# -*- coding: utf-8 -*-
# vim: tabstop=4 shiftwidth=4 softtabstop=4
#
# Copyright (C) 2013-2020 GEM Foundation
#
# OpenQuake is free software: you can redistribute it and/or modify it
# under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# OpenQuake is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with OpenQuake. If not, see <http://www.gnu.org/licenses/>.
"""
Module :mod:`openquake.hazardlib.geo.surface.multi` defines
:class:`MultiSurface`.
"""
import numpy
from copy import deepcopy
from scipy.spatial.distance import pdist, squareform
from openquake.hazardlib.geo.surface.base import BaseSurface, downsample_trace
from openquake.hazardlib.geo.mesh import Mesh
from openquake.hazardlib.geo import utils
from openquake.hazardlib.geo.surface import (
PlanarSurface, SimpleFaultSurface, ComplexFaultSurface)
from openquake.hazardlib.geo.surface.gridded import GriddedSurface
[docs]class MultiSurface(BaseSurface):
"""
Represent a surface as a collection of independent surface elements.
:param surfaces:
List of instances of subclasses of
:class:`~openquake.hazardlib.geo.surface.base.BaseSurface`
each representing a surface geometry element.
:param edge_set:
Retains list of upper edges from all of the surfaces, with each
edge given as a numpy array of [longitude, latitude, depth]
:param cartesian_edges:
For GC2, this holds the list of edge sets in an orthographic projection
such that the coordinates are all cartesian.
:param cartesian_endpoints:
For GC2, this hold the list of end-points of the edges in an
orthographic projection
:param proj:
For GC2, instance of :class:
`~openquake.hazardlib.geo.utils.OrthographicProjection` instantiated
with the bounding box limits of the fault
:param length_set:
List of lengths of upper edges of each surface
:param cum_length_set:
List of cumulative lengths of edges along fault
:param gc2_config:
For GC2, dictionary holding fault specific parameters for GC2
configuration
:param p0:
For GC2, reference origin point of the fault
:param gc2t:
GC2 T-coordinate
:param gc2u:
GC2 U-coordinate
:param tmp_mesh:
If fed with the same mesh twice (e.g. calling get_rx_distance and
then get_ry0_distance in sequence) does not repeat GC2 calculations,
this hold the last mesh it was fed with
:param gc_length:
For GC2, determines the length of the fault (km) in its own GC2
configuration
"""
@property
def surface_nodes(self):
"""
:returns:
a list of surface nodes from the underlying single node surfaces
"""
return [surf.surface_nodes[0] for surf in self.surfaces]
@property
def mesh(self):
"""
:returns: mesh corresponding to the whole multi surface
"""
meshes = [surface.mesh for surface in self.surfaces]
lons = numpy.concatenate([m.lons for m in meshes])
lats = numpy.concatenate([m.lats for m in meshes])
depths = numpy.concatenate([m.depths for m in meshes])
return Mesh(lons, lats, depths)
def __init__(self, surfaces, tol=0.1):
"""
Instantiate object with list of surfaces
:param float tol:
If surfaces contains an instance of :class:
`~openquake.hazardlib.geo.surface.simple_fault.SimpleFaultSurface`
or of :class:
`~openquake.hazardlib.geo.surface.complex_fault.ComplexFaultSurface`
then the surface is downsampled so that only the points
representing changes in strike are retained. This parameter sets
the tolerance (in degrees) for defining a change in strike
direction
"""
self.surfaces = surfaces
self.areas = None
self.edge_set = self._get_edge_set(tol)
self.cartesian_edges = []
self.cartesian_endpoints = []
self.proj = None
self.length_set = []
self.cum_length_set = []
self.gc2_config = None
self.p0 = None
self.gc2t = None
self.gc2u = None
self.tmp_mesh = None
self.gc_length = None
def _get_edge_set(self, tol=0.1):
"""
Retrieve set of top edges from all of the individual surfaces,
downsampling the upper edge based on the specified tolerance
"""
edges = []
for surface in self.surfaces:
if isinstance(surface, GriddedSurface):
return edges.append(surface.mesh)
elif isinstance(surface, PlanarSurface):
# Top edge determined from two end points
edge = []
for pnt in [surface.top_left, surface.top_right]:
edge.append([pnt.longitude, pnt.latitude, pnt.depth])
edges.append(numpy.array(edge))
elif isinstance(surface,
(ComplexFaultSurface, SimpleFaultSurface)):
# Rectangular meshes are downsampled to reduce their
# overall size
edges.append(downsample_trace(surface.mesh, tol))
else:
raise ValueError("Surface %s not recognised" % str(surface))
return edges
[docs] def get_min_distance(self, mesh):
"""
For each point in ``mesh`` compute the minimum distance to each
surface element and return the smallest value.
