Evaluation of ECMWF IFS-AER (CAMS)operational forecasts during cycle 41r1 - 46r1 withcalibrated ceilometer profiles over Germany

Abstract. Aerosol forecasts by the European Center for Medium Range Weather Forecasts (ECMWF) Integrated Forecasting System IFS-AER for years 2016–2019 (cycle 41r1 - 46r1) are compared to vertical profiles of particle backscatter from the Deutscher Wetterdienst (DWD) ceilometer network. The system has been developed in the Copernicus Atmosphere Monitoring Service (CAMS) and its precursors. The focus of this article is to evaluate the realism of the vertical aerosol distribution from 0.3 to 8 km above ground, coded in the shape, bias and temporal variation of the profiles. The common physical quantity, the attenuated backscatter β*(z), is directly measured and calculated from the model mass mixing ratios of the different aerosol types using the model's inherent aerosol microphysical properties. Pearson correlation coefficients of daily average simulated and observed vertical profiles between r = 0.6–0.8 in summer and 0.7–0.95 in winter indicate that most of the vertical structure is captured. It is governed by larger β*(z) in the mixing-layer and comparably well captured with the successive model versions. The aerosol load tends to be high-biased near the surface, be underestimated in the mixing layer and realistic at small background values in the undisturbed free troposphere. A seasonal cycle of the bias below 1 km height indicates that aerosol sources and/or lifetimes are overestimated in summer and pollution episodes not fully resolved in winter. Long-range transport of Saharan dust or fire smoke is captured and timely, only the dispersion to smaller scales is not resolved in detail. Over Germany β*(z) from Saharan dust and sea salt are considerably overestimated. Differences between model and ceilometer profiles are investigated using observed in-situ mass concentrations of organic, black carbon, SO4, NO3, NH4 and proxys for mineral dust and sea-salt near the surface. Accordingly, SO4 and OM sources as well as gas-to-particle partitioning of the NO3-NH4-system are too strong. The top of the mixing layer on average appears too smooth and few 100 m too low in the model. Finally, a discussion is included of the considerable uncertainties in the observations, the conversion from modeled to observed physical quantities, and from necessary adaptions of varying resolutions and definitions.