Rotational Seismology
Estimate Event Backazimuth



In [1]:
%matplotlib inline
import matplotlib.pyplot as plt'ggplot')
plt.rcParams['figure.figsize'] = 15, 8

1 Load pre-processed waveforms and event-xml

Waveforms of transverse acceleration and vertical rotation rate are loaded from the repository.
They were pre-processed as described in the Data Download + Pre-processing-Tutorial.
That means they are:

  • instrument corrected
  • resampled (decimation factor: 4)
  • filtered (lowpass: $f_{cutoff}$ = 2 Hz)
  • </ul>
    We also need the locations of station and event epicenter to obtain the theoretical backazimuth.

In [2]:
from obspy import read, read_events
from obspy.geodetics.base import gps2dist_azimuth
import numpy as np

AC = read('data/acc_Turkey_preproc.mseed')
RLAS = read('data/rot_Turkey_preproc.mseed')
cat = read_events('data/xml_Turkey.xml')
event = cat[0]

# event location from event info
source_latitude =[0].latitude
source_longitude =[0].longitude

# station location (Wettzell)
station_latitude = 49.144001
station_longitude = 12.8782

print(event.event_descriptions[0]['type'], ': ',event.event_descriptions[0]['text'])
1 Event(s) in Catalog:
2011-10-23T10:41:27.500000Z | +38.640,  +43.400 | 7.1 MW
Flinn-Engdahl region :  TURKEY

Obtain the theoretical backazimuth via gps2distazimuth

In [3]:
# theoretical backazimuth and distance
baz = gps2dist_azimuth(source_latitude, source_longitude, station_latitude, station_longitude)

print('Epicentral distance [m]: ',np.round(baz[0],1))
print('Theoretical azimuth [deg]: ', np.round(baz[1],1))
print('Theoretical backazimuth [deg]: ', np.round(baz[2],1))
Epicentral distance [m]:  2691458.3
Theoretical azimuth [deg]:  305.6
Theoretical backazimuth [deg]:  104.1

2 Estimate correlation coefficients and backazimuth for each time window

The estimation of the backazimuth (BAz = source direction angle) is based on Igel et al.(2007) and illustrated
as a simple example here:

We define the length of the time windows, we want to use and calculate the number of time windows for
our downloaded and pre-processed waveform data.

For each time-window, we rotate the horizontal acceleration components through all possible
source directions (BAz, 0-360°) in 10°-steps. In each of the steps, we check the similarity (phase-match) between
vertical rotation rate and the rotated acceleration by cross-correlation (X-corr).

The rotation angle/ BAz-value yielding the highest correlation coefficient for the specific time window rotates the
horizontal acceleration components to radial/ transverse (= into the angular direction of the strongest signal source).
That's exactly what McLeod et al. (1998) and Pancha et al. (2000) showed for Love waves:

Vertical rotation rate and transverse acceleration are in phase!

In [4]:
from obspy.signal.cross_correlation import xcorr
from obspy.signal.rotate import rotate_ne_rt

sampling_rate = int(RLAS[0].stats.sampling_rate)
sec = 60  # window length for correlation
num_windows = len(RLAS[0]) // (int(sampling_rate * sec))

# estimate the Backazimuth for each time window
step = 10
backas = np.linspace(0, 360 - step, 360 / step)
corrbaz = []
for i_deg in range(0, len(backas)):
    for i_win in range(0, num_windows):
        corrbazz = xcorr(RLAS[0][sampling_rate * sec * i_win : sampling_rate * sec * (i_win + 1)],
                                'E')[0].data, backas[i_deg])
                             [1][sampling_rate * sec * i_win : sampling_rate * sec * (i_win + 1)],0)
corrbaz = np.asarray(corrbaz)
corrbaz = corrbaz.reshape(len(backas), num_windows)

maxcorr = []
for l1 in range(0, num_windows):
    maxcor_r = backas[corrbaz[:, l1].argmax()]
maxcorr = np.asarray(maxcorr)
X, Y = np.meshgrid(np.arange(0, sec * num_windows, sec), backas)
/Users/lion/miniconda3/envs/seismo_live/lib/python3.7/site-packages/ DeprecationWarning: object of type <class 'float'> cannot be safely interpreted as an integer.
  # Remove the CWD from sys.path while we load stuff.
/Users/lion/miniconda3/envs/seismo_live/lib/python3.7/site-packages/obspy/signal/ ObsPyDeprecationWarning: Call to deprecated function xcorr(). Please use the correlate and xcorr_max functions.
  warnings.warn(msg, ObsPyDeprecationWarning)

