Network Structure of Atmospheric Perturbations

Abstract

Abstract Accurate atmospheric-state analysis is essential for understanding and prediction of the atmosphere and is a difficult scientific problem due to the chaotic nature of the atmosphere; namely, small atmospheric perturbations (APs) grow rapidly in nonlinear processes. The key to atmospheric-state analysis is knowing the structure of the APs. We analyzed the AP structure in terms of network theory using a 192-member AP ensemble. The AP ensemble was generated by an ensemble of variational data assimilation (DA) with the perturbed observation method using the operational numerical weather prediction system at the Japan Meteorological Agency. The generated APs captured flow-dependent AP structures corresponding to atmospheric normal modes, and their use in DA improved accuracy of atmospheric-state predictions. These show the usefulness of the APs. The network property of the APs are as follows. The atmosphere has a small average network distance compared with the square root of the number of nodes, and a large clustering coefficient (about 0.6). These show that the APs have small-world network properties. The degree distribution of APs shows the heavy-tailed structure. These three properties of APs are common to various complex networks in other systems. Hubs in the AP network correspond to atmospheric disturbances. The network community detection using network modularity shows about 18 communities with 0.8 modularity. These basic network properties of APs represent efficient information exchange in the atmosphere, which provides a complementary atmospheric picture to its traditional physical picture based on fluid dynamics and thermodynamics, and would be basic information for atmospheric sciences and extension of application target of the network theory. Significance Statement The purpose of this paper is to analyze properties of atmospheric perturbations using the network theory and an accurate ensemble of the atmospheric perturbations. This is important for atmospheric-state estimation and forecasting because the atmosphere is chaotic and small perturbations grow rapidly. Our results show that the atmosphere has a small average network distance and a large clustering coefficient, which are the properties of a small-world network, and a heavy-tailed degree distribution. These three properties of atmospheric perturbations are common to various complex networks in other systems.