Optical and Compositional Biases in ECMWF/CAMS and POLDER/GRASP: Lessons from Aerosol Products Comparison

As the MODIS era concludes, new missions like 3MI/EPS-SG and PACE/NASA are now delivering extensive multi-angular, multi-viewing polarimetric data on aerosols. This shift necessitates an evolution in chemistry-transport models (CTMs). Specifically, the successful assimilation of sophisticated aerosol optical properties into the European Centre for Medium-Range Weather Forecasts (ECMWF) CAMS model depends on precise microphysical definitions. This study addresses the existing microphysical inconsistencies that currently prevent the seamless assimilation of remote sensing observations into CTMs.

We compare two generations of the CAMS model using observations from 2008. The first is the Cy42R1 Reanalysis, which incorporates MODIS Aerosol Optical Depth (AOD) at 550 nm through assimilation. The second is the Cy49R2 Forecast, an unconstrained forward simulation. Our analysis uses aerosol properties retrieved from both POLDER measurements and AERONET ground-based data, employing the GRASP algorithm. To ensure a consistent and fair comparison with CAMS assumptions, we adopt a chemical component approach. This involves decomposing the total aerosol loading into its specific components—Black Carbon (BC), Organic Matter (OM), Dust (DU), Sulphate (SU), and Sea Salt (SS)—and fixing their refractive indices and size parameters during the retrieval process.

The CAMS model consistently underestimates AOD compared to both POLDER/GRASP and AERONET, a negative bias present in both its reanalysis and forecast products. Analysis of the Ångström Exponent indicates that the model frequently miscategorizes fine and coarse mode particles. This confusion results in a significant negative bias in CAMS's coarse mode AOD. Additionally, the Single Scattering Albedo (SSA) in CAMS lacks the spectral and spatial variability evident in the POLDER/GRASP retrievals. Further analysis reveals that optical discrepancies in the model's performance are rooted in chemical component-specific errors. The model significantly underrepresents the total column volume concentrations of fine-mode aerosols, especially BC and OM. Furthermore, modeled volume concentrations of SU and SS are negligibly low compared to observational data. For DU, a distinct shift in model strategy is apparent: the older Cy42R1 version underestimates DU volume concentration, while the newer Cy49R2 overestimates it, indicating a transition towards modeling coarser particle emissions. Validation against AERONET data confirms that POLDER/GRASP retrievals offer a reliable benchmark for these comparative assessments.

Current CTMs and retrieval algorithms evidently operate under differing microphysical assumptions. This strongly implies underlying inaccuracies in the assumed aerosol size distributions and refractive indices within these models. The improved DU representation in Cy49R2 is a step forward, but the persistent underestimation of other components limits the model's application for radiative forcing calculations. The subsequent essential step involves harmonization. It is necessary that ECMWF update CAMS aerosol microphysics to incorporate effective radii, size distribution, and refractive indices consistent with state-of-the-art polarimetric retrievals. Establishing a unified definition for aerosol components will enable the next generation of CAMS reanalysis to effectively assimilate data from 3MI and PACE, consequently mitigating uncertainties in global aerosol forcing.