Detection of Phenology-Defined Data Acquisition Time Frames For Crop Type Mapping

Agricultural monitoring and assessment based on satellite data increasingly gains importance due to the growing number of available satellite sensors with high geometric and temporal resolution. Such tasks often require multiple images acquired on specific dates that among others account for inter-annual phenological variations to provide accurate results. This contribution presents an approach that links peaks of spectral separability profiles to crop phenological phases. The phases are spatially interpolated using a phenological model and ground observations. The profiles show the respective temporal development of the F-measure which is used as indicator for class-wise separability. It originates from binary classifications of vegetation indices computed for each set of a satellite data archive covering multiple years. Acquisition dates, which repeatedly show a separability maximum define phenological indicator phases. Potential alternative phases can be also defined. Experiments based on multi-temporal RapidEye satellite imagery were performed for three crops at two German test sites under different environmental conditions. The results showed that the phases yellow ripeness, heading and flowering can function as indicator phases for high spectral separability of winter barley, winter wheat and winter rapeseed. We could identify at least two identical, stable indicator phases per crop type for both test sites, which suggests the transferability and robustness of the presented approach.