Spectral-entropy and amplitude-based event detectors compared on a noisy local seismic network

Entropy-based detectors have been proposed as an alternative to amplitude triggers for seismic event detection, with the Instantaneous Spectral Entropy Detection (ISED) method reported capable of detecting microseismic events where amplitude triggers underperform. Whether this advantage persists under a protocol combining an identically tuned classical baseline, an event-level test split, and an explicit false-alarm rate on continuous noise had not previously been established. We evaluated ISED and its non-extensive generalizations, the Tsallis and Rényi spectral entropies (q, α = 0.5–2), against a tuned amplitude STA/LTA on 800 manually reviewed events from a local network in Uzbekistan, tuned on 20% of events and evaluated on the remaining 80% against manual P arrivals (±3 s tolerance). The tuned STA/LTA attained F1 = 0.55; two entropy variants exceeded this value — Tsallis q = 0.5 (0.60) and Rényi α = 0.5 (0.56) — through higher recall at lower precision, whereas classical ISED did not. Tsallis q = 0.5 showed a higher false-alarm rate (3.5 min −1 ), while Rényi α = 0.5 achieved a comparable gain at a lower rate. Both entropy variants localized onset times more accurately than the baseline. The advantage of spectral-entropy detection is real, but parameter-dependent.

Hali tarjima qilinmagan