def remove_resp(self, trace, paz=None):
# removing response...
if paz:
# ...using paz:
if trace.stats.sampling_rate > 10.0:
# decimating large trace, else fft crashes
factor = int(np.ceil(trace.stats.sampling_rate / 10))
trace.decimate(factor=factor, no_filter=True)
trace.simulate(paz_remove=paz,
paz_simulate=obspy.signal.cornFreq2Paz(0.01),
remove_sensitivity=True,
simulate_sensitivity=True,
nfft_pow2=True)
else:
# ...using StationXML:
# first band-pass to downsample data before removing response
# (else remove_response() method is slow or even hangs)
trace.filter(type="bandpass",
freqmin=self.freqmin,
freqmax=self.freqmax,
corners=self.corners,
zerophase=self.zerophase)
psutils.resample(trace, dt_resample=self.period_resample)
trace.remove_response(output="VEL", zero_mean=True)
return trace
def remove_resp(self, trace, paz=None):
# removing response...
if paz:
# ...using paz:
if trace.stats.sampling_rate > 10.0:
# decimating large trace, else fft crashes
factor = int(np.ceil(trace.stats.sampling_rate / 10))
trace.decimate(factor=factor, no_filter=True)
trace.simulate(paz_remove=paz,
paz_simulate=obspy.signal.cornFreq2Paz(0.01),
remove_sensitivity=True,
simulate_sensitivity=True,
nfft_pow2=True)
else:
# ...using StationXML:
# first band-pass to downsample data before removing response
# (else remove_response() method is slow or even hangs)
trace.filter(type="bandpass",
freqmin=self.freqmin,
freqmax=self.freqmax,
corners=self.corners,
zerophase=self.zerophase)
psutils.resample(trace, dt_resample=self.period_resample)
trace.remove_response(output="VEL", zero_mean=True)
return trace
def time_norm(self, trace, trcopy):
if self.onebit_norm:
# one-bit normalization
trace.data = np.sign(trace.data)
else:
# normalization of the signal by the running mean
# in the earthquake frequency band
trcopy.filter(type="bandpass",
freqmin=self.freqmin_earthquake,
freqmax=self.freqmax_earthquake,
corners=self.corners,
zerophase=self.zerophase)
# decimating trace
psutils.resample(trcopy, self.period_resample)
# Time-normalization weights from smoothed abs(data)
# Note that trace's data can be a masked array
halfwindow = int(
round(self.window_time * trcopy.stats.sampling_rate / 2))
mask = ~trcopy.data.mask if np.ma.isMA(trcopy.data) else None
tnorm_w = psutils.moving_avg(np.abs(trcopy.data),
halfwindow=halfwindow,
mask=mask)
if np.ma.isMA(trcopy.data):
# turning time-normalization weights into a masked array
s = "[warning: {}.{} trace's data is a masked array]"
print s.format(trace.stats.network, trace.stats.station),
tnorm_w = np.ma.masked_array(tnorm_w, trcopy.data.mask)
if np.any((tnorm_w == 0.0) | np.isnan(tnorm_w)):
# illegal normalizing value -> skipping trace
raise pserrors.CannotPreprocess("Zero or NaN normalization weight")
# time-normalization
trace.data /= tnorm_w
return trace
def time_norm(self, trace, trcopy):
if self.onebit_norm:
# one-bit normalization
trace.data = np.sign(trace.data)
else:
# normalization of the signal by the running mean
# in the earthquake frequency band
trcopy.filter(type="bandpass",
freqmin=self.freqmin_earthquake,
freqmax=self.freqmax_earthquake,
corners=self.corners,
zerophase=self.zerophase)
# decimating trace
psutils.resample(trcopy, self.period_resample)
# Time-normalization weights from smoothed abs(data)
# Note that trace's data can be a masked array
halfwindow = int(round(self.window_time*trcopy.stats.sampling_rate/2))
mask = ~trcopy.data.mask if np.ma.isMA(trcopy.data) else None
tnorm_w = psutils.moving_avg(np.abs(trcopy.data),
halfwindow=halfwindow,
mask=mask)
if np.ma.isMA(trcopy.data):
# turning time-normalization weights into a masked array
s = "[warning: {}.{} trace's data is a masked array]"
print s.format(trace.stats.network, trace.stats.station),
tnorm_w = np.ma.masked_array(tnorm_w, trcopy.data.mask)
if np.any((tnorm_w == 0.0) | np.isnan(tnorm_w)):
# illegal normalizing value -> skipping trace
raise pserrors.CannotPreprocess("Zero or NaN normalization weight")
# time-normalization
trace.data /= tnorm_w
return trace
def preprocess_trace(self, trace, paz=None, verbose=False):
"""
Preprocesses a trace (so that it is ready to be cross-correlated),
by applying the following steps:
- removal of instrument response, mean and trend
- band-pass filtering between *freqmin*-*freqmax*
- downsampling to *period_resample* secs
- time-normalization (one-bit normalization or normalization
by the running mean in the earthquake frequency band)
- spectral whitening (if running mean normalization)
Raises CannotPreprocess exception if:
- trace only contains 0 (happens sometimes...)
