def make_trigger(origin_folder, filename, dest_folder, trig_on, trig_of,
trigger_type, nsta, nlta):
stream = obspy.read(os.path.join(origin_folder, filename))
freqmin = bandpass[1]
freqmax = bandpass[0]
print freqmin, freqmax
filtered = stream.filter("bandpass",
freqmin=bandpass[1],
freqmax=bandpass[0])
fm = float(filtered[0].stats.sampling_rate)
to_process = filtered.copy()
merged = to_process.merge(method=0)
if np.isnan(merged).any():
to_process = filtered.copy()
data = merge_numpy(to_process)
print type(data)
print type(nsta)
print type(nlta)
cft = recursive_sta_lta(data, int(float(nsta) * fm),
np.int(float(nlta) * fm))
on_of = trigger_onset(cft, trig_on, trig_off)
else:
# we can process the whole day.
data = merged[0]
cft = recursive_sta_lta(data, int(nsta * fm), int(nlta * fm))
on_of = trigger_onset(cft, trig_on, trig_off)
# we can process the whole day.
cft = recursive_sta_lta(data, int(nsta * fm), int(nlta * fm))
on_of = trigger_onset(cft, trig_on, trig_of)
save_stalta(dest_folder, filename, data, cft, on_of, trig_on, trig_of, fm)
def predict(dtfl, ev_list, dataOperator):
for c, evi in enumerate(ev_list):
try:
if c % 1000 == 0:
print(c)
dataset = dtfl.get('data/'+str(evi))
data = np.array(dataset)
pre_E = trigger_onset(recursive_sta_lta(
data[:, 0], config['sta_window'], config['lta_window']), config['on_trigger'], config['off_trigger'])
pre_N = trigger_onset(recursive_sta_lta(
data[:, 1], config['sta_window'], config['lta_window']), config['on_trigger'], config['off_trigger'])
# pre_Z = trigger_onset(recursive_sta_lta(
# data[:, 2], config['sta_window'], config['lta_window']), config['on_trigger'], config['off_trigger'])
N_end_time, E_end_time = 6000, 6000
if len(pre_E) == 0 and len(pre_N) == 0:
dataOperator.data_writer(dataset.attrs['trace_name'], dataset.attrs['p_arrival_sample'],
dataset.attrs['s_arrival_sample'], dataset.attrs['coda_end_sample'],
-1, -1, -1, dataset.attrs['trace_category'], "noise")
continue
if dataset.attrs['trace_category'] == 'noise':
dataOperator.data_writer(dataset.attrs['trace_name'], dataset.attrs['p_arrival_sample'],
dataset.attrs['s_arrival_sample'], dataset.attrs['coda_end_sample'],
-1, -1, -1, dataset.attrs['trace_category'], "earthquake_local")
continue
if len(pre_E):
E_end_time = pre_E[-1][1]
if len(pre_N):
N_end_time = pre_N[-1][1]
end_time = (E_end_time + N_end_time) / 2
p_pick, s_pick = ar_pick(data[:, 0], data[:, 1], data[:, 2], 100,
1.0, 20.0, 1.0, 0.1, 4.0, 1.0, 2, 8, 0.1, 0.2)
p_pick, s_pick = p_pick*100, s_pick*100
# y_true = [float(dataset.attrs['p_arrival_sample']),
# float(dataset.attrs['s_arrival_sample']), float(dataset.attrs['coda_end_sample'][0][0])]
# y_pred = [p_pick, s_pick, end_time]
# p_true = np.zeros(shape=(6000,))
# p_true[p_pick-20:p_pick+21] = 1
# a = np.array(y_true)
# b = np.array(y_pred)
# print(a * b)
# break
dataOperator.data_writer(dataset.attrs['trace_name'], dataset.attrs['p_arrival_sample'],
dataset.attrs['s_arrival_sample'], dataset.attrs['coda_end_sample'],
int(p_pick), int(s_pick), int(end_time), dataset.attrs['trace_category'], "earthquake_local")
except:
continue
return
def get_triggers_stalta(st):
len_sta, len_lta, trig_on, trig_off, freqmin, freqmax = load_stalta_parameters(
)
stations_list = []
triggers_list = []
if len(st) > 0:
for tr in st:
try:
tr.detrend('demean')
tr.detrend('linear')
tr.taper(max_percentage=0.015, type='hann')
tr.filter("bandpass",
freqmin=freqmin,
freqmax=freqmax,
corners=2)
cft_rec = recursive_sta_lta(
tr.data, int(len_sta * tr.stats.sampling_rate),
int(len_lta * tr.stats.sampling_rate))
on_off = trigger_onset(cft_rec, trig_on, trig_off)
triggers = []
for trig in on_off:
on = tr.times("utcdatetime")[trig[0]]
off = tr.times("utcdatetime")[trig[1]]
triggers.append([on, off])
stations_list.append(tr.stats.station)
triggers_list.append(triggers)
except:
continue
return stations_list, triggers_list
def _init_first_arrivals(self):
def find_fa(file, channel):
for fa in self.genie.current_inversion_cfg.first_arrivals:
if fa.file == file and fa.channel == channel:
return fa
return None
new_first_arrivals = []
for meas in self._measurements:
if meas.data is not None:
for i, trace in enumerate(meas.data["data"]):
cft = recursive_sta_lta(trace.data, 40, 60)
t = np.argmax(cft) / trace.stats.sampling_rate
fa = find_fa(meas.file, i)
if fa is not None:
fa.time_auto = t
new_first_arrivals.append(fa)
else:
new_first_arrivals.append(
FirstArrival(file=meas.file,
channel=i,
time_auto=t))
self.genie.current_inversion_cfg.first_arrivals = new_first_arrivals
def predictinput(sacfile, algo):
#print(input_path)
bhnfile = read("/home/shilpa/Desktop/earthquakeproject/files/" + sacfile)
bhn_tr = bhnfile[0]
df = bhn_tr.stats.sampling_rate
bhn_trigger = recursive_sta_lta(bhn_tr.data, int(5 * df), int(10 * df))
bhnonoff = trigger_onset(bhn_trigger, 1.2, 0.5)
p_pick, s_pick = ar_pick(bhnfile[0].data, bhnfile[0].data, bhnfile[0].data,
df, 1.0, 20.0, 1.0, 0.1, 4.0, 1.0, 2, 8, 0.1, 0.2)
data = []
temp = []
temp.append(200)
temp.append(p_pick)
temp.append(s_pick)
temp.append(int(bhnonoff[0][1]) - int(bhnonoff[0][0]))
data.append(temp)
print(data)
if algo == "decisiontree":
prediction = d_model.predict(data)
elif algo == "randomforest":
