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)
rctx = RuptureContext()
rctx.mag = 6.0
dctx = DistancesContext()
# Test values at the given distances and those outside range
dctx.rjb = np.array([0.5, 1.0, 10.0, 100.0, 500.0])
sctx = SitesContext()
sctx.vs30 = 1000. * np.ones(5)
stddevs = [const.StdDev.TOTAL]
expected_mean = np.array([2.0, 2.0, 1.0, 0.5, 1.0E-20])
expected_sigma = 0.25 * np.ones(5)
# PGA
mean, sigma = gsim.get_mean_and_stddevs(sctx, rctx, dctx,
imt_module.PGA(), stddevs)
np.testing.assert_array_almost_equal(np.exp(mean), expected_mean, 5)
np.testing.assert_array_almost_equal(sigma[0], expected_sigma, 5)
# SA
mean, sigma = gsim.get_mean_and_stddevs(sctx, rctx, dctx,
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)
# PGV
mean, sigma = gsim.get_mean_and_stddevs(sctx, rctx, dctx,
imt_module.PGV(), stddevs)
np.testing.assert_array_almost_equal(np.exp(mean), 10. * expected_mean,
5)
np.testing.assert_array_almost_equal(sigma[0], expected_sigma, 5)
def test_get_mean_and_stddevs(self):
"""
Tests mean and standard deviations without amplification
"""
gsim = GMPETable(gmpe_table=self.TABLE_FILE)
rctx = RuptureContext()
rctx.mag = 6.0
dctx = DistancesContext()
# Test values at the given distances and those outside range
dctx.rjb = np.array([0.5, 1.0, 10.0, 100.0, 500.0])
sctx = SitesContext()
stddevs = [const.StdDev.TOTAL]
expected_mean = np.array([2.0, 2.0, 1.0, 0.5, 1.0E-20])
# PGA
mean, sigma = gsim.get_mean_and_stddevs(sctx, rctx, dctx,
imt_module.PGA(), stddevs)
np.testing.assert_array_almost_equal(np.exp(mean), expected_mean, 5)
np.testing.assert_array_almost_equal(sigma[0], 0.5 * np.ones(5), 5)
# SA
mean, sigma = gsim.get_mean_and_stddevs(sctx, rctx, dctx,
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], 0.8 * np.ones(5), 5)
# PGV
mean, sigma = gsim.get_mean_and_stddevs(sctx, rctx, dctx,
imt_module.PGV(), stddevs)
np.testing.assert_array_almost_equal(np.exp(mean), 10. * expected_mean,
5)
np.testing.assert_array_almost_equal(sigma[0], 0.5 * np.ones(5), 5)
def test_get_amplification_factors(self):
"""
Tests the amplification tables
"""
ctx = RuptureContext()
ctx.rake = 45.0
ctx.mag = 6.0
# Takes distances at the values found in the table (not checking
# distance interpolation)
ctx.rjb = np.copy(self.amp_table.distances[:, 0, 0])
# Test Vs30 is 700.0 m/s midpoint between the 400 m/s and 1000 m/s
# specified in the table
stddevs = [const.StdDev.TOTAL]
expected_mean = np.ones_like(ctx.rjb)
# Check PGA and PGV
mean_amp, sigma_amp = self.amp_table.get_amplification_factors(
imt_module.PGA(), ctx, ctx.rjb, stddevs)
np.testing.assert_array_almost_equal(
mean_amp,
midpoint(1.0, 1.5) * expected_mean)
np.testing.assert_array_almost_equal(sigma_amp[0], 0.9 * expected_mean)
mean_amp, sigma_amp = self.amp_table.get_amplification_factors(
imt_module.PGV(), ctx, ctx.rjb, stddevs)
np.testing.assert_array_almost_equal(
mean_amp,
