Analysis of Seasonal Dual Pol TerraSAR-X Time Series Data for the Classification of Grassland Types in Southern Bavaria, Germany

Kurzfassung

Manmade changes of landscapes and the loss of biodiversity through an on-going intensification of agricultural land use demand for the monitoring of natural habitats, vegetation types and their changes. EU Policies like the Flora-Fauna-Habitat (FFH) directive or the High Nature Value (HNV) farmland (EU rural policy framework) request the observation and conservation of natural vegetation with a reporting duty on the conservation status of these areas every six and four years. The FFH directive includes natural and semi-natural vegetation identified as Natura2000 conservation areas and the HNV farmlands are areas in which the major land use is agriculture but also include a high species and habitat diversity or species of European conservation concern. In the context of this major demand of reporting, remote sensing has great advantages as basis for monitoring of vegetation types and habitats, as it is able to provide large-scale up-to-date geo-information for these monitoring and management activities. However, the monitoring of vegetation demands for data acquisitions at specific times during the growing season. SAR-based systems have the great advantage of acquiring data of the Earth surface within these crucial time frames weather and daytime independent. TerraSAR-X (TS-X) has the potential to acquire images with a high spatial resolution of less than 10 m and at the same time a high temporal resolution of 11 days repetition rate. In addition, textural information and the potential to analyse dielectric characteristics (e.g. moisture) enable the differentiation of various vegetation types and the deduction of biophysical characteristics. Within the project MSAVE, the applicability of temporally and geometrically high resolution data of optical, SAR and hyperspectral sensors for the monitoring of vegetation is evaluated and new methods for the analysis of multi-seasonal remote sensing data are developed, which are suited for the identification and quantification of habitats. The herein presented work as part of the project centres the evaluation of the differentiability of NATURA2000 habitats and HNV grassland with TS-X time series data. Thereby, a TS-X time series consisting of 11 dual-pol (VV/VH) acquisitions in 2011 for the test area located in Southern Bavaria as well as reference data collected during several field surveys within the SAR data collection period are used for the classification of different grassland types, which include the FFH habitat types 6410 (Molinia meadows), 7120 (Degraded raised bogs still capable of natural regeneration), 7140 (Transition mires and quaking bogs) and 7230 (Alkaline fens), as well as HNV grassland. Up-to-now, the TS-X time series has been pre-processed including multi-temporal filtering and the Kennaugh decomposition developed by Schmitt, A. (2012) has been applied to the data. Furthermore, texture features were derived. After a separability analysis for defining the best suitable features for discriminating the different grassland types, the classification is performed with the Maximum-Entropy algorithm. First results show that the high temporal and geometric resolution time series of TS-X allows for a detailed analysis on the usability of polarimetric SAR data for the classification of different vegetation types. The distribution of FFH habitats and HNV grassland can be modelled by means of Kennaugh elements of cross-polar TS-X time series data. The discrimination of the different grassland types is further improved including texture features in the classification.