# -*- coding: utf-8 -*-
# vim: tabstop=4 shiftwidth=4 softtabstop=4
#
# LICENSE
#
# Copyright (C) 2010-2019 GEM Foundation, G. Weatherill, M. Pagani,
# D. Monelli.
#
# The Hazard Modeller's Toolkit 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.
#
# You should have received a copy of the GNU Affero General Public License
# along with OpenQuake. If not, see <http://www.gnu.org/licenses/>
#
# DISCLAIMER
#
# The software Hazard Modeller's Toolkit (hmtk) provided herein
# is released as a prototype implementation on behalf of
# scientists and engineers working within the GEM Foundation (Global
# Earthquake Model).
#
# It is distributed for the purpose of open collaboration and in the
# hope that it will be useful to the scientific, engineering, disaster
# risk and software design communities.
#
# The software is NOT distributed as part of GEM’s OpenQuake suite
# (https://www.globalquakemodel.org/tools-products) and must be considered as a
# separate entity. The software provided herein is designed and implemented
# by scientific staff. It is not developed to the design standards, nor
# subject to same level of critical review by professional software
# developers, as GEM’s OpenQuake software suite.
#
# Feedback and contribution to the software is welcome, and can be
# directed to the hazard scientific staff of the GEM Model Facility
# (hazard@globalquakemodel.org).
#
# The Hazard Modeller's Toolkit (hmtk) is therefore distributed WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
# FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
# for more details.
#
# The GEM Foundation, and the authors of the software, assume no
# liability for use of the software.
'''
Utility functions for seismicity calculations
'''
import numpy as np
from shapely import geometry
from openquake.hazardlib.pmf import PRECISION
try:
from scipy.stats._continuous_distns import (truncnorm_gen,
_norm_cdf, _norm_sf,
_norm_ppf, _norm_isf)
[docs] class hmtk_truncnorm_gen(truncnorm_gen):
"""
At present, the scipy.stats.truncnorm.rvs object does not support
vector inputs for the bounds - this piece of duck punching changes that
"""
def _argcheck(self, a, b):
self.a = a
self.b = b
self._nb = _norm_cdf(b)
self._na = _norm_cdf(a)
self._sb = _norm_sf(b)
self._sa = _norm_sf(a)
self._delta = self._nb - self._na
idx = self.a > 0
self._delta[idx] = -(self._sb[idx] - self._sa[idx])
self._logdelta = np.log(self._delta)
return (a != b)
def _ppf(self, q, a, b):
output = np.zeros_like(self.a)
idx = self.a > 0
if np.any(idx):
output[idx] = _norm_isf(q[idx] * self._sb[idx] +
self._sa[idx] * (-q[idx] + 1.0))
idx = np.logical_not(idx)
if np.any(idx):
output[idx] = _norm_ppf(q[idx] * self._nb[idx] +
self._na[idx] * (-q[idx] + 1.0))
return output
hmtk_truncnorm = hmtk_truncnorm_gen(name="hmtk_truncnorm")
except Exception:
print("Continuous distributions not available on Scipy version < 0.15\n")
print("Bootstrap sampling of the depth distribution will raise an error")
hmtk_truncnorm = None
MARKER_NORMAL = np.array([0, 31, 59, 90, 120, 151, 181,
212, 243, 273, 304, 334])
MARKER_LEAP = np.array([0, 31, 60, 91, 121, 152, 182,
213, 244, 274, 305, 335])
SECONDS_PER_DAY = 86400.0
[docs]def decimal_year(year, month, day):
"""
Allows to calculate the decimal year for a vector of dates
(TODO this is legacy code kept to maintain comparability with previous
declustering algorithms!)
:param year: year column from catalogue matrix
:type year: numpy.ndarray
:param month: month column from catalogue matrix
:type month: numpy.ndarray
:param day: day column from catalogue matrix
:type day: numpy.ndarray
:returns: decimal year column
:rtype: numpy.ndarray
"""
marker = np.array([0., 31., 59., 90., 120., 151., 181.,
212., 243., 273., 304., 334.])
tmonth = (month - 1).astype(int)
day_count = marker[tmonth] + day - 1.
dec_year = year + (day_count / 365.)
