Understanding the cause and effect of climatic interactions such as the causes of extreme weather events and weather forecasting is highly relevant for society. At last year’s conference on Neural Information Processing Systems (NeurIPS) in Vancouver, a competition called “Causality for Climate” was held to address climate change challenges from a data perspective: Sebastian Weichwald, a recent graduate of both the Max Planck Institute for Intelligent Systems in Tübingen and ETH Zurich, is part of the winning team. Weichwald describes himself as an advocate of pragmatic causal modelling who aims to bring statistical causal modelling from theory to application.
The research field of causality focuses on understanding the influence of one event – the cause – on the emergence of another event – the effect – whereby the cause is partly responsible for the effect, and the effect is partly dependent on the cause. In turn, causal inference is the process of drawing a conclusion about a causal connection based on empirical data.
Understanding the causes of climate change calls for an understanding of the ways in which different climate variables interact, and on the ability to make predictions about these interactions. Finding causal relations in climate research relies mostly on expensive and time-consuming model simulations. Fortunately, with the explosion in the availability of large-scale climate data and increasing computational power via the cloud, there are new, complementary ways to use machine learning and causal inference to understand relationships in climate data. This understanding can help improve weather forecasting and identify the causes of extreme events.
The Causality 4 Climate competition focused on the causal discovery and development of new methodologies to understand climate data and was one of 12 accepted NeurIPS 2019 competitions. The winning team, Martin Jakobsen, Phillip Mogensen, Lasse Petersen, Nikolaj Thams, Gherardo Varando and Sebastian Weichwald from the Copenhagen Causality Lab at the University of Copenhagen’s Department of Mathematical Sciences, worked with 34 different datasets with the aim of identifying the causal relationships among the variables.
The team started with simple baseline approaches and closely monitored the results as they introduced new variations to identify the methods with the best performance. Overall, because the climate data was blind and participants were not aware what measurements different time series corresponded to, they could optimize for the best methodologies without being influenced by preconceived hypotheses or assumptions. Among 190 competitors, 40 of which were very active, Weichwald and his team from the Copenhagen Causality Lab (where he is currently a postdoc) took first place in 18 of 34 cateogories. They took second place in all remaining 16 categories, and won the overall competition by achieving an average AUC-ROC score of 0.917 (2nd and 3rd place achieved 0.722 and 0.676, respectively).
The NeurIPS competition can be re-watched here.
The C4C competition platform and the CauseMe platform, which offers a standing benchmark leader board of causal discovery algorithms, are available at https://causeme.uv.es/neurips2019 at https://causeme.net, respectively. (At the time of writing the servers were overloaded by requests.)
Code of the team's winning algorithms:
Paper describing winning method (forthcoming; accepted for publication at PMLR NeurIPS PostProceedings)