API src

Found 8 results.

Other language confidence: 0.5357000405276159

Übertragbarkeit von Modellen des Maschinellen Lernens in der digitalen Bodenkartierung

Maschinelle Lernmodelle haben große Erfolge beim Lernen komplexer Muster wie zum Beispiel die räumliche Verbreitung von Bodeneigenschaften gezeigt, die es erlauben Vorhersagen über nicht erfasste Bereiche zu treffen. Die Fähigkeit, das Gelernte auf andere Gebiete anzuwenden ist dagegen wenig entwickelt und bislang können die Modelle nur sehr eingeschränkt auf Bereiche außerhalb der unmittelbaren Lernumgebung übertragen werden. Ähnlich empirischen Regressionen gelten die Regelwerke, z.B. bei Entscheidungsbaumverfahren wie Random Forest, nur für den von Trainingsdaten abgedeckten Wertebereich. Für jedes weitere Gebiet werden erneut möglichst hochwertige und umfangreiche Trainingsdaten benötigt. Fortschritte im Bereich des Deep Learning (DL), z.B. Convolutional Neural Networks, des Transfer Learnings und kombinierte Ansätze im Bereich Feature Selection (FS) bieten hier erweiterte Möglichkeiten, um die Dimensionalität gerade bei kleineren Datensätzen einzuschränken, die Überanpassung an die Trainingsdaten zu minimieren und die Übertragung auf angrenzende Gebiete zu verbessern. Im vorliegenden Antrag nehmen wir diese Entwicklungen auf und versuchen Bodeneigenschaften auch für Bereiche außerhalb der Lernumgebung vorherzusagen. Dazu erstellen wir zunächst mit Umweltfaktoren eine gebietsspezifische Parametrisierung maschineller Lernmodelle anhand von geomorphometrischen, geologischen, landschaftsökologischen und klimatischen Parametern. Welche Parameter dies im Einzelnen sind und wie sie untereinander im Verhältnis stehen, wird exemplarisch für verschiedene Testdatensätze in Deutschland (humides Klima) und im Iran (semi-arid bis arides Klima) durch die Kombination von Methoden des DL und der FS berechnet. Im Folgeschritt werden die mit den ausgewählten Parametern der Umweltfaktoren und den Bodenprofildaten trainierten Modelle auf nicht trainierte Gebiete übertragen und an unabhängigen Bodendaten validiert. Die nicht trainierten Gebiete werden anhand von Distanz- und Ähnlichkeitsmaßen hinsichtlich ihrer Vergleichbarkeit mit den ursprünglichen Trainingsgebieten charakterisiert, um die Transferleistung der maschinellen Lernmodelle zu beurteilen. Abschließend ist vorgesehen, für die unbekannten Gebiete schrittweise Trainingsdaten zuzufügen, um die Entwicklung der Vorhersagegenauigkeit zu quantifizieren und die Transfereigenschaften verschiedener ML-Verfahren zu beurteilen. Als Trainingsdaten dienen LUCAS-Daten für Deutschland und Bodenprofildaten aus der nationalen SPDB Datenbank für den Iran. Die Umweltparameter werden aus Satellitendaten, digitalen Höhenmodellen, Weltklimadaten sowie Landnutzungskarten und geologischen Kartenwerken abgeleitet. Bodeneigenschaften sind Bodenkohlenstoffgehalt, Bodentextur, Carbonatgehalt und Kationenaustauschkapazität. Es werden 12 maschinelle Lernverfahren vergleichend angewendet.

H2020-EU.2.1. - Industrial Leadership - Leadership in enabling and industrial technologies - (H2020-EU.2.1. - Führende Rolle der Industrie - Führende Rolle bei grundlegenden und industriellen Technologien), ENCOMPASS ENgineering COMPASS

