Detecting causality in multivarate hydrochemical time series
Betreuer: Holger Lange
Detecting causality in multivariate hydrochemical time series Holger Lange When dealing with environmental time series from a given ecosystem, e.g. a forested watershed, it is standard to quantify correlations between them. Every textbook on statistics warns that “correlation is not causation” – that conclusions on causal connections based on high correlations coefficients are risky and often simply wrong. Examples abound in the literature. However, since rather recently, statisticians and modelers are more ambitious and develop methods to positively detect causal relationships between variables in (long) time series. An overview of the existing approaches for causal inference is provided by (Runge et al., 2019). For this master thesis, we will explore one specific causal discovery method, the PCMCI (Krich et al., 2020), constructing causal networks of multivariate datasets as implemented in the TIGRAMITE package (https://github.com/jakobrunge/tigramite/ ). The dataset we use consist of long (> 50 years) time series of hydrochemical ion concentrations from three small catchments located in the Bramke valley in the Harz Mountains, Germany. Focus will be on causal dependencies between different ions from the same catchment on one hand, and for the same groups of ions from different catchments. Changes in the strength, significance or even direction of causality might be detected since the observation period runs through a sequence of nonstationary periods, like afforestation, acid deposition, increasing warming and decreasing precipitation. The method has never been applied to this type of data before. The thesis will be supported and supervised in a collaboration between the University of Bayreuth (B. Aufgebauer, M. Hauhs, H. Lange) and the University of Leipzig (M. Mahecha). References Krich, C. et al., 2020. Estimating causal networks in biosphere–atmosphere interaction with the PCMCI approach. Biogeosciences, 17(4): 1033-1061. Runge, J. et al., 2019. Inferring causation from time series in Earth system sciences. Nature Communications, 10(1): 2553.