def test_zero_distance(self):
# test the calculation in case of zero rrup distance (for rrup=0
# the slab correction term has a singularity). In this case the
# method should return values equal to the ones obtained by
# replacing 0 values with 0.1
ctx = RuptureContext()
ctx.sids = [0, 1]
ctx.vs30 = numpy.array([800.0, 800.0])
ctx.mag = 5.0
ctx.rake = 0.0
ctx.hypo_depth = 0.0
ctx.rrup = numpy.array([0.0, 0.2])
ctx.occurrence_rate = .0001
mean_0, stds_0 = self.GSIM_CLASS().get_mean_and_stddevs(
ctx, ctx, ctx, PGA(), [StdDev.TOTAL])
ctx.rrup = numpy.array([0.1, 0.2])
mean_01, stds_01 = self.GSIM_CLASS().get_mean_and_stddevs(
ctx, ctx, ctx, PGA(), [StdDev.TOTAL])
numpy.testing.assert_array_equal(mean_0, mean_01)
numpy.testing.assert_array_equal(stds_0, stds_01)
def test_get_mean_and_stddevs_good(self):
"""
Tests the full execution of the GMPE tables for valid data
"""
gsim = GMPETable(gmpe_table=self.TABLE_FILE)
ctx = RuptureContext()
ctx.mag = 6.0
ctx.rake = 90.0
# Test values at the given distances and those outside range
ctx.rjb = np.array([0.5, 1.0, 10.0, 100.0, 500.0])
ctx.sids = np.arange(5)
stddevs = [const.StdDev.TOTAL]
expected_mean = np.array([20.0, 20.0, 10.0, 5.0, 1.0E-19])
# PGA
mean, sigma = gsim.get_mean_and_stddevs(ctx, ctx, ctx,
imt_module.PGA(), stddevs)
np.testing.assert_array_almost_equal(np.exp(mean), expected_mean, 5)
np.testing.assert_array_almost_equal(sigma[0], 0.25 * np.ones(5), 5)
# SA
mean, sigma = gsim.get_mean_and_stddevs(ctx, ctx, ctx,
imt_module.SA(1.0), stddevs)
np.testing.assert_array_almost_equal(np.exp(mean), expected_mean, 5)
np.testing.assert_array_almost_equal(sigma[0], 0.4 * np.ones(5), 5)
def trim_multiple_events(
st,
origin,
catalog,
travel_time_df,
pga_factor,
pct_window_reject,
gmpe,
site_parameters,
rupture_parameters,
):
"""
Uses a catalog (list of ScalarEvents) to handle cases where a trace might
contain signals from multiple events. The catalog should contain events
down to a low enough magnitude in relation to the events of interest.
Overall, the algorithm is as follows:
1) For each earthquake in the catalog, get the P-wave travel time
and estimated PGA at this station.
2) Compute the PGA (of the as-recorded horizontal channels).
3) Select the P-wave arrival times across all events for this record
that are (a) within the signal window, and (b) the predicted PGA is
greater than pga_factor times the PGA from step #1.
4) If any P-wave arrival times match the above criteria, then if any of
the arrival times fall within in the first pct_window_reject*100%
of the signal window, then reject the record. Otherwise, trim the
record such that the end time does not include any of the arrivals
selected in step #3.
Args:
st (StationStream):
Stream of data.
origin (ScalarEvent):
ScalarEvent object associated with the StationStream.
catalog (list):
List of ScalarEvent objects.
travel_time_df (DataFrame):
A pandas DataFrame that contains the travel time information
(obtained from
gmprocess.waveform_processing.phase.create_travel_time_dataframe).
The columns in the DataFrame are the station ids and the indices
are the earthquake ids.
pga_factor (float):
A decimal factor used to determine whether the predicted PGA
from an event arrival is significant enough that it should be
considered for removal.
pct_window_reject (float):
A decimal from 0.0 to 1.0 used to determine if an arrival should
be trimmed from the record, or if the entire record should be
rejected. If the arrival falls within the first
pct_window_reject * 100% of the signal window, then the entire
record will be rejected. Otherwise, the record will be trimmed
appropriately.
gmpe (str):
Short name of the GMPE to use. Must be defined in the modules file.
site_parameters (dict):
Dictionary of site parameters to input to the GMPE.
rupture_parameters:
Dictionary of rupture parameters to input to the GMPE.
Returns:
StationStream: Processed stream.