midpoint(1.0, 0.5) * expected_mean)
np.testing.assert_array_almost_equal(sigma_amp[0], 0.9 * expected_mean)
# Sa (0.5)
mean_amp, sigma_amp = self.amp_table.get_amplification_factors(
imt_module.SA(0.5), ctx, ctx.rjb, stddevs)
np.testing.assert_array_almost_equal(
mean_amp,
midpoint(1.0, 2.0) * expected_mean)
np.testing.assert_array_almost_equal(sigma_amp[0], 0.9 * expected_mean)
def test_retreival_tables_good_no_interp(self):
"""
Tests the retreival of the IML tables for 'good' conditions without
applying magnitude interpolations
"""
gsim = GMPETable(gmpe_table=self.TABLE_FILE)
# PGA
np.testing.assert_array_almost_equal(
gsim._return_tables(6.0, imt_module.PGA(), "IMLs"),
np.array([2., 1., 0.5]))
# PGV
np.testing.assert_array_almost_equal(
gsim._return_tables(6.0, imt_module.PGV(), "IMLs"),
np.array([20., 10., 5.]), 5)
# SA(1.0)
np.testing.assert_array_almost_equal(
gsim._return_tables(6.0, imt_module.SA(1.0), "IMLs"),
np.array([2.0, 1., 0.5]))
# Also for standard deviations
np.testing.assert_array_almost_equal(
gsim._return_tables(6.0, imt_module.PGA(), "Total"),
0.5 * np.ones(3))
np.testing.assert_array_almost_equal(
gsim._return_tables(6.0, imt_module.SA(1.0), "Total"),
0.8 * np.ones(3))
def test_retreival_tables_good_interp(self):
"""
Tests the retreival of the IML tables for 'good' conditions with
magnitude interpolations
"""
gsim = GMPETable(gmpe_table=self.TABLE_FILE)
expected_table_pgv = np.array([midpoint(20., 40.),
midpoint(10., 20.),
midpoint(5., 10.)])
np.testing.assert_array_almost_equal(
gsim._return_tables(6.5, imt_module.PGV(), "IMLs"),
expected_table_pgv,
5)
expected_table_sa1 = np.array([midpoint(2., 4.),
midpoint(1., 2.),
midpoint(0.5, 1.)])
np.testing.assert_array_almost_equal(
gsim._return_tables(6.5, imt_module.SA(1.0), "IMLs"),
expected_table_sa1)
from shakelib.gmpe.nga_east import NGAEast
home_dir = os.path.dirname(os.path.abspath(__file__))
data_dir = os.path.join(home_dir, 'nga_east_data')
stddev_types = [StdDev.TOTAL]
gmpe = NGAEast()
dx = base.DistancesContext()
dx.rrup = np.logspace(-1, np.log10(2000), 100)
rx = base.RuptureContext()
sx = base.SitesContext()
IMTS = [imt.PGA(), imt.PGV(), imt.SA(0.3), imt.SA(1.0), imt.SA(3.0)]
MAGS = [3, 5, 6, 7]
VS30 = [180, 380, 760, 2000]
def update_results():
# To build the data for testing
result = {}
for i in IMTS:
ikey = i.__str__()
result[ikey] = {}
for mag in MAGS:
rx.mag = mag
result[ikey][str(mag)] = {}
def _get_extent_from_multigmpe(rupture, config=None):
"""
Use MultiGMPE to determine extent
"""
(clon, clat) = _rupture_center(rupture)
origin = rupture.getOrigin()
if config is not None:
gmpe = MultiGMPE.from_config(config)
gmice = get_object_from_config('gmice', 'modeling', config)
if imt.SA in gmice.DEFINED_FOR_INTENSITY_MEASURE_TYPES:
default_imt = imt.SA(1.0)
elif imt.PGV in gmice.DEFINED_FOR_INTENSITY_MEASURE_TYPES:
default_imt = imt.PGV()
else:
default_imt = imt.PGA()
else:
