Das Projekt "Data Mining in der Oekologie" wird vom Umweltbundesamt gefördert und von GMD-Forschungszentrum Informationstechnik GmbH, Institut für Systementwurfstechnik durchgeführt. Die Arbeiten konzentrieren sich bisher auf die Extraktionen von Standortanspruechen von europaeischen Pflanzenarten aus einer Datenbank mit Standortaufnahmen (TERRA BOTANICA). Es konnte gezeigt werden, dass mit einfachen Data Mining Verfahren automatisch Beschreibungen erzeugt werden koennen, die den manuell erstellten hinsichtlich Genauigkeit nicht nachstehen. Die Erweiterung der Arbeiten auf einen groesseren Datenbestand ist in Vorbereitung.
Das Projekt "Mountain biodiversity in the Caucasus and its functional significance" wird vom Umweltbundesamt gefördert und von Universität Basel, Botanisches Institut, Abteilung Pflanzenökologie durchgeführt. Module 1: Slope stability and biodiversity in the Caucasus, a synthesis of available knowledge and recommendations. Grazing and erosion affect biodiversity on steep mountain terrain. In a previous SCOPES research project, species had been identified and ranked which are particularly robust against erosion and thus useful for re-vegetation activities. A keystone species had been discovered (Festuca valesiaca) that excerts slope engineer functions as soon as erosion leads to gully formation and creates erosion edges. Surprisingly, the very same species had been found to play a similar role in the Swiss Central Alps (paper submitted by Caprez et al.; parallel project in Switzerland by Huck et al. in the uppermost Reuss valley). Each of the winner and looser species (a total of c. 50 species) will be now characterized by range limits, centers of abundance in terms of elevation, geology and topography, to explore their wider role in a greater geographic framework (see module 2). These data will help to predict effects of global change and to identify indicator taxa (or groups of taxa) for sustainable rangeland management. The module offers synergies with ongoing research in the Alps (similar works in Innsbruck and Grenoble), following the research agenda of the Global Mountain Biodiversity Assessment (GMBA) of DIVERSITAS. Module 2: Biodiversity in the Great Caucasus: open access species database for improved biodiversity management and projections of trends under global change. This module proposes an electronic biodiversity archive initiative, aiming at building an electronic database that includes both, archive data (herbarium vouchers) as well as observational data (relevé data) of more than 40 years of field work in steep mountain terrain by the Georgian team (many hundreds of 'relevés', each consisting of exactly geo-referenced species lists). This database will open the possibility to link the Georgian field ecology community with the GBIF (Copenhagen) international biodiversity data portal, in particular to contribute to the Global Mountain Biodiversity Assessment (GMBA) mountain portal with GBIF (online in 2010). Open access to Great Caucasus data will permit a much larger comparison of typical settings of plant communities in steep mountains worldwide. In addition, the project will distill large scale patterns of species diversity in the Great Caucasus with respect to land use and erosion aspects (see Module 1), climatic and topographic affiliations of certain taxa and elevational trends. A particular task will be identifying the environmental envelope of species that had been found to be key stone species for slope stability.
Das Projekt "GeoKernels: Kernel-Based Methods for Geo- and Environmental Sciences - Phase2" wird vom Umweltbundesamt gefördert und von Universite de Lausanne, Institut de geomatique et d'analyse du risque durchgeführt. This proposal is a continuation of the project 'GeoKernels': Kernel-Based Methods for Geo- and Environmental Sciences (200021-113944/1). The projects deal with the fundamental developments in the field of intelligent geospatial data analysis and modelling using Machine Learning algorithms. The first phase of the GeoKernels project provided a general methodology for using the state-of-the-art models in machine learning (where kernel methods establish one of the main successful areas) for spatio-temporal data analysis and modelling. Real life data lie on some lower-dimensional manifolds in the original high-dimensional geo-feature space. For environmental data these natural low-dimensional geo-manifolds are induced by rivers, relief features, urban structures, hydro-geological formations, etc. During PhaseI in the GeoKernels methodology, semi-supervised learning was applied to the stated problems in an efficient and elegant manner. The continuation of the project (Phase2) is aimed at advancing the data-driven GeoKernels modelling methodology, bringing it closer to the need of real-life operational use, from one side, and developing new methods concerned with geomanifold modelling by feature extraction and interpretable predictions with multiple kernel learning. The new developments will provide more transparency to the data-driven methods and will bring more flexibility for modelling complex environmental processes. The methods are particularly targeted at applications in natural hazards assessments and forecasting, topo-climatic modelling and renewable resources evaluation. Due to the established collaboration, the results of this multidisciplinary project will improve spatial data collection and management process in different scientific fields, will develop new procedures of environmental pattern recognition and modelling approaches using recent achievements in machine learning. The main results will be presented at the international conferences and workshops and published in scientific journals and books. The results, including the software modules (Machine Learning Office) and online interactive case studies will be available at the website of the project for the research and educational purposes (www.geokernels.org).