See :meth:`superclass method
<.base.BaseSurface.get_min_distance>`
for spec of input and result values.
"""
dists = [surf.get_min_distance(mesh) for surf in self.surfaces]
return numpy.min(dists, axis=0)
[docs] def get_closest_points(self, mesh):
"""
For each point in ``mesh`` find the closest surface element, and return
the corresponding closest point.
See :meth:`superclass method
<.base.BaseSurface.get_closest_points>`
for spec of input and result values.
"""
# first, for each point in mesh compute minimum distance to each
# surface. The distance matrix is flattend, because mesh can be of
# an arbitrary shape. By flattening we obtain a ``distances`` matrix
# for which the first dimension represents the different surfaces
# and the second dimension the mesh points.
dists = numpy.array(
[surf.get_min_distance(mesh).flatten() for surf in self.surfaces]
)
# find for each point in mesh the index of closest surface
idx = dists == numpy.min(dists, axis=0)
# loop again over surfaces. For each surface compute the closest
# points, and associate them to the mesh points for which the surface
# is the closest. Note that if a surface is not the closest to any of
# the mesh points then the calculation is skipped
lons = numpy.empty_like(mesh.lons.flatten())
lats = numpy.empty_like(mesh.lats.flatten())
depths = None if mesh.depths is None else \
numpy.empty_like(mesh.depths.flatten())
for i, surf in enumerate(self.surfaces):
if not idx[i, :].any():
continue
cps = surf.get_closest_points(mesh)
lons[idx[i, :]] = cps.lons.flatten()[idx[i, :]]
lats[idx[i, :]] = cps.lats.flatten()[idx[i, :]]
if depths is not None:
depths[idx[i, :]] = cps.depths.flatten()[idx[i, :]]
lons = lons.reshape(mesh.lons.shape)
lats = lats.reshape(mesh.lats.shape)
if depths is not None:
depths = depths.reshape(mesh.depths.shape)
return Mesh(lons, lats, depths)
[docs] def get_joyner_boore_distance(self, mesh):
"""
For each point in mesh compute the Joyner-Boore distance to all the
surface elements and return the smallest value.
See :meth:`superclass method
<.base.BaseSurface.get_joyner_boore_distance>`
for spec of input and result values.
"""
# for each point in mesh compute the Joyner-Boore distance to all the
# surfaces and return the shortest one.
dists = [
surf.get_joyner_boore_distance(mesh) for surf in self.surfaces]
return numpy.min(dists, axis=0)
[docs] def get_top_edge_depth(self):
"""
Compute top edge depth of each surface element and return area-weighted
average value (in km).
"""
areas = self._get_areas()
depths = numpy.array(
[surf.get_top_edge_depth() for surf in self.surfaces])
return numpy.sum(areas * depths) / numpy.sum(areas)
[docs] def get_strike(self):
"""
Compute strike of each surface element and return area-weighted average
value (in range ``[0, 360]``) using formula from:
http://en.wikipedia.org/wiki/Mean_of_circular_quantities
Note that the original formula has been adapted to compute a weighted
rather than arithmetic mean.
"""
areas = self._get_areas()
strikes = numpy.array([surf.get_strike() for surf in self.surfaces])
v1 = (numpy.sum(areas * numpy.sin(numpy.radians(strikes))) /
numpy.sum(areas))
v2 = (numpy.sum(areas * numpy.cos(numpy.radians(strikes))) /
numpy.sum(areas))
return numpy.degrees(numpy.arctan2(v1, v2)) % 360
[docs] def get_dip(self):
"""
Compute dip of each surface element and return area-weighted average
value (in range ``(0, 90]``).
Given that dip values are constrained in the range (0, 90], the simple
formula for weighted mean is used.
"""
areas = self._get_areas()
dips = numpy.array([surf.get_dip() for surf in self.surfaces])
return numpy.sum(areas * dips) / numpy.sum(areas)
[docs] def get_width(self):
"""
Compute width of each surface element, and return area-weighted
average value (in km).
"""
areas = self._get_areas()
widths = numpy.array([surf.get_width() for surf in self.surfaces])
return numpy.sum(areas * widths) / numpy.sum(areas)
[docs] def get_area(self):
"""
Return sum of surface elements areas (in squared km).
