#!/usr/bin/python # -*- coding: utf-8 -*- from __future__ import print_function import argparse import datetime import glob import os import pylot.core.loc.focmec as focmec import pylot.core.loc.hash as hash import pylot.core.loc.hypo71 as hypo71 import pylot.core.loc.hypodd as hypodd import pylot.core.loc.hyposat as hyposat import pylot.core.loc.nll as nll import pylot.core.loc.velest as velest from obspy import read_events from obspy.core.event import ResourceIdentifier # from PySide.QtGui import QWidget, QInputDialog from pylot.core.analysis.magnitude import MomentMagnitude, LocalMagnitude from pylot.core.io.data import Data from pylot.core.io.inputs import PylotParameter from pylot.core.pick.autopick import autopickevent, iteratepicker from pylot.core.util.dataprocessing import restitute_data, read_metadata from pylot.core.util.defaults import SEPARATOR from pylot.core.util.event import Event from pylot.core.util.structure import DATASTRUCTURE from pylot.core.util.utils import real_None, remove_underscores, trim_station_components, check4gaps, check4doubled from pylot.core.util.version import get_git_version as _getVersionString __version__ = _getVersionString() def autoPyLoT(input_dict=None, parameter=None, inputfile=None, fnames=None, eventid=None, savepath=None, savexml=True, station='all', iplot=0, ncores=0): """ Determine phase onsets automatically utilizing the automatic picking algorithms by Kueperkoch et al. 2010/2012. :param inputfile: path to the input file containing all parameter information for automatic picking (for formatting details, see. `~pylot.core.io.inputs.PylotParameter` :type inputfile: str :return: .. rubric:: Example """ if ncores == 1: sp_info = 'autoPyLoT is running serial on 1 cores.' else: if ncores == 0: ncores_readable = 'all available' else: ncores_readable = ncores sp_info = 'autoPyLoT is running in parallel on {} cores.'.format(ncores_readable) splash = '''************************************\n *********autoPyLoT starting*********\n The Python picking and Location Tool\n Version {version} 2017\n \n Authors:\n L. Kueperkoch (BESTEC GmbH, Landau i. d. Pfalz)\n M. Paffrath (Ruhr-Universitaet Bochum)\n S. Wehling-Benatelli (Ruhr-Universitaet Bochum)\n {sp} ***********************************'''.format(version=_getVersionString(), sp=sp_info) print(splash) parameter = real_None(parameter) inputfile = real_None(inputfile) eventid = real_None(eventid) locflag = 1 if input_dict and isinstance(input_dict, dict): if 'parameter' in input_dict: parameter = input_dict['parameter'] if 'fig_dict' in input_dict: fig_dict = input_dict['fig_dict'] if 'station' in input_dict: station = input_dict['station'] if 'fnames' in input_dict: fnames = input_dict['fnames'] if 'eventid' in input_dict: eventid = input_dict['eventid'] if 'iplot' in input_dict: iplot = input_dict['iplot'] if 'locflag' in input_dict: locflag = input_dict['locflag'] if 'savexml' in input_dict: savexml = input_dict['savexml'] if not parameter: if inputfile: parameter = PylotParameter(inputfile) #iplot = parameter['iplot'] else: infile = os.path.join(os.path.expanduser('~'), '.pylot', 'pylot.in') print('Using default input file {}'.format(infile)) parameter = PylotParameter(infile) else: if not type(parameter) == PylotParameter: print('Wrong input type for parameter: {}'.format(type(parameter))) return if inputfile: print('Parameters set and input file given. Choose either of both.') return evt = None # reading parameter file if parameter.hasParam('datastructure'): # getting information on data structure datastructure = DATASTRUCTURE[parameter.get('datastructure')]() dsfields = {'root': parameter.get('rootpath'), 'dpath': parameter.get('datapath'), 'dbase': parameter.get('database')} exf = ['root', 'dpath', 'dbase'] if parameter['eventID'] is not '*' and fnames == 'None': dsfields['eventID'] = parameter['eventID'] exf.append('eventID') datastructure.modifyFields(**dsfields) datastructure.setExpandFields(exf) # check if default location routine NLLoc is available if parameter['nllocbin'] and locflag: # get NLLoc-root path nllocroot = parameter.get('nllocroot') # get path to NLLoc executable nllocbin = parameter.get('nllocbin') nlloccall = '%s/NLLoc' % nllocbin # get