The Helmholtz Centre for Environmental Research (UFZ) with its 1,100 employees has gained an excellent reputation as an international competence centre for environmental sciences. We are part of the largest scientific organisation in Germany, the Helmholtz association. Our mission: Our research seeks to find a balance between social development and the long-term protection of our natural resources.
Discovering relevant patterns in large volumes of data and extracting useful information for further application is an increasingly important problem in today´s data rich world.
The majority of users of complex computer models in the environmental sciences face the same challenge: the transformation / condensation of the huge amount of simulation data into knowledge. At the very core of this transformation process three components have been identified: uncertainty quantification, development of fast surrogate models and the identification of key parameters and dependencies. Within the interdisciplinary Helmholtz project “Reduced Complexity Models – explore advanced data science techniques to create models of reduced complexity“, several Helmholtz centers aim at addressing and developing the three components simultaneously and in a unified manner. The UFZ is therefore seeking for a
Subject: Development of surrogate hydrologic models with reduced complexityWorking time 100% (39 h / week), limited for 2 years (24 months)
Typically, science uses partial and/or ordinary differential equation based mathematical models to describe processes in nature like weather, turbulence or plasmas. Often, the mathematical model is highly complex with many parameters. The resulting computer code is huge and expensive in terms of the computer time.
Exploring a multidimensional parameter space requires an efficient simulation design. To explore this multidimensional space it is often necessary to develop a simpler surrogate model to substitute the (expensive) computer model.
The accuracy of a surrogate model needs to be verified, e.g. by design augmentation.
Fully autonomous and self-adaptive approaches for the surrogate model building process are essential, i.e. closed-loop sequential or batch-optimization. Crucial steps to this goal will be the development of efficient adaptive algorithms based on marginalized or a posteriori error estimates that efficiently distribute computational resources to minimize the turn-around time.
Bayesian modeling evidence techniques will be used to evaluate and find the most robust reduced complexity model surrogate structure.
The successful candidate will hence develop transfer functions based on data-driven function approximation techniques and copula-based method techniques to establish the optimal causation and dependence predictor network. Competing surrogate models will be evaluated with Bayesian Model evidence techniques to identify the most robust one.
- PhD in environmental informatics with strong expertise in mathematics
- Strong knowledge in High Performance Computing
- Good knowledge in data science and Bayesian methods
- Excellent English skills
- The candidate should be motivated to work in a highly interdisciplinary team
- Excellent technical facilities which are without parallel
- The freedom you need to bridge the difficult gap between basic research and close to being ready for application
- Work in inter-disciplinary, multinational teams and excellent links with national and international research networks
- A vibrant region with a high quality of life and a wide cultural offering for a balance between family and professional life
- Interesting career opportunities and an extensive range of training and further education courses
- Remuneration up to TVöD public-sector pay grade 14 including attractive public-sector social security benefits