# Put in some default values for conf
config = {
'extent': {
'mmi': {
'threshold': 4.5,
'mindist': 100,
'maxdist': 1000
}
}
}
# Generic GMPEs choices based only on active vs stable
# as defaults...
stable = is_stable(origin.lon, origin.lat)
if not stable:
ASK14 = AbrahamsonEtAl2014()
CB14 = CampbellBozorgnia2014()
CY14 = ChiouYoungs2014()
gmpes = [ASK14, CB14, CY14]
site_gmpes = None
weights = [1/3.0, 1/3.0, 1/3.0]
gmice = WGRW12()
else:
Fea96 = FrankelEtAl1996MwNSHMP2008()
Tea97 = ToroEtAl1997MwNSHMP2008()
Sea02 = SilvaEtAl2002MwNSHMP2008()
C03 = Campbell2003MwNSHMP2008()
TP05 = TavakoliPezeshk2005MwNSHMP2008()
AB06p = AtkinsonBoore2006Modified2011()
Pea11 = PezeshkEtAl2011()
Atk08p = Atkinson2008prime()
Sea01 = SomervilleEtAl2001NSHMP2008()
gmpes = [Fea96, Tea97, Sea02, C03,
TP05, AB06p, Pea11, Atk08p, Sea01]
site_gmpes = [AB06p]
weights = [0.16, 0.0, 0.0, 0.17, 0.17, 0.3, 0.2, 0.0, 0.0]
gmice = AK07()
gmpe = MultiGMPE.from_list(
gmpes, weights, default_gmpes_for_site=site_gmpes)
default_imt = imt.SA(1.0)
min_mmi = config['extent']['mmi']['threshold']
sd_types = [const.StdDev.TOTAL]
# Distance context
dx = DistancesContext()
# This imposes minimum/ maximum distances of:
# 80 and 800 km; could make this configurable
d_min = config['extent']['mmi']['mindist']
d_max = config['extent']['mmi']['maxdist']
dx.rjb = np.logspace(np.log10(d_min), np.log10(d_max), 2000)
# Details don't matter for this; assuming vertical surface rupturing fault
# with epicenter at the surface.
dx.rrup = dx.rjb
dx.rhypo = dx.rjb
dx.repi = dx.rjb
dx.rx = np.zeros_like(dx.rjb)
dx.ry0 = np.zeros_like(dx.rjb)
dx.rvolc = np.zeros_like(dx.rjb)
# Sites context
sx = SitesContext()
# Set to soft soil conditions
sx.vs30 = np.full_like(dx.rjb, 180)
sx = MultiGMPE.set_sites_depth_parameters(sx, gmpe)
sx.vs30measured = np.full_like(sx.vs30, False, dtype=bool)
sx = Sites._addDepthParameters(sx)
sx.backarc = np.full_like(sx.vs30, False, dtype=bool)
# Rupture context
rx = RuptureContext()
rx.mag = origin.mag
rx.rake = 0.0
# From WC94...
rx.width = 10**(-0.76 + 0.27*rx.mag)
rx.dip = 90.0
rx.ztor = origin.depth
rx.hypo_depth = origin.depth
gmpe_imt_mean, _ = gmpe.get_mean_and_stddevs(
sx, rx, dx, default_imt, sd_types)
# Convert to MMI
gmpe_to_mmi, _ = gmice.getMIfromGM(gmpe_imt_mean, default_imt)
