def test_pga(self):
sa = SA(period=1e-50, damping=5)
pga = PGA()
cormo = JB2009CorrelationModel(vs30_clustering=False)
corma = cormo._get_correlation_matrix(self.SITECOL, sa)
corma2 = cormo._get_correlation_matrix(self.SITECOL, pga)
self.assertTrue((corma == corma2).all())
cormo = JB2009CorrelationModel(vs30_clustering=True)
corma = cormo._get_correlation_matrix(self.SITECOL, sa)
corma2 = cormo._get_correlation_matrix(self.SITECOL, pga)
self.assertTrue((corma == corma2).all())
def test_correlation_with_total_stddev(self):
mean1 = 10
mean2 = 14
inter = 1e-300
intra1 = 0.2
intra2 = 1.6
p1 = Point(0, 0)
p2 = Point(0, 0.3)
sites = [
Site(p1, mean1, False, inter, intra1),
Site(p2, mean2, False, inter, intra2)
]
self.sites = SiteCollection(sites)
numpy.random.seed(41)
cormo = JB2009CorrelationModel(vs30_clustering=False)
gsim = FakeGSIMTotalStdDev(self)
gsim.expect_same_sitecol = False
with self.assertRaises(CorrelationButNoInterIntraStdDevs):
ground_motion_fields(self.rupture,
self.sites, [self.imt1],
gsim,
truncation_level=None,
realizations=6000,
correlation_model=cormo)
def test_rupture_site_filtering(self):
mean = 10
inter = 2
intra = 3
points = [Point(0, 0), Point(0, 0.05)]
sites = [Site(point, mean, False, inter, intra) for point in points]
self.sites = SiteCollection(sites)
def rupture_site_filter(rupture_sites):
[(rupture, sites)] = rupture_sites
yield rupture, sites.filter(sites.mesh.lats == 0)
self.gsim.expect_same_sitecol = False
numpy.random.seed(37)
cormo = JB2009CorrelationModel(vs30_clustering=False)
gmfs = ground_motion_fields(self.rupture,
self.sites, [self.imt1],
self.gsim,
truncation_level=None,
realizations=1,
correlation_model=cormo,
rupture_site_filter=rupture_site_filter)
s1gmf, s2gmf = gmfs[self.imt1]
numpy.testing.assert_array_equal(s2gmf, 0)
numpy.testing.assert_array_almost_equal(s1gmf, 11.1852253)
def test_no_correlation_mean_and_intra_respected(self):
mean1 = 10
mean2 = 14
inter = 1e-300
intra1 = 0.2
intra2 = 1.6
p1 = Point(0, 0)
p2 = Point(0, 0.3)
sites = [
Site(p1, mean1, False, inter, intra1),
Site(p2, mean2, False, inter, intra2)
]
self.sites = SiteCollection(sites)
numpy.random.seed(41)
cormo = JB2009CorrelationModel(vs30_clustering=False)
s1_intensity, s2_intensity = ground_motion_fields(
self.rupture,
self.sites,
[self.imt1],
self.gsim,
truncation_level=None,
realizations=6000,
correlation_model=cormo,
)[self.imt1]
self.assertAlmostEqual(s1_intensity.mean(), mean1, delta=1e-3)
self.assertAlmostEqual(s2_intensity.mean(), mean2, delta=1e-3)
self.assertAlmostEqual(s1_intensity.std(), intra1, delta=2e-3)
self.assertAlmostEqual(s2_intensity.std(), intra2, delta=1e-2)
def test_no_truncation(self):
mean = 10
inter = 1e-300
intra = 3
points = [
Point(0, 0),
Point(0, 0.05),
Point(0.06, 0.025),
Point(0, 1.0),
Point(-10, -10)
]
sites = [Site(point, mean, False, inter, intra) for point in points]
self.sites = SiteCollection(sites)
numpy.random.seed(23)
cormo = JB2009CorrelationModel(vs30_clustering=False)
corma = cormo._get_correlation_matrix(self.sites, self.imt1)
gmfs = ground_motion_fields(self.rupture,
self.sites, [self.imt1],
self.gsim,
truncation_level=None,
realizations=6000,
correlation_model=cormo)
