Analysis of NDVI Time Series for Classification of Grasslands and Detection of Their Land use Practices
Halada, žuboš; Halabuk, Andrej

Semi-natural grasslands in agricultural landscape bear high biodiversity values and there is still lack of precise information on their spatial extent and status at pan European scale. One of the main criteria for good status of the semi-natural grasslands is their extensive usage, e.g. regular cutting and/or grazing. In order to get such information, detection of site management is needed for longer periods (e.g. 10 years) in order to reveal trends for possible abandonment or intensification. Because of gradual availability of time series products from sensors such as AVHRR, MODIS and under a great expectation of upcoming PROBA-V and Sentinel3 missions, we analyzed suitability of a 12-year MODIS NDVI time series at 250m spatial resolution for detection of grassland management and its monitoring on annual basis in Slovak heterogeneous landscape. Particularly, we focused on detection of cutting practices, overgrowing, flooding, overgrazing, which are all considered as important drivers of grassland functioning. Grassland seasonal pattern of NDVI varied substantially reflecting not only different vegetation type but also land use, management practices or site hydrology. We demonstrated that when some knowledge on grassland occurrence exists (e.g. from CLC), classification based on temporal NDVI profile brings valuable information on grassland status. In general, productivity (which relate to amplitude of NDVI curve) and seasonality (variance within season) represent main distinctive characteristics of grasslands. Especially, timing of NDVI peak, rate of increase of NDVI in spring, bimodal shape of NDVI, or negative anomaly in spring can be used for distinguishing of extensive meadows, pastures, flooded meadows and abandon overgrown grasslands. However, more regional specific knowledge from grassland experts needs to be used in order to derive consistent grassland classification system for broad scale mapping and classification.