return dec_year
[docs]def leap_check(year):
"""
Returns logical array indicating if year is a leap year
"""
return np.logical_and((year % 4) == 0,
np.logical_or((year % 100 != 0), (year % 400) == 0))
[docs]def decimal_time(year, month, day, hour, minute, second):
"""
Returns the full time as a decimal value
:param year:
Year of events (integer numpy.ndarray)
:param month:
Month of events (integer numpy.ndarray)
:param day:
Days of event (integer numpy.ndarray)
:param hour:
Hour of event (integer numpy.ndarray)
:param minute:
Minute of event (integer numpy.ndarray)
:param second:
Second of event (float numpy.ndarray)
:returns decimal_time:
Decimal representation of the time (as numpy.ndarray)
"""
tmo = np.ones_like(year, dtype=int)
tda = np.ones_like(year, dtype=int)
tho = np.zeros_like(year, dtype=int)
tmi = np.zeros_like(year, dtype=int)
tse = np.zeros_like(year, dtype=float)
#
# Checking inputs
if any(month < 1) or any(month > 12):
raise ValueError('Month must be in [1, 12]')
if any(day < 1) or any(day > 31):
raise ValueError('Day must be in [1, 31]')
if any(hour < 0) or any(hour > 24):
raise ValueError('Hour must be in [0, 24]')
if any(minute < 0) or any(minute > 60):
raise ValueError('Minute must be in [0, 60]')
if any(second < 0) or any(second > 60):
raise ValueError('Second must be in [0, 60]')
#
# Initialising values
if any(month):
tmo = month
if any(day):
tda = day
if any(hour):
tho = hour
if any(minute):
tmi = minute
if any(second):
tse = second
#
# Computing decimal
tmonth = tmo - 1
day_count = MARKER_NORMAL[tmonth] + tda - 1
id_leap = leap_check(year)
leap_loc = np.where(id_leap)[0]
day_count[leap_loc] = MARKER_LEAP[tmonth[leap_loc]] + tda[leap_loc] - 1
year_secs = ((day_count.astype(float) * SECONDS_PER_DAY) + tse +
(60. * tmi.astype(float)) + (3600. * tho.astype(float)))
dtime = year.astype(float) + (year_secs / (365. * 24. * 3600.))
dtime[leap_loc] = year[leap_loc].astype(float) + \
(year_secs[leap_loc] / (366. * 24. * 3600.))
return dtime
[docs]def haversine(lon1, lat1, lon2, lat2, radians=False, earth_rad=6371.227):
"""
Allows to calculate geographical distance
using the haversine formula.
:param lon1: longitude of the first set of locations
:type lon1: numpy.ndarray
:param lat1: latitude of the frist set of locations
:type lat1: numpy.ndarray
:param lon2: longitude of the second set of locations
:type lon2: numpy.float64
:param lat2: latitude of the second set of locations
:type lat2: numpy.float64
:keyword radians: states if locations are given in terms of radians
:type radians: bool
:keyword earth_rad: radius of the earth in km
:type earth_rad: float
:returns: geographical distance in km
:rtype: numpy.ndarray
"""
if not radians:
cfact = np.pi / 180.
lon1 = cfact * lon1
lat1 = cfact * lat1
lon2 = cfact * lon2
lat2 = cfact * lat2
# Number of locations in each set of points
if not np.shape(lon1):
nlocs1 = 1
lon1 = np.array([lon1])
lat1 = np.array([lat1])
else:
nlocs1 = np.max(np.shape(lon1))
if not np.shape(lon2):
nlocs2 = 1
lon2 = np.array([lon2])
lat2 = np.array([lat2])
else:
nlocs2 = np.max(np.shape(lon2))
# Pre-allocate array
distance = np.zeros((nlocs1, nlocs2))
i = 0
while i < nlocs2:
# Perform distance calculation
dlat = lat1 - lat2[i]
dlon = lon1 - lon2[i]
aval = (np.sin(dlat / 2.) ** 2.) + (np.cos(lat1) * np.cos(lat2[i]) *
(np.sin(dlon / 2.) ** 2.))
