I’m seeing our STA/LTA trigger fire about 18% more often overnight after switching to a 2 Hz high-pass, while the manually verified event count stayed flat (n=12, last 24h). What quick, data-grounded checks do you run to keep predictive models from amplifying noise in edge conditions before alerts go live?
With the 2 Hz high-pass causing an 18% overnight bump, I’d plot STA/LTA peak vs pre-trigger RMS by hour and gate alerts when RMS is below the 10–20th percentile of your daytime true-event distribution. ‘run to keep predictive models from amplifying noise in edge conditions before’ — add a cheap multi-station coherence check (e.g., 2 of 3 stations within 3 s) as a guardrail; did that trim your night FPs over the last 24 h (n=12)?
I’d add a 2–3 station, 3–5s coincidence gate for the 18% bump; caveat: clustered aftershocks can hide real bursts, @tchen_83.