Observational Constrained Attribution of Regional Aerosol Simulation Biases in the AerChemMIP models

     Regional aerosol simulation biases in climate models have been noted since the CMIP5 era. The biases can cause noticeable error in the radiative forcing estimations. In this research, we investigate the aerosol optical depth (AOD) biases over China from 2002 to 2015 in nine climate models that participate the Aerosol and Chemistry Model Intercomparison Project (AerChemMIP) of CMIP6. The AerChemMIP ensemble mean is high biased over four populated regions in winter and low biased in two populated regions compared to the MODIS satellite retrievals. The patterns of model biases were persistent over years. Large inter-model spread is found in the high AOD regions. We decompose AOD to the product of emission rate, lifetime and mass extinction coefficient such that the AOD biases can be attributed to the errors of each term and their cross error term. The error of each term is analyzed by first regressing to several observable predictors, such as precipitation, Angström exponent, and relative humidity, followed by constraining the predictors by observational or reanalysis data. The results show that error due to emission dominates for many models, followed by lifetime and MEC errors. Furthermore, we argue that for regional analysis, due to imbalance between emission and removal fluxes, the removal/emission ratio should be further constrained by observations. This study provides a diagnosis for climate models to improve their simulation in aerosol loading on regional scale by optimizing the modeling of meteorology as well as aerosol properties and life cycle.