Reducing Uncertainty in Estimates of Environmental Parameters From Ambient Noise Using Statistical Array Processing

In recent years, extracting environmental information from diffuse ambient noise has become an increasingly viable alternative to traditional active source methods. Due to uncontrollable factors such as noise field directionality, presence of spatially compact sources and unknown medium properties, results from ambient noise processing are often biased and need careful interpretation. Thus it is important to develop robust approaches that can perform well in the presence of such detriments. The first part of the dissertation focuses on interpreting the coherence and attenuation estimates from seismic arrays. Adaptive array processing using stations from the Southern California Seismic Network is used to identify the presence of multiple seismic waves, namely the fundamental and first mode Rayleigh wave, and body waves. The spatial coherence function (SCF) is modeled as a linear superposition of these waves, with the proportions estimated from data. The SCF shows beating and phase cancellation effects due to the interactions between wavenumbers, which could be misinterpreted as attenuation. The array geometry is also shown to limit the ranges at which the coherence can be estimated well. The second part of the dissertation focuses on developing statistical techniques to mitigate the effects of spatially compact sources on the noise processing. Analytical expressions are derived for the asymptotic eigenvalues of the true spatial covariance matrix (CM) for a uniform line array in three and two dimensional isotropic noise fields with and without attenuation. Using random matrix theory, the asymptotic probability density of the eigenvalues of the sample covariance matrix (SCM) also is derived in each of these scenarios. These analytical results provide upperbounds for the noise eigenvalues of the SCM. In the third part of the dissertation, the analytical results are combined with a sequential hypothesis testing framework. This then is used to identify the outliers (which correspond to strong and spatially compact sources) in shallow water ocean acoustic data. The cross-correlation results after rejecting these outliers are shown to be unbiased and converge faster with a higher signal-to-noise ratio. The performance of the eigenvalue rejection technique under different noise model assumptions also is investigated.