sampled_corma = numpy.corrcoef(gmfs[self.imt1])
assert_allclose(corma, sampled_corma, rtol=0, atol=0.02)
def test_with_intra_and_inter(self):
mean = 10
inter = 2
intra = 3
sites = SiteCollection([Site(Point(0, 0), mean, False, inter, intra)])
self.gsim.expect_same_sitecol = False
numpy.random.seed(37)
cormo = JB2009CorrelationModel(vs30_clustering=False)
distribution = scipy.stats.norm()
eps_intra = distribution.rvs(size=(1, 1))
eps_intra = numpy.array(
cormo.apply_correlation(sites, self.imt1, eps_intra))[0]
eps_inter = distribution.rvs(size=(1, 1))
gmf = ground_motion_field_with_residuals(
self.rupture,
sites,
self.imt1,
self.gsim,
truncation_level=None,
intra_residual_epsilons=eps_intra,
inter_residual_epsilons=eps_inter)
numpy.testing.assert_array_almost_equal(gmf, 11.1852253)
def test_period_one_and_above(self):
cormo = JB2009CorrelationModel(vs30_clustering=False)
cormo2 = JB2009CorrelationModel(vs30_clustering=True)
imt = SA(period=1.0, damping=5)
corma = cormo._get_correlation_matrix(self.SITECOL, imt)
aaae(corma, [[1, 0.2730787, 1, 0.2730787],
[0.2730787, 1, 0.2730787, 0.07457198],
[1, 0.2730787, 1, 0.2730787],
[0.2730787, 0.07457198, 0.2730787, 1]])
corma2 = cormo2._get_correlation_matrix(self.SITECOL, imt)
self.assertTrue((corma == corma2).all())
imt = SA(period=10.0, damping=5)
corma = cormo._get_correlation_matrix(self.SITECOL, imt)
aaae(corma, [[1, 0.56813402, 1, 0.56813402],
[0.56813402, 1, 0.56813402, 0.32277627],
[1, 0.56813402, 1, 0.56813402],
[0.56813402, 0.32277627, 0.56813402, 1]])
corma2 = cormo2._get_correlation_matrix(self.SITECOL, imt)
self.assertTrue((corma == corma2).all())
def test_filtered_sitecol(self):
filtered = self.SITECOL.filtered([0, 2])
numpy.random.seed(13)
cormo = JB2009CorrelationModel(vs30_clustering=False)
intra_residuals_sampled = numpy.random.normal(size=(2, 5))
intra_residuals_correlated = cormo.apply_correlation(
filtered, PGA(), intra_residuals_sampled)
aaae(intra_residuals_correlated,
[[-0.71239066, 0.75376638, -0.04450308, 0.45181234, 1.34510171],
[0.51816327, 1.36481251, 0.86016437, 1.48732124, -1.01860545]],
decimal=6)
def test_clustered(self):
cormo = JB2009CorrelationModel(vs30_clustering=True)
imt = SA(period=0.001, damping=5)
corma = cormo._get_correlation_matrix(self.SITECOL, imt)
aaae(corma, [[1, 0.44046654, 1, 0.44046654],
[0.44046654, 1, 0.44046654, 0.19401077],
[1, 0.44046654, 1, 0.44046654],
[0.44046654, 0.19401077, 0.44046654, 1]])
imt = SA(period=0.5, damping=5)
corma = cormo._get_correlation_matrix(self.SITECOL, imt)
aaae(corma, [[1, 0.36612758, 1, 0.36612758],
[0.36612758, 1, 0.36612758, 0.1340494],
[1, 0.36612758, 1, 0.36612758],
[0.36612758, 0.1340494, 0.36612758, 1]])
def test_no_clustering(self):
cormo = JB2009CorrelationModel(vs30_clustering=False)
imt = SA(period=0.1, damping=5)
corma = cormo._get_correlation_matrix(self.SITECOL, imt)
aaae(corma, [[1, 0.03823366, 1, 0.03823366],
[0.03823366, 1, 0.03823366, 0.00146181],
[1, 0.03823366, 1, 0.03823366],
[0.03823366, 0.00146181, 0.03823366, 1]])
imt = SA(period=0.95, damping=5)
corma = cormo._get_correlation_matrix(self.SITECOL, imt)
aaae(corma, [[1, 0.26107857, 1, 0.26107857],
[0.26107857, 1, 0.26107857, 0.06816202],
[1, 0.26107857, 1, 0.26107857],
