Statistical Analysis Of The Seasonal Variability Of Atmospheric Composition Using Eof And Fourier Analysis
Presenter: John McKinnon1
Co-Author(s): Chayan Roychoudhury, Amin Mirrezaei, Yafang Guo, Benjamin Gaubert
Advisor(s): Avelino F. Arellano
1Department of Hydrology and Atmospheric Sciences, University of Arizona
Understanding the patterns in the temporal and spatial variability of CO, NOx, H2O and CH4 aids us in separating sources and sinks of these gases for the use of improving model biases in atmospheric chemistry and climate simulations. This study focuses on matrix decomposition and spectral methods to accomplish this, building on our previous results where we use EOF (Empirical Orthogonal Function) analysis to separate the modes of variability for satellite retrieved MOPITT CO into long-term, seasonal, and residual modes. Because the dataset is non-Gaussian, these modes of variability are linearly dependent in time showing that each mode cannot be used to separate individual CO sources/sinks based on their sector. In our new results we use EOF analysis and Fourier analysis to compare normal PDFs (probability density functions) of the power spectra for satellite retrieved MOPITT and IASI CO with CAMS reanalysis and CAM-chem modeled simulations. By examining the mean and variance in the PDFs we show regional differences in their spatiotemporal distributions and how these differences relate to the dominant time scales for each mode. We determine that the seasonal variability from MOPITT is strongest and the most localized in space, which we postulate is because of its near infrared enhanced retrieval. The spectral distributions of CAM-chem tags show that the anthropogenic signal of CO can be statistically separated from the biogenic signal in the Northern Hemisphere while the signal from fires can be separated in the Southern Hemisphere.