We present a novel approach for detailed crop type and tree species classification exploiting high spatio-temporal resolution Sentinel-2A data. Here, we address the challenge of frequent cloud coverage in satellite observations by defining, purely data driven, adaptable time periods to create image composites from training pixels. A set of composites from all established time periods within the classification year serve as multi-temporal input data for an adaptable classification approach that uses multiple random forest models for a Germany-wide pixel-based land-use classification. For crop types, we achieved an overall classification accuracy of 88.5 %. The same method will be applied for nationwide tree species mapping. A pre-study on tree species identification showed an an overall accuracy of 74.7 %. The map was used to define tree species composition to better determing conservation status of forests.