Workshop "Uncertainty in Data-Driven Environmental Modelling"

Thursday, 29 August 2019, 9:30–17:00
ETH Zurich, HG D1.2

Background. "Big data" is an umbrella term for ongoing developments in collecting, processing, and analysing large volumes of data. While new sources of data and analytic tools such as machine learning have become common in environmental sciences, the resulting uncertainties arising in data-driven modelling have not been analysed thoroughly. Since environmental scientists have long experience with technical and conceptual uncertainty analysis, they are in a good position to contribute to a better understanding of uncertainties arising in data-driven modelling.

Aims. This workshop explored representational uncertainties in large environmental datasets as well as in datadriven modelling practices in environmental sciences. Presentations covered case studies as well as technical and conceptual aspects of uncertainty analysis and their implications for decision-making. The outcomes of the workshop could be helpful for the planning of future research, for taking a critical look at an emerging trend in environmental research, and for identifying potential for collaborations and synergies. Questions that were also addressed in the workshop include: (1) What are the main sources of uncertainty when working with new data sources and new analytic tools? (2) Are these uncertainties qualitatively different from uncertainties that environmental scientists faced traditionally? (3) How does the lack of transparency of data-driven models affect uncertainties? (4) What are appropriate methods to quantify and handle these uncertainties in research? (5) What are the implications of these uncertainties for policy analysis and decision support?

Programme
09:30 Welcome by Reto Knutti (IAC)
09:40 Konrad Schindler (D-BAUG): High-resolution mapping of vegetation height by deep regression from satellite images
10:10 Eniko Szekely (SDSC): Uncertainty quantification in machine learning and data science
10:50 Coffee break
11:10 Nicolai Meinshausen (D-MATH): Sampling, model and distributional uncertainties
11:50 Marius Zumwald (IED/IAC): Uncertain data: Using crowdsensing data for urban temperature prediction
12:30 Lunch break
13:50 Rachael Garrett (IED): Uncertainty in land change data and its influence on policy analysis
14:30 Benedikt Knüsel (IED/IAC): Data-driven climate predictions: uncertainty, opacity, and confidence
15:10 Coffee break
15:30 Sven Ove Hansson (KTH Stockholm): Values and uncertainties in toxicology
16:10 David Bresch (IED): Preliminary synthesis and open questions
16:30 Concluding discussion
17:00 End of workshop

Organisers:
Benedikt Knüsel (IED), Marius Zumwald (IED), Christoph Baumberger (IED), David Bresch (IED), Reto Knutti (IAC)