A network approach to determine ecosystem vulnerability to extremes

Abstract:

Ecosystems evolve due to complex interactions over various space and time scales. Process networks, in which nodes are time-series variables and links are information theoretic measures, allow us to analyze a system in terms of unique and redundant time dependencies. It has been shown that process networks composed of measured and modeled ecohydrologic time-series variables exhibit breakdowns in feedbacks during periods of climate extremes such as drought. In this study, we use an information decomposition approach to partition shared information between ecohydrologic time-series nodes into redundant and unique components. Redundant information is information shared between multiple source nodes with a target node, while unique shared information is only attributable to a single source node. In an ecohydrologic network, unique shared information between two nodes is likely to represent a critical ecosystem link, and redundant shared information indicates synchronization or internally induced feedbacks between variables. We form networks using flux tower, weather station, and ecohydrologic model output variables over a range of natural and intensively managed ecosystem types. Methods to compute information transfer components with short datasets are used in order to observe shifts in network behavior that vary with weather conditions and extreme events on hourly to weekly timescales. We compare network properties with satellite derived vegetation indices, and evaluate how links shift in terms of strength, uniqueness, or redundancy as ecosystems respond to environmental conditions. This analysis shows that a network approach can detect critical linkages that dictate ecosystem vulnerability to extreme events.