distance[:, i] = (2. * earth_rad * np.arctan2(np.sqrt(aval),
np.sqrt(1 - aval))).T
i += 1
return distance
[docs]def greg2julian(year, month, day, hour, minute, second):
"""
Function to convert a date from Gregorian to Julian format
:param year:
Year of events (integer numpy.ndarray)
:param month:
Month of events (integer numpy.ndarray)
:param day:
Days of event (integer numpy.ndarray)
:param hour:
Hour of event (integer numpy.ndarray)
:param minute:
Minute of event (integer numpy.ndarray)
:param second:
Second of event (float numpy.ndarray)
:returns julian_time:
Julian representation of the time (as float numpy.ndarray)
"""
year = year.astype(float)
month = month.astype(float)
day = day.astype(float)
timeut = hour.astype(float) + (minute.astype(float) / 60.0) + \
(second / 3600.0)
julian_time = ((367.0 * year) -
np.floor(
7.0 * (year + np.floor((month + 9.0) / 12.0)) / 4.0) -
np.floor(3.0 *
(np.floor((year + (month - 9.0) / 7.0) / 100.0) +
1.0) / 4.0) +
np.floor((275.0 * month) / 9.0) +
day + 1721028.5 + (timeut / 24.0))
return julian_time
[docs]def piecewise_linear_scalar(params, xval):
'''Piecewise linear function for a scalar variable xval (float).
:param params:
Piecewise linear parameters (numpy.ndarray) in the following form:
[slope_i,... slope_n, turning_point_i, ..., turning_point_n, intercept]
Length params === 2 * number_segments, e.g.
[slope_1, slope_2, slope_3, turning_point1, turning_point_2, intercept]
:param xval:
Value for evaluation of function (float)
:returns:
Piecewise linear function evaluated at point xval (float)
'''
n_params = len(params)
n_seg, remainder = divmod(n_params, 2)
if remainder:
raise ValueError(
'Piecewise Function requires 2 * nsegments parameters')
if n_seg == 1:
return params[1] + params[0] * xval
gradients = params[0:n_seg]
turning_points = params[n_seg: -1]
c_val = np.array([params[-1]])
for iloc in range(1, n_seg):
c_val = np.hstack(
[c_val, (c_val[iloc - 1] + gradients[iloc - 1] *
turning_points[iloc - 1]) - (gradients[iloc] *
turning_points[iloc - 1])])
if xval <= turning_points[0]:
return gradients[0] * xval + c_val[0]
elif xval > turning_points[-1]:
return gradients[-1] * xval + c_val[-1]
else:
select = np.nonzero(turning_points <= xval)[0][-1] + 1
return gradients[select] * xval + c_val[select]
[docs]def sample_truncated_gaussian_vector(data, uncertainties, bounds=None):
'''
Samples a Gaussian distribution subject to boundaries on the data
:param numpy.ndarray data:
Vector of N data values
:param numpy.ndarray uncertainties:
Vector of N data uncertainties
:param int number_bootstraps:
Number of bootstrap samples
:param tuple bounds:
(Lower, Upper) bound of data space
'''
nvals = len(data)
if bounds:
# if bounds[0] or (fabs(bounds[0]) < PRECISION):
if bounds[0] is not None:
lower_bound = (bounds[0] - data) / uncertainties
else:
lower_bound = -np.inf * np.ones_like(data)
# if bounds[1] or (fabs(bounds[1]) < PRECISION):
if bounds[1] is not None:
upper_bound = (bounds[1] - data) / uncertainties
else:
upper_bound = np.inf * np.ones_like(data)
sample = hmtk_truncnorm.rvs(lower_bound, upper_bound, size=nvals)
else:
sample = np.random.normal(0., 1., nvals)
return data + uncertainties * sample
[docs]def hmtk_histogram_1D(values, intervals, offset=1.0E-10):
"""
So, here's the problem. We tend to refer to certain data (like magnitudes)
rounded to the nearest 0.1 (or similar, i.e. 4.1, 5.7, 8.3 etc.). We also
like our tables to fall on on the same interval, i.e. 3.1, 3.2, 3.3 etc.
We usually assume that the counter should correspond to the low edge,
i.e. 3.1 is in the group 3.1 to 3.2 (i.e. L <= M < U).
Floating point precision can be a bitch! Because when we read in magnitudes
from files 3.1 might be represented as 3.0999999999 or as 3.1000000000001
and this is seemingly random. Similarly, if np.arange() is used to generate
the bin intervals then we see similar floating point problems emerging. As
we are frequently encountering density plots with empty rows or columns
where data should be but isn't because it has been assigned to the wrong
group.
Instead of using numpy's own historgram function we use a slower numpy
version that allows us to offset the intervals by a smaller amount and
ensure that 3.0999999999, 3.0, and 3.10000000001 would fall in the group
3.1 - 3.2!
