First-break shaping, denoising, & picking

First-break picking is not the trendiest of research topics, but can be one of the most expensive, time-consuming, and frustrating steps for land processors. And the situation may be getting worse. Many seismic surveys now use simultaneous shooting with single small vibrators as their source, producing noisier first breaks than before. For this reason, I have dedicated considerable development time to the issue of first-break picking.

I have focused on three areas:

• Vibroseis wavelet shaping
• Noise removal
• Spatially continuous picking

The result of these three technologies is more accurate picks generated quicker and at less cost.

Vibroseis wavelet shaping

This novel wavelet-shaping filter significantly improves the results of picking vibroseis data by:

• Removing ringyness, reducing the chances of cycle skipping.
• Focusing the first-break energy, reducing the chance of first breaks being overwhelmed by noise.
• Locating the initial peak of the first break at the true theoretical first-break time.

Read my recent Geophysical Prospecting paper (Trickett, 2022 in the references below).

Noise removal

First breaks are often noisy due to cultural and environmental noise, poor penetration of the seismic signal through the near surface, and simultaneous shooting, so that many millions of dollars are spent every year manually correcting automatic picks. And with the advent of simultaneous high-density shooting using single-point sources and receivers, the problem of first-break noise may become worse.

There are few published methods for removing noise from first breaks before picking. It’s a difficult problem, as the noise is often strong and short-wavelength statics must be preserved. Juniper Bay’s unique solution exploits the “locally surface-consistent” property of cross spreads and the power of robust statistics (Trickett, 2019). It has the following features:

One weakness is that it does not handle geometry errors well, such as significantly mispositioned sources and receivers.

Read my SEG conference paper (Trickett, 2019 in the references below).

Spatially continuous picking

Many automatic first-arrival pickers, including ProMAX’s module First Break Picking, determine picks on a trace-by-trace basis, meaning each trace is picked independently of the others. Such pickers generally try to find a time point where some trace attribute or group of attributes changes abruptly in character. The most common and successful of these attributes is the energy level of the samples, but more exotic ones like fractal dimension can be used as well.

Trace-by-trace pickers work fairly well for dynamite data where the first arrivals tend to be clean and there is an abrupt onset of source energy on each trace. But on vibroseis data these pickers often work poorly. This is because vibroseis first arrivals tend to be noisier and have a broader and non-causal seismic wavelet, resulting in a less abrupt change in trace attributes near the first-arrival time.

To pick vibroseis data well, a picker must consider more than one trace simultaneously. One strategy is to encourage spatially continuity of the picks – that is, after linear-moveout correction, there should be few or no large time differences between nearby traces. There are surprizingly few papers that describe such strategies. Juniper Bay’s picker uses a shortest-path algorithm to select final picks from numerous candidate picks for each trace. Attributes such as energy ratio are incorporated into the distances used in the shortest-path algorithm.

Juniper Bay’s first-break picker also works on cross spreads, and is typically run immediately after first-break noise removal in the same job.  It generates highly consistent picks even in noisy areas by using a shortest-path algorithm in two spatial dimensions (common source and receiver).

References