC-Earth3-AerChem, GFDL-ESM4, MRI-ESM2-0, NorESM2-LM and NorESM2-MM) present a false high-value center in western China. The directly outputted total PM2.5 concentrations from eight models are generally underestimated, and show a too strong increasing trend over 20002014 by up to 3.4yr-1 in the areas of high anthropogenic emissions such as eastern and central China. Compared with the observational data, the OC concentrations in each model are underestimated, although central regional mean concentrations of HadGEM3-GC31-LL (11.8ppb) and UKESM1-0-LL (11.7ppb) are closer to the observations (13.0ppb). The simulated BC concentrations are less underestimated in eastern and central China. The summertime concentrations of nitrate are underestimated in all models, while EC-Earth3-AerChem in wintertime (14.7ppb) is closer to the observations (15.8ppb). Of the five models that include explicit ammonium simulations, the results in four models (EC-Earth3-AerChem, GFDL-ESM4, GISS-E2-1-G and GISS-E2-1-H) are pretty close, whereas CESM2-WACCM shows lower concentration. Our study sheds light on the performance of climate models in simulations of PM2.5 and its compositions over China.