:param numpy.ndarray values:
Values of data
:param numpy.ndarray intervals:
Data bins
:param float offset:
Small amount to offset the bins for floating point precision
:returns:
Count in each bin (as float)
"""
nbins = len(intervals) - 1
counter = np.zeros(nbins, dtype=float)
x_ints = intervals - offset
for i in range(nbins):
idx = np.logical_and(values >= x_ints[i], values < x_ints[i + 1])
counter[i] += float(np.sum(idx))
return counter
[docs]def hmtk_histogram_2D(xvalues, yvalues, bins, x_offset=1.0E-10,
y_offset=1.0E-10):
"""
See the explanation for the 1D case - now applied to 2D.
:param numpy.ndarray xvalues:
Values of x-data
:param numpy.ndarray yvalues:
Values of y-data
:param tuple bins:
Tuple containing bin intervals for x-data and y-data (as numpy arrays)
:param float x_offset:
Small amount to offset the x-bins for floating point precision
:param float y_offset:
Small amount to offset the y-bins for floating point precision
:returns:
Count in each bin (as float)
"""
xbins, ybins = (bins[0] - x_offset, bins[1] - y_offset)
n_x = len(xbins) - 1
n_y = len(ybins) - 1
counter = np.zeros([n_y, n_x], dtype=float)
for j in range(n_y):
y_idx = np.logical_and(yvalues >= ybins[j], yvalues < ybins[j + 1])
x_vals = xvalues[y_idx]
for i in range(n_x):
idx = np.logical_and(x_vals >= xbins[i], x_vals < xbins[i + 1])
counter[j, i] += float(np.sum(idx))
return counter.T
[docs]def bootstrap_histogram_1D(
values, intervals, uncertainties=None,
normalisation=False, number_bootstraps=None, boundaries=None):
'''
Bootstrap samples a set of vectors
:param numpy.ndarray values:
The data values
:param numpy.ndarray intervals:
The bin edges
:param numpy.ndarray uncertainties:
The standard deviations of each observation
:param bool normalisation:
If True then returns the histogram as a density function
:param int number_bootstraps:
Number of bootstraps
:param tuple boundaries:
(Lower, Upper) bounds on the data
:param returns:
1-D histogram of data
'''
if not number_bootstraps or np.all(np.fabs(uncertainties < PRECISION)):
# No bootstraps or all uncertaintes are zero - return ordinary
# histogram
#output = np.histogram(values, intervals)[0]
output = hmtk_histogram_1D(values, intervals)
if normalisation:
output = output / float(np.sum(output))
else:
output = output
return output
else:
temp_hist = np.zeros([len(intervals) - 1, number_bootstraps],
dtype=float)
for iloc in range(0, number_bootstraps):
sample = sample_truncated_gaussian_vector(values,
uncertainties,
boundaries)
#output = np.histogram(sample, intervals)[0]
output = hmtk_histogram_1D(sample, intervals)
temp_hist[:, iloc] = output
output = np.sum(temp_hist, axis=1)
if normalisation:
output = output / float(np.sum(output))
else:
output = output / float(number_bootstraps)
return output
[docs]def bootstrap_histogram_2D(
xvalues, yvalues, xbins, ybins,
boundaries=[None, None], xsigma=None, ysigma=None,
normalisation=False, number_bootstraps=None):
'''
Calculates a 2D histogram of data, allowing for normalisation and
bootstrap sampling
:param numpy.ndarray xvalues:
Data values of the first variable
:param numpy.ndarray yvalues:
Data values of the second variable
:param numpy.ndarray xbins:
Bin edges for the first variable
:param numpy.ndarray ybins:
Bin edges for the second variable
:param list boundaries:
List of (Lower, Upper) tuples corresponding to the bounds of the
two data sets
:param numpy.ndarray xsigma:
Error values (standard deviatons) on first variable
:param numpy.ndarray ysigma:
Error values (standard deviatons) on second variable
:param bool normalisation:
If True then returns the histogram as a density function
:param int number_bootstraps:
Number of bootstraps
:param returns:
2-D histogram of data
'''
if (xsigma is None and ysigma is None) or not number_bootstraps:
# No sampling - return simple 2-D histrogram
#output = np.histogram2d(xvalues, yvalues, bins=[xbins, ybins])[0]
output = hmtk_histogram_2D(xvalues, yvalues, bins=(xbins, ybins))
if normalisation:
output = output / float(np.sum(output))
return output
else:
if xsigma is None:
xsigma = np.zeros(len(xvalues), dtype=float)
if ysigma is None:
ysigma = np.zeros(len(yvalues), dtype=float)
temp_hist = np.zeros(
[len(xbins) - 1, len(ybins) - 1, number_bootstraps],
dtype=float)
for iloc in range(0, number_bootstraps):
xsample = sample_truncated_gaussian_vector(xvalues, xsigma,
boundaries[0])
ysample = sample_truncated_gaussian_vector(yvalues, ysigma,
boundaries[0])
# temp_hist[:, :, iloc] = np.histogram2d(xsample,
# ysample,
# bins=[xbins, ybins])[0]
temp_hist[:, :, iloc] = hmtk_histogram_2D(xsample,
ysample,
bins=(xbins, ybins))
if normalisation:
output = np.sum(temp_hist, axis=2)
output = output / np.sum(output)
else:
output = np.sum(temp_hist, axis=2) / float(number_bootstraps)
return output
# Parameters of WGS84 projection (in km)
WGS84 = {"a": 6378.137, "e": 0.081819191, "1/f": 298.257223563}
WGS84["e2"] = WGS84["e"] ** 2.