# Minimum distance that exceeds threshold MMI?
dists_exceed_mmi = dx.rjb[gmpe_to_mmi > min_mmi]
if len(dists_exceed_mmi):
mindist_km = np.max(dists_exceed_mmi)
else:
mindist_km = d_min
# Get a projection
proj = OrthographicProjection(clon - 4, clon + 4, clat + 4, clat - 4)
if isinstance(rupture, (QuadRupture, EdgeRupture)):
ruptx, rupty = proj(
rupture.lons[~np.isnan(rupture.lons)],
rupture.lats[~np.isnan(rupture.lats)]
)
else:
ruptx, rupty = proj(clon, clat)
xmin = np.nanmin(ruptx) - mindist_km
ymin = np.nanmin(rupty) - mindist_km
xmax = np.nanmax(ruptx) + mindist_km
ymax = np.nanmax(rupty) + mindist_km
# Put a limit on range of aspect ratio
dx = xmax - xmin
dy = ymax - ymin
ar = dy / dx
if ar > 1.2:
# Inflate x
dx_target = dy / 1.2
ddx = dx_target - dx
xmax = xmax + ddx / 2
xmin = xmin - ddx / 2
if ar < 0.83:
# Inflate y
dy_target = dx * 0.83
ddy = dy_target - dy
ymax = ymax + ddy / 2
ymin = ymin - ddy / 2
lonmin, latmin = proj(np.array([xmin]), np.array([ymin]), reverse=True)
lonmax, latmax = proj(np.array([xmax]), np.array([ymax]), reverse=True)
#
# Round coordinates to the nearest minute -- that should make the
# output grid register with common grid resolutions (60c, 30c,
# 15c, 7.5c)
#
logging.debug("Extent: %f, %f, %f, %f" %
(lonmin, lonmax, latmin, latmax))
return _round_coord(lonmin[0]), _round_coord(lonmax[0]), \
_round_coord(latmin[0]), _round_coord(latmax[0])
def get_mean_and_stddevs(self, sites, rx, dists, imt, stddev_types):
# List of GMPE weights, which is the product of the the branch weights
# for the seed models vs the NGA East resampled models as well as the
# weights for the indivudual GMPES as defined by Petersen et al. (2019)
#
# Note that the NGA East resampled models are a function of spectral
# period.
#
# NGA East Seeds (1/3)
# ├── B_bca10d (0.06633), wts = 0.333 * 0.06633 = 0.02208789
# ├── B_ab95 (0.02211), wts = 0.333 * 0.02211 = 0.00736263
# ...
# NGA East Resampled or "USGS" (2/3)
# ├── Model 1 (0.1009 for PGA), wts = 0.667 * 0.1009 = 0.0673003
# ├── Model 2 (0.1606 for PGA), wts = 0.667 * 0.1606 = 0.1071202
# ...
#
wts = [0] * len(self.gmpes)
# Is IMT PGA or PGV?
is_pga = imt == IMT.PGA()
is_pgv = imt == IMT.PGV()
# Is magnitude less than 4? If so, we will need to set it to 4.0 and
# then extrapolate the tables at the end.
# But... in the brave new world of new OpenQuake, sites, rx, and
# dists are all exactly the same object, so when we copy rx to
# rup, and change it, as well as when we change the distances
# below, we need to do it all in rup, then pass rup as all three
# contexts when we call the gmpe.get_mean_and_stddevs()
rup = copy.deepcopy(rx)
if rup.mag < 4.0:
is_small_mag = True
delta_mag = rup.mag - 4.0
rup.mag = 4.0
else:
is_small_mag = False
for i, tp in enumerate(self.ALL_TABLE_PATHS):
if 'usgs' in tp:
# Get model number from i-th path using regex
mod_num = int(re.search(r'\d+', tp).group())
coefs = np.array(self.NGA_EAST_USGS.iloc[mod_num - 1])
# Is the IMT PGA, PGA, or SA?
if is_pga:
iweight = coefs[-2]
elif is_pgv:
iweight = coefs[-1]
else:
# For SA, need to interpolate; we'll use log-period and
# linear-weight interpolation.
iweight = np.interp(
np.log(imt.period), np.log(self.per_array),
coefs[self.per_idx_start:self.per_idx_end])
wts[i] = self.NGA_EAST_USGS_WEIGHT * iweight
else:
# Strip off the cruft to get the string we need to match
str_match = tp.replace('nga_east_', '').replace('.hdf5', '')
matched = self.NGA_EAST_SEEDS[self.NGA_EAST_SEEDS['model'] ==
str_match]
if len(matched):
iweight = self.NGA_EAST_SEEDS[self.NGA_EAST_SEEDS['model']
== str_match].iloc[0, 1]
wts[i] = self.NGA_EAST_SEED_WEIGHT * iweight
total_gmpe_weights = self.sigma_weights * wts
if not np.allclose(np.sum(total_gmpe_weights), 1.0):
raise ValueError('Weights must sum to 1.0.')