"""
if not st.passed:
return st
# Check that we know the signal split for each trace in the stream
for tr in st:
if not tr.hasParameter("signal_split"):
return st
signal_window_starttime = st[0].getParameter("signal_split")["split_time"]
arrivals = travel_time_df[st[0].stats.network + "." + st[0].stats.station]
arrivals = arrivals.sort_values()
# Filter by any arrival times that appear in the signal window
arrivals = arrivals[(arrivals > signal_window_starttime)
& (arrivals < st[0].stats.endtime)]
# Make sure we remove the arrival that corresponds to the event of interest
if origin.id in arrivals.index:
arrivals.drop(index=origin.id, inplace=True)
if arrivals.empty:
return st
# Calculate the recorded PGA for this record
stasum = StationSummary.from_stream(st, ["ROTD(50.0)"], ["PGA"])
recorded_pga = stasum.get_pgm("PGA", "ROTD(50.0)")
# Load the GMPE model
gmpe = load_model(gmpe)
# Generic context
rctx = RuptureContext()
# Make sure that site parameter values are converted to numpy arrays
site_parameters_copy = site_parameters.copy()
for k, v in site_parameters_copy.items():
site_parameters_copy[k] = np.array([site_parameters_copy[k]])
rctx.__dict__.update(site_parameters_copy)
# Filter by arrivals that have significant expected PGA using GMPE
is_significant = []
for eqid, arrival_time in arrivals.items():
event = next(event for event in catalog if event.id == eqid)
# Set rupture parameters
rctx.__dict__.update(rupture_parameters)
rctx.mag = event.magnitude
# TODO: distances should be calculated when we refactor to be
# able to import distance calculations
rctx.repi = np.array([
gps2dist_azimuth(
st[0].stats.coordinates.latitude,
st[0].stats.coordinates.longitude,
event.latitude,
event.longitude,
)[0] / 1000
])
rctx.rjb = rctx.repi
rctx.rhypo = np.sqrt(rctx.repi**2 + event.depth_km**2)
rctx.rrup = rctx.rhypo
rctx.sids = np.array(range(np.size(rctx.rrup)))
pga, sd = gmpe.get_mean_and_stddevs(rctx, rctx, rctx, imt.PGA(), [])
# Convert from ln(g) to %g
predicted_pga = 100 * np.exp(pga[0])
if predicted_pga > (pga_factor * recorded_pga):
is_significant.append(True)
else:
is_significant.append(False)
significant_arrivals = arrivals[is_significant]
if significant_arrivals.empty:
return st
# Check if any of the significant arrivals occur within the
signal_length = st[0].stats.endtime - signal_window_starttime
cutoff_time = signal_window_starttime + pct_window_reject * (signal_length)
if (significant_arrivals < cutoff_time).any():
for tr in st:
tr.fail("A significant arrival from another event occurs within "
"the first %s percent of the signal window" %
(100 * pct_window_reject))
# Otherwise, trim the stream at the first significant arrival
else:
for tr in st:
signal_end = tr.getParameter("signal_end")
signal_end["end_time"] = significant_arrivals[0]
signal_end["method"] = "Trimming before right another event"
tr.setParameter("signal_end", signal_end)
cut(st)
return st
def signal_end(
st,
event_time,
event_lon,
event_lat,
event_mag,
method=None,
vmin=None,
floor=None,
model=None,
epsilon=2.0,
):
"""
Estimate end of signal by using a model of the 5-95% significant
duration, and adding this value to the "signal_split" time. This probably
only works well when the split is estimated with a p-wave picker since
the velocity method often ends up with split times that are well before
signal actually starts.
Args:
st (StationStream):
Stream of data.
event_time (UTCDateTime):
Event origin time.
event_mag (float):
Event magnitude.
event_lon (float):
Event longitude.
event_lat (float):
Event latitude.
method (str):
Method for estimating signal end time. Either 'velocity'
or 'model'.
vmin (float):
Velocity (km/s) for estimating end of signal. Only used if
method="velocity".
floor (float):
Minimum duration (sec) applied along with vmin.
model (str):
Short name of duration model to use. Must be defined in the
gmprocess/data/modules.yml file.
epsilon (float):
Number of standard deviations; if epsilon is 1.0, then the signal
window duration is the mean Ds + 1 standard deviation. Only used
for method="model".
Returns:
trace with stats dict updated to include a
stats['processing_parameters']['signal_end'] dictionary.
"""
# Load openquake stuff if method="model"
if method == "model":
dmodel = load_model(model)
# Set some "conservative" inputs (in that they will tend to give
# larger durations).
rctx = RuptureContext()
rctx.mag = event_mag
rctx.rake = -90.0
rctx.vs30 = np.array([180.0])
rctx.z1pt0 = np.array([0.51])
dur_imt = imt.from_string("RSD595")
stddev_types = [const.StdDev.TOTAL]
for tr in st:
if not tr.hasParameter("signal_split"):
continue
if method == "velocity":
if vmin is None:
raise ValueError('Must specify vmin if method is "velocity".')
if floor is None:
raise ValueError('Must specify floor if method is "velocity".')
epi_dist = (gps2dist_azimuth(
lat1=event_lat,
lon1=event_lon,
lat2=tr.stats["coordinates"]["latitude"],
lon2=tr.stats["coordinates"]["longitude"],
)[0] / 1000.0)
end_time = event_time + max(floor, epi_dist / vmin)
elif method == "model":
if model is None:
raise ValueError('Must specify model if method is "model".')
epi_dist = (gps2dist_azimuth(
lat1=event_lat,
lon1=event_lon,
lat2=tr.stats["coordinates"]["latitude"],
lon2=tr.stats["coordinates"]["longitude"],
)[0] / 1000.0)
# Repi >= Rrup, so substitution here should be conservative
# (leading to larger durations).
rctx.rrup = np.array([epi_dist])
rctx.sids = np.array(range(np.size(rctx.rrup)))
lnmu, lnstd = dmodel.get_mean_and_stddevs(rctx, rctx, rctx,
dur_imt, stddev_types)
duration = np.exp(lnmu + epsilon * lnstd[0])
# Get split time
split_time = tr.getParameter("signal_split")["split_time"]
end_time = split_time + float(duration)
else:
raise ValueError('method must be either "velocity" or "model".')
# Update trace params
end_params = {
"end_time": end_time,
"method": method,
"vsplit": vmin,
"floor": floor,
"model": model,
"epsilon": epsilon,
}
tr.setParameter("signal_end", end_params)
return st