Predictive Model Development for Complex Geosystems
Predictability, Uncertainty Quantification and Model Selection
When developing computational models for complex processes our ultimate goal is to predict their future behaviour, for example the potential impact area of a landslides, or the evolution of the biogeochemical regime in subglacial habitats. It is hence crucial to also assess the computational model's predictive power, hence its readiness for predictions. The latter requires a holistic analysis of the computational model at stake. In our group, we therefore investigate concepts and computational tools needed for such analyses, including theoretical systems analyses, investigations based on synthetic examples, model calibration and model selection based on available data (e.g. Bayesian techniques or machine learning) and forward uncertainty propagation. Recently, we also started to investigate how computational models can be used to optimize data acquisition itself following concepts known in experimental design. All of the above efforts are computationally demanding as they typically require a large number of forward simulation runs. We investigate both conceptual approaches (model hierarchies / theoretical complexity reduction) as well as computational techniques (high performance computing / efficient data structures) to increase the computational feasibility.
Geohazards and Climate Change Impact
Geohazards such as landslides and avalanches pose risks to society and infrastructure worldwide. In many regions these risks are likely to even increase due to climate change and continued urbanization. Data-driven and physics-based models are being developed that enable hazard practitioners to conduct risk assessment and to plan mitigation measures. In our group, we develop computational methods to simulate the impact of rapid gravity-driven mass movements such as landslides and avalanches. We investigate fundamental process models and develop predictive simulation techniques for these in complex terrain. We are furthermore investigating and developing computationally feasible approaches to uncertainty quantification, sensitivity analysis, parameters estimation, model selection and optimal design in the context of geohazards.
Cryosphere Exploration on Earth and Beyond
The discovery of subglacial lakes in Antarctica, Canada and Greenland led to a number of subglacial exploration efforts in recent years. Today we know that these water reservoirs, which are covered by thousands of meters of ice, still constitute vivid ecosystems. This fuelled further speculation that life is also possible underneath ice shells on the Ocean Worlds of our Solar System, such as Jovian Europa or Saturnian Enceladus. Designing ice exploration missions with autonomous robots, analysing biogeochemical subglacial processes, as well as processing and analysing data acquired during such missions all require advanced computational modelling and data integration, which we develop in our group. In that context, we are involved in a number of national and international cryosphere exploration projects, e.g. the DLR Explorer Initiative and the JPL-led Cryobot Project.
Image SourcesIn: Predictive Model Development for Complex Geosystems:
Result modified from Kowalski J., McElwaine J.N., J. Fluid Mech 714, 434-462 (2013)
In: Predictability, Uncertainty Quantification and Model Selection:
Result modified from Zhao H., Kowalski J., accepted for publication in NHESS (2020)
In: Geohazards and Climate Change Impact:
Foreground: Aerial photo of an Austrian debris flow (iStock: #840304108)
Background (visible when hovered): Simulation result from Kowalski J., PhD thesis, (2009)
In: Cryosphere Exploration on Earth and Beyond
Photograph by Peter Winandy (2017)