mean = np.full_like(sites.vs30, 0)
stddevs = []
for i in range(len(stddev_types)):
stddevs.append(np.full_like(sites.vs30, 0))
# Apply max distance to dists.rrup -->> now rup.rrup
np.clip(rup.rrup, 0, MAX_RRUP)
#
# Some models don't have PGV terms, so we will make PSA for them
# and then use the conditional conversion to get PGV
#
if is_pgv:
ab2020 = AbrahamsonBhasin2020(rup.mag)
vimt = IMT.SA(ab2020.getTref())
# Loop over gmpes
for i, gm in enumerate(self.gmpes):
if is_pgv:
# Is PGV and also not available for gm?
try:
_ = _return_tables(gm, rup.mag, imt, "IMLs")
except KeyError:
#
# No table for PGV, compute vimt, then convert to PGV
#
vmean, vstddevs = gm.get_mean_and_stddevs(
rup, rup, rup, vimt, stddev_types)
tmean, tstddevs = ab2020.getPGVandSTDDEVS(
vmean, vstddevs, stddev_types, rup.rrup, rup.vs30)
except Exception:
logging.error("Unexpected error:", sys.exc_info()[0])
else:
#
# Table exists for PGV, proceed normally
#
tmean, tstddevs = gm.get_mean_and_stddevs(
rup, rup, rup, imt, stddev_types)
else:
tmean, tstddevs = gm.get_mean_and_stddevs(
rup, rup, rup, imt, stddev_types)
mean += tmean * total_gmpe_weights[i]
for j, sd in enumerate(tstddevs):
stddevs[j] += sd * total_gmpe_weights[i]
# Zero out values at distances beyond the range for which NGA East
# was defined. -->> was dists.rrup, now rup.rrup
mean[rup.rrup > MAX_RRUP] = -999.0
# Do we need to extrapolate for small magnitude factor?
if is_small_mag:
if is_pga:
slopes = np.interp(np.log(rup.rrup), np.log(self.SMALL_M_DIST),
self.SMALL_M_SLOPE_PGA)
elif is_pgv:
slopes = np.interp(np.log(rup.rrup), np.log(self.SMALL_M_DIST),
self.SMALL_M_SLOPE_PGV)
else:
interp_obj = RectBivariateSpline(np.log(self.SMALL_M_DIST),
np.log(self.SMALL_M_PER),
self.SMALL_M_SLOPE,
kx=1,
ky=1)
slopes = interp_obj.ev(np.log(rup.rrup), np.log(imt.period))
mean = mean + slopes * delta_mag
return mean, stddevs
def get_mean_and_stddevs(self, sites, rup, dists, imt, stddev_types):
# List of GMPE weights, which is the product of the the branch weights
# for the seed models vs the NGA East resampled models as well as the
# weights for the indivudual GMPES as defined by Petersen et al. (2019)
#
# Note that the NGA East resampled models are a function of spectral
# period.
#
# NGA East Seeds (1/3)
# ├── B_bca10d (0.06633), wts = 0.333 * 0.06633 = 0.02208789
# ├── B_ab95 (0.02211), wts = 0.333 * 0.02211 = 0.00736263
# ...
# NGA East Resampled or "USGS" (2/3)
# ├── Model 1 (0.1009 for PGA), wts = 0.667 * 0.1009 = 0.0673003
# ├── Model 2 (0.1606 for PGA), wts = 0.667 * 0.1606 = 0.1071202
# ...
#
wts = [0] * len(self.gmpes)
# Is IMT PGA or PGV?
is_pga = imt == IMT.PGA()
is_pgv = imt == IMT.PGV()
# Is magnitude less than 4? If so, we will need to set it to 4.0 and
# then extrapolate the tables at the end.
if rup.mag < 4.0:
is_small_mag = True
delta_mag = rup.mag - 4.0
rup.mag = 4.0
else:
is_small_mag = False
for i, tp in enumerate(self.ALL_TABLE_PATHS):
if 'usgs' in tp:
# Get model number from i-th path using regex
mod_num = int(re.search(r'\d+', tp).group())
coefs = np.array(
self.NGA_EAST_USGS.iloc[mod_num - 1]
)
# Is the IMT PGA, PGA, or SA?
if is_pga:
iweight = coefs[-2]
elif is_pgv:
iweight = coefs[-1]
else:
