[update] refactoring, added type hints
This commit is contained in:
parent
452f2a2e18
commit
a068bb8457
@ -21,7 +21,7 @@ from joblib import Parallel, delayed
|
||||
from obspy import read, Stream, UTCDateTime, Trace
|
||||
from obspy.taup import TauPyModel
|
||||
from obspy.core.event.base import WaveformStreamID, ResourceIdentifier
|
||||
from obspy.core.event.origin import Pick
|
||||
from obspy.core.event.origin import Pick, Origin
|
||||
from obspy.geodetics.base import gps2dist_azimuth
|
||||
from obspy.signal.cross_correlation import correlate, xcorr_max
|
||||
|
||||
@ -34,6 +34,12 @@ from pylot.core.util.event import Event as PylotEvent
|
||||
from pylot.correlation.utils import (get_event_id, get_event_pylot, get_event_obspy_dmt, get_picks, write_json,
|
||||
get_metadata)
|
||||
|
||||
DEBUG_LEVELS = {'debug': logging.DEBUG,
|
||||
'info': logging.INFO,
|
||||
'warn': logging.WARNING,
|
||||
'error': logging.ERROR,
|
||||
'critical': logging.CRITICAL}
|
||||
|
||||
|
||||
class CorrelationParameters:
|
||||
"""
|
||||
@ -210,7 +216,7 @@ class XCorrPickCorrection:
|
||||
Returns:
|
||||
array: A numpy array representing the time axis.
|
||||
"""
|
||||
return np.linspace(0,trace.stats.endtime -trace.stats.starttime, trace.stats.npts)
|
||||
return np.linspace(0, trace.stats.endtime - trace.stats.starttime, trace.stats.npts)
|
||||
|
||||
def plot_cc(figure_output_dir: str, plot_filename: str) -> None:
|
||||
"""
|
||||
@ -316,7 +322,7 @@ class XCorrPickCorrection:
|
||||
peak_index = cc.argmax()
|
||||
first_sample = peak_index
|
||||
# XXX this could be improved..
|
||||
while first_sample > 0 and cc_curvature[first_sample - 1] <= 0:
|
||||
while first_sample > 0 >= cc_curvature[first_sample - 1]:
|
||||
first_sample -= 1
|
||||
last_sample = peak_index
|
||||
while last_sample < len(cc) - 1 and cc_curvature[last_sample + 1] <= 0:
|
||||
@ -337,7 +343,7 @@ class XCorrPickCorrection:
|
||||
|
||||
# quadratic fit for small subwindow
|
||||
coeffs, cov_mat = np.polyfit(cc_t[first_sample:last_sample + 1], cc[first_sample:last_sample + 1], deg=2,
|
||||
full=False, cov=True)[:2]
|
||||
full=False, cov=True)[:2]
|
||||
|
||||
a, b, c = coeffs
|
||||
|
||||
@ -380,9 +386,10 @@ class XCorrPickCorrection:
|
||||
return pick2_corr, coeff, uncert, fwfm
|
||||
|
||||
|
||||
def correlation_main(database_path_dmt, pylot_infile_path, params, channel_config, istart=0, istop=1e9, update=False,
|
||||
event_blacklist=None):
|
||||
'''
|
||||
def correlation_main(database_path_dmt: str, pylot_infile_path: str, params: dict, channel_config: dict,
|
||||
istart: int = 0, istop: int = 1e9, update: bool = False,
|
||||
event_blacklist: str = None, select_events: list = None) -> None:
|
||||
"""
|
||||
Main function of this program, correlates waveforms around theoretical, or other initial (e.g. automatic cf) picks
|
||||
on one or more reference stations.
|
||||
All stations with a correlation higher "min_corr_stacking" are stacked onto the station with the highest mean
|
||||
@ -390,12 +397,21 @@ def correlation_main(database_path_dmt, pylot_infile_path, params, channel_confi
|
||||
Finally, all other stations will be correlated with the re-picked, stacked seismogram around their theoretical (or
|
||||
initial) onsets and the shifted pick time of the stacked seismogram will be assigned as their new pick time.
|
||||
|
||||
:param database_path_dmt: obspy_dmt databse directory
|
||||
:param pylot_infile_path: path containing input file for autoPyLoT (automatic picking parameters)
|
||||
:param params: dictionary with parameter class objects
|
||||
Args:
|
||||
database_path_dmt (str): The path to the obspydmt database.
|
||||
pylot_infile_path (str): The path to the Pylot infile for autoPyLoT (automatic picking parameters).
|
||||
params (dict): Parameters for the correlation script.
|
||||
channel_config (dict): Configuration for channels.
|
||||
istart (int, optional): The starting index for events. Defaults to 0.
|
||||
istop (float, optional): The stopping index for events. Defaults to 1e9.
|
||||
update (bool, optional): Whether to update. Defaults to False.
|
||||
event_blacklist (str, optional): Path to the event blacklist file. Defaults to None.
|
||||
select_events (list, optional): List of selected events. Defaults to None.
|
||||
Returns:
|
||||
None
|
||||
|
||||
:return:
|
||||
'''
|
||||
"""
|
||||
assert os.path.isdir(database_path_dmt), 'Unrecognized directory {}'.format(database_path_dmt)
|
||||
|
||||
tstart = datetime.now()
|
||||
@ -405,39 +421,38 @@ def correlation_main(database_path_dmt, pylot_infile_path, params, channel_confi
|
||||
logging.info(50 * '#')
|
||||
|
||||
for phase_type in params.keys():
|
||||
if params[phase_type]['plot_detailed']: params[phase_type]['plot'] = True
|
||||
if params[phase_type]['plot_detailed']:
|
||||
params[phase_type]['plot'] = True
|
||||
|
||||
eventdirs = glob.glob(os.path.join(database_path_dmt, '*.?'))