[0.26107857, 0.06816202, 0.26107857, 1]])
def test(self):
numpy.random.seed(13)
cormo = JB2009CorrelationModel(vs30_clustering=False)
intra_residuals_sampled = numpy.random.normal(size=(3, 100000))
intra_residuals_correlated = cormo.apply_correlation(
self.SITECOL, PGA(), intra_residuals_sampled)
inferred_corrcoef = numpy.corrcoef(intra_residuals_correlated)
mean = intra_residuals_correlated.mean()
std = intra_residuals_correlated.std()
self.assertAlmostEqual(mean, 0, delta=0.002)
self.assertAlmostEqual(std, 1, delta=0.002)
actual_corrcoef = cormo._get_correlation_matrix(self.SITECOL, PGA())
numpy.testing.assert_almost_equal(inferred_corrcoef,
actual_corrcoef,
decimal=2)
def setUp(self):
self.imt = "SA(0.15)"
points = [Point(0, 0), Point(10, 10), Point(20, 20)]
sites = [Site(point, 10, False, 2, 3, id=i)
for i, point in enumerate(points)]
self.sites = models.SiteCollection(sites)
assets = [mock.Mock] * 5
self.sites_assets = ((0, assets[0:1]),
(1, assets[1:]),
(2, assets[2:]))
self.gsims = mock.Mock()
self.gsims.__getitem__ = mock.Mock(return_value=mock.Mock())
self.cormo = JB2009CorrelationModel(vs30_clustering=False)
from collections import OrderedDict
from shapely import wkt
from openquake.hazardlib.geo.point import Point
from openquake.hazardlib.geo.surface import PlanarSurface
from openquake.hazardlib.geo.geodetic import geodetic_distance
from openquake.hazardlib.correlation import JB2009CorrelationModel
from openquake.hazardlib.site import Site, SiteCollection
from openquake.hazardlib.imt import from_string
from openquake.hazardlib.gsim import get_available_gsims
from openquake.hazardlib.gsim.base import ContextMaker
from openquake.hazardlib.sourceconverter import RuptureConverter
from openquake.hazardlib import nrml
from smtk.residuals.gmpe_residuals import Residuals
DEFAULT_CORRELATION = JB2009CorrelationModel(False)
GSIM_LIST = get_available_gsims()
def build_planar_surface(geometry):
"""
Builds the planar rupture surface from the openquake.nrmllib.models
instance
"""
# Read geometry from wkt
geom = wkt.loads(geometry.wkt)
top_left = Point(geom.xy[0][0],
geom.xy[1][0],
geometry.upper_seismo_depth)
top_right = Point(geom.xy[0][1],
geom.xy[1][1],
def test(self):
cormo = JB2009CorrelationModel(vs30_clustering=False)
lt = cormo.get_lower_triangle_correlation_matrix(self.SITECOL, PGA())
aaae(lt, [[1.0, 0.0, 0.0], [1.97514806e-02, 9.99804920e-01, 0.0],
[1.97514806e-02, 5.42206860e-20, 9.99804920e-01]])
def main(var, r, voi, rand, cor_model, vs_corr):
"""
Main program for computing spatial correlation
INPUTS:
var- variables dictionary from initialize function. Contains
M,N,K,site_collection_SM, site_collection_station
uncertaintydata, data, location_lon/lat_g
r - radius
voi- variable of interest, i.e. PGA
rand- array of random variables
cor_model- JB2009 or GA2010
vs_corr- boolean to determine if Vs30 are correlated. See
JB2009
intensity_factor- factor for non-native data
OUT: cor- grid of spatially correlated epsilon
data- grid of ShakeMap data
data_new- data with added spatial correlation
grid_arr- array for storing grid indices for multiple realizations
mu_arr- array for storing Sig21.T*Sig11inv for multiple realizations
sigma_arr- array for storing sigma for multiple realizations
