To assess and mitigate PM2.5 pollution, it is imperative to accurately characterize its spatiotemporal variability and distribution. As station measurements are spatially selective, additional data sources such as satellite observations or chemical transport models (CTMs) are indispensable for providing essential complementary information. With the increasing accessibility of high-quality reanalysis datasets on air pollutants from CTMs, the use of satellite-based observations with their limited spatial and temporal availability is more and more put into question. Model simulations present a clear advantage concerning spatiotemporal continuity. However, how reliable are these model simulations and how does the information they provide differ from satellite-derived data? By applying a random forest (RF) approach we derive PM2.5 concentrations for Germany based on observations of aerosol optical depth (AOD) and a variety of other atmospheric parameters and systematically compare the satellite-derived PM2.5 dataset to results from the CAMS regional model system. We discern the differences in the information they provide to demonstrate the added value of the satellite AOD-derived data for the investigation of long-term PM2.5 variability. The development in PM2.5 variability is analyzed on a yearly and seasonal base over the time period from 2018 to 2022, including the COVID-19 lockdown in spring 2020. We can demonstrate a better accuracy of the RF-based AOD model compared to CAMS reanalysis data with R2 values of 0.71 and 0.57, respectively. Moreover, we find that the AOD-based data provides more detail in spatial gradients and can better depict local and short-term events of enhanced PM2.5 pollution.