Merging with very small differences in each sample.

[krischer]

Should we support merging the following to Traces?

|          |         |          |                Trace 1
                       |          |          |   Trace 2

It currently works. What should be the endtime? Currenly each sample of Trace 2 is shifted to fit Trace 1.

[megies]

It would certainly be better to use a linear interpolation of trace2 on the sampling "grid" of trace1. Would this be too slow/expensive? Anyway, it's a time error of subsample order so this might be overkill... But if the samples are shifted it should at least print a warning I think...

[anonymous]

• merge with method=0 should ofc mark this data as gap
• using method>0 should allow interpolation - but mark somehow the timeshift within the stats attribute of the trace (e.g. quality flag) - a warning alone won't be enough for automatic processing

[beyreuth]

• If the offset is more than 5% of the sampling interval we should refuse to merge such traces. The result is just wrong.
• There is no way to automatically find which timestamp is the correct one.

[krischer]

• By default libmseed has a time tolerance between two records of 1/2 sampling period. So we should either change the default value or adhere to the same tolerance.

• Interpolation would probably also be pretty cheap, e.g. in the example above assuming 5% offset something like

  trace2 = 0.05*trace2[:-1] + 0.95*trace2[1:]
  

  The first sample needs seperate treatment.

[barsch]

>>> from obspy.core import UTCDateTime, Trace
>>> t1=Trace(data=np.arange(100))
>>> t2=Trace(data=np.arange(100))
>>> t1.stats.starttime = UTCDateTime(100)
>>> t2.stats.starttime = UTCDateTime(100)

>>> print t1
... | 1970-01-01T00:01:40.000000Z - 1970-01-01T00:03:19.000000Z | 1.0 Hz, 100 samples
>>> print t2
... | 1970-01-01T00:03:20.500000Z - 1970-01-01T00:04:59.500000Z | 1.0 Hz, 100 samples

>>> muh = t1 + t2
>>> print(muh)
... | 1970-01-01T00:01:40.000000Z - 1970-01-01T00:04:59.000000Z | 1.0 Hz, 200 samples

>>> muh.data[99]
99
>>> muh.data[100]
0
>>> muh.data[101]
1

[barsch]

group decision: raise a warning within merge() - needs test case

[barsch]

This should now be invalid by using UTCDateTime.DEFAULT_PRECISION - see  http://docs.obspy.org/packages/autogen/obspy.core.utcdatetime.UTCDateTime.html#precision.