CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: A graphical lasso model for Hermitian matrices to detect global time-lagged teleconnections Authors:  Indranil Sahoo - Virginia Commonwealth University (United States) [presenting]
Joseph Guinness - NC State University (United States)
Brian Reich - North Carolina State University (United States)
Abstract: Teleconnections refer to spatially and temporally connected large-scale anomalies that influence the variability of atmospheric phenomena. Since teleconnections influence the global climate system, it is important to understand the abnormal behavior and interactions of these phenomena and identify them accurately. We provide a mathematical definition of teleconnections based on a spatio-temporal model using spherical needlet functions. Spherical needlets are exactly localized at several overlapping intervals corresponding to different frequencies in the frequency domain and form a tight frame. This ensures the perfect reconstruction property of an orthonormal basis. We also extend the famous graphical Lasso algorithm to incorporate Hermitian matrices and use it to estimate the inverse covariance matrix of needlet coefficients after projecting them onto the Fourier domain. The proposed method is demonstrated by simulation studies and detection of possible global teleconnections in the HadCM3 model output air temperature data.