name of phase file phasef = parameter.get('phasefile') phasefile = '%s/obs/%s' % (nllocroot, phasef) # get name of NLLoc-control file ctrf = parameter.get('ctrfile') ctrfile = '%s/run/%s' % (nllocroot, ctrf) # pattern of NLLoc ttimes from location grid ttpat = parameter.get('ttpatter') # pattern of NLLoc-output file nllocoutpatter = parameter.get('outpatter') maxnumit = 3 # maximum number of iterations for re-picking else: locflag = 0 print(" !!! ") print("!!No location routine available, autoPyLoT is running in non-location mode!!") print("!!No source parameter estimation possible!!") print(" !!! ") if not input_dict: # started in production mode datapath = datastructure.expandDataPath() if fnames == 'None' and parameter['eventID'] is '*': # multiple event processing # read each event in database events = [events for events in glob.glob(os.path.join(datapath, '*')) if os.path.isdir(events)] elif fnames == 'None' and parameter['eventID'] is not '*' and not type(parameter['eventID']) == list: # single event processing events = glob.glob(os.path.join(datapath, parameter['eventID'])) elif fnames == 'None' and type(parameter['eventID']) == list: # multiple event processing events = [] for eventID in parameter['eventID']: events.append(os.path.join(datapath, eventID)) else: # autoPyLoT was initialized from GUI events = [] events.append(eventid) evID = os.path.split(eventid)[-1] locflag = 2 else: # started in tune mode datapath = os.path.join(parameter['rootpath'], parameter['datapath']) events = [] events.append(os.path.join(datapath, parameter['database'], eventid)) if not events: print('autoPyLoT: No events given. Return!') return # transform system path separator to '/' for index, eventpath in enumerate(events): eventpath = eventpath.replace(SEPARATOR, '/') events[index] = eventpath for eventpath in events: evID = os.path.split(eventpath)[-1] fext = '.xml' filename = os.path.join(eventpath, 'PyLoT_' + evID + fext) try: data = Data(evtdata=filename) data.get_evt_data().path = eventpath print('Reading event data from filename {}...'.format(filename)) except Exception as e: print('Could not read event from file {}: {}'.format(filename, e)) data = Data() pylot_event = Event(eventpath) # event should be path to event directory data.setEvtData(pylot_event) if fnames == 'None': data.setWFData(glob.glob(os.path.join(datapath, eventpath, '*'))) station = 'all' if savepath == None or savepath == 'None': savepath = eventpath # the following is necessary because within # multiple event processing no event ID is provided # in autopylot.in try: parameter.get('eventID') except: now = datetime.datetime.now() eventID = '%d%02d%02d%02d%02d' % (now.year, now.month, now.day, now.hour, now.minute) parameter.setParam(eventID=eventID) else: data.setWFData(fnames) eventpath = events[0] # now = datetime.datetime.now() # evID = '%d%02d%02d%02d%02d' % (now.year, # now.month, # now.day, # now.hour, # now.minute) parameter.setParam(eventID=eventid) wfdat = data.getWFData() # all available streams if not station == 'all': wfdat = wfdat.select(station=station) if not wfdat: print('Could not find station {}. STOP!'.format(station)) return wfdat = remove_underscores(wfdat) # trim components for each station to avoid problems with different trace starttimes for one station wfdat = check4gaps(wfdat) wfdat = check4doubled(wfdat) wfdat = trim_station_components(wfdat, trim_start=True, trim_end=False) metadata = read_metadata(parameter.get('invdir')) corr_dat = None if locflag: print("Restitute data ...") corr_dat = restitute_data(wfdat.copy(), *metadata, ncores=ncores) if not corr_dat and locflag: locflag = 2 print('Working on event %s. Stations: %s' % (eventpath, station)) print(wfdat) ########################################################## # !automated picking starts here! if input_dict: if 'fig_dict' in input_dict: fig_dict = input_dict['fig_dict'] picks = autopickevent(wfdat, parameter, iplot=iplot, fig_dict=fig_dict, ncores=ncores, metadata=metadata, origin=data.get_evt_data().origins) else: picks = autopickevent(wfdat, parameter, iplot=iplot, ncores=ncores, metadata=metadata, origin=data.get_evt_data().origins) ########################################################## # locating