prediction = r_model.predict(data)
elif algo == "svm":
prediction = r_model.predict(data)
result = " "
if (prediction[0] == 1):
result = "an Earthquake"
else:
result = "No Earthquake"
return result
def extract_filter_params(trace):
samp_rate = trace.stats.sampling_rate
trigs = []
p_arrivals = model.get_travel_times_geo(trace.stats.event_depth,
trace.stats.event_latitude,
trace.stats.event_longitude,
trace.stats.station_latitude,
trace.stats.station_longitude,
phase_list=[
'P',
])
mean_parrival = np.mean([parr.time for parr in p_arrivals])
for band in [(1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7), (7, 8),
(8, 9), (1.5, 2.5), (2.5, 3.5), (3.5, 4.5), (4.5, 5.5),
(5.5, 6.5), (6.5, 7.5), (7.5, 8.5), (8.5, 9.5)]:
tr_copy = trace.copy()
clean_trace(tr_copy,
tr_copy.stats.starttime,
tr_copy.stats.endtime,
freqmin=band[0],
freqmax=band[1])
cft = recursive_sta_lta(tr_copy.data, int(5 * samp_rate),
int(20 * samp_rate))
upper, lower = find_best_bounds(cft, samp_rate)
trigs.extend([
(onset, tr_copy.stats.channel, upper - lower, band, upper)
for onset in trigger_onset(cft,
upper,
lower,
max_len=(60 *
tr_copy.stats.sampling_rate),
max_len_delete=True)
])
trigs = [
trig for trig in trigs
if abs(trace.stats.event_time + mean_parrival - trace.stats.starttime -
(trig[0][0] / samp_rate)) < pphase_search_margin
]
if len(trigs) > 0:
mintrigdiff = abs(trace.stats.event_time + mean_parrival -
trace.stats.starttime - trigs[0][0][0])
besttrig = trigs[0][0][0]
best_trig_margin = trigs[0][2]
best_band = trigs[0][3]
best_upper = trigs[0][4]
for trig in trigs:
if abs(trace.stats.event_time + mean_parrival - trace.stats.starttime - trig[0][0]) <= mintrigdiff and \
trig[2] >= best_trig_margin and trig[4] >= best_upper:
mintrigdiff = abs(trace.stats.event_time + mean_parrival -
trace.stats.starttime - trig[0][0])
besttrig = trig[0][0]
best_trig_margin = trig[2]
best_band = trig[3]
best_upper = trig[4]
return (best_band, best_trig_margin, best_upper)
else:
print('Something went wrong ... ')
return None, None, None
def picker(data, rate, nsta=10, nlta=100, uncer=0.5):
stalta = recursive_sta_lta(data, nsta, nlta)
n_max = np.argmax(stalta)
quality = stalta[n_max]
n_onset = int(uncer * rate)
n_diff_max = np.argmax(np.diff(stalta[n_max - n_onset:n_max + n_onset]))
n_pick = n_max - n_onset + n_diff_max
arr = n_pick / rate
return arr, quality
def _filter(self): ''' Filters the stream associated with this class. ''' if self.filt: if self.filt in 'bandpass': self.stalta = recursive_sta_lta( self.stream[0].copy().filter(type=self.filt, freqmin=self.freqmin, freqmax=self.freqmax), int(self.sta * self.sps), int(self.lta * self.sps)) else: self.stalta = recursive_sta_lta( self.stream[0].copy().filter(type=self.filt, freq=self.freq), int(self.sta * self.sps), int(self.lta * self.sps)) else: self.stalta = recursive_sta_lta(self.stream[0], int(self.sta * self.sps), int(self.lta * self.sps))
def _channel_loop(tr, parameters, max_trigger_length=60, despike=False):
"""
Internal loop for parellel processing.
:type tr: obspy.core.trace
:param tr: Trace to look for triggers in.
:type parameters: list
:param parameters: List of TriggerParameter class for trace.
:type max_trigger_length: float
:type despike: bool
:return: trigger
:rtype: list
"""
for par in parameters:
if par['station'] == tr.stats.station and \
par['channel'] == tr.stats.channel:
parameter = par
break
else:
Logger.warning('No parameters set for station ' +
str(tr.stats.station))
return []
triggers = []
Logger.debug(tr)
tr.detrend('simple')
if despike:
median_filter(tr)
if parameter['lowcut'] and parameter['highcut']:
tr.filter('bandpass',
freqmin=parameter['lowcut'],
freqmax=parameter['highcut'])
elif parameter['lowcut']:
tr.filter('highpass', freq=parameter['lowcut'])
elif parameter['highcut']:
tr.filter('lowpass', freq=parameter['highcut'])
# find triggers for each channel using recursive_sta_lta
df = tr.stats.sampling_rate
cft = recursive_sta_lta(tr.data, int(parameter['sta_len'] * df),
int(parameter['lta_len'] * df))
if max_trigger_length:
trig_args = {'max_len_delete': True}
trig_args['max_len'] = int(max_trigger_length * df + 0.5)
tmp_trigs = trigger_onset(cft, float(parameter['thr_on']),
float(parameter['thr_off']), **trig_args)
for on, off in tmp_trigs:
cft_peak = tr.data[on:off].max()
cft_std = tr.data[on:off].std()
on = tr.stats.starttime + \
float(on) / tr.stats.sampling_rate
off = tr.stats.starttime + \
float(off) / tr.stats.sampling_rate
triggers.append(
(on.timestamp, off.timestamp, tr.id, cft_peak, cft_std))
return triggers
def compute_sta_lta(data,
fm,
trigger_type,
nlta=10.0,
nsta=5.0,
trig_on=1.2,
trig_off=0.5):
"""
Function that handles the building of STA/LTA event picking: classic, recursive and delayed. It follows Obspy
implementation of these algorithms and can be interfaced with the main GUI to plot the results, or with the
CLI to other analysis routines. A detailed comparison of STA/LTA techniques algorithms are included in:
Withers, M., Aster, R., Young, C., Beiriger, J., Harris, M., Moore, S., and Trujillo, J. (1998),
A comparison of select trigger algorithms for automated global seismic phase and event detection,
Bulletin of the Seismological Society of America, 88 (1), 95-106.
