ENAC Seminar Series by Dr. J. Wegner, Prof. N. Vercauteren & Dr. L. Gilarranz
Event details
| Date | 15.01.2019 |
| Hour | 09:00 › 12:15 |
| Speaker | Dr. J. Wegner, Prof. N. Vercauteren & Dr. L. Gilarranz |
| Location | |
| Category | Conferences - Seminars |
9:00 – 10:00 – Dr. Jan Wegner
Senior Researcher and Lecturer, Dept. of Civil, Environmental and Geomatic Engineering, ETHZ
Deep Machine Learning for Environmental Sciences
Dr. Wegner will present data-driven methods at the interface of computer science, ecology, and engineering that estimate a rich set of environmental variables at very large scale. After a short introduction to deep learning, he will talk about projects on vegetation monitoring, sustainability, and hydrology to highlight the great potential of modern machine learning and point at possible pitfalls.
He will demonstrate how we can combine deep learning with publicly available data to map hundreds of thousands of street trees. A similar approach can estimate tree defoliation of forest trees across Switzerland to investigate the impact of climate change on vegetation in high alpine regions. He will further show a highly automated approach to stop deforestation in tropical regions using satellite images and how we can count individual trees of specific species at country scale with deep semantic density estimation. Finally, he will explain how we can use computer vision for quantifying urban flood events using social media images. Partially submerged objects like cars, bicycles or humans are automatically identified in the images to estimate their submersion level.
10:15 – 11:15 – Prof. Nikki Vercauteren
Assistant Professor, Dept. for Mathematics and Computer Sciences, Freie Universitaet Berlin, Germany
Towards stochastic modeling of turbulence in the stably stratified atmospheric boundary layer
Atmospheric boundary layers with thermally stable stratification are the least understood type of boundary layers due to suppressed turbulence and the presence of myriads of processes on multiple spatiotemporal scales that modulate the turbulence. Stable boundary layers (SBLs) are however the norm in Polar and winter alpine environments, and more generally at nighttime. Complex alpine terrain results in even more scale interactions due to orographic effects on small-scale atmospheric dynamics. In such environments, turbulence is typically unsteady and intermittent. Classical approaches to turbulence parameterization fail to reproduce turbulent dissipation in SBL context and this is a known source of errors in larger scale atmospheric models, including climate models. In this presentation Prof. Vercauteren will approach the question of intermittency of turbulence and its partial modulation by non-turbulent motions based on multiscale data analysis and statistical clustering methods. The multitude of small-scale non-turbulent motions affecting the SBL is poorly understood, and even state-of-the-art Large Eddy Simulation (LES) tools cannot generate those sources of non-stationarity of turbulence. She will show how analyzing turbulence data based on statistically classified flow regimes helps unravel organizing principles in complex the dynamics of near-surface SBL turbulent flows. She will suggest a novel framework to include stochastic inflow structures representing characteristics of classified field data in a LES tool. Such a framework will enable simulations of intermittent flows by LES, and will additionally serve as a computational method to study and derive new types of stochastic parameterizations for weather and climate models.
11:15 – 12:15 – Dr. Luis Gilarranz
Postdoctoral researcher, Swiss Federal Institute of Aquatic Science and Technology (EAWAG)
Community Ecology in Light of Complex Networks: biodiversity across space and time
Ecosystems worldwide are experiencing an unprecedented rate of degradation. This not only has tremendous consequences for wildlife but our lives and economies as well. After decades of research, we wonder if we have a good enough understanding of ecological systems to revert the situation. Such understanding should come from a dialogue between theoretical advances and experiments and synthesis that may support or debunk such theories.
In this talk, Dr. Gilarranz will contrast theory against data to show that species interactions, perturbations, and dispersal routes play an important role in determining the health of ecological communities. Complex networks emerge as powerful tools to understand the relationships between species and community dynamics. They allow us to unveil previously undocumented effects of anthropogenic activities, to understand the geographical factors that determine the number of species coexisting at a certain location, or understanding how perturbations spread. However, most networks are static descriptions of the systems they represent. In the absence of time series, our observations are hidden in the invisible present. The perspectives for advancing the field have never been so exciting.