Get theoretical P- and S-Wave first arrivals from a 1D velocity model using TauPy:

In [5]:
from obspy.taup import TauPyModel

TauPy_model = TauPyModel('ak135')
arrivals_p = TauPy_model.get_travel_times(distance_in_degree=0.001 * baz[0] / 111.11,
arrivals_s = TauPy_model.get_travel_times(distance_in_degree=0.001 * baz[0] / 111.11,
tiemp = []
tiems = []
for i in range(0,len(arrivals_p)): tiemp.append(arrivals_p[i].time)
for ii in range(0,len(arrivals_s)): tiems.append(arrivals_s[ii].time)

# first arrivals
arriv_p = min(tiemp)
arriv_s = min(tiems)
/Users/lion/miniconda3/envs/seismo_live/lib/python3.7/site-packages/obspy/taup/ UserWarning: Resizing a TauP array inplace failed due to the existence of other references to the array, creating a new array. See Obspy #2280.

3 Plot backazimuth estimation

Vertical rotation rate

Transverse acceleration (rotated from N-E component acceleration, see above)

Bottom: Backazimuth estimation

  • Green displays anti-correlation, red shows positive correlation/ phase match.
  • Yellow indicates no correlation.
  • Black dots highlight the BAz values for the maximum correlation coefficient in that time window.
  • The theoretical backazimuth is plotted for comparison.
In [6]:
import matplotlib as mpl

time = np.linspace(0, len(AC[0].data)/sampling_rate,len(AC[0].data)) ## Time-array
# vertical rotation rate
plt.subplot2grid((3, 30), (0, 0), colspan=29)
plt.plot(time, RLAS[0].data,label='vertical rotation rate')
plt.xlim(0, time[-1])
plt.ylabel('vert. rot. rate \n[nrad/s]')

# add P- and S-wave arrivals
plt.axvline(arriv_p);plt.annotate('P-arrival', xy=(arriv_p+20,np.max(RLAS[0].data)),xycoords='data');
plt.axvline(arriv_s);plt.annotate('S-arrival', xy=(arriv_s+20,np.max(RLAS[0].data)),xycoords='data');

# transverse acceleration
plt.subplot2grid((3, 30), (1, 0), colspan=29)
plt.plot(time, AC[0].data, 'k',label='transverse acceleration')
plt.xlim(0, time[-1])
plt.ylabel('transv. acc. \n[$nm/s^2$]')
plt.ticklabel_format(axis='y', style='sci', scilimits=(-2,2))

# backazimuth estimation plot
plt.subplot2grid((3, 30), (2, 0), colspan=29)
im = plt.pcolor(X, Y, corrbaz,, vmin=-1,vmax=1)
plt.plot(np.arange(sec/2., sec * num_windows, sec), maxcorr, '.k')
plt.xlim(0, time[-1])
plt.ylim(0, 360)
plt.ylabel(u'estimated \nbackazimuth [°]')
plt.xlabel('time [s]')

# plot theoretical Backazimuth for comparison
xx = np.arange(0, sec * num_windows + 1, sec)
tba = np.ones(len(xx)) * baz[2]
plt.plot(xx, tba, c='.5', lw=1.5)
plt.text(2600, 106, u'Theor. BAz = '+str(round(baz[2],2))+'°', color='k',fontsize=12,backgroundcolor='.9')

# add colorbar
fig = plt.subplot2grid((3, 30), (2, 29))
norm = mpl.colors.Normalize(vmin=-1, vmax=1)
cb1 = mpl.colorbar.ColorbarBase(fig,, norm=norm, orientation='vertical')