http://www.bssaonline.org/content/88/1/95.abstract
Args:
data : Numpy Array
The seismic data we want to apply our STA/LTA routine
fm : float
The sampling frequency of the main trace
trigger_type : str
A string identifiying which trigger type we want (Recursive, Delayed, Classic)
nlta : float
Length of the Long Time Average Window (LTA)
nsta : float
Length of the Short Time Average Window (STA)
trig_on : float
Value of the CF to consider as an activation trigger
trig_off : float
Value of the CF to consider as a de-activation trigger
Returns:
cft: Numpy Array
The characteristic function result of the
on_of: Tuple
A data tuple containing the on/ofs times of the even picking
"""
if np.isnan(data).any():
data = merge_numpy(data)
try:
if trigger_type == "Recursive":
cft = recursive_sta_lta(data, int(nsta * fm), int(nlta * fm))
elif trigger_type == "Delayed":
cft = delayed_sta_lta(data, int(nsta * fm), int(nlta * fm))
else:
cft = classic_sta_lta(data, int(nsta * fm), int(nlta * fm))
on_of = trigger_onset(cft, trig_on, trig_off)
return cft, on_of
except ArithmeticError:
print "Problem whilst computing the trigger"
def test_rec_sta_lta_c(self):
"""
Test case for ctypes version of recursive_sta_lta
"""
nsta, nlta = 5, 10
c1 = recursive_sta_lta(self.data, nsta, nlta)
self.assertAlmostEqual(c1[99], 0.80810165)
self.assertAlmostEqual(c1[100], 0.75939449)
self.assertAlmostEqual(c1[101], 0.91763978)
self.assertAlmostEqual(c1[102], 0.97465004)
def getTrigger(sac, short=2, long=30):
df = sac.stats.sampling_rate
# get cft
cft = recursive_sta_lta(sac.data, int(short * df), int(long * df))
# set threshold
threshold = np.mean(cft) + (np.max(cft) - np.mean(cft)) / 4
# get on
on_of = trigger_onset(cft, threshold, threshold)
if len(on_of) != 0:
return on_of[:, 0]
else:
return np.array([])
def getTrigger(sac, short=2, long=30):
df = sac.stats.sampling_rate
# get cft
cft = recursive_sta_lta(sac.data, int(short * df), int(long * df))
# set threshold
threshold = np.mean(cft) + (np.max(cft) - np.mean(cft))/4
# get on
on_of = trigger_onset(cft, threshold*1.1, threshold*0.9)
if len(on_of) != 0:
return on_of[:, 0]
else:
return np.array([])
def _doubleChecking(station_list, detections, preprocessed_dir, moving_window, thr_on=3.7, thr_of=0.5):
'this function perform traditional detection (STA/LTA) and picker (AIC) to double check for events on the remaining stations when an event has been detected on more than two stations'
for stt in station_list:
sttt = stt.split('_')[0]
# print(sttt)
if sttt not in detections['station'].to_list():
new_picks = {}
if platform.system() == 'Windows':
file_name = preprocessed_dir+"\\"+sttt+".hdf5"
file_csv = preprocessed_dir+"\\"+sttt+".csv"
else:
file_name = preprocessed_dir+"/"+sttt+".hdf5"
file_csv = preprocessed_dir+"/"+sttt+".csv"
df = pd.read_csv(file_csv)
df['start_time'] = pd.to_datetime(df['start_time'])
mask = (df['start_time'] > detections.iloc[0]['event_start_time']-timedelta(seconds = moving_window)) & (df['start_time'] < detections.iloc[0]['event_start_time']+timedelta(seconds = moving_window))
df = df.loc[mask]
dtfl = h5py.File(file_name, 'r')
dataset = dtfl.get('data/'+df['trace_name'].to_list()[0])
data = np.array(dataset)
cft = recursive_sta_lta(data[:,2], int(2.5 * 100), int(10. * 100))
on_of = trigger_onset(cft, thr_on, thr_of)
if len(on_of) >= 1:
p_pick, s_pick = ar_pick(data[:,2], data[:,1], data[:,0], 100, 1.0, 20.0, 1.0, 0.1, 4.0, 1.0, 2, 8, 0.1, 0.2)
if (on_of[0][1]+100)/100 > p_pick > (on_of[0][0]-100)/100:
# print('got one')
new_picks['traceID'] = df['trace_name'].to_list()[0]
new_picks['network'] = dataset.attrs["network_code"]
new_picks['station'] = sttt
new_picks['instrument_type'] = df['trace_name'].to_list()[0].split('_')[2]
new_picks['stlat'] = round(dataset.attrs["receiver_latitude"], 4)
new_picks['stlon'] = round(dataset.attrs["receiver_longitude"], 4)
new_picks['stelv'] = round(dataset.attrs["receiver_elevation_m"], 2)
new_picks['event_start_time'] = datetime.strptime(str(UTCDateTime(dataset.attrs['trace_start_time'].replace(' ', 'T')+'Z')+(on_of[0][0]/100)).replace('T', ' ').replace('Z', ''), '%Y-%m-%d %H:%M:%S.%f')
new_picks['event_end_time'] = datetime.strptime(str(UTCDateTime(dataset.attrs['trace_start_time'].replace(' ', 'T')+'Z')+(on_of[0][1]/100)).replace('T', ' ').replace('Z', ''), '%Y-%m-%d %H:%M:%S.%f')
new_picks['detection_prob'] = 0.3
new_picks['detection_unc'] = 0.6
new_picks['p_arrival_time'] = datetime.strptime(str(UTCDateTime(dataset.attrs['trace_start_time'].replace(' ', 'T')+'Z')+p_pick).replace('T', ' ').replace('Z', ''), '%Y-%m-%d %H:%M:%S.%f')
new_picks['p_prob'] = 0.3
new_picks['p_unc'] = 0.6
new_picks['p_snr'] = None
new_picks['s_arrival_time'] = None
new_picks['s_prob'] = 0.0
new_picks['s_unc'] = None
new_picks['s_snr'] = None
new_picks['amp'] = None
detections = detections.append(new_picks , ignore_index=True)