# For SA, need to interpolate; we'll use log-period and
# linear-weight interpolation.
iweight = np.interp(
np.log(imt.period),
np.log(self.per_array),
coefs[self.per_idx_start:self.per_idx_end]
)
wts[i] = self.NGA_EAST_USGS_WEIGHT * iweight
else:
# Strip off the cruft to get the string we need to match
str_match = tp.replace('nga_east_', '').replace('.hdf5', '')
matched = self.NGA_EAST_SEEDS[
self.NGA_EAST_SEEDS['model'] == str_match]
if len(matched):
iweight = self.NGA_EAST_SEEDS[
self.NGA_EAST_SEEDS['model'] == str_match].iloc[0, 1]
wts[i] = self.NGA_EAST_SEED_WEIGHT * iweight
total_gmpe_weights = self.sigma_weights * wts
if not np.allclose(np.sum(total_gmpe_weights), 1.0):
raise ValueError('Weights must sum to 1.0.')
mean = np.full_like(sites.vs30, 0)
stddevs = []
for i in range(len(stddev_types)):
stddevs.append(np.full_like(sites.vs30, 0))
# Apply max distance to dists.rrup
np.clip(dists.rrup, 0, MAX_RRUP)
# Since we will be dropping the models that don't have PGV,
# we now also need to track the total sum of weights for when
# the imt is PGV so that we can re-distribute the weights.
if is_pgv:
twts = []
# Loop over gmpes
for i, gm in enumerate(self.gmpes):
if is_pgv:
# Is PGV and also not available for gm?
try:
gm._return_tables(rup.mag, imt, "IMLs")
except KeyError:
continue
except:
logging.error("Unexpected error:", sys.exc_info()[0])
tmean, tstddevs = gm.get_mean_and_stddevs(
sites, rup, dists, imt, stddev_types)
mean += tmean * total_gmpe_weights[i]
for j, sd in enumerate(tstddevs):
stddevs[j] += sd * total_gmpe_weights[i]
if is_pgv:
twts.append(total_gmpe_weights[i])
if is_pgv:
# Rescale the PGV wieghts so that they sum to 1 after dropping
# the models that are not defined for PGV.
mean = mean / np.sum(twts)
for j, sd in enumerate(stddevs):
stddevs[j] = stddevs[j] / np.sum(twts)
# Zero out values at distances beyond the range for which NGA East
# was defined.
mean[dists.rrup > MAX_RRUP] = -999.0
# Do we need to extrapolate for small magnitude factor?
if is_small_mag:
if is_pga:
slopes = np.interp(
np.log(dists.rrup),
np.log(self.SMALL_M_DIST),
self.SMALL_M_SLOPE_PGA)
elif is_pgv:
slopes = np.interp(
np.log(dists.rrup),
np.log(self.SMALL_M_DIST),
self.SMALL_M_SLOPE_PGV)
else:
interp_obj = RectBivariateSpline(
np.log(self.SMALL_M_DIST),
np.log(self.SMALL_M_PER),
self.SMALL_M_SLOPE, kx=1, ky=1)
slopes = interp_obj.ev(
np.log(dists.rrup),
np.log(imt.period)
)
mean = mean + slopes * delta_mag
return mean, stddevs
# Print output
print(out)
print(err)
log = str(out) + str(err)
## Read in the output file:
vs30_data = np.genfromtxt('tmp2', usecols=2)
pd.DataFrame(vs30_data).to_csv(
'/home/eking/Documents/internship/data/Kappa/vs30.csv')
kappa_data = fullfile[' tstar(s) ']
############# GMPEs ###################
imt_pga = imt.PGA()
imt_pgv = imt.PGV()
imt_arias = imt.IA()
uncertaintytype = const.StdDev.TOTAL
## Set GMPEs:
zhao2006 = ZhaoEtAl2006SInter()
travasarou = TravasarouEtAl2003()
bssa14 = BooreEtAl2014()
## Set the empty arrays:
median_zhao2006 = np.array([])
median_travasarou = np.array([])
median_bssa14 = np.array([])
sd_zhao2006 = np.array([])
sd_travasarou = np.array([])
