UFZ Metadata Catalog for Spatial Resources
Classification
Type of resources
Available actions
Topics
Keywords
Contact for the resource
Provided by
Formats
Representation types
Service types
Resolution
-
Based on Sentinel-2A imageries from 2016, the map shows a detailed classification of Germany's arable land, distinguishing 19 land-use classes. "Forests", "Waters", "Urban" and "Other Vegetation" are predefined classes and were not classified in this study. The map is the result of a highly automated adaptable approach for pixel-based compositing and classification, called APiC. The novel APiC concept is designed for the use of high-resolution spatio-temporal space-borne data. Within the scope of extensive land-use mapping we define composite periods purely data-driven based on cloud-free training pixels. Since these adapted periods (APs) are not static nor dependent on expert knowledge, they support the creation of training pixel composites (TPCs) in regions with different weather conditions, species composition and phenological behaviour. We aimed to maximize the AP number within the classification year in order to best represent land-use phenology. To account for the cloud cover distribution of all pixels to be classified, over 10,000 predictive models were needed to create this map. Multiple classification models allow to contrast the classified land-use with the associated model performance at pixel level. Due to different climatic and geomorphological gradients in Germany that have an impact on plant phenology, six landscape regions have been independently classified. The individual classification results have been combined for the map shown.
-
Quantification of grassland land-use intensity and its components livestock density, number of mowing events and fertiliser application using machine learning algorithms and time series of Copernicus Sentinel-2 optical satellite data with 20 m spatial resolution. Land-use intensity was inferred using the index proposed by [Blüthgen 2012]. Input variables (livestock, mowing and fertilisation) were derived from Sentinel-2 optical satellite data using Convolutional Neural Networks trained on data from the DFG Biodiversity Exploratories. More details will be made available soon in [Lange 2021]. Grassland pixels were chosen according to the digital landscape model (DLM) of the official topographic-cartographic information system ATKIS [BKG 2015]. Livestock densities are livestock units (calculated from animal number, species and age) per ha and day and were aggregated into four livestock classes (see [Lange 2021]). Mowing number represents the number of cuts in the respective year. Fertiliser application is given as boolean, fertilised or not fertilised. [Blüthgen 2012]: Blüthgen N, Dormann C, Prati D, Klaus V, Kleinebecker T, Hölzel N, Alt F, Boch S, Gockel S, Andreas H, Müller J, Nieschulze J, Renner S, Schöning I, Schumacher U, Socher S, Wells K, Birkhofer K, Buscot F, Weisser W (2012). A quantitative index of land-use intensity in grasslands: Integrating mowing, grazing and fertilization. Basic and Applied Ecology. 13. 207-220. 10.1016/j.baae.2012.04.001. [Lange 2021]: Lange M, Feilhauer H, Kühn I, Doktor D (2021): Mapping land-use intensities of grasslands in Germany with machine learning and Sentinel-2 time series. Manuscript submitted to Remote Sensing of Environment. [BKG 2015]: BKG (2015): Digitales Basis-Landschaftsmodell (AAA-Modellierung). GeoBasis-DE. Geodaten der deutschen Landesvermessung. Bundesamt für Kartographie und Geodäsie.