The ENCOMPASS project principally aims to create a fully digital integrated design decision support (IDDS) system to cover the whole manufacturing chain for a laser powder bed fusion (L-PBF) process encompassing all individual processes within in. The ENCOMPASS concept takes a comprehensive view of the L-PBF process chain through synergising and optimising the key stages. The integration at digital level enables numerous synergies between the steps in the process chain and in addition, the steps themselves are being optimised to improve the capability and efficiency of the overall manufacturing chain. ENCOMPASS addresses the three key steps in the process chain: component design, build process, and post-build process steps (post-processing and inspection). The links between these stages are being addressed by the following five interrelations: 1. Between the design process and both the build and post-build processes in terms of manufacturing constraints / considerations to optimise overall component design 2. Between the design process and build process component-specific L-PBF scanning strategies and parameters to optimise processing and reduce downstream processing 3. Between the design process and the build and post-build processes in terms of adding targeted feature quality tracking to the continuous quality monitoring throughout the process chain 4. Between the build and post-build processes by using build specific processing strategies and adaptation based on actual quality monitoring data (for inspection and post-processing) 5. Between all stages and the data management system with the integrated design decision support (IDDS) system By considering the entire AM process chain, rather than the AM machine in isolation, ENCOMPASS will integrate process decision making tools and produce substantial increases in AM productivity, with clear reductions in change over times and re-design, along with increased 'right-first time', leading to overall reductions in production costs, materials wastage, and over-processing. This will lead to higher economic and environmental sustainability of manufacturing, and re-inforce the EU's position in industrial leadership in laser based AM.

Teilprojekt 3: Synthese der pollen-basierten Landoberflächen und Klimakarten, Teilprojekt 2: Globale Zeitserien von Äolischem Staub, Aridität und Humidität

Für jeden Kontinent und jeden Ozean sollen höchstaufgelöste und bestdatierte Datenreihen der Staubaktivität und der Vegetation auf einer gemeinsamen Zeitskala über die letzten 40.000 (130.000) Jahre zusammengestellt werden. Diese gemeinsame Zeitachse wird aus Stalagmit-Datierungen für jede Klimaregion separat aufgebaut und dann zu den Eiskerndatensätzen (CO2, Temperatur) und der orbitalen Insolationskurve in Bezug gesetzt. Gemeinsam können all diese Informationen zur Landoberflächenalbedo und Aridität/Humidität der PalMod Initiative als Grundlage für Validierung oder Forcing dienen. In der ersten Phase werden alle häufig zitierten Datensätze der Staub-/Lössaktivität für jeden Kontinent und Ozean aus Datenbanken und der publizierten Literatur zusammengestellt. Dabei werden insbesondere die Altersmodelle kritisch evaluiert. In vielen Fällen wird der Antragsteller direkten Kontakt mit den Produzenten der Daten suchen müssen, da Altersmodelle nur auf Basis der Erfahrung mit dem Kernmaterial (d.h. von den Erstbearbeitern) aufgebaut werden können. Neben der Zusammenstellung von Proxies zur Staubaktivität und Vegetationsbedeckung werden aus allen publizierten U/Th-Datierungen die Wachstumsphasen von Stalagmiten für die verschiedenen Klimazonen (Aridität/Humidität) zusammengestellt. Im ersten Jahr findet das 'DataMining' statt. Alle statistischen Informationen werden auf einer gemeinsamen Zeitachse in einer einheitlichen Datenmatrix zusammengeführt. Die geowissenschaftlich-klimatologische Auswertung und Interpretation und die 'modellfähigen Daten' der Trübung und Albedo wird an die Modellierergruppen weitergegeben.

Stream network expansion and contraction - patterns, controls and importance (STREAMEC)

Temporary streams are very common, also in humid climates, and often represent a substantial portion of total stream length. However, temporary streams are generally not included in stream monitoring networks and hydrologists have, so far, largely ignored them. As a result, little is known about the hydrological responses of temporary streams and the factors that control the occurrence of flow in temporary streams. Connectivity of previously disconnected stream segments can result in sudden increases in streamflow and significantly alter stream water quality in downstream perennial streams. For the latter, it is especially important to note that organic material and other substances that might have accumulated in the channel during periods of no flow, might be flushed out at the onset of flow in temporary streams, causing a water chemistry dynamic that is different from that of constantly flowing streams. It is thus important to study how and when connectivity between flowing sections of the temporary stream network is established and what controls the occurrence of flow in temporary streams. The overall aim of the proposed research is to explore these aqua temporaria incognita. This will provide a better understanding of temporary streams and their importance for spatial and temporal variations in runoff and stream water quality in headwater catchments. This will be done by detailed field experiments in two pre-Alpine headwater catchments. We will map the presence of flowing water in temporary streams using an array of low costs sensors, time-lapse cameras and traditional mapping techniques to study how connectivity of flowing stream reaches is established and what topographic and landscape factors control the occurrence of flow seasonally and during different events. We will use the occurrence of flow in the stream network as an indicator of subsurface flow and hillslope-stream connectivity in order to better understand the precipitation and groundwater level thresholds that control runoff generation. This is a novel way to obtain information on the spatial variability of the temporal controls on subsurface flow generation. Finally, by monitoring the onset of flow and connectivity of stream reaches and simultaneous sampling of water chemistry in temporary streams during events, we will determine how connectivity of the temporary stream network affects downstream water quality.

