Merge branch 'develop' of ariadne.geophysik.ruhr-uni-bochum.de:/data/git/pylot into develop

This commit is contained in:
Marcel Paffrath 2015-09-30 13:55:01 +02:00
commit 2308695fa8
2 changed files with 81 additions and 21 deletions

View File

@ -9,6 +9,7 @@ import matplotlib.pyplot as plt
import numpy as np
from obspy.core import Stream
from pylot.core.pick.utils import getsignalwin
from scipy.optimize import curve_fit
class Magnitude(object):
'''
@ -166,20 +167,68 @@ class DCfc(Magnitude):
L = (N - 1) / tr.stats.sampling_rate
f = np.arange(0, fny, 1/L)
# remove zero-frequency and frequencies above
# corner frequency of seismometer (assumed
# to be 100 Hz)
fi = np.where((f >= 1) & (f < 100))
F = f[fi]
YY = Y[fi]
# get plateau (DC value) and corner frequency
# initial guess of plateau
DCin = np.mean(YY[0:100])
# initial guess of corner frequency
# where spectral level reached 50% of flat level
iin = np.where(YY >= 0.5 * DCin)
Fcin = F[iin[0][np.size(iin) - 1]]
fit = synthsourcespec(F, DCin, Fcin)
[optspecfit, pcov] = curve_fit(synthsourcespec, F, YY.real, [DCin, Fcin])
self.w0 = optspecfit[0]
self.fc = optspecfit[1]
print ("DCfc: Determined DC-value: %e m/Hz, \n" \
"Determined corner frequency: %f Hz" % (self.w0, self.fc))
if self.getiplot() > 1:
f1 = plt.figure(1)
f1 = plt.figure()
plt.subplot(2,1,1)
plt.plot(t, np.multiply(tr, 1000), 'k') # show displacement in mm
plt.plot(t[iwin], np.multiply(xdat, 1000), 'g') # show displacement in mm
# show displacement in mm
plt.plot(t, np.multiply(tr, 1000), 'k')
plt.plot(t[iwin], np.multiply(xdat, 1000), 'g')
plt.title('Seismogram and P pulse, station %s' % tr.stats.station)
plt.xlabel('Time since %s' % tr.stats.starttime)
plt.ylabel('Displacement [mm]')
plt.subplot(2,1,2)
plt.semilogy(f, Y.real)
plt.title('Source Spectrum from P Pulse')
plt.loglog(f, Y.real, 'k')
plt.loglog(F, YY.real)
plt.loglog(F, fit, 'g')
plt.title('Source Spectrum from P Pulse, DC=%e m/Hz, fc=%4.1f Hz' \
% (self.w0, self.fc))
plt.xlabel('Frequency [Hz]')
plt.ylabel('Amplitude [m/Hz]')
plt.grid()
plt.show()
raw_input()
plt.close(f1)
def synthsourcespec(f, omega0, fcorner):
'''
Calculates synthetic source spectrum from given plateau and corner
frequency assuming Akis omega-square model.
:param: f, frequencies
:type: array
:param: omega0, DC-value (plateau) of source spectrum
:type: float
:param: fcorner, corner frequency of source spectrum
:type: float
'''
#ssp = omega0 / (pow(2, (1 + f / fcorner)))
ssp = omega0 / (1 + pow(2, (f / fcorner)))
return ssp

View File

@ -313,7 +313,6 @@ def autopickstation(wfstream, pickparam):
##############################################################
# get DC value (w0) and corner frequency (fc) of source spectrum
# from P pulse
# restitute streams
# initialize Data object
data = Data()
[corzdat, restflag] = data.restituteWFData(invdir, zdat)
@ -323,33 +322,45 @@ def autopickstation(wfstream, pickparam):
# class needs stream object => build it
z_copy = zdat.copy()
z_copy[0].data = corintzdat
# largest detectable period == window length
# after P pulse for calculating source spectrum
winzc = (1 / bpz2[0]) * z_copy[0].stats.sampling_rate
impickP = mpickP * z_copy[0].stats.sampling_rate
wfzc = z_copy[0].data[impickP : impickP + winzc]
# calculate spectrum using only first cycles of
# waveform after P onset!
zc = crossings_nonzero_all(wfzc)
if np.size(zc) == 0:
print ("Something is wrong with the waveform, " \
"no zero crossings derived!")
print ("Cannot calculate source spectrum!")
else:
calcwin = (zc[3] - zc[0]) * z_copy[0].stats.delta
# calculate source spectrum and get w0 and fc
calcwin = 1 / bpz2[0] # largest detectable period == window length
# around P pulse for calculating source spectrum
specpara = DCfc(z_copy, mpickP, calcwin, iplot)
w0 = specpara.getw0()
fc = specpara.getfc()
print 'autopickstation: P-weight: %d, SNR: %f, SNR[dB]: %f, ' \
'Polarity: %s' % (Pweight, SNRP, SNRPdB, FM)
print ("autopickstation: P-weight: %d, SNR: %f, SNR[dB]: %f, " \
"Polarity: %s" % (Pweight, SNRP, SNRPdB, FM))
Sflag = 1
else:
print 'Bad initial (AIC) P-pick, skipping this onset!'
print ("Bad initial (AIC) P-pick, skipping this onset!")
print 'AIC-SNR=', aicpick.getSNR(), 'AIC-Slope=', aicpick.getSlope(), 'counts/s'
print '(min. AIC-SNR=', minAICPSNR, ', min. AIC-Slope=', minAICPslope, 'counts/s)'
Sflag = 0
else:
print 'autopickstation: No vertical component data available!, ' \
'Skipping station!'
print ("autopickstation: No vertical component data available!, " \
"Skipping station!")
if edat is not None and ndat is not None and len(edat) > 0 and len(
ndat) > 0 and Pweight < 4:
print 'Go on picking S onset ...'
print '##################################################'
print 'Working on S onset of station %s' % edat[0].stats.station
print 'Filtering horizontal traces ...'
print ("Go on picking S onset ...")
print ("##################################################")
print ("Working on S onset of station %s" % edat[0].stats.station)
print ("Filtering horizontal traces ...")
# determine time window for calculating CF after P onset
cuttimesh = [round(max([mpickP + sstart, 0])),