"""
return numpy.sum(self._get_areas())
[docs] def get_bounding_box(self):
"""
Compute bounding box for each surface element, and then return
the bounding box of all surface elements' bounding boxes.
:return:
A tuple of four items. These items represent western, eastern,
northern and southern borders of the bounding box respectively.
Values are floats in decimal degrees.
"""
lons = []
lats = []
for surf in self.surfaces:
west, east, north, south = surf.get_bounding_box()
lons.extend([west, east])
lats.extend([north, south])
return utils.get_spherical_bounding_box(lons, lats)
[docs] def get_middle_point(self):
"""
If :class:`MultiSurface` is defined by a single surface, simply
returns surface's middle point, otherwise find surface element closest
to the surface's bounding box centroid and return corresponding
middle point.
Note that the concept of middle point for a multi surface is ambiguous
and alternative definitions may be possible. However, this method is
mostly used to define the hypocenter location for ruptures described
by a multi surface
(see :meth:`openquake.hazardlib.source.characteristic.CharacteristicFaultSource.iter_ruptures`).
This is needed because when creating fault based sources, the rupture's
hypocenter locations are not explicitly defined, and therefore an
automated way to define them is required.
"""
if len(self.surfaces) == 1:
return self.surfaces[0].get_middle_point()
west, east, north, south = self.get_bounding_box()
longitude, latitude = utils.get_middle_point(west, north, east, south)
dists = []
for surf in self.surfaces:
dists.append(
surf.get_min_distance(Mesh(numpy.array([longitude]),
numpy.array([latitude]),
None)))
dists = numpy.array(dists).flatten()
idx = dists == numpy.min(dists)
return numpy.array(self.surfaces)[idx][0].get_middle_point()
[docs] def get_surface_boundaries(self):
lons = []
lats = []
for surf in self.surfaces:
lons_surf, lats_surf = surf.get_surface_boundaries()
lons.extend(lons_surf)
lats.extend(lats_surf)
return lons, lats
[docs] def get_surface_boundaries_3d(self):
lons = []
lats = []
deps = []
for surf in self.surfaces:
lons_surf, lats_surf, deps_surf = surf.get_surface_boundaries_3d()
lons.extend(lons_surf)
lats.extend(lats_surf)
deps.extend(deps_surf)
return lons, lats, deps
def _get_areas(self):
"""
Return surface elements area values in a numpy array.
"""
if self.areas is None:
self.areas = []
for surf in self.surfaces:
self.areas.append(surf.get_area())
self.areas = numpy.array(self.areas)
return self.areas
def _get_cartesian_edge_set(self):
"""
For the GC2 calculations a set of cartesian representations of the
fault edges are needed. In this present case we use a common cartesian
framework for all edges, as opposed to defining a separate orthographic
projection per edge
"""
# Get projection space for cartesian projection
edge_sets = numpy.vstack(self.edge_set)
west, east, north, south = utils.get_spherical_bounding_box(
edge_sets[:, 0],
edge_sets[:, 1])
self.proj = utils.OrthographicProjection(west, east, north, south)
for edges in self.edge_set:
# Project edges into cartesian space
px, py = self.proj(edges[:, 0], edges[:, 1])
# Store the two end-points of the trace
self.cartesian_endpoints.append(
numpy.array([[px[0], py[0], edges[0, 2]],
[px[-1], py[-1], edges[-1, 2]]]))
self.cartesian_edges.append(numpy.column_stack([px, py,
edges[:, 2]]))
# Get surface length vector for the trace - easier in cartesian
lengths = numpy.sqrt((px[:-1] - px[1:]) ** 2. +
(py[:-1] - py[1:]) ** 2.)
