Application of generalized autoregressive conditionally heteroskedastic modeling to wind noise for enhancing detection of transient infrasound

Abstract

Autoregressive moving average (ARMA) process models are often used to represent stationary random processes. In applications where the process of interest is not stationary or has samples from a heavy-tailed distribution, extension to a generalized autoregressive conditionally heteroskedastic (GARCH) model can be used to improve the process representation. This modeling approach was developed by the financial industry and is frequently used in econometrics to represent financial time series possessing volatility clustering and to improve return forecasts. Recently, ARMA/GARCH models have been used to improve surface-to-surface radar detector performance where the clutter due to sea state possesses a heavy-tailed distribution. In this presentation, we will demonstrate the application of vector ARMA/GARCH models to the modeling of wind noise measured on infrasound arrays and to detection of transient infrasound in the presence of this noise.