|
||||
|
||||
pylot_parameter = PylotParameter(pylot_infile_path)
|
||||
|
||||
if event_blacklist:
|
||||
with open(event_blacklist, 'r') as infile:
|
||||
event_blacklist = [line.split('\n')[0] for line in infile.readlines()]
|
||||
with open(event_blacklist, 'r') as fid:
|
||||
event_blacklist = [line.split('\n')[0] for line in fid.readlines()]
|
||||
|
||||
# iterate over all events in "database_path_dmt"
|
||||
for eventindex, eventdir in enumerate(eventdirs):
|
||||
if not istart <= eventindex < istop:
|
||||
continue
|
||||
|
||||
# MP MP testing +++
|
||||
# ids_filter = ['20181229_033914.a']
|
||||
# if not os.path.split(eventdir)[-1] in ids_filter:
|
||||
# continue
|
||||
# MP MP testing ---
|
||||
if select_events and not os.path.split(eventdir)[-1] in select_events:
|
||||
continue
|
||||
|
||||
logging.info('\n' + 100 * '#')
|
||||
logging.info('Working on event {} ({}/{})'.format(eventdir, eventindex + 1, len(eventdirs)))
|
||||
if event_blacklist and get_event_id(eventdir) in event_blacklist:
|
||||
logging.info('Event on blacklist. Continue')
|
||||
|
||||
correlate_event(eventdir, pylot_parameter, params=params, channel_config=channel_config, update=update)
|
||||
correlate_event(eventdir, pylot_parameter, params=params, channel_config=channel_config,
|
||||
update=update)
|
||||
|
||||
logging.info('Finished script after {} at {}'.format(datetime.now() - tstart, datetime.now()))
|
||||
|
||||
|
||||
def get_estimated_inclination(stations_dict, origin, phases, model='ak135'):
|
||||
''' calculate a mean inclination angle for all stations for seismometer rotation to spare computation time'''
|
||||
def get_estimated_inclination(stations_dict: dict, origin: Origin, phases: str, model: str = 'ak135') -> float:
|
||||
""" calculate a mean inclination angle for all stations for seismometer rotation to spare computation time"""
|
||||
model = TauPyModel(model)
|
||||
phases = [*phases.split(',')]
|
||||
avg_lat = np.median([item['latitude'] for item in stations_dict.values()])
|
||||
@ -448,9 +463,9 @@ def get_estimated_inclination(stations_dict, origin, phases, model='ak135'):
|
||||
return arrivals[0].incident_angle
|
||||
|
||||
|
||||
def modify_horizontal_name(wfdata):
|
||||
''' This function only renames (does not rotate!) numeric channel IDs to be able to rotate them to LQT. It is
|
||||
therefore not accurate and only used for picking with correlation. Returns a copy of the original stream. '''
|
||||
def modify_horizontal_name(wfdata: Stream) -> Stream:
|
||||
""" This function only renames (does not rotate!) numeric channel IDs to be able to rotate them to LQT. It is
|
||||
therefore not accurate and only used for picking with correlation. Returns a copy of the original stream. """
|
||||
# wfdata = wfdata.copy()
|
||||
stations = np.unique([tr.stats.station for tr in wfdata])
|
||||
for station in stations:
|
||||
@ -460,13 +475,14 @@ def modify_horizontal_name(wfdata):
|
||||
st = wfdata.select(station=station, location=location)
|
||||
channels = [tr.stats.channel for tr in st]
|
||||
check_numeric = [channel[-1].isnumeric() for channel in channels]
|
||||
check_nonnumeric = [not (cn) for cn in check_numeric]
|
||||
check_nonnumeric = [not cn for cn in check_numeric]
|
||||
if not any(check_numeric):
|
||||
continue
|
||||
numeric_channels = np.array(channels)[check_numeric].tolist()
|
||||
nonnumeric_channels = np.array(channels)[check_nonnumeric].tolist()
|
||||
if not 'Z' in [nc[-1] for nc in nonnumeric_channels]:
|
||||
logging.warning('Modify_horizontal_name failed: Only implemented for existing Z component! Return original data.')
|
||||
if 'Z' not in [nc[-1] for nc in nonnumeric_channels]:
|
||||
logging.warning(
|
||||
'Modify_horizontal_name failed: Only implemented for existing Z component! Return original data.')
|
||||
return wfdata
|
||||
numeric_characters = sorted([nc[-1] for nc in numeric_channels])
|
||||
if numeric_characters == ['1', '2']:
|
||||
@ -474,7 +490,8 @@ def modify_horizontal_name(wfdata):
|
||||
elif numeric_characters == ['2', '3']:
|
||||
channel_dict = {'2': 'N', '3': 'E'}
|
||||
else:
|
||||
logging.warning('Modify_horizontal_name failed: Channel order not implemented/unknown. Return original data.')
|
||||
logging.warning(
|
||||
'Modify_horizontal_name failed: Channel order not implemented/unknown. Return original data.')
|
||||
return wfdata
|
||||
for tr in st:
|
||||
channel = tr.stats.channel
|
||||
@ -486,10 +503,10 @@ def modify_horizontal_name(wfdata):
|
||||
return wfdata
|
||||
|
||||
|
||||
def cut_stream_to_same_length(wfdata):
|
||||
def remove(wfdata, st):
|
||||
for tr in st:
|
||||
wfdata.remove(tr)
|
||||
def cut_stream_to_same_length(wfdata: Stream) -> None:
|
||||
def remove(data, stream):
|
||||
for tr in stream:
|
||||
data.remove(tr)
|
||||
|
||||
stations = np.unique([tr.stats.station for tr in wfdata])
|
||||
for station in stations:
|
||||
@ -515,7 +532,7 @@ def cut_stream_to_same_length(wfdata):
|
||||
st.trim(tstart, tend, pad=True, fill_value=0.)
|
||||
|
||||
|
||||
def get_unique_phase(picks, rename_phases=None):
|
||||
def get_unique_phase(picks: list, rename_phases: dict = None) -> str:
|
||||
phases = [pick.phase_hint for pick in picks]
|
||||
if rename_phases:
|
||||
phases = [rename_phases[phase] if phase in rename_phases else phase for phase in phases]
|
||||
@ -523,10 +540,10 @@ def get_unique_phase(picks, rename_phases=None):
|
||||
if len(set(phases)) == 1:
|
||||
return phases[0]
|
||||
|
||||
return False
|
||||
|
||||
|
||||
def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
# TODO: simplify this function
|
||||
def correlate_event(eventdir: str, pylot_parameter: PylotParameter, params: dict, channel_config: dict,
|
||||
update: bool) -> bool:
|
||||
rename_phases = {'Pdiff': 'P'}
|
||||
|
||||
# create ObsPy event from .pkl file in dmt eventdir
|
||||
@ -588,7 +605,8 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
wfdata = wfdata_raw.copy()
|
||||
# resample and filter
|
||||
if params[phase_type]['sampfreq']:
|
||||