return detections
def semblancestalta(sembmaxvaluevector, sembmaxlatvector, sembmaxlonvector):
data = np.array(sembmaxvaluevector, dtype=np.float64)
tr = Trace(data, header=None)
sta = 0.5
lta = 4
cft = recursive_sta_lta(tr, int(sta * tr.stats.sampling_rate),
int(lta * tr.stats.sampling_rate))
thrOn = 0.5
thrOff = 1.5
plotTrigger(tr, cft, thrOn, thrOff)
def phase_trigger(self, time_before, time_after, stl=5, ltl=10):
t1, t2 = self._get_time(time_before, time_after)
self.st_pick = self.st.copy().trim(t1, t2)
if len(self.st_pick) == 0:
return
if self.phase[-1] == 'P':
tr = self.st_pick.select(channel='*Z')[0]
else:
tr = self.st_pick.select(channel='*T')[0]
df = tr.stats.sampling_rate
cft = recursive_sta_lta(tr.data, int(stl * df), int(ltl * df))
n_trigger = np.argmax(np.diff(cft)[int(ltl * df):]) + int(ltl * df)
self.t_trigger = t1 + n_trigger / df
self.trigger_shift = n_trigger / df - time_before
def standard_trigger_finder(trace, channel_name):
global std_on
global std_off
t = trace.copy()
max_triggers = 30
max_trigger_length = 20000
ctf_start = 300 # avoids triggering on initial spike
if "PDB" in channel_name:
#trace.filter('highpass', freq=1500)
sta = 20
lta = 60
ctf = recursive_sta_lta(t.data, sta, lta)
ctf = ctf[ctf_start:]
std_on = ctf[find_index_of_best_val(ctf, max_triggers)] * 0.98
std_off = std_on * 0.8
trigger_indices = trigger_onset(ctf, std_on, std_off,
max_trigger_length)
if "OT" in channel_name:
#t.filter('bandpass', freqmin=1000, freqmax=15000)
sta = 10
lta = 50
ctf = recursive_sta_lta(t.data, sta, lta)
ctf = ctf[ctf_start:]
#std_on = ctf[find_index_of_max_val(ctf, max_triggers)] * 0.94
std_on = ctf[find_index_of_best_val(ctf, max_triggers)] * 0.92
if (std_on < 1):
std_on += (1 - std_on) * 1.1
#print("std_on: " + str(std_on))
std_off = 1 #std_on * 0.92
trigger_indices = trigger_onset(ctf, std_on, std_off,
max_trigger_length)
return trigger_indices, ctf
def test_trigger_onset_issue_2891(self):
"""
Regression test for issue 2891
This used to raise an error if a trigger was activated near the end of
the trace, and all sample values after that trigger on threshold are
above the designated off threshold. So basically this can only happen
if the on threshold is below the off threshold, which is kind of
unusual, but we fixed it nevertheless, since people can run into this
playing around with different threshold settings
"""
tr = read(os.path.join(self.path,
'BW.UH1._.EHZ.D.2010.147.a.slist.gz'))[0]
cft = recursive_sta_lta(tr.data, 5, 30)
trigger_onset(cft, 2.5, 3.2)
def update_cft(prev_val, selected=None):
print(ticker_alg.value)
if ticker_alg.value == 'Classic STA/LTA':
from obspy.signal.trigger import classic_sta_lta, trigger_onset
on = trigger_slider.value[1]; off=trigger_slider.value[0]
cft = classic_sta_lta(st[0].data,
int(stalta_slider.value[0] * st[0].stats.sampling_rate),
int(stalta_slider.value[1] * st[0].stats.sampling_rate))
on_off = np.array(trigger_onset(cft, on, off))
source_stalta.data = dict(times=st[0].times(), cft=cft)
source_triggers.data = dict(ontimes=st[0].times()[on_off[:,0]], y=np.zeros(on_off[:,0].shape))
#source_triggers.data = dict(offtimes=st[0].times()[on_off[:,1]], y=np.zeros(on_off[:,1].shape))
sta_on.location = on
sta_off.location = off
elif ticker_alg.value == 'Recursive STA/LTA':
from obspy.signal.trigger import recursive_sta_lta, trigger_onset
on = trigger_slider.value[1]; off=trigger_slider.value[0]
cft = recursive_sta_lta(st[0].data,
int(stalta_slider.value[0] * st[0].stats.sampling_rate),
int(stalta_slider.value[1] * st[0].stats.sampling_rate))
on_off = np.array(trigger_onset(cft, on, off))
source_stalta.data = dict(times=st[0].times(), cft=cft)
source_triggers.data = dict(ontimes=st[0].times()[on_off[:,0]], y=np.zeros(on_off[:,0].shape))
#source_triggers.data = dict(offtimes=st[0].times()[on_off[:,1]], y=np.zeros(on_off[:,1].shape))
sta_on.location = on
sta_off.location = off
elif ticker_alg.value == 'Carl-Sta-Trig [Not Yet Implemented]':
from obspy.signal.trigger import carl_sta_trig, trigger_onset
on = 3000; off=-500
cft = carl_sta_trig(st[0].data, int(5 * st[0].stats.sampling_rate), int(10 * st[0].stats.sampling_rate), 0.8, 0.8)
on_off = np.array(trigger_onset(cft, on, off))
source_stalta.data = dict(times=st[0].times(), cft=cft)
source_triggers.data = dict(ontimes=st[0].times()[on_off[:,0]], y=np.zeros(on_off[:,0].shape))
sta_on.location = on
sta_off.location = off
else:
print(ticker_alg.value + ' is not yet implemented.')
ticker_alg.value = prev_val
def pick_arrival(trace,
nsta_seconds,
nlta_seconds,
df,
origin_time,
pick_threshold,
plot_flag=False):
"""
P wave arrival is picked using a recursive sta/lta algorithm.
Parameters
----------
trace: obspy trace
Seimic data.
nsta_seconds, nlta_seconds, pick_threshold: float
parameters for sta/lta
df: int, float
Data sampling rate
origin_time: obspy UTCDateTime
Earthquake occurrence time
.