Vegetation and Soil Erosion in the Mountain Forest of Cochabamba (Bolivia)

Within an integrated project of the Universities of La Paz (Bolivia) and Zurich, ecological investigations are realized since 1996. In the upper mountain forests near the cities of La Paz (in the so called yungas') and Cochabamba (the chapare') vegetation cover, soil erosion and pollen influx are studied in order to get local climatic information at altitudes of 1500-3200 m a.s.l. A decrease of forest areas is observed - as well as an increase of agricultural areas - even with slope inclinations of 40%, with many damages by soil erosion as a consequence. The humid climate with annual precipitation of 100-200 cm, however, is somehow protecting the area from large ecological impacts, because a secondary forest is coming up rapidly. On the other hand, the nearby region above 3200 m a.s.l. is semihumid to semiarid and strongly affected by soil degradation. The deforestation areas of the past centuries were localized by pollen analysis of near-surface soil samples.

Water Harvesting for Rainfed Africa: investing in dryland agriculture for growth and resilience (WAHARA)

Objective: WAHARA will take a transdisciplinary approach to develop innovative, locally adapted water harvesting solutions with wider relevance for rain-fed Africa. Water harvesting technologies play a key role in bringing about an urgently needed increase in agricultural productivity, and to improve food and water security in rural areas. Water harvesting technologies enhance water buffering capacity, contributing to the resilience of African dry lands to climate variability and climate change, as well as to socio-economic changes such as population growth and urbanisation. To ensure the continental relevance of project results, research will concentrate on four geographically dispersed study sites in Tunisia, Burkina Faso, Ethiopia and Zambia, covering diverse socio-economic conditions and a range from arid to sub-humid climates. The project emphasizes: i) participatory technology design, i.e. selecting and adapting technologies that have synergies with existing farming systems and that are preferred by local stakeholders, yet tap from a global repertoire of innovative options; ii) sustainable impact, i.e. technologies that combine multiple uses of water, green and blue water management, and integrated water and nutrient management. Using models, water harvesting systems will be designed for maximum impact without compromising downstream water-users, contributing to sustainable regional development; iii) integration and adaptability, i.e. paying attention to the generic lessons to be learned from local experiences, and developing guidelines on how technologies can be adapted to different conditions; and iv) learning and action, i.e. a strategy will be developed to enable learning and action from successes achieved locally: a. within a region, to upscale from water harvesting technologies to water harvesting systems, and b. across regions, promoting knowledge exchange at continental scale.

Changes in carbon uptake and emissions by oceans in a changing climate (CARBOCHANGE)

Objective: CARBOCHANGE will provide the best possible process-based quantification of net ocean carbon uptake under changing climate conditions using past and present ocean carbon cycle changes for a better prediction of future ocean carbon uptake. We will improve the quantitative understanding of key biogeochemical and physical processes through a combination of observations and models. We will upscale new process understanding to large-scale integrative feedbacks of the ocean carbon cycle to climate change and rising carbon dioxide concentrations. We will quantify the vulnerability of the ocean carbon sources and sinks in a probabilistic sense using cutting edge coupled Earth system models under a spectrum of emission scenarios including climate stabilisation scenarios as required for the 5th IPCC assessment report. The drivers for the vulnerabilities will be identified. The most actual observations of the changing ocean carbon sink will be systematically integrated with the newest ocean carbon models, a coupled land-ocean model, an Earth system model of intermediate complexity, and fully fledged Earth system models through a spectrum of data assimilation methods as well as advanced performance assessment tools. Results will be optimal process descriptions and most realistic error margins for future ocean carbon uptake quantifications with models under the presently available observational evidence. The project will deliver calibrated future evolutions of ocean pH and carbonate saturation as required by the research community on ocean acidification in the EU project EPOCA and further projects in this field. The time history of atmosphere-ocean carbon fluxes past, present, and future will be synthesised globally as well as regionally for the transcontinental RECCAP project. Observations and model results will merge into GEOSS/GEO through links with the European coordination action COCOS and will prepare the marine branch of the European Research Infrastructure ICOS.

Thermal perceptions of humans under climate change conditions in moderate and tropical-humid climate

Comparison of methods concerning the quantification of thermal comfort and thermal stress in regions with different climate. The assessment of thermal bioclimate based on the human energy balance requires the estimation of thermal indices by the use of measurements and modeling but also the verification of the results by people based on questionnaires.

1