wfdata = resample_parallel(wfdata, params[phase_type]['sampfreq'], ncores=params[phase_type]['ncores'])
|
||||
wfdata = resample_parallel(wfdata, params[phase_type]['sampfreq'],
|
||||
ncores=params[phase_type]['ncores'])
|
||||
else:
|
||||
logging.warning('Resampling deactivated! '
|
||||
'Make sure that the sampling rate of all input waveforms is identical')
|
||||
@ -602,7 +620,8 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
else:
|
||||
method = 'auto'
|
||||
# get all picks from PyLoT *.xml file
|
||||
if not pfe.startswith('_'): pfe = '_' + pfe
|
||||
if not pfe.startswith('_'):
|
||||
pfe = '_' + pfe
|
||||
picks = get_picks(eventdir, extension=pfe)
|
||||
# take picks of selected phase only
|
||||
picks = [pick for pick in picks if pick.phase_hint == phase_type]
|
||||
@ -630,7 +649,7 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
model='ak135')
|
||||
# rotate ZNE -> LQT (also cut to same length)
|
||||
wfdata = rotate_stream(wfdata, metadata=metadata, origin=origin, stations_dict=stations_dict,
|
||||
channels=channels[phase_type], inclination=inclination, ncores=ncores)
|
||||
channels=CHANNELS[phase_type], inclination=inclination, ncores=ncores)
|
||||
|
||||
# make copies before filtering
|
||||
wfdata_lowf = wfdata.copy()
|
||||
@ -644,7 +663,8 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
|
||||
# make directory for correlation output
|
||||
correlation_out_dir = create_correlation_output_dir(eventdir, filter_options_final, true_phase_type)
|
||||
fig_dir = create_correlation_figure_dir(correlation_out_dir) if params[phase_type]['save_fig'] == True else ''
|
||||
fig_dir = create_correlation_figure_dir(correlation_out_dir) \
|
||||
if params[phase_type]['save_fig'] is True else ''
|
||||
|
||||
if not params[phase_type]['use_stacked_trace']:
|
||||
# first stack mastertrace by using stations with high correlation on that station in station list with the
|
||||
@ -666,13 +686,14 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
if params[phase_type]['plot']:
|
||||
# plot correlations of traces used to generate stacked trace
|
||||
plot_mastertrace_corr(nwst_id, correlations_dict, wfdata_lowf, stations_dict, picks, trace_master, method,
|
||||
min_corr=params[phase_type]['min_corr_stacking'], title=eventdir + '_before_stacking',
|
||||
fig_dir=fig_dir)
|
||||
min_corr=params[phase_type]['min_corr_stacking'],
|
||||
title=eventdir + '_before_stacking', fig_dir=fig_dir)
|
||||
|
||||
# continue if there is not enough traces for stacking
|
||||
if nstack < params[phase_type]['min_stack']:
|
||||
logging.warning('Not enough traces to stack: {} (min_stack = {}). Skip this event'.format(nstack,
|
||||
params[phase_type][
|
||||
params[
|
||||
phase_type][
|
||||
'min_stack']))
|
||||
continue
|
||||
|
||||
@ -680,8 +701,8 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
trace_master.write(os.path.join(correlation_out_dir, '{}_stacked.mseed'.format(trace_master.id)))
|
||||
|
||||
# now pick stacked trace with PyLoT for a more precise pick (use raw trace, gets filtered by autoPyLoT)
|
||||
pick_stacked = repick_mastertrace(wfdata_lowf, trace_master, pylot_parameter, event, event_id, metadata,
|
||||
phase_type, correlation_out_dir)
|
||||
pick_stacked = repick_master_trace(wfdata_lowf, trace_master, pylot_parameter, event, event_id, metadata,
|
||||
phase_type, correlation_out_dir)
|
||||
if not pick_stacked:
|
||||
continue
|
||||
|
||||
@ -696,7 +717,7 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
trace_master_highf.filter(filter_type, **filter_options_final)
|
||||
|
||||
input_list = prepare_correlation_input(wfdata_lowf, picks, channels_list, trace_master_lowf, pick_stacked,
|
||||
params=params[phase_type], plot=params[phase_type]['plot'],
|
||||
phase_params=params[phase_type], plot=params[phase_type]['plot'],
|
||||
fig_dir=fig_dir_traces, ncorr=2, wfdata_highf=wfdata_highf,
|
||||
trace_master_highf=trace_master_highf)
|
||||
|
||||
@ -712,8 +733,8 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
export_picks(eventdir=eventdir,
|
||||
correlations_dict=get_stations_min_corr(correlations_dict_stacked,
|
||||
params[phase_type]['min_corr_export']),
|
||||
params=params[phase_type], picks=picks, taupypicks=taupypicks_orig, phase_type=true_phase_type,
|
||||
pf_extension=pfe)
|
||||
params=params[phase_type], picks=picks, taupypicks=taupypicks_orig,
|
||||
phase_type=true_phase_type, pf_extension=pfe)
|
||||
|
||||
# plot results
|
||||
if params[phase_type]['plot']:
|
||||
@ -748,10 +769,12 @@ def correlate_event(eventdir, pylot_parameter, params, channel_config, update):
|
||||
|
||||
if len(correlations_dict_stacked) > 0:
|
||||
return True
|
||||
else:
|
||||
return False
|
||||
|
||||
|
||||
def remove_outliers(picks, corrected_taupy_picks, threshold):
|
||||
''' delete a pick if difference to corrected taupy pick is larger than threshold'''
|
||||
""" delete a pick if difference to corrected taupy pick is larger than threshold"""
|
||||
|
||||
n_picks = len(picks)
|
||||
n_outl = 0
|
||||
@ -780,7 +803,7 @@ def remove_outliers(picks, corrected_taupy_picks, threshold):
|
||||
|
||||
|
||||
def remove_invalid_picks(picks):
|
||||
''' Remove picks without uncertainty (invalid PyLoT picks)'''
|
||||
""" Remove picks without uncertainty (invalid PyLoT picks)"""
|
||||
count = 0
|
||||
deleted_picks_ids = []
|
||||
for index, pick in list(reversed(list(enumerate(picks)))):
|
||||
@ -803,7 +826,7 @@ def get_picks_mean(picks):
|
||||
|
||||
|
||||
def get_corrected_taupy_picks(picks, taupypicks, all_available=False):
|
||||
''' get mean/median from picks taupy picks, correct latter for the difference '''
|
||||
""" get mean/median from picks taupy picks, correct latter for the difference """
|
||||
|
||||
def nwst_id_from_wfid(wfid):
|
||||
return '{}.{}'.format(wfid.network_code if wfid.network_code else '',
|
||||
@ -824,8 +847,9 @@ def get_corrected_taupy_picks(picks, taupypicks, all_available=False):
|
||||
mean_diff = taupy_mean - picks_mean
|
||||
|
||||
logging.info(f'Calculated {len(taupypicks_new)} TauPy theoretical picks.')
