Remote sensing data are an important source for the derivation of land use information. In this work, an optimised classification approach is presented that employs the potentials of increasing temporal availability of optical satellite data and the integration of phenological ground observations. A method for the automated optimisation of spectral indices was developed, which could be used to derive temporal profiles of field crop spectral separability. Further, a geostatistical model for spatial modelling of observations of plant phenological phases was developed. Finally, these approaches are combined to detect phenological time frames for the selection of the most valuable acquisition dates for classification purposes out of dense satellite image time series. The optimised classification framework is demonstrated in a study site in Central Germany and compared to an established multi-temporal classification approach.