- a normalization weight is 0 or NaN
- a Nan appeared in trace data
Note that the processing steps are performed in-place.
@type trace: L{Trace}
@param paz: poles and zeros of instrumental response
(set None if response is directly attached to trace)
"""
# =======================================
# Check if any data to work with to begin
# =======================================
if np.all(trace.data == 0.0):
# no data -> skipping trace
raise pserrors.CannotPreprocess("Only zeros")
# ==========================
# Remove instrument response
# ==========================
if RESP_REMOVE:
t1 = dt.datetime.now()
trace = self.remove_resp(trace, paz=paz)
delta = (dt.datetime.now() - t1).total_seconds()
if verbose:
print "\nRemoved response in {:.1f} seconds".format(delta)
#Initialise the AutoTrigger class for the current trace
if HIGHAMP_REMOVE:
# this check finds high amplitude noise creates a list of where it is located in the signal
TRIGGER = AutoTrigger(trace, freqmin=FREQMIN, freqmax=FREQMAX)
UTC_events = TRIGGER.trigger_times(check=True)
if UTC_events is not None:
# set the application to remove these parts from the trace!
trace_new = None
for event in UTC_events:
# event is a list of length 2, start and end times
# use trace.trim(starttime=x, endtime=y)
# trace 1 is from trace start until the event starttime
trace1 = trace.trim(starttime=trace.stats.starttime,
endtime=event[0])
trace2 = trace.trim(starttime=event[1],
endtime=trace.stats.endtime)
if trace_new is None:
trace_new = trace1 + trace2
else:
trace_new += (trace1 + trace2)
# trace_new has removed all high amp. noise from trace original
trace = trace_new
print trace.data
# ==============================================
# Check trace data completeness is above MINFILL
# ==============================================
if COMPLETENESS:
# Data fill for current TIME INTERVAL!
# CHECK THAT THE get_fill_trace function works!
fill = psutils.get_fill_trace(trace)
if fill < MINFILL:
# not enough data
raise pserrors.CannotPreprocess("{:.0f}% fill"\
.format(fill*100))
# ========================
# Trim, demean and detrend
# ========================
if TDD:
t1 = dt.datetime.now()
midt = trace.stats.starttime + (trace.stats.endtime -
trace.stats.starttime) / 2.0
t0 = UTCDateTime(midt.date)# date of trace, at time 00h00m00s
trace.trim(starttime=t0, endtime=t0 + dt.timedelta(days=1))
trace.detrend(type='constant')
trace.detrend(type='linear')
delta = (dt.datetime.now() - t1).total_seconds()
if verbose:
print "\nProcessed trim in {:.1f} seconds".format(delta)
# =========
# Band-pass
# =========
if BANDPASS:
t0 = dt.datetime.now()
trace.filter(type="bandpass",
freqmin=self.freqmin,
freqmax=self.freqmax,
corners=self.corners,
zerophase=self.zerophase)
delta = (dt.datetime.now() - t0).total_seconds()
if verbose:
print "\nProcessed filters in {:.1f} seconds".format(delta)
# ============
# Downsampling
# ============
if DOWNSAMPLE:
if abs(1.0 / trace.stats.sampling_rate - self.period_resample)>EPS:
psutils.resample(trace, dt_resample=self.period_resample)
# ==================
# Time normalization
# ==================
if TIME_NOMALISATION:
t0 = dt.datetime.now()
# keeping a copy of the trace for weights of time-normalization
trcopy = trace.copy()
trace = self.time_norm(trace, trcopy)
delta = (dt.datetime.now() - t0).total_seconds()
if verbose:
print "\nProcessed time-normalisation in {:.1f} seconds"\
.format(delta)
# ==================
# Spectral whitening
# ==================
if SPEC_WHITENING:
t0 = dt.datetime.now()
trace = self.spectral_whitening(trace)
delta = (dt.datetime.now() - t0).total_seconds()
if verbose:
print "\nProcessed spectral whitening in {:.1f} seconds".\
format(delta)
# ==============================================
# Verifying that we don't have nan in trace data
# ==============================================
if np.any(np.isnan(trace.data)):
raise pserrors.CannotPreprocess("Got NaN in trace data")
def preprocess_trace(self, trace, paz=None, verbose=False):
"""
Preprocesses a trace (so that it is ready to be cross-correlated),
by applying the following steps:
- removal of instrument response, mean and trend
- band-pass filtering between *freqmin*-*freqmax*
- downsampling to *period_resample* secs
- time-normalization (one-bit normalization or normalization
by the running mean in the earthquake frequency band)
- spectral whitening (if running mean normalization)
Raises CannotPreprocess exception if:
- trace only contains 0 (happens sometimes...)
