"""
Module to simplify importing the OQ implementation of the
NGA-East GMPE suite.
"""
import os
import re
import sys
import logging
import copy
import pandas as pd
import numpy as np
from scipy.interpolate import RectBivariateSpline
from openquake.hazardlib.gsim.base import GMPE
from openquake.hazardlib import const
from openquake.hazardlib import imt as IMT
from openquake.hazardlib.gsim.usgs_ceus_2019 import NGAEastUSGSGMPE
from openquake.hazardlib.gsim.gmpe_table import _return_tables
from shakelib.multiutils import gmpe_gmas
from shakelib.conversions.imt.abrahamson_bhasin_2020 import AbrahamsonBhasin2020
# Max distance for evaluating NGAEast. This *should* be 1500, but due to what
# appears to be a floating point precision issue, we get a division by zero
# error at a value of 1500 returning nans. So we have to cap the distance at a
# value slightly smaller.
# This is also used to zero out the results at greater distances.
MAX_RRUP = 1499.9
[docs]class NGAEast(GMPE):
"""
Returns NGA East GMPE that combines all of the individual NGAEastUSGSGMPE
GMPEs.
"""
DEFINED_FOR_TECTONIC_REGION_TYPE = const.TRT.STABLE_CONTINENTAL
DEFINED_FOR_INTENSITY_MEASURE_COMPONENT = const.IMC.RotD50
DEFINED_FOR_INTENSITY_MEASURE_TYPES = set([IMT.PGA, IMT.PGV, IMT.SA])
DEFINED_FOR_STANDARD_DEVIATION_TYPES = set(
[const.StdDev.TOTAL, const.StdDev.INTER_EVENT, const.StdDev.INTRA_EVENT]
)
REQUIRES_SITES_PARAMETERS = set(("vs30",))
REQUIRES_RUPTURE_PARAMETERS = set(("mag",))
REQUIRES_DISTANCES = set(("rrup",))
TABLE_PATHS = os.listdir(NGAEastUSGSGMPE.PATH)
this_module = os.path.dirname(__file__)
NGA_BASE_PATH = os.path.join(
this_module, "..", "..", "shakemap", "data", "nga_east_tables"
)
NGA_EAST_USGS_WEIGHT = 0.667
NGA_EAST_SEED_WEIGHT = 0.333
NGA_EAST_USGS = pd.read_csv(
os.path.join(NGA_BASE_PATH, "nga-east-usgs-weights.dat")
)
NGA_EAST_SEEDS = pd.read_csv(
os.path.join(NGA_BASE_PATH, "nga-east-seed-weights.dat")
)
# Sigma models and their weights
SIGMA_MODS = ["EPRI", "PANEL"]
SIGMA_WEIGHTS = [0.8, 0.2]
# For small magnitude extrapolation
PATH = os.path.join(os.path.dirname(__file__), "nga_east_small_mag")
SMALL_M_SLOPE = np.loadtxt(os.path.join(PATH, "nga-east-smallM_slopes.txt"))
SMALL_M_SLOPE_PGA = np.loadtxt(os.path.join(PATH, "nga-east-smallM_slope_pga.txt"))
SMALL_M_SLOPE_PGV = np.loadtxt(os.path.join(PATH, "nga-east-smallM_slope_pgv.txt"))
SMALL_M_DIST = np.loadtxt(os.path.join(PATH, "nga-east-smallM_slope_distances.txt"))
SMALL_M_PER = np.loadtxt(os.path.join(PATH, "nga-east-smallM_slope_periods.txt"))
# -------------------------------------------------------------------------
# To simplify, use the COLLAPSED branch, but cannot get inter and intra
# event standard deviations in this case.
# SIGMA_MODS = ["COLLAPSED"]
# SIGMA_WEIGHTS = [1.0]
# Parse the periods of the columns
per_idx_start = 1
per_idx_end = -2
per_list_str = NGA_EAST_USGS.keys().tolist()[per_idx_start:per_idx_end]
per_array = np.array(
[float(p.replace("SA", "").replace("P", ".")) for p in per_list_str]
)
def __init__(self):
gmpes = []
sigma_wts = []
all_table_paths = []
for i, sigma_mod in enumerate(self.SIGMA_MODS):
for table_path in self.TABLE_PATHS:
gmpe = NGAEastUSGSGMPE(gmpe_table=table_path, sigma_model=sigma_mod)
gmpes.append(gmpe)
sigma_wts.append(self.SIGMA_WEIGHTS[i])
all_table_paths.append(table_path)
self.gmpes = gmpes
self.sigma_weights = np.array(sigma_wts)
self.ALL_TABLE_PATHS = all_table_paths
[docs] 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 np.any(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 = gmpe_gmas(gm, 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 = gmpe_gmas(gm, rup, imt, stddev_types)
else:
tmean, tstddevs = gmpe_gmas(gm, 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