|
||||
logging.info('Calculated median difference from TauPy theoretical picks of {} seconds. (mean: {})'.format(median_diff,
|
||||
mean_diff))
|
||||
logging.info(
|
||||
'Calculated median difference from TauPy theoretical picks of {} seconds. (mean: {})'.format(median_diff,
|
||||
mean_diff))
|
||||
|
||||
# return all available taupypicks corrected for median difference to autopicks
|
||||
if all_available:
|
||||
@ -851,7 +875,8 @@ def load_stacked_trace(eventdir, min_corr_stack):
|
||||
if not os.path.isfile(corr_dict_fpath):
|
||||
logging.warning('No correlations_for_stacking dict found for event {}!'.format(eventdir))
|
||||
return
|
||||
corr_dict = json.load(corr_dict_fpath) # TODO: Check this line
|
||||
with open(corr_dict_fpath) as fid:
|
||||
corr_dict = json.load(fid)
|
||||
|
||||
# get stations for stacking and nstack
|
||||
stations4stack = get_stations_min_corr(corr_dict, min_corr_stack)
|
||||
@ -866,10 +891,10 @@ def load_stacked_trace(eventdir, min_corr_stack):
|
||||
return corr_dict, nwst_id, stacked_trace, nstack
|
||||
|
||||
|
||||
def repick_mastertrace(wfdata, trace_master, pylot_parameter, event, event_id, metadata, phase_type, corr_out_dir):
|
||||
rename_LQT = phase_type == 'S'
|
||||
def repick_master_trace(wfdata, trace_master, pylot_parameter, event, event_id, metadata, phase_type, corr_out_dir):
|
||||
rename_lqt = phase_type == 'S'
|
||||
# create an 'artificial' stream object which can be used as input for PyLoT
|
||||
stream_master = modify_master_trace4pick(trace_master.copy(), wfdata, rename_LQT=rename_LQT)
|
||||
stream_master = modify_master_trace4pick(trace_master.copy(), wfdata, rename_lqt=rename_lqt)
|
||||
stream_master.write(os.path.join(corr_out_dir, 'stacked_master_trace_pylot.mseed'.format(trace_master.id)))
|
||||
try:
|
||||
picksdict = autopickstation(stream_master, pylot_parameter, verbose=True, metadata=metadata,
|
||||
@ -894,10 +919,10 @@ def repick_mastertrace(wfdata, trace_master, pylot_parameter, event, event_id, m
|
||||
return pick_stacked
|
||||
|
||||
|
||||
def create_correlation_output_dir(eventdir, fopts, phase_type):
|
||||
def create_correlation_output_dir(eventdir: str, fopts: dict, phase_type: str) -> str:
|
||||
folder = 'correlation'
|
||||
folder = add_fpath_extension(folder, fopts, phase_type)
|
||||
export_path = os.path.join(eventdir, folder)
|
||||
export_path = str(os.path.join(eventdir, folder))
|
||||
if not os.path.isdir(export_path):
|
||||
os.mkdir(export_path)
|
||||
return export_path
|
||||
@ -927,12 +952,12 @@ def write_correlation_output(export_path, correlations_dict, correlations_dict_s
|
||||
write_json(correlations_dict_stacked, os.path.join(export_path, 'correlation_results.json'))
|
||||
|
||||
|
||||
def modify_master_trace4pick(trace_master, wfdata, rename_LQT=True):
|
||||
'''
|
||||
def modify_master_trace4pick(trace_master, wfdata, rename_lqt=True):
|
||||
"""
|
||||
Create an artificial Stream with master trace instead of the trace of its corresponding channel.
|
||||
This is done to find metadata for correct station metadata which were modified when stacking (e.g. loc=ST)
|
||||
USE THIS ONLY FOR PICKING!
|
||||
'''
|
||||
"""
|
||||
# fake ZNE coordinates for autopylot
|
||||
lqt_zne = {'L': 'Z', 'Q': 'N', 'T': 'E'}
|
||||
|
||||
@ -949,7 +974,7 @@ def modify_master_trace4pick(trace_master, wfdata, rename_LQT=True):
|
||||
# take location from old trace and overwrite
|
||||
trace_master.stats.location = trace.stats.location
|
||||
trace = trace_master
|
||||
if rename_LQT:
|
||||
if rename_lqt:
|
||||
channel = trace.stats.channel
|
||||
component_new = lqt_zne.get(channel[-1])
|
||||
if not component_new:
|
||||
@ -1088,7 +1113,7 @@ def get_pickfile_name_correlated(event_id, fopts, phase_type):
|
||||
# return trace_this
|
||||
|
||||
|
||||
def prepare_correlation_input(wfdata, picks, channels, trace_master, pick_master, params, plot=None, fig_dir=None,
|
||||
def prepare_correlation_input(wfdata, picks, channels, trace_master, pick_master, phase_params, plot=None, fig_dir=None,
|
||||
ncorr=1, wfdata_highf=None, trace_master_highf=None):
|
||||
# prepare input for multiprocessing worker for all 'other' picks to correlate with current master-trace
|
||||
input_list = []
|
||||
@ -1096,7 +1121,7 @@ def prepare_correlation_input(wfdata, picks, channels, trace_master, pick_master
|
||||
assert (ncorr in [1, 2]), 'ncorr has to be 1 or 2'
|
||||
|
||||
for pick_other in picks:
|
||||
trace_other_highf = None
|
||||
stream_other = stream_other_high_f = trace_other_high_f = channel = None
|
||||
network_other = pick_other.waveform_id.network_code
|
||||
station_other = pick_other.waveform_id.station_code
|
||||
nwst_id_other = '{nw}.{st}'.format(nw=network_other, st=station_other)
|
||||
@ -1106,7 +1131,7 @@ def prepare_correlation_input(wfdata, picks, channels, trace_master, pick_master
|
||||
for channel in channels:
|
||||
stream_other = wfdata.select(network=network_other, station=station_other, channel=channel)
|
||||
if ncorr == 2:
|
||||
stream_other_highf = wfdata_highf.select(network=network_other, station=station_other, channel=channel)
|
||||
stream_other_high_f = wfdata_highf.select(network=network_other, station=station_other, channel=channel)
|
||||
if stream_other:
|
||||
break
|
||||
if not stream_other:
|
||||
@ -1114,37 +1139,39 @@ def prepare_correlation_input(wfdata, picks, channels, trace_master, pick_master
|
||||
continue
|
||||
trace_other = stream_other[0]
|
||||
if ncorr == 2:
|
||||
trace_other_highf = stream_other_highf[0]
|
||||
trace_other_high_f = stream_other_high_f[0]
|
||||
if trace_other == stream_other:
|
||||
continue
|
||||
|
||||
input_dict = {'nwst_id': nwst_id_other, 'trace1': trace_master, 'pick1': pick_master, 'trace2': trace_other,
|
||||
'pick2': pick_other, 'channel': channel, 't_before': params['t_before'],
|
||||
't_after': params['t_after'], 'cc_maxlag': params['cc_maxlag'],
|
||||
'cc_maxlag2': params['cc_maxlag2'], 'plot': plot, 'fig_dir': fig_dir, 'ncorr': ncorr,
|
||||
'trace1_highf': trace_master_highf, 'trace2_highf': trace_other_highf,
|
||||
'min_corr': params['min_corr_export']}
|
||||
'pick2': pick_other, 'channel': channel, 't_before': phase_params['t_before'],
|
||||
't_after': phase_params['t_after'], 'cc_maxlag': phase_params['cc_maxlag'],
|
||||
'cc_maxlag2': phase_params['cc_maxlag2'], 'plot': plot, 'fig_dir': fig_dir, 'ncorr': ncorr,
|
||||
'trace1_highf': trace_master_highf, 'trace2_highf': trace_other_high_f,
|
||||
'min_corr': phase_params['min_corr_export']}
|
||||
input_list.append(input_dict)
|
||||
|
||||
return input_list
|
||||
|
||||
|
||||
def stack_mastertrace(wfdata_lowf, wfdata_highf, wfdata_raw, picks, params, channels, method, fig_dir):
|
||||
'''
|
||||
def stack_mastertrace(wfdata_lowf: Stream, wfdata_highf: Stream, wfdata_raw: Stream, picks: list, params: dict,
|
||||
channels: list, method: str, fig_dir: str):
|
||||
"""
|
||||
Correlate all stations with the first available station given in station_list, a list containing central,
|
||||
permanent, long operating, low noise stations with descending priority.