- a normalization weight is 0 or NaN
- a Nan appeared in trace data
Note that the processing steps are performed in-place.
@type trace: L{Trace}
@param paz: poles and zeros of instrumental response
(set None if response is directly attached to trace)
"""
# =======================================
# Check if any data to work with to begin
# =======================================
#print type(trace)
#print "trace: ", trace
if np.all(trace.data == 0.0):
# no data -> skipping trace
raise pserrors.CannotPreprocess("Only zeros")
# ==========================
# Remove instrument response
# ==========================
if RESP_REMOVE:
t1 = dt.datetime.now()
trace = self.remove_resp(trace, paz=paz)
delta = (dt.datetime.now() - t1).total_seconds()
if verbose:
print "\nRemoved response in {:.1f} seconds".format(delta)
#Initialise the AutoTrigger class for the current trace
if HIGHAMP_REMOVE:
# this check finds high amplitude noise creates a list of where it is located in the signal
TRIGGER = AutoTrigger(trace, freqmin=FREQMIN, freqmax=FREQMAX)
UTC_events = TRIGGER.trigger_times(check=True)
if UTC_events is not None:
# set the application to remove these parts from the trace!
trace_new = None
for event in UTC_events:
# event is a list of length 2, start and end times
# use trace.trim(starttime=x, endtime=y)
# trace 1 is from trace start until the event starttime
trace1 = trace.trim(starttime=trace.stats.starttime,
endtime=event[0])
trace2 = trace.trim(starttime=event[1],
endtime=trace.stats.endtime)
if trace_new is None:
trace_new = trace1 + trace2
else:
trace_new += (trace1 + trace2)
# trace_new has removed all high amp. noise from trace original
trace = trace_new
print trace.data
# ==============================================
# Check trace data completeness is above MINFILL
# ==============================================
if COMPLETENESS:
# Data fill for current TIME INTERVAL!
# CHECK THAT THE get_fill_trace function works!
fill = psutils.get_fill_trace(trace)
if fill < MINFILL:
# not enough data
raise pserrors.CannotPreprocess("{:.0f}% fill"\
.format(fill*100))
# ========================
# Trim, demean and detrend
# ========================
if TDD:
t1 = dt.datetime.now()
midt = trace.stats.starttime + (trace.stats.endtime -
trace.stats.starttime) / 2.0
t0 = UTCDateTime(midt.date) # date of trace, at time 00h00m00s
trace.trim(starttime=t0, endtime=t0 + dt.timedelta(days=1))
trace.detrend(type='constant')
trace.detrend(type='linear')
delta = (dt.datetime.now() - t1).total_seconds()
if verbose:
print "Processed trim in {:.1f} seconds".format(delta)
#plt.figure()
#plt.title("TDD")
#plt.plot(trace.data)
#plt.show()
#plt.clf()
# =========
# Band-pass
# =========
if BANDPASS:
t0 = dt.datetime.now()
trace.filter(type="bandpass",
freqmin=self.freqmin,
freqmax=self.freqmax,
corners=self.corners,
zerophase=self.zerophase)
delta = (dt.datetime.now() - t0).total_seconds()
if verbose:
print "Processed filters in {:.1f} seconds".format(delta)
#plt.figure()
#plt.title("Bandpass")
#plt.plot(trace.data)
#plt.show()
#plt.clf()
# ============
# Downsampling
# ============
if DOWNSAMPLE:
if abs(1.0 / trace.stats.sampling_rate -
self.period_resample) > EPS:
psutils.resample(trace, dt_resample=self.period_resample)
#plt.figure()
#plt.title("DOWNSAMPLE")
#plt.plot(trace.data)
#plt.show()
#plt.clf()
# ==================
# Time normalization
# ==================
if TIME_NOMALISATION:
t0 = dt.datetime.now()
# keeping a copy of the trace for weights of time-normalization
trcopy = trace.copy()
trace = self.time_norm(trace, trcopy)
delta = (dt.datetime.now() - t0).total_seconds()
if verbose:
print "Processed time-normalisation in {:.1f} seconds"\
.format(delta)
#plt.figure()
#plt.title("NORMALISATION")
#plt.plot(trace.data)
#plt.show()
#plt.clf()
# ==================
# Spectral whitening
# ==================
if SPEC_WHITENING:
t0 = dt.datetime.now()
trace = self.spectral_whitening(trace)
delta = (dt.datetime.now() - t0).total_seconds()
if verbose:
print "Processed spectral whitening in {:.1f} seconds".\
format(delta)
#plt.figure()
#plt.title("SPECTRAL WHITENING")
#plt.plot(trace.data)
#plt.show()
#plt.clf()
# ==============================================
# Verifying that we don't have nan in trace data
# ==============================================
if np.any(np.isnan(trace.data)):
raise pserrors.CannotPreprocess("Got NaN in trace data")
