Rule-based classification of multi-temporal satellite imagery for habitat and agricultural land cover mapping

Aim: To evaluate the use of time-series of Landsat sensor data acquired over an annual cycle for mapping semi-natural habitats and agricultural land cover. Location: Berwyn Mountains, North Wales, United Kingdom. Methods: Using eCognition Expert, segmentation of the Landsat sensor data was undertaken for actively managed agricultural land based on Integrated Administration and Control System (IACS) land parcel boundaries, whilst a per-pixel level segmentation was undertaken for all remaining areas. Numerical decision rules based on fuzzy logic that coupled knowledge of ecology and the information content of single and multi-date remotely sensed data and derived products (e.g., vegetation indices) were developed to discriminate vegetation types based primarily on inferred differences in phenology, structure, wetness and productivity. Results: The rule-based classification gave a good representation of the distribution of habitats and agricultural land. The more extensive, contiguous and homogeneous habitats could be mapped with accuracies exceeding 80%, although accuracies were lower for more complex environments (e.g., upland mosaics) or those with broad definition (e.g., semi-improved grasslands). Main conclusions: The application of a rule-based classification to temporal imagery acquired over selected periods within an annual cycle provides a viable approach for mapping and monitoring of habitats and agricultural land in the United Kingdom that could be employed operationally.