The theory of statistical hypothesis testing is used to develop and apply a seismic signal detection filter. The filter, herein named the sign filter, scans a stacked section and designates a linear segment as “signal” or “noise” based on the value of the sign test statistic evaluated over the amplitudes within the segment; only the signals are passed. The sign test statistic is nonparametric, so that probabilistic calculations related to the filtering process do not require rigid assumptions regarding the noise distribution. Consequently, it is possible to calculate both the probability that the filter will pass a segment containing only noise, and the expected number of noise‐only segments to be passed. These numbers may be adjusted by changing the tunable parameters of the filter. The detector was tested on both synthetic and field data. For synthetic data, all of the signals present in the data were identified, and the output did not contain any spurious signals, even for a signal‐to‐noise ratio smaller than 1. For field data, the events chosen by the filter, for the most part, agree closely with those visible in the input section; and much of the spatially incoherent energy is suppressed. A few of the passed segments were not visually coherent in the input stack; we suggest a method by which such segments might be identified and removed. The method is fairly general and may be modified for different definitions of signal. The case of linear alignments is the easiest to implement, and the detector promises to be useful in both the processing (automatic picking of first arrivals in source gathers) and interpretation (identification of primary reflections in stacked sections) phases of seismic data analysis.
Appendix A6: Sign Test Statistic
R