"""
start = time.time()
OL_time = 0
IL_time = 0
M = var['M']
N = var['N']
K = var['K']
if cor_model == 'JB2009':
CM = JB2009CorrelationModel(vs_corr)
else:
CM = GA2010CorrelationModel()
# Initialize vector where data_new will be stored
X = np.zeros([M*N,1])
# Initialize vectors for storing data for multiple realizations
grid_arr = [None] * (M*N)
mu_arr = [None] * (M*N)
sigma_arr = np.zeros([M*N,1])
rand_arr = np.zeros([M*N,1])
# Get spcing of horozontal and vertical points
ld = set_up_grid_dist(M,N,var['site_collection_SM'])
pre_loop_time = time.time() - start
for i in range(0,M):
OL_start = time.time()
# Find the number of points in radius horozontally and vertically for each row
vhva = calc_vert_hor(i, r, ld['l'], ld['d'])
# Calculate the full distance matrix for each row
dist = calc_full_dist(i, vhva['vert'], vhva['hor'], N, var['site_collection_SM'])
first_time_per_row = 1
OL_time += time.time() - OL_start
for j in range(0,N):
IL_start = time.time()
num = i*N+j
# Find the reduced distance matrix
dist_calc = reduce_distance(j, vhva['vert'], vhva['hor'], vhva['added_vert'], N, dist['distance_matrix'], dist['grid_indices'])
# Include stations in distance matrix and find the indices of the points within the radius
out = inc_stations(j, i, N, K, r, var['site_collection_SM'], var['site_collection_station'],
dist_calc['dist_mat'], X, dist_calc['inc_ind'])
if np.size(dist_calc['inc_indices']) == 1:
# no conditioning points, correlation value is random
X[num] = rand[num]
# Store for multiple realizations
grid_arr [num] = np.zeros(0)
mu_arr [num] = np.zeros(0)
sigma_arr[num] = 1
rand_arr [num] = X[num]
else:
# Check if reduced distance matrix is full distance matrix
if ((vhva['vert'] == 1 and dist_calc['num_indices'] == vhva['hor']+1)or(vhva['vert'] != 1 and \
dist_calc['num_indices'] == 2*vhva['hor']+1)) and (np.size(out['inc_sta_indices']) == 0):
# If this is the first full distance matrix per row, calculate base case
if first_time_per_row == 1:
base = calculate_corr(out['dist_mat'], voi, CM)
first_time_per_row = 0
mu = base['Sig12'].T*base['Sig11inv']*out['x']
rand_num = rand[num]
X[num] = mu+rand_num*base['R']
# Store for multiple realizations
grid_arr [num] = dist_calc['inc_ind'][0:-1]
mu_arr [num] = base['Sig12'].T*base['Sig11inv']
sigma_arr[num] = base['R']
rand_arr [num] = rand_num
else:
other = calculate_corr(out['dist_mat'], voi, CM)
mu = other['Sig12'].T*other['Sig11inv']*out['x']
rand_num = rand[num]
X[num] = mu+rand_num*other['R']
# Store for multiple realizations
grid_arr [num] = dist_calc['inc_ind'][0:-1]
mu_arr [num] = other['Sig12'].T*other['Sig11inv']
sigma_arr[num] = other['R']
rand_arr [num] = rand_num
IL_time += time.time() - IL_start
if np.mod(i*N+j,5000) == 0:
print 'Finishing step:', i*N+j
DATA = var['data']
COR = np.reshape(X, [M,N]) #units epsilon
X = np.multiply(COR, var['uncertaintydata']) # ln(pctg)
DATA_NEW = np.multiply(DATA,np.exp(X))
end = time.time() - start
print 'Total Time', end
print 'Pre loop Time', pre_loop_time
print 'Inner loop time', IL_time
print 'Outer loop time', OL_time
return {'cor':COR, 'data':DATA, 'data_new':DATA_NEW, 'grid_arr':grid_arr, 'mu_arr':mu_arr, 'sigma_arr':sigma_arr}