if locflag > 0: # write phases to NLLoc-phase file nll.export(picks, phasefile, parameter) # For locating the event the NLLoc-control file has to be modified! nllocout = '%s_%s' % (evID, nllocoutpatter) # create comment line for NLLoc-control file nll.modify_inputs(ctrf, nllocroot, nllocout, phasef, ttpat) # locate the event nll.locate(ctrfile, inputfile) # !iterative picking if traces remained unpicked or occupied with bad picks! # get theoretical onset times for picks with weights >= 4 # in order to reprocess them using smaller time windows around theoretical onset # get stations with bad onsets badpicks = [] for key in picks: if picks[key]['P']['weight'] >= 4 or picks[key]['S']['weight'] >= 4: badpicks.append([key, picks[key]['P']['mpp']]) # TODO keep code DRY (Don't Repeat Yourself) the following part is written twice # suggestion: delete block and modify the later similar block to work properly if len(badpicks) == 0: print("autoPyLoT: No bad onsets found, thus no iterative picking necessary!") # get NLLoc-location file locsearch = '%s/loc/%s.????????.??????.grid?.loc.hyp' % (nllocroot, nllocout) if len(glob.glob(locsearch)) > 0: # get latest NLLoc-location file if several are available nllocfile = max(glob.glob(locsearch), key=os.path.getctime) evt = read_events(nllocfile)[0] # calculate seismic moment Mo and moment magnitude Mw moment_mag = MomentMagnitude(corr_dat, evt, parameter.get('vp'), parameter.get('Qp'), parameter.get('rho'), True, \ iplot) # update pick with moment property values (w0, fc, Mo) for station, props in moment_mag.moment_props.items(): picks[station]['P'].update(props) evt = moment_mag.updated_event() net_mw = moment_mag.net_magnitude() print("Network moment magnitude: %4.1f" % net_mw.mag) # calculate local (Richter) magntiude WAscaling = parameter.get('WAscaling') magscaling = parameter.get('magscaling') local_mag = LocalMagnitude(corr_dat, evt, parameter.get('sstop'), WAscaling, True, iplot) for station, amplitude in local_mag.amplitudes.items(): picks[station]['S']['Ao'] = amplitude.generic_amplitude print("Local station magnitudes scaled with:") print("log(Ao) + %f * log(r) + %f * r + %f" % (WAscaling[0], WAscaling[1], WAscaling[2])) evt = local_mag.updated_event(magscaling) net_ml = local_mag.net_magnitude(magscaling) print("Network local magnitude: %4.1f" % net_ml.mag) print("Network local magnitude scaled with:") print("%f * Ml + %f" % (magscaling[0], magscaling[1])) else: print("autoPyLoT: No NLLoc-location file available!") print("No source parameter estimation possible!") locflag = 9 else: # get theoretical P-onset times from NLLoc-location file locsearch = '%s/loc/%s.????????.??????.grid?.loc.hyp' % (nllocroot, nllocout) if len(glob.glob(locsearch)) > 0: # get latest file if several are available nllocfile = max(glob.glob(locsearch), key=os.path.getctime) nlloccounter = 0 while len(badpicks) > 0 and nlloccounter <= maxnumit: nlloccounter += 1 if nlloccounter > maxnumit: print("autoPyLoT: Number of maximum iterations reached, stop iterative picking!") break print("autoPyLoT: Starting with iteration No. %d ..." % nlloccounter) if input_dict: if 'fig_dict' in input_dict: fig_dict = input_dict['fig_dict'] picks = iteratepicker(wfdat, nllocfile, picks, badpicks, parameter, fig_dict=fig_dict) else: picks = iteratepicker(wfdat, nllocfile, picks, badpicks, parameter) # write phases to NLLoc-phase file nll.export(picks, phasefile, parameter) # remove actual NLLoc-location file to keep only the last os.remove(nllocfile) # locate the event nll.locate(ctrfile, inputfile) print("autoPyLoT: Iteration No. %d finished." % nlloccounter) # get updated NLLoc-location file nllocfile = max(glob.glob(locsearch), key=os.path.getctime) # check for bad picks badpicks = [] for key in picks: if picks[key]['P']['weight'] >= 4 or picks[key]['S']['weight'] >= 4: badpicks.append([key, picks[key]['P']['mpp']]) print("autoPyLoT: After iteration No. %d: %d bad onsets found ..." % (nlloccounter, \ len(badpicks))) if len(badpicks) == 0: print("autoPyLoT: No more bad onsets found, stop iterative picking!") nlloccounter = maxnumit evt = read_events(nllocfile)[0] if locflag < 2: # calculate seismic moment Mo and moment magnitude Mw moment_mag = MomentMagnitude(corr_dat, evt, parameter.get('vp'), parameter.get('Qp'), parameter.get('rho'), True, \ iplot) # update pick with moment property values (w0, fc, Mo) for station, props in moment_mag.moment_props.items(): picks[station]['P'].update(props) evt = moment_mag.updated_event() net_mw = moment_mag.net_magnitude() print("Network moment magnitude: %4.1f" % net_mw.mag) # calculate local (Richter) magntiude WAscaling = parameter.get('WAscaling') magscaling = parameter.get('magscaling') local_mag = LocalMagnitude(corr_dat, evt, parameter.get('sstop'), WAscaling, True, iplot) for station, amplitude in local_mag.amplitudes.items(): picks[station]['S']['Ao'] = amplitude.generic_amplitude print("Local station magnitudes scaled with:") print("log(Ao) + %f * log(r) + %f * r + %f" % (WAscaling[0], WAscaling[1], WAscaling[2])) evt = local_mag.updated_event(magscaling) net_ml = local_mag.net_magnitude(magscaling) print("Network local magnitude: %4.1f" % net_ml.mag) print("Network local magnitude scaled with:") print("%f * Ml + %f" % (magscaling[0], magscaling[1])) else: print("autoPyLoT: No NLLoc-location file available! Stop iteration!") locflag = 9 ########################################################## # write phase files for various location # and fault mechanism calculation routines # ObsPy event object if evt is not None: event_id = eventpath.split('/')[-1] evt.resource_id = ResourceIdentifier('smi:local/' + event_id) data.applyEVTData(evt, 'event') data.applyEVTData(picks) if savexml: if savepath == 'None' or savepath == None: savepath = eventpath fnqml = '%s/PyLoT_%s' % (savepath, evID) data.exportEvent(fnqml, fnext='.xml', fcheck=['auto', 'magnitude', 'origin']) if locflag == 1: # HYPO71 hypo71file = '%s/PyLoT_%s_HYPO71_phases' % (eventpath, evID) hypo71.export(picks, hypo71file, parameter) # HYPOSAT hyposatfile = '%s/PyLoT_%s_HYPOSAT_phases' % (eventpath, evID) hyposat.export(picks, hyposatfile, parameter) # VELEST velestfile = '%s/PyLoT_%s_VELEST_phases.cnv' % (eventpath, evID) velest.export(picks, velestfile, evt, parameter) # hypoDD hypoddfile = '%s/PyLoT_%s_hypoDD_phases.pha' % (eventpath, evID) hypodd.export(picks, hypoddfile, parameter, evt) # FOCMEC focmecfile = '%s/PyLoT_%s_FOCMEC.in' % (eventpath, evID) focmec.export(picks, focmecfile, parameter, evt) # HASH hashfile = '%s/PyLoT_%s_HASH' % (eventpath, evID) hash.export(picks, hashfile, parameter, evt) endsplash = '''------------------------------------------\n' -----Finished event %s!-----\n' ------------------------------------------'''.format \ (version=_getVersionString()) % evID print(endsplash) if locflag == 0: print("autoPyLoT was running in non-location mode!") endsp = '''####################################\n ************************************\n *********autoPyLoT terminates*******\n The Python picking and Location Tool\n ************************************'''.format(version=_getVersionString()) print(endsp) return picks if __name__ == "__main__": # parse arguments parser = argparse.ArgumentParser( description='''autoPyLoT automatically picks phase onset times using higher order statistics, autoregressive prediction and AIC followed by locating the seismic events using NLLoc''') parser.add_argument('-i', '-I', '--inputfile', type=str, action='store', help='''full path to the file containing the input parameters for autoPyLoT''') parser.add_argument('-p', '-P', '--iplot', type=int, action='store', help='''optional, logical variable for plotting: 0=none, 1=partial, 2=all''') parser.add_argument('-f', '-F', '--fnames', type=str, action='store', help='''optional, list of data file names''') parser.add_argument('-e', '--eventid', type=str, action='store', help='''optional, event path incl. event ID''') parser.add_argument('-s', '-S', '--spath', type=str, action='store', help='''optional, save path for autoPyLoT output''') parser.add_argument('-c', '-C', '--ncores', type=int, action='store', default=0, help='''optional, number of CPU cores used for parallel processing (default: all available(=0))''') cla = parser.parse_args() picks = autoPyLoT(inputfile=str(cla.inputfile), fnames=str(cla.fnames), eventid=str(cla.eventid), savepath=str(cla.spath), ncores=cla.ncores, iplot=str(cla.iplot))