Fine particulate matter (PM2.5) has strong and adverse effects on the environment and human health. To estimate health risks and environmental impacts it is very important to know about current and prospective amounts of ground-level PM2.5 concentrations on regional scales. The in-situ station network in Germany is well developed but still provides only selective spatial information on air pollution. For a gapless monitoring additional data sets are required. Satellite data provides area-wide measurements of air pollutants and can depict their synoptic distribution in an adequate way. Chemical-transport models are moreover able to predict the amount and dispersion of aerosols in a very high temporal and spatial resolution, which makes them the key tool in monitoring air quality on regional and local scales. Modelling of aerosols is still very uncertain due to the complexity of accurately including aerosol properties and transfer processes, but also because of inaccurate emission data bases. In our study we use satellite data to produce detailed maps of PM2.5 distributions for Germany with the objective of using them as input for the air quality forecast system POLYPHEMUS/DLR. We want to improve the general performance of the model by adjusting the PM2.5 amounts in the model with observation data from satellites in terms of data assimilation. PM2.5 concentrations cannot be measured directly by satellites. This paper presents a semi-empirical linear regression approach to estimate ground-level PM2.5 concentrations using satellite observations of aerosol optical depth (AOD). The method was applied to different satellite sensor products, namely MODIS and SLSTR. For both sensors the resulting PM2.5 concentrations showed good correlations with in-situ station measurements with R-values of 0.83 for MODIS and 0.81 for SLSTR for the considered year 2018. Differences in the spatial coverage of the two satellite sensors induced us to combine the data sets to an ensemble product. We found major benefits using this ensemble, primarily regarding the data amount for the calculation of local mean values of PM2.5 concentrations. We could produce detailed maps of ground-level PM2.5 concentrations which can be used for the identification of high polluted areas, the monitoring of transnational pollution patterns and the localization of specific emission sources. The assimilation of the produced datasets into the air quality model POLYPHEMUS/DLR will be the next step in our study.