# -*- coding: utf-8 -*-
"""
Banimahd et al. (2026) Turkiye ANN-based Ground-Motion Model
=============================================================
This module implements the ANN-based, region-specific ground-motion model
for Turkiye by Banimahd et al. (2026) as an OpenQuake GSIM/GMPE.
Banimahd A, Karimzadeh S, et al. (2026).
Artificial neural network-based non-parametric ground motion models for
multiple intensity measures in Turkiye.
Engineering Applications of Artificial Intelligence.
Model overview
--------------
- Trained on strong-motion data from Turkiye.
- ML regressor: ensemble of 10 feed-forward neural networks.
- Inputs (in the order used for training and ONNX):
1. fd : focal depth (km)
2. fm : fault mechanism (1=Normal, 2=Reverse, 3=StrikeSlip)
3. mw : moment magnitude
4. rjb : Joyner-Boore distance (km)
5. vs30 : averaged shear-wave velocity in the top 30 m (m/s)
Outputs (25 values, in ln-space):
- PGA, PSA(T): ln(g)
- PGV : ln(cm/s)
"""
import os
import csv
import gzip
import numpy as np
from openquake.baselib.onnx import PicklableInferenceSession
from openquake.hazardlib.gsim.base import GMPE
from openquake.hazardlib import const
from openquake.hazardlib.imt import PGA, PGV, SA
# Paths
_DATA_DIR = os.path.join(os.path.dirname(__file__), "banimahd_turkiye_2026_data")
_ONNX_FILE = os.path.join(_DATA_DIR, "onnx_models", "GMM_Turkiye_2026.onnx.gz")
_STDS_FILE = os.path.join(_DATA_DIR, "stds.csv")
# GSIM class
[docs]class Banimahd2026Turkiye(GMPE):
"""
ANN-based Ground-Motion Model for Turkiye (Banimahd et al., 2026).
This GSIM wraps an ensemble of 10 feed-forward neural networks exported
to a single ONNX file. It returns the mean ln(IM) and standard deviations
(intra-event, inter-event, total) for each requested IMT.
NOTE: This implementation only supports PGA, PGV, and SA, and not the
other IMTs (IA, RSD575, RSD595, CAV) supported by the original GMM.
"""
DEFINED_FOR_TECTONIC_REGION_TYPE = const.TRT.ACTIVE_SHALLOW_CRUST
DEFINED_FOR_INTENSITY_MEASURE_TYPES = {PGA, PGV, SA}
DEFINED_FOR_INTENSITY_MEASURE_COMPONENT = const.IMC.GEOMETRIC_MEAN
DEFINED_FOR_STANDARD_DEVIATION_TYPES = {
const.StdDev.INTRA_EVENT,
const.StdDev.INTER_EVENT,
const.StdDev.TOTAL,
}
DEFINED_FOR_REFERENCE_VELOCITY = 760.0
REQUIRES_RUPTURE_PARAMETERS = {"mag", "hypo_depth", "rake"}
REQUIRES_DISTANCES = {"rjb"}
REQUIRES_SITES_PARAMETERS = {"vs30"}
_PERIODS = [0.03, 0.05, 0.075, 0.1, 0.15, 0.2, 0.25, 0.3, 0.4,
0.5, 0.75, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0]
def __init__(self):
self.sigma_intra = {}
self.tau_inter = {}
self.phi_total = {}
if not os.path.exists(_STDS_FILE):
raise IOError(f"Cannot find stds.csv at {_STDS_FILE}")
with open(_STDS_FILE, newline="") as f:
reader = csv.DictReader(f)
for row in reader:
key = row["ID"]
self.sigma_intra[key] = float(row["Sigma"])
self.tau_inter[key] = float(row["Tau"])
self.phi_total[key] = float(row["Phi"])
with gzip.open(_ONNX_FILE, "rb") as f:
model_bytes = f.read()
self.session = PicklableInferenceSession(model_bytes)
[docs] def compute(self, ctx: np.recarray, imts, mean, sig, tau, phi):
"""
Compute mean and standard deviations for all requested IMTs at once.
Parameters
----------
ctx : np.recarray
Context object with rupture, distance, and site parameters.
imts : list
List of intensity measure types.
mean : np.ndarray
Output array for mean values, shape (M, N).
sig : np.ndarray
Output array for total stddev, shape (M, N).
tau : np.ndarray
Output array for inter-event stddev, shape (M, N).
phi : np.ndarray
Output array for intra-event stddev, shape (M, N).
"""
# Number of sites
N = len(ctx)
mw = ctx.mag.astype(float)
rjb = ctx.rjb.astype(float)
vs30 = ctx.vs30.astype(float)
fd = ctx.hypo_depth.astype(float)
rake = ctx.rake.astype(float)
# Fault mechanism (vectorized)
# Normal: rake close to -90
# Reverse: rake close to +90
# Strike-slip: rake close to 0 or ±180
fm = np.ones(N, dtype=np.float32)
fm[(rake > -30.0) & (rake < 30.0)] = 3.0
fm[rake > 30.0] = 2.0
fm[(rake > 150.0) | (rake < -150.0)] = 3.0
# Build input matrix (fd, fm, mw, rjb, vs30)
X = np.column_stack([fd, fm, mw, rjb, vs30]).astype(np.float32)
# Run ONNX inference once for all 25 outputs
input_name = self.session.get_inputs()[0].name
out = self.session.run(None, {input_name: X})[0]
# Fill mean/sig/tau/phi for each requested IMT
for m, imt in enumerate(imts):
imt_str = imt.string
if imt_str == "PGA":
out_idx = 0
key = "ln(PGA)"
elif imt_str == "PGV":
out_idx = 1
key = "ln(PGV)"
elif imt_str.startswith("SA(") and imt_str.endswith(")"):
period = imt.period
# The ONNX model returns 25 outputs in this order:
# 0: PGA, 1: PGV, 2: Ia, 3: D5-75, 4: D5-95, 5: Tm, 6: CAV
# 7-24: SA at 18 periods (see _PERIODS)
out_idx = 7 + self._PERIODS.index(period)
key = f"ln(PSA={period})"
else:
raise ValueError(f"IMT {imt_str} not supported")
mean[m, :] = out[:, out_idx]
sig[m, :] = self.phi_total[key] # TOTAL
tau[m, :] = self.tau_inter[key] # INTER-EVENT
phi[m, :] = self.sigma_intra[key] # INTRA-EVENT
