Causal inference for time series

Abstract

Many research questions in Earth and environmental sciences are inherently causal, requiring robust analyses to establish whether and how changes in one variable cause changes in another. Causal inference provides the theoretical foundations to use data and qualitative domain knowledge to quantitatively answer these questions, complementing statistics and machine learning techniques. However, there is still a broad language gap between the methodological and domain science communities. In this Technical Review, we explain the use of causal inference frameworks with a focus on the challenges of time series data. Domain-adapted explanations, method guidance and practical case studies provide an accessible summary of methods for causal discovery and causal efect estimation. Examples from climate and biogeosciences illustrate typical challenges, such as contemporaneous causation, hidden confounding and non-stationarity, and some strategies to address these challenges. Integrating causal thinking into data-driven science will facilitate process understanding and more robust machine learning and statistical models for Earth and environmental sciences, enabling the tackling of many open problems with relevant environmental, economic and societal implications.