SCCS Colloquium - Sep 16, 2020
|Date:||Sep 16, 2020|
|Room:||Online (password: SCCS)|
|Time:||15:00 - 16:00|
Valentina Schüller: Monitoring Numerical Climate Simulations - A Tool for EC-Earth
Climate model experiments are time consuming numerical simulations. Real time monitoring of experiments allows to spot problems in the model performance early on: Unrealistic results or significant decreases in the model speed can be detected at runtime, thus potentially saving computational resources. While model-independent analysis tools exist for finalized and post-processed output, monitoring is model-specific and uses raw output data. For the newest version of the European community Earth System Model EC-Earth, a Python based monitoring application has been developed. It tracks the physical and computational performance based on established metrics and diagnostics in climate science. Although the concrete implementation is optimized for EC-Earth 4, the software design and choice of monitoring diagnostics are of broader relevance. This talk presents the central design decisions as well as the status quo of the tool's feature set. This Bachelor's thesis was written at the Swedish Meteorological and Hydrological Institute (SMHI).
Keywords: climate simulation, software development, monitoring
Willem van Hove: Identifying Predictors for Energy Poverty in Europe using Machine Learning
Master's thesis submission talk. Willem is advised by Prof. Hans-Joachim Bungartz, Prof. Van Der Zwaan, and Francesco Dalla Longa.
Energy poverty is defined as a household's inability to afford its energy bills, something that has become an issue in Europe. The transition to a more sustainable energy supply is expected to radically transform energy infrastructure in all European countries. The cost of this transformation needs to be distributed in such a way that all households are still able to pay their energy bills. This thesis investigates the use of unconventional computational science methods in order to understand this complex societal issue. We attempt to identify drivers for energy poverty in Europe by interpreting a machine learning model. A gradient boosting classifier is trained on a European data set and its internal model is analyzed, thus providing new insights into the underlying intricacies of energy poverty.
Keywords: Energy Poverty, Machine Learning, Gradient Boosting, Interpretable Artificial Intelligence