Returns
-------
P_pick: array-like
Picked arrivals in samples.
Reference:
Withers, M., Aster, R., Young, C., Beiriger, J., Harris, M., Moore, S., and Trujillo, J. (1998),
A comparison of select trigger algorithms for automated global seismic phase and event detection,
Bulletin of the Seismological Society of America, 88 (1), 95-106.
"""
cft = recursive_sta_lta(trace, int(nsta_seconds * df),
int(nlta_seconds * df))
arrivals = trigger_onset(cft, pick_threshold, 0.5)
if plot_flag:
plot_trigger(trace, cft, pick_threshold, 0.5, show=True)
P_pick = check_arrival_time(arrivals, trace.stats.starttime, origin_time,
df)
return P_pick
def getTrigger(sac, short=2, long=25): # 1.75 1.25 4.wan perfect
df = sac.stats.sampling_rate
# print 'sampling_rate = '
# print df
# get cft
cft = recursive_sta_lta(sac.data, int(short * df), int(long * df))
# set threshold
threshold = np.mean(cft) + (np.max(cft) - np.mean(cft)) / 4
if np.isnan(threshold) == 1:
print 'thre = nan'
threshold = 3.2
# get on
# gk change
# on_of = trigger_onset(cft, threshold, threshold)
on_of = trigger_onset(cft, threshold * 1.38, threshold * 0.92)
if len(on_of) != 0:
return on_of[:, 0]
else:
return np.array([])
def network_detection(st, cft_return=True):
# TODO: Dynamic threshold method of Akram 2013
fs = st[0].stats.sampling_rate
sta_len_sec = 2.5 * DOM_PERIOD # 2-3 times dominant period
lta_len_sec = 7.5 * sta_len_sec # 5-10 times STA
nsta = int(sta_len_sec * fs)
nlta = int(lta_len_sec * fs)
on_thresh = 3.0 # 3.5
off_thresh = 0.5
numsta = len(list(set([tr.stats.station for tr in st])))
min_chans = numsta * 2 # Minimum number of channels to log network detection
cft_stream = Stream()
if cft_return:
for i, tr in enumerate(st.traces):
cft = recursive_sta_lta(tr.data, nsta=nsta, nlta=nlta)
# cft = eps_smooth(cft, w=int(EPS_WINLEN * Fs))
cft_stream += Trace(data=cft, header=tr.stats)
detection_list = coincidence_trigger(None,
on_thresh,
off_thresh,
cft_stream,
thr_coincidence_sum=min_chans,
max_trigger_length=2.0,
delete_long_trigger=True,
details=True)
else:
detection_list = coincidence_trigger('recstalta',
on_thresh,
off_thresh,
st,
sta=sta_len_sec,
lta=lta_len_sec,
thr_coincidence_sum=min_chans,
max_trigger_length=2.0,
delete_long_trigger=True,
details=True)
# Dictionary keys:
# time, stations, trace_ids, coincidence_sum, cft_peaks, cft_stds, duration, cft_wmean, cft_std_wmean
return detection_list, cft_stream
def pick_arrival(trace,
nsta_seconds,
nlta_seconds,
df,
origin_time,
pick_threshold,
plot_flag=False):
"""
P wave arrival is picked using a recursive sta/lta algorithm.
"""
cft = recursive_sta_lta(trace, int(nsta_seconds * df),
int(nlta_seconds * df))
arrivals = trigger_onset(cft, pick_threshold, 0.5)
if plot_flag:
plot_trigger(trace, cft, pick_threshold, 0.5, show=True)
P_pick = check_arrival_time(arrivals, trace.stats.starttime, origin_time,
df)
return P_pick
def pick(self,
path_to_file=None,
from_file=False,
method='sta_lta',
trigger_on=2.0,
sta=1.0,
lta=5.0):
'''
Pick onsets
'''
from obspy.signal.trigger import recursive_sta_lta
from numpy import zeros, where, nan, genfromtxt, array
if from_file == False:
self.picks = zeros(len(self.filtered_stations))
for k in range(len(self.filtered_stations)):
df = self.waveforms_up[k].stats.sampling_rate
cft = recursive_sta_lta(self.waveforms_up[k], int(sta * df),
int(lta * df))
i = where(cft > trigger_on)[0]
if len(i) > 0:
self.picks[k] = self.waveforms_up[k].times()[i[0]]
else:
i = where(cft > trigger_on / 2.)[0]
self.picks[k] = self.waveforms_up[k].times()[i[0]]
else:
picks = genfromtxt(path_to_file + self.ID + '.pick',
usecols=1,
dtype='S')
for k in range(len(picks)):
if picks[k] == 'nan':
picks[k] = float(nan)
else:
picks[k] = float(picks[k])
picks = array(picks).astype('float')
self.picks = array(picks)
def LTASTAtr(tr,thres1, thres2,STA, LTA,plotSTA):
""" return the cut on and off of the LTA/STA list [[cuton, cutoff], [cuton, cutoff]]
* input :
- tr : type : trace , stream to filnd STA, LTA
- thres1 : type; float : cut on limit of STA/LTA values : after tjis value the cut on is defined
- thres2 : type, float : cut off limit of STA/LTA values : after this value the cut off is defined
- STA : type int : size of the LTA windows in second : STA = the trace average on this time windows
- LTA : type int : size of the LTA windows in second : LTA = the trace average on this time windows
- plotSTA: type, bool; it true, the trace and it's characteristic function are plotted
RQ: AFTER TESTING IT'S SEEEMS GOOD TO HAVE A RATIO WSTA/WLTA > 1/4 AND A CUT OFF HIGHER THAN CUT ON
* outputs
- L_onoff: type np, array : 2D array of cut on and cut_off time in number of sample ie time* df where df is the sampling rate [[cuton,cutoff], [cuton1, cutoff1]]
exemple:
st = Read_event('15','206','15','1', '1', True)
stcorrec = Stream_Correction(st, '1', False)
stfiltered= Stream_PBfilter(stcorrec,0.5, 20,False)
LTASTA(stfiltered,2, 2.5,300,1400,True)
"""
#0. sampling rate ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
df = tr.stats.sampling_rate
#1. characteristic function of the trace following classical LTA ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#cft = classicSTALTA(tr.data, int(STA * df), int(LTA* df))
#2. characteristic function of the trace following recursive LTA
cft2 =recursive_sta_lta(tr.data, int(STA * df), int(LTA* df))
#3. list of [cuton, cutoff] time in number of samples~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#max_len = maximum lenght of the triggered event in sample,
#max_len_delete = Do not write events longer than max_len into report file.