self.length_set.append(lengths)
# Get cumulative surface length vector
self.cum_length_set.append(
numpy.hstack([0., numpy.cumsum(lengths)]))
return edge_sets
def _setup_gc2_framework(self):
"""
This method establishes the GC2 framework for a multi-segment
(and indeed multi-typology) case based on the description in
Spudich & Chiou (2015) - see section on Generalized Coordinate
System for Multiple Rupture Traces
"""
# Generate cartesian edge set
edge_sets = self._get_cartesian_edge_set()
self.gc2_config = {}
# Determine furthest two points apart
endpoint_set = numpy.vstack([cep for cep in self.cartesian_endpoints])
dmat = squareform(pdist(endpoint_set))
irow, icol = numpy.unravel_index(numpy.argmax(dmat), dmat.shape)
# Join further points to form a vector (a_hat in Spudich & Chiou)
# According to Spudich & Chiou, a_vec should be eastward trending
if endpoint_set[irow, 0] > endpoint_set[icol, 0]:
# Row point is to the east of column point
beginning = endpoint_set[icol, :2]
ending = endpoint_set[irow, :2]
else:
# Column point is to the east of row point
beginning = endpoint_set[irow, :2]
ending = endpoint_set[icol, :2]
# Convert to unit vector
a_vec = ending - beginning
self.gc2_config["a_hat"] = a_vec / numpy.linalg.norm(a_vec)
# Get e_j set
self.gc2_config["ejs"] = []
for c_edges in self.cartesian_edges:
self.gc2_config["ejs"].append(
numpy.dot(c_edges[-1, :2] - c_edges[0, :2],
self.gc2_config["a_hat"]))
# A "total E" is defined as the sum of the e_j values
self.gc2_config["e_tot"] = sum(self.gc2_config["ejs"])
sign_etot = numpy.sign(self.gc2_config["e_tot"])
b_vec = numpy.zeros(2)
self.gc2_config["sign"] = []
for i, c_edges in enumerate(self.cartesian_edges):
segment_sign = numpy.sign(self.gc2_config["ejs"][i]) * sign_etot
self.gc2_config["sign"].append(segment_sign)
if segment_sign < 0:
# Segment is discordant - reverse the points
c_edges = numpy.flipud(c_edges)
self.cartesian_edges[i] = c_edges
self.cartesian_endpoints[i] = numpy.flipud(
self.cartesian_endpoints[i])
b_vec += (c_edges[-1, :2] - c_edges[0, :2])
# Get unit vector
self.gc2_config["b_hat"] = b_vec / numpy.linalg.norm(b_vec)
if numpy.dot(a_vec, self.gc2_config["b_hat"]) >= 0.0:
self.p0 = beginning
else:
self.p0 = ending
# To later calculate Ry0 it is necessary to determine the maximum
# GC2-U coordinate for the fault
self._get_gc2_coordinates_for_rupture(edge_sets)
def _get_gc2_coordinates_for_rupture(self, edge_sets):
"""
Calculates the GC2 coordinates for the nodes of the upper edge of the
fault
"""
# Establish GC2 length - for use with Ry0
rup_gc2t, rup_gc2u = self.get_generalised_coordinates(
edge_sets[:, 0], edge_sets[:, 1])
# GC2 length should be the largest positive GC2 value of the edges
self.gc_length = numpy.max(rup_gc2u)
def _get_ut_i(self, seg, sx, sy):
"""
Returns the U and T coordinate for a specific trace segment
:param seg:
End points of the segment edge
:param sx:
Sites longitudes rendered into coordinate system
:param sy:
Sites latitudes rendered into coordinate system
"""
p0x, p0y, p1x, p1y = seg[0, 0], seg[0, 1], seg[1, 0], seg[1, 1]
# Unit vector normal to strike
t_i_vec = [p1y - p0y, -(p1x - p0x), 0.0]
t_i_hat = t_i_vec / numpy.linalg.norm(t_i_vec)
# Unit vector along strike
u_i_vec = [p1x - p0x, p1y - p0y, 0.0]
u_i_hat = u_i_vec / numpy.linalg.norm(u_i_vec)
# Vectors from P0 to sites
rsite = numpy.column_stack([sx - p0x, sy - p0y])
return numpy.sum(u_i_hat[:-1] * rsite, axis=1),\
numpy.sum(t_i_hat[:-1] * rsite, axis=1)
[docs] def get_generalised_coordinates(self, lons, lats):
"""
Transforms the site positions into the generalised coordinate form
described by Spudich and Chiou (2015) for the multi-rupture and/or
discordant case
Spudich, Paul and Chiou, Brian (2015) Strike-parallel and strike-normal
coordinate system around geometrically complicated rupture traces —
Use by NGA-West2 and further improvements: U.S. Geological Survey
Open-File Report 2015-1028
"""