|
||||
A master trace will be created by stacking well correlating traces onto this station.
|
||||
'''
|
||||
"""
|
||||
|
||||
def get_best_station4stack(station_results):
|
||||
''' return station with maximum mean_ccc'''
|
||||
ccc_means = {nwst_id: value['mean_ccc'] for nwst_id, value in station_results.items() if
|
||||
def get_best_station4stack(sta_result):
|
||||
""" return station with maximum mean_ccc"""
|
||||
ccc_means = {nwst_id: value['mean_ccc'] for nwst_id, value in sta_result.items() if
|
||||
not np.isnan(value['mean_ccc'])}
|
||||
if len(ccc_means) < 1:
|
||||
logging.warning('No valid station found for stacking! Return.')
|
||||
return
|
||||
best_station_id = max(ccc_means, key=ccc_means.get)
|
||||
logging.info('Found highest mean correlation for station {} ({})'.format(best_station_id, max(ccc_means.values())))
|
||||
logging.info(
|
||||
'Found highest mean correlation for station {} ({})'.format(best_station_id, max(ccc_means.values())))
|
||||
return best_station_id
|
||||
|
||||
station_results = iterate_correlation(wfdata_lowf, wfdata_highf, channels, picks, method, params, fig_dir=fig_dir)
|
||||
@ -1153,7 +1180,7 @@ def stack_mastertrace(wfdata_lowf, wfdata_highf, wfdata_raw, picks, params, chan
|
||||
# in case no stream with a valid pick is found
|
||||
if not nwst_id_master:
|
||||
logging.info('No mastertrace found! Will skip this event.')
|
||||
return False
|
||||
return None
|
||||
|
||||
trace_master = station_results[nwst_id_master]['trace']
|
||||
stations4stack = station_results[nwst_id_master]['stations4stack']
|
||||
@ -1163,22 +1190,27 @@ def stack_mastertrace(wfdata_lowf, wfdata_highf, wfdata_raw, picks, params, chan
|
||||
|
||||
dt_pre, dt_post = params['dt_stacking']
|
||||
trace_master, nstack = apply_stacking(trace_master, stations4stack, wfdata_raw, picks, method=method,
|
||||
check_RMS=params['check_RMS'], plot=params['plot'], fig_dir=fig_dir,
|
||||
check_rms=params['check_RMS'], plot=params['plot'], fig_dir=fig_dir,
|
||||
dt_pre=dt_pre, dt_post=dt_post)
|
||||
|
||||
return correlations_dict, nwst_id_master, trace_master, nstack
|
||||
|
||||
|
||||
def iterate_correlation(wfdata_lowf, wfdata_highf, channels, picks, method, params, fig_dir=None):
|
||||
'''iterate over possible stations for master-trace store them and return a dictionary'''
|
||||
def iterate_correlation(wfdata_lowf: Stream, wfdata_highf: Stream, channels: list, picks: list, method: str,
|
||||
params: dict, fig_dir: str = None) -> dict:
|
||||
"""iterate over possible stations for master-trace store them and return a dictionary"""
|
||||
|
||||
station_results = {nwst_id: {'mean_ccc': np.nan, 'correlations_dict': None, 'stations4stack': None, 'trace': None}
|
||||
station_results = {nwst_id: {'mean_ccc': np.nan, 'correlations_dict': dict(), 'stations4stack': dict(),
|
||||
'trace': None}
|
||||
for nwst_id in params['station_list']}
|
||||
|
||||
for nwst_id_master in params['station_list']:
|
||||
logging.info(20 * '#')
|
||||
logging.info('Starting correlation for station: {}'.format(nwst_id_master))
|
||||
nw, st = nwst_id_master.split('.')
|
||||
|
||||
# get master-trace
|
||||
stream_master_lowf = stream_master_highf = None
|
||||
for channel in channels:
|
||||
stream_master_lowf = wfdata_lowf.select(network=nw, station=st, channel=channel)
|
||||
stream_master_highf = wfdata_highf.select(network=nw, station=st, channel=channel)
|
||||
@ -1211,8 +1243,8 @@ def iterate_correlation(wfdata_lowf, wfdata_highf, channels, picks, method, para
|
||||
|
||||
input_list = prepare_correlation_input(wfdata_lowf, picks, channels, trace_master_lowf, pick_master,
|
||||
trace_master_highf=trace_master_highf, wfdata_highf=wfdata_highf,
|
||||
params=params, plot=params['plot_detailed'], fig_dir=fig_dir_traces,
|
||||
ncorr=2)
|
||||
phase_params=params, plot=params['plot_detailed'],
|
||||
fig_dir=fig_dir_traces, ncorr=2)
|
||||
|
||||
if params['plot_detailed'] and not fig_dir_traces:
|
||||
# serial
|
||||
@ -1231,9 +1263,11 @@ def iterate_correlation(wfdata_lowf, wfdata_highf, channels, picks, method, para
|
||||
return station_results
|
||||
|
||||
|
||||
def apply_stacking(trace_master, stations4stack, wfdata, picks, method, check_RMS, dt_pre=250., dt_post=250.,
|
||||
plot=False, fig_dir=None):
|
||||
def check_trace_length_and_cut(trace, correlated_midtime, dt_pre, dt_post):
|
||||
def apply_stacking(trace_master: Trace, stations4stack: dict, wfdata: Stream, picks: list, method: str,
|
||||
check_rms: bool, dt_pre: float = 250., dt_post: float = 250., plot: bool = False,
|
||||
fig_dir: str = None) -> tuple:
|
||||
|
||||
def check_trace_length_and_cut(trace: Trace):
|
||||
starttime = correlated_midtime - dt_pre
|
||||
endtime = correlated_midtime + dt_post