L_onoff = trigger_onset(cft2, thres1, thres2, max_len=9e+99, max_len_delete=False)
#4. plot~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
if plotSTA==True :
##plotTrigger(tr[0], cft, thres1,thres2)
plotTrigger(tr[0], cft2, thres1,thres2)
plt.title('recursive')
return np.array(L_onoff)
def STA_LTA(self,
dataset,
batch_size=100,
short_window=30,
long_window=200):
"""
STA/LTA method
:param dataset: Dataset name.
:param batch_size: Model directory name.
"""
dataset_path, eval_path = self.get_eval_dir(dataset)
dataset = seisnn.io.read_dataset(dataset)
data_len = self.get_dataset_length(self.database)
progbar = tf.keras.utils.Progbar(data_len)
n = 0
for val in dataset.prefetch(100).batch(batch_size):
progbar.add(batch_size)
title = f"eval_{n:0>5}"
trace_len = val['trace'].shape[2]
batch_len = val['trace'].shape[0]
predict = np.zeros((batch_len, 1, 3008, 3))
for i in range(batch_len):
z_trace = val['trace'].numpy()[i, :, :, 0].reshape(trace_len)
cft = recursive_sta_lta(z_trace, short_window, long_window)
predict[i, :, :, 0] = cft
val['predict'] = predict
val['id'] = tf.convert_to_tensor(title.encode('utf-8'),
dtype=tf.string)[tf.newaxis]
example = next(seisnn.example_proto.batch_iterator(val))
instance = Instance(example)
instance.to_tfrecord(os.path.join(eval_path, title + '.tfrecord'))
n += 1
0].stats.starttime + 21 * 60 * 60 + 10 * 60 #+18*60*60 +20*60 #time window start
end = start + 70 * 60
#end=sample[0].stats.endtime
trs = trace.slice(
starttime=start, endtime=end
) #cut out sample waveform with same window length as chosen event
#trs_e = obspy.signal.filter.envelope(trs.data)
#print('reference waveform')
trs.plot(type='relative', color='b') #, starttime=start , endtime=end)
sr = trace.stats.sampling_rate
nsta = int(2 * sr) #2
nlta = int(10 * sr) #20
stream = trs.data
cft = recursive_sta_lta(stream, nsta, nlta)
trig_on = 6 #8
trig_off = 0.2 #0.2
plot_trigger(trs, cft, trig_on, trig_off)
on_off = trigger_onset(cft, trig_on, trig_off)
for x in range(0, len(on_off)):
tr = trace.slice(starttime=start + (on_off[x, 0] / sr) - 10,
endtime=start + (on_off[x, 1] / sr) + 10)
tr_e = obspy.signal.filter.envelope(tr.data)
#%% frequency info
tr_data = tr.data
m = np.mean(tr_data)
tr_data = tr_data - m
import matplotlib.pyplot as plt
from obspy import read
from obspy.signal.trigger import recursive_sta_lta, trigger_onset
from seisnn.io import get_dir_list
predict_pkl_dir = "/mnt/tf_data/pkl/2017_02"
predict_pkl_list = get_dir_list(predict_pkl_dir)
for i, pkl in enumerate(predict_pkl_list):
trace = read(pkl).traces[0]
df = trace.stats.sampling_rate
cft = recursive_sta_lta(trace.data, int(0.5 * df), int(1. * df))
on_of = trigger_onset(cft, 1, 0.5)
# Plotting the results
ax = plt.subplot(211)
plt.plot(trace.data, 'k')
ymin, ymax = ax.get_ylim()
plt.vlines(on_of[:, 0], ymin, ymax, color='r', linewidth=2)
plt.vlines(on_of[:, 1], ymin, ymax, color='b', linewidth=2)
plt.subplot(212, sharex=ax)
plt.plot(cft, 'k')
plt.hlines([3.5, 0.5], 0, len(cft), color=['r', 'b'], linestyle='--')
plt.axis('tight')
plt.show()
import obspy
from obspy.signal.trigger import plot_trigger, recursive_sta_lta
trace = obspy.read("https://examples.obspy.org/ev0_6.a01.gse2")[0]
df = trace.stats.sampling_rate
cft = recursive_sta_lta(trace.data, int(5 * df), int(10 * df))
plot_trigger(trace, cft, 1.2, 0.5)
import obspy
from obspy.clients.arclink import Client
from obspy.signal.trigger import recursive_sta_lta, trigger_onset
# Retrieve waveforms via ArcLink
client = Client(host="erde.geophysik.uni-muenchen.de", port=18001,
user="******")
t = obspy.UTCDateTime("2009-08-24 00:19:45")
st = client.get_waveforms('BW', 'RTSH', '', 'EHZ', t, t + 50)
# For convenience
tr = st[0] # only one trace in mseed volume
df = tr.stats.sampling_rate
# Characteristic function and trigger onsets
cft = recursive_sta_lta(tr.data, int(2.5 * df), int(10. * df))
on_of = trigger_onset(cft, 3.5, 0.5)
# Plotting the results
ax = plt.subplot(211)
plt.plot(tr.data, 'k')
ymin, ymax = ax.get_ylim()
plt.vlines(on_of[:, 0], ymin, ymax, color='r', linewidth=2)
plt.vlines(on_of[:, 1], ymin, ymax, color='b', linewidth=2)
plt.subplot(212, sharex=ax)
plt.plot(cft, 'k')
plt.hlines([3.5, 0.5], 0, len(cft), color=['r', 'b'], linestyle='--')
plt.axis('tight')
plt.show()
def main():
outputdir = './output/'
data = load_data()
X_train = data['X_train']
y_train = data["y_train"]
X_test = data['X_test']
y_test = data["y_test"]
ridge = load_json(outputdir + 'ridge.json')
ada_rf_exp = load_json(outputdir + 'ada_rf_exp.json')
#rf = load_json(outputdir + 'rf.json')
# Baseline model (linear fit for log10(Z.disp.max_amp))
z = np.polyfit(X_train['Z.disp.max_amp'].values, y_train.values, 1)
y_pred_base = X_test['Z.disp.max_amp'].values * z[0] + z[1]
#####################################################
#Figure 1:
#Empirical measurements used in literature