Senior Researcher and Lecturer, Dept. of Civil, Environmental and Geomatic Engineering, ETHZ
Deep Machine Learning for Environmental Sciences
Dr. Wegner will present data-driven methods at the interface of computer science, ecology, and engineering that estimate a rich set of environmental variables at very large scale. After a short introduction to deep learning, he will talk about projects on vegetation monitoring, sustainability, and hydrology to highlight the great potential of modern machine learning and point at possible pitfalls.
He will demonstrate how we can combine deep learning with publicly available data to map hundreds of thousands of street trees. A similar approach can estimate tree defoliation of forest trees across Switzerland to investigate the impact of climate change on vegetation in high alpine regions. He will further show a highly automated approach to stop deforestation in tropical regions using satellite images and how we can count individual trees of specific species at country scale with deep semantic density estimation. Finally, he will explain how we can use computer vision for quantifying urban flood events using social media images. Partially submerged objects like cars, bicycles or humans are automatically identified in the images to estimate their submersion level.
10:15 – 11:15 – Prof. Nikki Vercauteren
Assistant Professor, Dept. for Mathematics and Computer Sciences, Freie Universitaet Berlin, Germany
Towards stochastic modeling of turbulence in the stably stratified atmospheric boundary layer
Atmospheric boundary layers with thermally stable stratification are the least understood type of boundary layers due to suppressed turbulence and the presence of myriads of processes on multiple spatiotemporal scales that modulate the turbulence. Stable boundary layers (SBLs) are however the norm in Polar and winter alpine environments, and more generally at nighttime. Complex alpine terrain results in even more scale interactions due to orographic effects on small-scale atmospheric dynamics. In such environments, turbulence is typically unsteady and intermittent. Classical approaches to turbulence parameterization fail to reproduce turbulent dissipation in SBL context and this is a known source of errors in larger scale atmospheric models, including climate models. In this presentation Prof. Vercauteren will approach the question of intermittency of turbulence and its partial modulation by non-turbulent motions based on multiscale data analysis and statistical clustering methods. The multitude of small-scale non-turbulent motions affecting the SBL is poorly understood, and even state-of-the-art Large Eddy Simulation (LES) tools cannot generate those sources of non-stationarity of turbulence. She will show how analyzing turbulence data based on statistically classified flow regimes helps unravel organizing principles in complex the dynamics of near-surface SBL turbulent flows. She will suggest a novel framework to include stochastic inflow structures representing characteristics of classified field data in a LES tool. Such a framework will enable simulations of intermittent flows by LES, and will additionally serve as a computational method to study and derive new types of stochastic parameterizations for weather and climate models.
11:15 – 12:15 – Dr. Luis Gilarranz
Postdoctoral researcher, Swiss Federal Institute of Aquatic Science and Technology (EAWAG)
Community Ecology in Light of Complex Networks: biodiversity across space and time
Ecosystems worldwide are experiencing an unprecedented rate of degradation. This not only has tremendous consequences for wildlife but our lives and economies as well. After decades of research, we wonder if we have a good enough understanding of ecological systems to revert the situation. Such understanding should come from a dialogue between theoretical advances and experiments and synthesis that may support or debunk such theories.
In this talk, Dr. Gilarranz will contrast theory against data to show that species interactions, perturbations, and dispersal routes play an important role in determining the health of ecological communities. Complex networks emerge as powerful tools to understand the relationships between species and community dynamics. They allow us to unveil previously undocumented effects of anthropogenic activities, to understand the geographical factors that determine the number of species coexisting at a certain location, or understanding how perturbations spread. However, most networks are static descriptions of the systems they represent. In the absence of time series, our observations are hidden in the invisible present. The perspectives for advancing the field have never been so exciting.
Practical information
- General public
- Free
Organizer
- ENAC
Contact
- Cristina Perez