|
||||
|
||||
@ -1248,7 +1282,7 @@ def apply_stacking(trace_master, stations4stack, wfdata, picks, method, check_RM
|
||||
|
||||
trace_master = trace_master.copy()
|
||||
trace_master.stats.location = 'ST'
|
||||
check_trace_length_and_cut(trace_master, pick.time, dt_pre=dt_pre, dt_post=dt_post)
|
||||
check_trace_length_and_cut(trace_master)
|
||||
|
||||
# empty trace so that it does not stack twice
|
||||
trace_master.data = np.zeros(len(trace_master.data))
|
||||
@ -1262,15 +1296,16 @@ def apply_stacking(trace_master, stations4stack, wfdata, picks, method, check_RM
|
||||
stream_other = wfdata.select(network=nw, station=st, channel=channel)
|
||||
correlated_midtime = pick_other.time - dpick
|
||||
trace_other = stream_other[0].copy()
|
||||
check_trace_length_and_cut(trace_other, correlated_midtime, dt_pre=dt_pre, dt_post=dt_post)
|
||||
check_trace_length_and_cut(trace_other)
|
||||
|
||||
if not len(trace_other) == len(trace_master):
|
||||
logging.warning('Can not stack trace on master trace because of different lengths: {}. Continue'.format(nwst_id))
|
||||
logging.warning(
|
||||
'Can not stack trace on master trace because of different lengths: {}. Continue'.format(nwst_id))
|
||||
continue
|
||||
|
||||
traces4stack.append(trace_other)
|
||||
|
||||
if check_RMS:
|
||||
if check_rms:
|
||||
traces4stack = checkRMS(traces4stack, plot=plot, fig_dir=fig_dir)
|
||||
|
||||
if plot:
|
||||
@ -1307,12 +1342,14 @@ def checkRMS(traces4stack, plot=False, fig_dir=None, max_it=10, ntimes_std=5.):
|
||||
traces4stack = sorted(traces4stack, key=lambda x: x.id)
|
||||
|
||||
for trace in traces4stack:
|
||||
rms_list.append(RMS(trace.data))
|
||||
rms_list.append(calc_rms(trace.data))
|
||||
trace_names.append(trace.id)
|
||||
|
||||
# iterative elimination of RMS outliers
|
||||
iterate = True
|
||||
count = 0
|
||||
std = 0
|
||||
mean = 0
|
||||
while iterate:
|
||||
count += 1
|
||||
if count >= max_it:
|
||||
@ -1341,7 +1378,8 @@ def checkRMS(traces4stack, plot=False, fig_dir=None, max_it=10, ntimes_std=5.):
|
||||
return traces4stack
|
||||
|
||||
|
||||
def plot_rms(rms_list, trace_names, mean, std, fig_dir, count, ntimes_std):
|
||||
def plot_rms(rms_list: list, trace_names: list, mean: float, std: float, fig_dir: str, count: int,
|
||||
ntimes_std: float) -> None:
|
||||
fig = plt.figure(figsize=(16, 9))
|
||||
ax = fig.add_subplot(111)
|
||||
ax.semilogy(rms_list, 'b.')
|
||||
@ -1359,7 +1397,7 @@ def plot_rms(rms_list, trace_names, mean, std, fig_dir, count, ntimes_std):
|
||||
plt.close(fig)
|
||||
|
||||
|
||||
def RMS(X):
|
||||
def calc_rms(X):
|
||||
"""
|
||||
Returns root mean square of a given array LON
|
||||
"""
|
||||
@ -1381,7 +1419,7 @@ def resample_parallel(stream, freq, ncores):
|
||||
|
||||
|
||||
def rotate_stream(stream, metadata, origin, stations_dict, channels, inclination, ncores):
|
||||
''' Rotate stream to LQT. To do this all traces have to be cut to equal lenths'''
|
||||
""" Rotate stream to LQT. To do this all traces have to be cut to equal lenths"""
|
||||
input_list = []
|
||||
cut_stream_to_same_length(stream)
|
||||
new_stream = Stream()
|
||||
@ -1424,7 +1462,7 @@ def rotate_stream(stream, metadata, origin, stations_dict, channels, inclination
|
||||
# return stream
|
||||
|
||||
|
||||
def correlate_parallel(input_list, ncores):
|
||||
def correlate_parallel(input_list: list, ncores: int) -> dict:
|
||||
parallel = Parallel(n_jobs=ncores)
|
||||
|
||||
logging.info('Correlate_parallel: Generated {} parallel jobs.'.format(ncores))
|
||||
@ -1435,7 +1473,7 @@ def correlate_parallel(input_list, ncores):
|
||||
return unpack_result(correlation_result)
|
||||
|
||||
|
||||
def correlate_serial(input_list, plot=False):
|
||||
def correlate_serial(input_list: list, plot: bool = False) -> dict:
|
||||
correlation_result = []
|
||||
for input_dict in input_list:
|
||||
input_dict['plot'] = plot
|
||||
@ -1443,7 +1481,7 @@ def correlate_serial(input_list, plot=False):
|
||||
return unpack_result(correlation_result)
|
||||
|
||||
|
||||
def taupy_parallel(input_list, ncores):
|
||||
def taupy_parallel(input_list: list, ncores: int) -> dict:
|
||||
parallel = Parallel(n_jobs=ncores)
|
||||
|
||||
logging.info('Taupy_parallel: Generated {} parallel jobs.'.format(ncores))
|
||||
@ -1454,19 +1492,19 @@ def taupy_parallel(input_list, ncores):
|
||||
return unpack_result(taupy_results)
|
||||
|
||||
|
||||
def unpack_result(result):
|
||||
def unpack_result(result: list) -> dict:
|
||||
result_dict = {item['nwst_id']: {key: item[key] for key in item.keys()} for item in result}
|
||||
nerr = 0
|
||||
for item in result_dict.values():
|
||||
if item['err']:
|
||||
logging.debug(f'Found error for {item["nwst_id"]}: {item["err"]}.')
