Experimental Setup Guide¶
Sensor selection and configuration¶
The unknown sensor response (URESP) must fit comfortably within the flat frequency response of the known sensor (KRESP).
For example, it is impossible to derive a proper frequency response for a broad band sensor like the Nanometric Trillium Compact (Low frequency corner at ~0.01 Hz) from a short-period sensor like OSOP’s Sixaola (low frequency cornerat ~0.5 hz) as the short-period sensor is blind to the lower frequencies past its low frequency corner.
Sensor installation¶
The theory behind respGen explained:
- “Two seismometers are placed close together so that it can be assumed that they record the same ground motion.” - Sleeman et al., 2006
- “The data used are obtained by placing two sets of sensors close enough together that we can assume they record the same ground motion. ” - Pavlis and Vernon, 1994
Sensors, therefore, must be co-located for the method to deliver meaningful results. It is also recommended that they are leveled well and aligned in the same direction (normally towards North). If you are only comparing date from the vertical components then the sensors do not need to be co-aligned.
To ensure good coherence between sensors at higher frequencies,
- Ensure the instruments are well coupled to a common, stable platform
- Reduce contamination from nearby acoustic and electronic sources
Data acquisition & duration¶
The duration of your experiment will depend on the instrument response of the instruments. The lower the flat frequency response of the “unknown” sensor, the longer the experiment must run.
For broadband sensors, it is best practice to let the sensors run for a few days to let the sensors settled down (Ringler and Hutt, 2010).
In reference to experiments with broad band sensors, Ringler et al. 2011 state “…for at least 24 hours so as to be likely to yield at least one 10-hr sample with a few or no detectable transient Earth signals or instrument pulses.”
We suggest you adhere to the 100x rule: Record data for 100x the longest period for which you want good results. Stated in another way, the required sample duration should be 100 times the longest period needed to be measured. For the short-period instruments with flat frequency responses from ~0.5 Hz to ~0.8*Nyquist, for example, plotting to 10 seconds period should cover any meaningful frequencies, so you would need to record only 1000 seconds (17 minutes) of data.
A support reference for the 100x rule: http://www.earthmode.org/procedure.htm (from http://www.earthmode.org/)
Warning
The sample rates of both sensors MUST match
Data selection¶
THIS IS IMPORTANT!!!!
A good signal-to-noise ratio is essential¶
Data selection is absolutely critical to a successful analysis with respGen. Remember:
garbage in = garbage out
The more perfect the coherency between sensors, the better the results. Coherency should be close to 1 from the lowest frequency of interest (or slightly less than the lower corner frequency of the “unknown sensor”) allow the way up to 0.8*Nyquist (where the antialias corner generally is).
As observed by Pavlis and Version (1994), Ground noise must be “of sufficiently high amplitude to guarantee that one is recording above the” instrument noise “across the entire frequency band of interest” and “[Data must be] of sufficiently high amplitude to guarantee that one is recording above the amplifier noise floor across the entire frequency band of interest.”. That is to say that your instrumentation MUST be able to record true ground motion and not random instrumental noise. If you are using a digitizer with << 24 true bits, beware! The instrumentation MUST be able to resolve the ambient seismic noise for respGen to work.
Should I wait for a seismic event or use background seismic noise?: “One could choose to be cautious and wait for distinct events and use those signals. However, as shown below, this is not at all necessary, and we have found excellent results can be obtained by simply recording background noise.” -Pavlis and Vernon, 1994
When selecting the data for respGen, try to avoid transients/ spikes/ obvious electronic noise.
For this reason, respGen is not as good in estimating the transfer coefficients in the low frequency corner as it is in the ‘flat’ region of the response. At the lower corner it is more likely that we are no longer recording true ground motion, but only instrument noise.
Note
We do attempt to throw out frequencies with poor coherence (COHE_LIMIT) that might occur at instrument noise dominated frequencies.
Alignment/ overlap¶
It is not a problem if the input signals start and end at different times. The program will cut them so that they overlap.
Gaps¶
The input signals (waveforms in miniSEED format) should not contain gaps. If the program detects gaps, it prints and error and exits.