Understanding the nature of causal information flow using directed acyclic graphs in land-atmosphere exchange using high-frequency flux tower data.

Abstract:

Non-linear dynamics of ecohydrological components at the land-atmosphere system can be inferred using high-frequency observations through a causal history approach in which a network of inter-dependencies is identified using directed acyclic graphs (DAG) from observed multivariate time series. Characterizing how the nature of the information flow is changing in time by considering the evolution of the system dynamics represented by DAGs remains an open question. In this study, we aim to quantify the evolutionary dynamics in the multivariate land-atmosphere system represented as a sequence of DAGs using two perspectives: functional and structural. We use high-frequency data at 10 Hz from a 25-meter eddy covariance flux tower in Central Illinois at the Midwestern U.S, to determine how the present state of a target variable results from the interaction of previous states of all interacting variables, referred to as the causal history. The functional comparison of the sequences of DAGs shows how the information flow is changing in time possibly as a result of structural differences. We aim to determine if (1) functional differences are a reflection of structural differences, or (2) in spite of structural differences there are no functional differences. The approach implemented here for investigating the causal evolution of multivariate dynamics when represented by DAGs for times series, combined with the novel approach of the analysis of the evolution of the system network structure will provide a methodological framework to understand how the land-atmosphere system evolves to determine system behavioral patterns.