|
||||
#logging.debug(f'Detailed traceback: {item["exc"]}')
|
||||
# logging.debug(f'Detailed traceback: {item["exc"]}')
|
||||
nerr += 1
|
||||
logging.info('Unpack results: Found {} errors after multiprocessing'.format(nerr))
|
||||
return result_dict
|
||||
|
||||
|
||||
def get_taupy_picks(origin, stations_dict, pylot_parameter, phase_type, ncores=None):
|
||||
def get_taupy_picks(origin: Origin, stations_dict, pylot_parameter, phase_type, ncores=None):
|
||||
input_list = []
|
||||
|
||||
taup_phases = pylot_parameter['taup_phases'].split(',')
|
||||
@ -1526,10 +1564,10 @@ def taupy_worker(input_dict):
|
||||
first_arrival = arrivals[0]
|
||||
output_dict = dict(nwst_id=input_dict['nwst_id'], phase_name=first_arrival.name,
|
||||
phase_time=source_time + first_arrival.time, phase_dist=first_arrival.distance, err=None,
|
||||
exc=None,)
|
||||
exc=None, )
|
||||
except Exception as e:
|
||||
exc = traceback.format_exc()
|
||||
output_dict = dict(nwst_id=input_dict['nwst_id'], phase_name=None, phase_time=None, err=str(e), exc=exc,)
|
||||
output_dict = dict(nwst_id=input_dict['nwst_id'], phase_name=None, phase_time=None, err=str(e), exc=exc, )
|
||||
return output_dict
|
||||
|
||||
|
||||
@ -1559,7 +1597,7 @@ def rotation_worker(input_dict):
|
||||
|
||||
|
||||
def correlation_worker(input_dict):
|
||||
'''worker function for multiprocessing'''
|
||||
"""worker function for multiprocessing"""
|
||||
|
||||
# unpack input dictionary
|
||||
nwst_id = input_dict['nwst_id']
|
||||
@ -1613,7 +1651,7 @@ def correlation_worker(input_dict):
|
||||
'channel': channel, }
|
||||
|
||||
|
||||
def get_pick4station(picks, network_code, station_code, method='auto'):
|
||||
def get_pick4station(picks: list, network_code: str, station_code: str, method: str = 'auto') -> Pick:
|
||||
for pick in picks:
|
||||
if pick.waveform_id.network_code == network_code:
|
||||
if pick.waveform_id.station_code == station_code:
|
||||
@ -1621,13 +1659,14 @@ def get_pick4station(picks, network_code, station_code, method='auto'):
|
||||
return pick
|
||||
|
||||
|
||||
def get_stations_min_corr(corr_results, min_corr):
|
||||
def get_stations_min_corr(corr_results: dict, min_corr: float) -> dict:
|
||||
corr_results = {nwst_id: result for nwst_id, result in corr_results.items() if result['ccc'] > min_corr}
|
||||
return corr_results
|
||||
|
||||
|
||||
def plot_mastertrace_corr(nwst_id, corr_results, wfdata, stations_dict, picks, trace_master, method, min_corr, title,
|
||||
fig_dir=None):
|
||||
def plot_mastertrace_corr(nwst_id: str, corr_results: dict, wfdata: Stream, stations_dict: dict, picks: list,
|
||||
trace_master: Trace, method: str, min_corr: float, title: str,
|
||||
fig_dir: str = None) -> None:
|
||||
nw, st = nwst_id.split('.')
|
||||
coords_master = stations_dict[nwst_id]
|
||||
pick_master = get_pick4station(picks, nw, st, method)
|
||||
@ -1650,7 +1689,7 @@ def plot_mastertrace_corr(nwst_id, corr_results, wfdata, stations_dict, picks, t
|
||||
plt.show()
|
||||
|
||||
|
||||
def make_figure_dirs(fig_dir, trace_master_id):
|
||||
def make_figure_dirs(fig_dir: str, trace_master_id: str) -> str:
|
||||
fig_dir_traces = ''
|
||||
if fig_dir and os.path.isdir(fig_dir):
|
||||
fig_dir_traces = os.path.join(fig_dir, 'corr_with_{}'.format(trace_master_id))
|
||||
@ -1659,15 +1698,18 @@ def make_figure_dirs(fig_dir, trace_master_id):
|
||||
return fig_dir_traces
|
||||
|
||||
|
||||
def plot_section(wfdata, trace_this, pick_this, picks, stations2compare, channels, corr_results, method, dt_pre=20.,
|
||||
dt_post=50, axes=None, max_stations=50.):
|
||||
'''Plot a section with all stations used for correlation on ax'''
|
||||
def plot_section(wfdata: Stream, trace_this: Trace, pick_this: Pick, picks: list, stations2compare: dict,
|
||||
channels: list, corr_results: dict, method: str, dt_pre: float = 20., dt_post: float = 50.,
|
||||
axes: dict = None, max_stations: int = 50) -> None:
|
||||
"""Plot a section with all stations used for correlation on ax"""
|
||||
ylabels = []
|
||||
yticks = []
|
||||
|
||||
trace_this = trace_this.copy()
|
||||
trace_this.trim(starttime=pick_this.time - dt_pre, endtime=pick_this.time + dt_post)
|
||||
|
||||
ax_sec = None
|
||||
|
||||
# iterate over all closest stations ("other" stations)
|
||||
for index, nwst_id in enumerate(stations2compare.keys()):
|
||||
if index >= max_stations:
|
||||
@ -1700,6 +1742,8 @@ def plot_section(wfdata, trace_this, pick_this, picks, stations2compare, channel
|
||||
if np.isnan(dpick) or not pick_other or not pick_other.time:
|
||||
continue
|
||||
|
||||
stream = None
|
||||
|
||||
# continue if there are no data for station for whatever reason
|
||||
for channel in channels:
|
||||
stream = wfdata.select(station=st, network=nw, channel=channel)
|
||||
@ -1718,13 +1762,14 @@ def plot_section(wfdata, trace_this, pick_this, picks, stations2compare, channel
|
||||
index + 0.5, color='r', lw=0.5)
|
||||
|
||||
# Plot desciption
|
||||
ax_sec.set_yticks(yticks)
|
||||
ax_sec.set_yticklabels(ylabels)
|
||||
ax_sec.set_title('Section with corresponding picks.')
|
||||
ax_sec.set_xlabel('Samples. Traces are shifted in time.')
|
||||
if ax_sec:
|
||||
ax_sec.set_yticks(yticks)
|
||||
ax_sec.set_yticklabels(ylabels)
|
||||
ax_sec.set_title('Section with corresponding picks.')
|
||||
ax_sec.set_xlabel('Samples. Traces are shifted in time.')
