A statistical technique for high amplitude noise detection: Application to swell noise attenuation

Abstract

High amplitude noise is a common problem in seismic acquisition, particularly for marine data. Current filtering techniques that target high amplitude, narrow band or impulsive noise, considers amplitudes above a specific threshold to be noisy. In practice, the threshold is set by trial‐and‐error and is often changed to match the varying noise power across the dataset. In this abstract a data‐driven method is proposed to compute the appropriate threshold. It uses the fact that noisy amplitudes exhibit different statistical properties, compared to signal amplitudes, which can be used to separate them from the rest of the data. A statistical model is fitted using the data and outputs the probability “likelihood” that each sample in the data is noisy. Only those samples with significant probability are considered to be of noise origin. The choice of a probability threshold is more objective and reflects the processor's “confidence”, that an amplitude sample can be considered to be noise. The proposed technique is generic, but here it is developed and implemented for swell noise attenuation. Initial results are very encouraging and show a better performance compared with a standard industry technique.