# If the GC2 configuration has not been setup already - do it!
if not self.gc2_config:
self._setup_gc2_framework()
# Initially the weights are set to zero
sx, sy = self.proj(lons, lats)
sum_w_i = numpy.zeros_like(lons)
sum_w_i_t_i = numpy.zeros_like(lons)
sum_wi_ui_si = numpy.zeros_like(lons)
# Find the cumulative length of the fault up until the given segment
# Essentially calculating s_i
general_t = numpy.zeros_like(lons)
general_u = numpy.zeros_like(lons)
on_segment = numpy.zeros_like(lons, dtype=bool)
# Loop over the traces
for j, edges in enumerate(self.cartesian_edges):
# Loop over segments in trace
# s_ij_total = 0.0
for i in range(edges.shape[0] - 1):
# Get u_i and t_i
u_i, t_i = self._get_ut_i(edges[i:(i + 2), :], sx, sy)
# If t_i is 0 and u_i is within the section length then site is
# directly on the edge - therefore general_t is 0
w_i = numpy.zeros_like(lons)
ti0_check = numpy.fabs(t_i) < 1.0E-3 # < 1 m precision
on_segment_range = numpy.logical_and(
u_i >= 0.0,
u_i <= self.length_set[j][i])
# Deal with the case in which t_i is 0 and the site is inside
# of the segment
idx0 = numpy.logical_and(ti0_check, on_segment_range)
# In this null case w_i is ignored - however, null sites on
# previous segments would not be null sites on this segment,
# so we update the list of null sites
on_segment[numpy.logical_or(on_segment, idx0)] = True
# Also take care of the U case this time using
# equation 12 of Spudich and Chiou
s_ij = self.cum_length_set[j][i] + numpy.dot(
(edges[0, :2] - self.p0), self.gc2_config["b_hat"])
general_u[idx0] = u_i[idx0] + s_ij
# In the first case, ti = 0, u_i is outside of the segment
# this implements equation 5
idx1 = numpy.logical_and(ti0_check,
numpy.logical_not(on_segment_range))
w_i[idx1] = ((1.0 / (u_i[idx1] - self.length_set[j][i])) -
(1.0 / u_i[idx1]))
# In the last case the site is not on the edge (t != 0)
# implements equation 4
idx2 = numpy.logical_not(ti0_check)
w_i[idx2] = ((1. / t_i[idx2]) * (numpy.arctan(
(self.length_set[j][i] - u_i[idx2]) / t_i[idx2]) -
numpy.arctan(-u_i[idx2] / t_i[idx2])))
idx = numpy.logical_or(idx1, idx2)
# Equation 3
sum_w_i[idx] += w_i[idx]
# Part of equation 2
sum_w_i_t_i[idx] += (w_i[idx] * t_i[idx])
# Part of equation 9
sum_wi_ui_si[idx] += (w_i[idx] * (u_i[idx] + s_ij))
# For those sites not on the segment edge itself
idx_t = numpy.logical_not(on_segment)
general_t[idx_t] = (1.0 / sum_w_i[idx_t]) * sum_w_i_t_i[idx_t]
general_u[idx_t] = (1.0 / sum_w_i[idx_t]) * sum_wi_ui_si[idx_t]
return general_t, general_u
[docs] def get_rx_distance(self, mesh):
"""
For each point determine the corresponding rx distance using the GC2
configuration.
See :meth:`superclass method
<.base.BaseSurface.get_rx_distance>`
for spec of input and result values.
"""
# If the GC2 calculations have already been computed (by invoking Ry0
# first) and the mesh is identical then class has GC2 attributes
# already pre-calculated
if not self.tmp_mesh or (self.tmp_mesh == mesh):
self.gc2t, self.gc2u = self.get_generalised_coordinates(mesh.lons,
mesh.lats)
# Update mesh
self.tmp_mesh = deepcopy(mesh)
# Rx coordinate is taken directly from gc2t
return self.gc2t
[docs] def get_ry0_distance(self, mesh):
"""
For each point determine the corresponding Ry0 distance using the GC2
configuration.
See :meth:`superclass method
<.base.BaseSurface.get_ry0_distance>`
for spec of input and result values.
"""
# If the GC2 calculations have already been computed (by invoking Ry0
# first) and the mesh is identical then class has GC2 attributes
# already pre-calculated
if not self.tmp_mesh or (self.tmp_mesh == mesh):
# If that's not the case, or the mesh is different then
# re-compute GC2 configuration
self.gc2t, self.gc2u = self.get_generalised_coordinates(mesh.lons,
mesh.lats)
# Update mesh
self.tmp_mesh = deepcopy(mesh)
# Default value ry0 (for sites within fault length) is 0.0
ry0 = numpy.zeros_like(self.gc2u, dtype=float)
# For sites with negative gc2u (off the initial point of the fault)
# take the absolute value of gc2u
neg_gc2u = self.gc2u < 0.0
ry0[neg_gc2u] = numpy.fabs(self.gc2u[neg_gc2u])
# Sites off the end of the fault have values shifted by the
# GC2 length of the fault
pos_gc2u = self.gc2u >= self.gc_length
ry0[pos_gc2u] = self.gc2u[pos_gc2u] - self.gc_length
return ry0