Detecting the Magnitude and Direction of Information Flow between Land and Atmosphere

Abstract:

The interactions between land and atmosphere play a significant role in the climate and weather system. Although several studies have provided a depiction of land-atmosphere (L-A) interactions based on model hindcasts, they have mainly focused on linear relationships among variables and have left nonlinearity unexamined. Furthermore, the lack of long-term globally distributed observations has hindered a robust, realistic identification of L-A interactions on a global scale. In this study, in addition to the traditional approaches based on linear statistics, we apply information-theory-based approaches that are useful for detecting linear and non-linear relationships between variables that are suitable for the analysis of causal relationships in nonlinear systems. We provide an improved insight into how land states (e.g., soil moisture) affect the atmosphere (e.g., precipitation) using a network of feedback loops that depicts the magnitude and direction of statistical information flow among the variables in the process chains linking land and atmosphere through the energy and water cycles. Recently developed global, gridded, observationally- and satellite-based data sets form the foundation of this study and are used to benchmark the performance of the Subseasonal Experiment (SubX) models.

Analyzing correlation patterns and information flows show the coupling strength and categories between pairs of variables. The similarities and discrepancies between observations and models in terms of strength and spatial patterns of L-A feedbacks are highlighted. The results enable us to take one step further in understanding the L-A interactions: from correlation to causation.