#####################################################
label_font = 20
tick_font = 20
alpha = 0.5
plt.figure(1, figsize=(24, 5))
fig = plt.subplot(131)
plt.plot(y_train, X_train['Z.disp.max_amp'], '.', alpha=alpha)
plt.plot(y_test, X_test['Z.disp.max_amp'], 'r.', alpha=alpha)
plt.plot(np.arange(2.5, 7.5, 1), (np.arange(2.5, 7.5, 1) - z[1]) / z[0],
'k--',
linewidth=2)
plt.xlabel('Magnitude', fontsize=label_font)
plt.ylabel(r'$log_{10}(P_d * r^2)$', fontsize=label_font)
fig.tick_params(labelsize=tick_font)
fig.legend(['Train', 'Test', 'Linear fit (train)'],
loc='lower right',
fontsize=15)
fig = plt.subplot(132)
plt.plot(y_train, X_train['Z.tau_c'], '.', alpha=alpha)
plt.plot(y_test, X_test['Z.tau_c'], 'r.', alpha=alpha)
plt.xlabel('Magnitude', fontsize=label_font)
plt.ylabel(r'$log_{10}(\tau_c)$', fontsize=label_font)
fig.tick_params(labelsize=tick_font)
fig = plt.subplot(133)
plt.plot(y_train, X_train['Z.tau_p_max'], '.', alpha=alpha)
plt.plot(y_test, X_test['Z.tau_p_max'], 'r.', alpha=alpha)
plt.xlabel('Magnitude', fontsize=label_font)
plt.ylabel(r'$log_{10}(\tau_p^{max})$', fontsize=label_font)
#plt.ylabel(r'$log_{10}(mean frequency)$', fontsize = label_font)
fig.tick_params(labelsize=tick_font)
plt.savefig(outputdir + 'empirical_measurements.png', format='png')
#####################################################
#Figure 2:
#Feature coefficients/importance
#####################################################
plt.figure(2, figsize=(12, 8))
ax = plt.subplot(141)
plot_feature_importance(ridge['coef'],
ridge['features'],
ax,
top_feature=30)
plt.xlabel('Coefficient', fontsize=15)
plt.title('Ridge', fontsize=15)
#plt.tick_params(labelsize = tick_font)
ax = plt.subplot(143)
plot_feature_importance(ada_rf_exp['coef'],
ada_rf_exp['features'],
ax,
top_feature=30)
plt.xlabel('Feature importance', fontsize=15)
plt.title('Adaboosted Random Forest', fontsize=15)
#plt.tick_params(labelsize = tick_font)
plt.savefig(outputdir + 'Feature_importance.png', format='png')
#####################################################
#Figure 3:
#True magnitude vs. prediction
#####################################################
label_font = 25
tick_font = 20
alpha = 0.5
plt.figure(3, figsize=(27, 15))
fig = plt.subplot(2, 3, 1)
plot_true_vs_prediction(y_test, y_pred_base, symbol='r.')
plt.text(2.2, 6.0, 'Test set', fontsize=label_font + 10)
fig.tick_params(labelsize=tick_font + 5)
plt.ylabel('Predicted magnitude', fontsize=label_font + 10)
plt.title('Baseline', fontsize=label_font + 10)
fig = plt.subplot(2, 3, 2)
plot_true_vs_prediction(y_test, ridge['y_test_pred'], symbol='r.')
fig.tick_params(labelsize=tick_font + 5)
plt.title('Ridge', fontsize=label_font + 10)
fig = plt.subplot(2, 3, 3)
plot_true_vs_prediction(y_test, ada_rf_exp['y_test_pred'], symbol='r.')
fig.tick_params(labelsize=tick_font + 5)
plt.title('Adaboosted Random Forest', fontsize=label_font + 10)
#####################################################
fig = plt.subplot(2, 3, 4)
error_base = error_large_eq(y_test.values, y_pred_base)
plot_true_vs_error(y_test.values, y_pred_base, error_base)
plt.xlabel('True Magnitude', fontsize=label_font + 10)
plt.ylabel('Test error', fontsize=label_font + 10)
fig.tick_params(labelsize=tick_font + 5)
fig = plt.subplot(2, 3, 5)
error_ridge = error_large_eq(y_test.values, ridge['y_test_pred'])
plot_true_vs_error(y_test.values, ridge['y_test_pred'], error_ridge)
plt.xlabel('True Magnitude', fontsize=label_font + 10)
fig.tick_params(labelsize=tick_font + 5)
fig = plt.subplot(2, 3, 6)
error_ada_rf = error_large_eq(y_test.values, ada_rf_exp['y_test_pred'])
plot_true_vs_error(y_test.values, ada_rf_exp['y_test_pred'], error_ada_rf)
plt.xlabel('True Magnitude', fontsize=label_font + 10)
fig.tick_params(labelsize=tick_font + 5)
plt.savefig(outputdir + 'prediction.png', format='png')
#####################################################
#Figure 4:
#Error distribution (> M4.5)
#####################################################
plt.figure(4, figsize=(8, 5.5))
bins = np.arange(-1.4, 0.8, 0.1)
plt.hist(error_base['error_large'], bins=bins, alpha=0.7, label="Baseline")
plt.hist(error_ridge['error_large'], bins=bins, alpha=0.7, label="Ridge")
plt.hist(error_ada_rf['error_large'],
bins=bins,
alpha=0.7,
label="Adaboosted RF")
plt.legend(loc='upper left', fontsize=15)
plt.plot([0, 0], [0, 50], 'k--')
plt.ylim([0, 50])
plt.xlabel('Error (> M4.5)', fontsize=label_font)
plt.tick_params(labelsize=tick_font)
plt.quiver(-0.05, 42, -1, 0)
plt.quiver(0.05, 42, 1, 0)
plt.text(-0.6, 45, 'Underestimate', fontsize=15)
plt.text(0.12, 45, 'Overestimate', fontsize=15)
plt.savefig(outputdir + 'error.png', format='png')
#####################################################
#Figure 5:
#Map view
#####################################################
station_loc = pd.read_csv('../data/station.csv')