|
||||
|
||||
|
||||
def plot_stacked_trace_pick(trace, pick, pylot_parameter):
|
||||
def plot_stacked_trace_pick(trace: Trace, pick: Pick, pylot_parameter: PylotParameter) -> plt.Figure:
|
||||
trace_filt = trace.copy()
|
||||
if pylot_parameter['filter_type'] and pylot_parameter['filter_options']:
|
||||
ftype = pylot_parameter['filter_type'][0]
|
||||
@ -1746,8 +1791,9 @@ def plot_stacked_trace_pick(trace, pick, pylot_parameter):
|
||||
return fig
|
||||
|
||||
|
||||
def plot_stations(stations_dict, stations2compare, coords_this, corr_result, trace, pick, window_title=None):
|
||||
''' Plot routine to check proximity algorithm. '''
|
||||
def plot_stations(stations_dict: dict, stations2compare: dict, coords_this: dict, corr_result: dict, trace: Trace,
|
||||
pick: Pick, window_title: str = None):
|
||||
""" Plot routine to check proximity algorithm. """
|
||||
|
||||
title = trace.id
|
||||
|
||||
@ -1812,15 +1858,15 @@ def plot_stations(stations_dict, stations2compare, coords_this, corr_result, tra
|
||||
return fig, axes
|
||||
|
||||
|
||||
def get_data(eventdir, data_dir, headonly=False):
|
||||
''' Read and return waveformdata read from eventdir/data_dir. '''
|
||||
def get_data(eventdir: str, data_dir: str, headonly: bool = False) -> Stream:
|
||||
""" Read and return waveformdata read from eventdir/data_dir. """
|
||||
wfdata_path = os.path.join(eventdir, data_dir)
|
||||
wfdata = read(os.path.join(wfdata_path, '*'), headonly=headonly)
|
||||
return wfdata
|
||||
|
||||
|
||||
def get_closest_stations(coords, stations_dict, n):
|
||||
''' Calculate distances and return n closest stations in stations_dict for station at coords. '''
|
||||
""" Calculate distances and return n closest stations in stations_dict for station at coords. """
|
||||
distances = {}
|
||||
for station_id, st_coords in stations_dict.items():
|
||||
dist = gps2dist_azimuth(coords['latitude'], coords['longitude'], st_coords['latitude'], st_coords['longitude'],
|
||||
@ -1836,7 +1882,7 @@ def get_closest_stations(coords, stations_dict, n):
|
||||
|
||||
|
||||
def get_random_stations(coords, stations_dict, n):
|
||||
''' Calculate distances and return n randomly selected stations in stations_dict for station at coords. '''
|
||||
""" Calculate distances and return n randomly selected stations in stations_dict for station at coords. """
|
||||
random_keys = random.sample(list(stations_dict.keys()), n)
|
||||
distances = {}
|
||||
for station_id, st_coords in stations_dict.items():
|
||||
@ -1851,11 +1897,41 @@ def get_random_stations(coords, stations_dict, n):
|
||||
return random_stations
|
||||
|
||||
|
||||
# Notes: if use_master_trace is set True, traces above a specified ccc threshold will be stacked onto one 'ideal'
|
||||
# station into a master-trace. This trace will be picked and used again for cross correlation with all other stations
|
||||
# to find a pick for these stations.
|
||||
def prepare_corr_params(parse_args, logger):
|
||||
with open(parse_args.params) as infile:
|
||||
parameters_dict = yaml.safe_load(infile)
|
||||
|
||||
if __name__ == "__main__":
|
||||
# logging
|
||||
logger.setLevel(DEBUG_LEVELS.get(parameters_dict['logging']))
|
||||
|
||||
# number of cores
|
||||
if parse_args.ncores is not None:
|
||||
ncores = int(parse_args.ncores)
|
||||
else:
|
||||
ncores = None
|
||||
|
||||
# plot options
|
||||
plot_dict = dict(plot=parse_args.plot, plot_detailed=parse_args.plot_detailed,
|
||||
save_fig=not parse_args.show_fig)
|
||||
|
||||
corr_params = {phase: CorrelationParameters(ncores=ncores) for phase in parameters_dict['pick_phases']}
|
||||
for phase, params_phase in corr_params.items():
|
||||
params_phase.add_parameters(**plot_dict)
|
||||
params_phase.add_parameters(**parameters_dict[phase])
|
||||
|
||||
return corr_params
|
||||
|
||||
|
||||
def init_logger():
|
||||
logger = logging.getLogger()
|
||||
handler = logging.StreamHandler(sys.stdout)
|
||||
fhandler = logging.FileHandler('pick_corr.out')
|
||||
logger.addHandler(handler)
|
||||
logger.addHandler(fhandler)
|
||||
return logger
|
||||
|
||||
|
||||
def setup_parser():
|
||||
parser = argparse.ArgumentParser(description='Correlate picks from PyLoT.')
|
||||
parser.add_argument('dmt_path', help='path containing dmt_database with PyLoT picks')
|
||||
parser.add_argument('pylot_infile', help='path to autoPyLoT inputfile (pylot.in)')
|
||||
@ -1881,44 +1957,22 @@ if __name__ == "__main__":
|
||||
parser.add_argument('-istart', default=0, help='first event index')
|
||||
parser.add_argument('-istop', default=1e9, help='last event index')
|
||||
|
||||
args = parser.parse_args()
|
||||
return parser.parse_args()
|
||||
|
||||
# PARAMETER DEFINITIONS ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
|
||||
|
||||
# plot options
|
||||
plot_dict = dict(plot=args.plot, plot_detailed=args.plot_detailed, save_fig=not (args.show_fig))
|
||||
if __name__ == "__main__":
|
||||
ARGS = setup_parser()
|
||||
|
||||
# initialize logging
|
||||
LOGGER = init_logger()
|
||||
|
||||
# alparray configuration: ZNE ~ 123
|
||||
channels = {'P': ['*Z', '*1'], 'S': ['*Q', '*T']} # '*N', '*E', '*2', '*3',
|
||||
debug_levels = {'debug': logging.DEBUG,
|
||||
'info': logging.INFO,
|
||||
'warn': logging.WARNING,
|
||||
'error': logging.ERROR,
|
||||
'critical': logging.CRITICAL}
|
||||
CHANNELS = {'P': ['*Z', '*1'], 'S': ['*Q', '*T']} # '*N', '*E', '*2', '*3',
|
||||
|
||||
# initialize parameters from yaml, set logging level
|
||||
CORR_PARAMS = prepare_corr_params(ARGS, LOGGER)
|
||||
|
||||
if args.ncores is not None:
|
||||
_ncores = int(args.ncores)
|
||||
else:
|
||||
_ncores = None
|
||||
|
||||
with open(args.params) as infile:
|
||||
parameters_dict = yaml.safe_load(infile)
|
||||
|
||||
params = {phase: None for phase in parameters_dict['pick_phases']}
|
||||
|
||||
logger = logging.getLogger()
|
||||
logger.setLevel(debug_levels.get(parameters_dict['logging']))
|
||||
handler = logging.StreamHandler(sys.stdout)
|
||||
fhandler = logging.FileHandler('pick_corr.out')
|
||||
logger.addHandler(handler)
|
||||
logger.addHandler(fhandler)
|
||||
|
||||
for phase in params.keys():
|
||||
params_phase = CorrelationParameters(ncores=_ncores)
|
||||
params_phase.add_parameters(**plot_dict)
|
||||
params_phase.add_parameters(**parameters_dict[phase])
|
||||
params[phase] = params_phase
|
||||
|
||||
correlation_main(args.dmt_path, args.pylot_infile, params=params, istart=int(args.istart), istop=int(args.istop),
|
||||
channel_config=channels, update=args.update, event_blacklist=args.blacklist)
|
||||
# MAIN +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
|
||||
correlation_main(ARGS.dmt_path, ARGS.pylot_infile, params=CORR_PARAMS, istart=int(ARGS.istart),
|
||||
istop=int(ARGS.istop),
|
||||
channel_config=CHANNELS, update=ARGS.update, event_blacklist=ARGS.blacklist)
|
||||
|
Loading…
Reference in New Issue
Block a user