The dead_channel_lin metric is calculated by fitting the PSD mean curve for the current day to a linear curve and calculating the standard deviation of the residuals. This is a closely-related alternative to the dead_channel_exp metric. The mean of a healthy set of broadband PSDs will have a very non-linear shape with large residuals near the microseism peaks, while a “dead channel” will typically mimic a linear shape as a function of log(period). Before fitting a linear curve, the dead_channel_lin metric trims a few frequency bins from either end of the PSD mean curve to eliminate the effects of sparse sampling at the channel’s shortest and longest periods.
Healthy channels are characterized by large measurements (the PSD mean will not fit a line in PSD mean-log(period) space very well, so the standard deviation of the fit residuals will be large). Channels that are dead or not returning expected seismic energy are characterized by small standard deviations. A value < 3 typically indicates a broadband channel that may have a problem.
Small values for this metric can flag dead (or otherwise unhealthy) channels caused by offline sensors, sensor malfunction, datalogger malfunction, and pegged sensor masses. Environmental noise or gaps in the data may also be the cause of small values.
Traces – one N.S.L.C (Network.Station.Location.Channel) per measurement
Window – up to 24 hours depending on the completeness of data
Data Source – IRIS SEED archive
SEED Channel Types – [BCDFH]H? | High Gain
- Retrieve the PSD mean as a function of period for the current day.
- Trim the period range to span (4/sample rate) to 100 seconds.
- Fit the [PSD mean vs. log(period)] to a line.
- Calculate the standard deviation of the residuals of the linear fit:
value = sqrt [ average( BestFitLine(Ti) - PSDmean(Ti) ) ] for i = 1,…,N
value – the standard deviation of the residuals of the log(PSD mean) vs. log(period) fit
target – the trace analyzed, labeled as N.S.L.C.Q (Network.Station.Location.Channel.Quality)
start – time of the first sample of time series data for that day in UTC
end – time of the last sample of time series data for that day in UTC
lddate – date/time the measurement was made and loaded into the MUSTANG database (UTC)