catalog = pd.read_csv('../data/source.csv')
plot_stations_and_events(station_loc, catalog, map_flag=True)
plt.savefig(outputdir + 'mapview.png', format='png')
plot_map_us()
plt.savefig(outputdir + 'mapview_us.png', format='png')
#####################################################
#Figure 6:
#Time series data visualization
#####################################################
st = read('../data/proc/2016-06-10T08:04:38.700000Z/CI.mseed')
trace = st.select(channel='BHE', station='BAR')
df = trace[0].stats.sampling_rate
cft = recursive_sta_lta(trace[0], int(0.05 * df), int(20 * df))
arrivals = trigger_onset(cft, 100, 0.5)
start = arrivals[0][0] / df - 60
plt.figure(5, figsize=(12, 6))
#plt.subplot(2,1,1)
plt.fill_between([start, start + 4],
-0.0075,
0.009,
facecolor=[0.7, 0.7, 0.7])
plt.plot(np.arange(0, 180, 1. / df) - 60, trace[0])
plt.xlim([60 - 60, 130 - 60])
plt.ylim([-0.0075, 0.009])
plt.text(start + 0.25, -0.006, '4 s', fontsize=25)
plt.tick_params(labelsize=tick_font)
plt.title('M5.19, 2016-06-10, 08:04:38.70', fontsize=label_font)
plt.xlabel('Time (s)', fontsize=label_font)
plt.ylabel('Amplitude (m/s)', fontsize=label_font)
plt.savefig(outputdir + 'time_series.png', format='png')
import matplotlib.pyplot as plt
from obspy import read
from obspy.signal.trigger import recursive_sta_lta, trigger_onset
from seisnn.io import get_dir_list
predict_pkl_dir = "/mnt/tf_data/dataset/2017_02"
predict_pkl_list = get_dir_list(predict_pkl_dir)
on = 3.5
off = 0.5
for i, pkl in enumerate(predict_pkl_list):
trace = read(pkl).traces[0]
start_time = trace.stats.starttime
df = trace.stats.sampling_rate
cft = recursive_sta_lta(trace.data, int(0.2 * df), int(2. * df))
on_of = trigger_onset(cft, on, off)
# Plotting the results
ax = plt.subplot(211)
plt.plot(trace.data, 'k')
ymin, ymax = ax.get_ylim()
try:
plt.vlines(on_of[:, 0], ymin, ymax, color='r', linewidth=2)
plt.vlines(on_of[:, 1], ymin, ymax, color='b', linewidth=2)
except TypeError:
pass
plt.subplot(212, sharex=ax)
plt.plot(cft, 'k')
plt.hlines([on, off], 0, len(cft), color=['r', 'b'], linestyle='--')
plt.xticks(range(0, 3001, 500), range(0, 31, 5))
#print(len(data_stream))
sr = 100
nsta = int(1 * sr)
nlta = int(10 * sr)
trig_on = 2.5
trig_off = 0.05
#for x in range(0,len(data_stream)):
for x in range(0, 10):
data_s = event_stream[x].data
max_a = data_s.max()
min_a = data_s.min()
p2p = max_a - min_a
cft = recursive_sta_lta(data_s, nsta, nlta)
# plot_trigger(sq_stream[x], cft, trig_on, trig_off)
on_off = trigger_onset(cft, trig_on, trig_off)
start = event_stream[x].stats.starttime
tr = event_stream[x].slice(starttime=start + (on_off[0, 0] / sr),
endtime=start + (on_off[0, 1] / sr))
print('event:', event_stream[x].stats.starttime,
'from station: ', stre[x].stats.station, ', has energy: ',
sum(np.square(tr.data)), ' and peak to peak:', p2p)
plt.figure(x)
plt.plot(tr)
plt.figure(x + 20)
plt.plot(np.square(tr.data))
#%% energy information
def _channel_loop(tr, parameters, max_trigger_length=60,
despike=False, debug=0):
"""
Internal loop for parellel processing.
:type tr: obspy.core.trace
:param tr: Trace to look for triggers in.
:type parameters: list
:param parameters: List of TriggerParameter class for trace.
:type max_trigger_length: float
:type despike: bool
:type debug: int
:return: trigger
:rtype: list
"""
from eqcorrscan.utils.despike import median_filter
from obspy.signal.trigger import trigger_onset, plot_trigger
from obspy.signal.trigger import recursive_sta_lta
import warnings
for par in parameters:
if par['station'] == tr.stats.station and \
par['channel'] == tr.stats.channel:
parameter = par
break
else:
msg = 'No parameters set for station ' + str(tr.stats.station)
warnings.warn(msg)
return []
triggers = []
if debug > 0:
print(tr)
tr.detrend('simple')
if despike:
median_filter(tr)
if parameter['lowcut'] and parameter['highcut']:
tr.filter('bandpass', freqmin=parameter['lowcut'],
freqmax=parameter['highcut'])
elif parameter['lowcut']:
tr.filter('highpass', freq=parameter['lowcut'])
elif parameter['highcut']:
tr.filter('lowpass', freq=parameter['highcut'])
# find triggers for each channel using recursive_sta_lta
df = tr.stats.sampling_rate
cft = recursive_sta_lta(tr.data, int(parameter['sta_len'] * df),
int(parameter['lta_len'] * df))
if max_trigger_length:
trig_args = {'max_len_delete': True}
trig_args['max_len'] = int(max_trigger_length *
df + 0.5)
if debug > 3:
plot_trigger(tr, cft, parameter['thr_on'], parameter['thr_off'])
tmp_trigs = trigger_onset(cft, float(parameter['thr_on']),
float(parameter['thr_off']),
**trig_args)
for on, off in tmp_trigs:
cft_peak = tr.data[on:off].max()
cft_std = tr.data[on:off].std()
on = tr.stats.starttime + \
float(on) / tr.stats.sampling_rate
off = tr.stats.starttime + \
float(off) / tr.stats.sampling_rate
triggers.append((on.timestamp, off.timestamp,
tr.id, cft_peak,
cft_std))
return triggers