To eliminate the upward bias of seismic magnitude caused by censoring of signal hidden by
noise, noise level at each station in a network must be estimated. Where noise levels are not
measured directly, the method of Kelly and Lacoss (1969) has been used to infer them from
bulletin data (Lilwall and Douglas 1984). To verify this estimate of noise level, noise
thresholds of International Monitoring System (IMS) stations inferred from the International
Data Centre (IDC) Reviewed Event Bulletin (REB) by the Kelly and Lacoss method for
2005-2013 are compared with direct measurements on (i) noise preceding first arrivals in
filtered (0.8-4.5 Hz) IMS seismic data, and (ii) noise preceding the expected time of
arrival of signals from events, where signal was not actually seen (values gathered by
the IDC for maximum-likelihood magnitude calculation). For most stations the
direct pre-signal noise measurements are ~0.25 units of log A/T lower than the
Kelly&Lacoss thresholds; because the IDC automatic system declares a detection only
when the short-term-average-to-long-term-average ratio threshold, which varies
with station and frequency band between ~3-6, is exceeded. The noise values at
expected times of non-observed signal arrival are ~0.15 units lower than the Kelly and
Lacoss thresholds. Exceptions are caused by faulty channels being used for the
direct noise or body-wave magnitude (mb) measurements or, for station ARCES and
possibly FINES, SPITS and HFS, the wider filter used for signal amplitude than
for signal detection admitting noise that swamped the signal. Abrupt changes in
thresholds might show mis-documented sensor sensitivity changes at individual stations. |