# Parameters of WGS84 projection (in m)
WGS84m = {"a": 6378137., "e": 0.081819191, "1/f": 298.2572221}
WGS84m["e2"] = WGS84m["e"] ** 2.
[docs]def TO_Q(lat): return (
(1.0 - WGS84["e2"]) * (
(np.sin(lat) / (1.0 - (WGS84["e2"] * (np.sin(lat) ** 2.))) -
((1. / (2.0 * WGS84["e"])) * np.log((1.0 - WGS84["e"] * np.sin(lat)) /
(1.0 + WGS84["e"] * np.sin(lat))))))
)
[docs]def TO_Qm(lat): return (
(1.0 - WGS84m["e2"]) * (
(np.sin(lat) / (1.0 - (WGS84m["e2"] * (np.sin(lat) ** 2.))) -
((1. / (2.0 * WGS84m["e"])) * np.log((1.0 - WGS84m["e"] * np.sin(lat)) /
(1.0 + WGS84m["e"] * np.sin(lat))))))
)
[docs]def lonlat_to_laea(lon, lat, lon0, lat0, f_e=0.0, f_n=0.0):
"""
Converts vectors of longitude and latitude into Lambert Azimuthal
Equal Area projection (km), with respect to an origin point
:param numpy.ndarray lon:
Longitudes
:param numpy.ndarray lat:
Latitude
:param float lon0:
Central longitude
:param float lat0:
Central latitude
:param float f_e:
False easting (km)
:param float f_e:
False northing (km)
:returns:
* easting (km)
* northing (km)
"""
lon = np.radians(lon)
lat = np.radians(lat)
lon0 = np.radians(lon0)
lat0 = np.radians(lat0)
q_0 = TO_Q(lat0)
q_p = TO_Q(np.pi / 2.)
q_val = TO_Q(lat)
beta = np.arcsin(q_val / q_p)
beta0 = np.arcsin(q_0 / q_p)
r_q = WGS84["a"] * np.sqrt(q_p / 2.)
dval = WGS84["a"] * (
np.cos(lat0) / np.sqrt(1.0 - (WGS84["e2"] * (np.sin(lat0) ** 2.))) /
(r_q * np.cos(beta0)))
bval = r_q * np.sqrt(
2. / (1.0 + (np.sin(beta0) * np.sin(beta)) + (np.cos(beta) *
np.cos(beta0) * np.cos(lon - lon0))))
easting = f_e + ((bval * dval) * (np.cos(beta) * np.sin(lon - lon0)))
northing = f_n + (bval / dval) * ((np.cos(beta0) * np.sin(beta)) -
(np.sin(beta0) * np.cos(beta) * np.cos(lon - lon0)))
return easting, northing
[docs]def area_of_polygon(polygon):
"""
Returns the area of an OpenQuake polygon in square kilometres
"""
lon0 = np.mean(polygon.lons)
lat0 = np.mean(polygon.lats)
# Transform to lamber equal area projection
x, y = lonlat_to_laea(polygon.lons, polygon.lats, lon0, lat0)
# Build shapely polygons
poly = geometry.Polygon(zip(x, y))
return poly.area