Hi Marc, I wrote to Brady Cox who very kindly and quickly replied to me.
This is the gist of his words:
As you may recall from my 2020 paper, n is the number of standard deviations. So, when using n = 2 we are incorporating approximately 95% of the f0 values obtained from the various time windows. When using n = 1 we are incorporating only about 68% of the f0 values from the various time windows. The choice of n depends on how noisy the site is. If the site is very noisy, there will be a lot of near-receiver noise sources that are imparting energy to the sensor that is not traveling upward through the ground. These contaminated time windows need to be rejected. In this case, one would need to set n lower to reject more of the contaminated windows. However, setting n too low will result in good windows being rejected, which will bias your f0 sigma values to low estimates. The median value of f0 may or may not be affected very much, but the sigma of f0 will definitely be affected. I typically start by using n = 2 and then decrease n if it looks like I am still not getting rid of some of the outlying f0 values. This is subjective, but I feel it is a good way to do it. The STA / LTA anti-triggering is doing something completely different. It is simply looking at which time windows have relatively high amplitudes and rejecting them. This is based on the premise that near-receiver sources of energy will have high amplitudes. In many cases they do. But not always. I found that the STA / LTA method was difficult to use and could not produce results that were statistically based. That’s why I developed the FWA approach.
I think this explanation might be useful for Geopsy users who want to experiment with FWA as an alternative to antitrigger.
1 post • Page 1 of 1