"""
This file is part of CLIMADA.
Copyright (C) 2017 ETH Zurich, CLIMADA contributors listed in AUTHORS.
CLIMADA is free software: you can redistribute it and/or modify it under the
terms of the GNU General Public License as published by the Free
Software Foundation, version 3.
CLIMADA 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 General Public License for more details.
You should have received a copy of the GNU General Public License along
with CLIMADA. If not, see <https://www.gnu.org/licenses/>.
---
Define StormEurope class.
"""
__all__ = ['StormEurope']
import bz2
import datetime as dt
import logging
from pathlib import Path
from typing import Optional
import numpy as np
import xarray as xr
import pandas as pd
import matplotlib.pyplot as plt
from scipy import sparse
from climada.util.config import CONFIG
from climada.hazard.base import Hazard
from climada.hazard.centroids.centr import Centroids
from climada.util.files_handler import get_file_names
from climada.util.dates_times import (datetime64_to_ordinal,
last_year,
first_year,
date_to_str
)
from climada.util.dwd_icon_loader import (download_icon_centroids_file,
download_icon_grib,
delete_icon_grib,
)
LOGGER = logging.getLogger(__name__)
HAZ_TYPE = 'WS'
"""Hazard type acronym for Winter Storm"""
N_PROB_EVENTS = 5 * 6
"""Number of events per historic event in probabilistic dataset"""
[docs]
class StormEurope(Hazard):
"""A hazard set containing european winter storm events. Historic storm
events can be downloaded at http://wisc.climate.copernicus.eu/ and read
with `from_footprints`. Weather forecasts can be automatically downloaded from
https://opendata.dwd.de/ and read with from_icon_grib(). Weather forecast
from the COSMO-Consortium http://www.cosmo-model.org/ can be read with
from_cosmoe_file().
Attributes
----------
ssi_wisc : np.array, float
Storm Severity Index (SSI) as recorded in
the footprint files; apparently not reproducible from the footprint
values only.
ssi : np.array, float
SSI as set by set_ssi; uses the Dawkins definition by default.
"""
intensity_thres = 14.7
"""Intensity threshold for storage in m/s; same as used by WISC SSI calculations."""
vars_opt = Hazard.vars_opt.union({'ssi_wisc', 'ssi', 'ssi_full_area'})
"""Name of the variables that aren't need to compute the impact."""
[docs]
def __init__(self,
units: str = 'm/s',
ssi: Optional[np.ndarray] = None,
ssi_wisc: Optional[np.ndarray] = None,
ssi_full_area: Optional[np.ndarray] = None,
**kwargs):
"""Initialize a StormEurope object
Parameters
----------
units : str
The units of the Hazard intensity. Defaults to 'm/s'
ssi : numpy.ndarray
The Storm Severity Index (SSI). 1d vector, same length as number of
storms. Defaults to an empty array.
ssi_wisc : numpy.ndarray
The Storm Severity Index (SSI) as reconded in the footprint files.
1d vector, same length as number of storms. Defaults to an empty
array.
ssi_full_area : numpy.ndarray
1d vector, same length as number of storms.
Used to conserve the the index that is derived from the area and windspeed
of the full geographic extent of a storm system. Needed for cropping the
`StormEurope` object into a smaller region`.
Defaults to an empty array.
"""
kwargs.setdefault('haz_type', HAZ_TYPE)
Hazard.__init__(self, units=units, **kwargs)
self.ssi = ssi if ssi is not None else np.array([], float)
self.ssi_wisc = ssi_wisc if ssi_wisc is not None else np.array([], float)
self.ssi_full_area = ssi_full_area if ssi_full_area is not None else np.array([], float)
@classmethod
def _read_one_nc(cls, file_name, centroids, intensity_thres):
"""Read a single WISC footprint. Assumes a time dimension of length 1.
Omits a footprint if another file with the same timestamp has already
been read.
Parameters
----------
file_name : str
Absolute or relative path to *.nc
centroids : Centroids
Centr. instance that matches the
coordinates used in the *.nc, only validated by size.
intensity_thres : float
Intensity threshold for storage in m/s.
Returns
-------
new_haz : StormEurope
Hazard instance for one single storm.
"""
ncdf = xr.open_dataset(file_name)
if centroids.size != (ncdf.sizes['latitude'] * ncdf.sizes['longitude']):
ncdf.close()
LOGGER.warning(('Centroids size doesn\'t match NCDF dimensions. '
'Omitting file %s.'), file_name)
return None
# xarray does not penalise repeated assignments, see
# http://xarray.pydata.org/en/stable/data-structures.html
stacked = ncdf.max_wind_gust.stack(
intensity=('latitude', 'longitude', 'time')
)
stacked = stacked.where(stacked > intensity_thres)
stacked = stacked.fillna(0)
# fill in values from netCDF
ssi_wisc = np.array([float(ncdf.ssi)])
intensity = sparse.csr_matrix(stacked)
new_haz = cls(ssi_wisc=ssi_wisc,
intensity=intensity,
event_name=[ncdf.storm_name],
date=np.array([datetime64_to_ordinal(ncdf.time.data[0])]),
# fill in default values
centroids=centroids,
event_id=np.array([1]),
frequency=np.array([1]),
orig=np.array([True]),)
ncdf.close()
return new_haz
[docs]
def read_cosmoe_file(self, *args, **kwargs):
"""This function is deprecated, use StormEurope.from_cosmoe_file instead."""
LOGGER.warning("The use of StormEurope.read_cosmoe_file is deprecated."
"Use StormEurope.from_cosmoe_file instead.")
self.__dict__ = StormEurope.from_cosmoe_file(*args, **kwargs).__dict__
[docs]
@classmethod
def from_cosmoe_file(cls, fp_file, run_datetime, event_date=None,
model_name='COSMO-2E', description=None, intensity_thres=None):
"""Create a new StormEurope object with gust footprint from weather forecast.
The funciton is designed for the COSMO ensemble model used by
the COSMO Consortium http://www.cosmo-model.org/ and postprocessed to
an netcdf file using fieldextra. One event is one full day in UTC.
Works for MeteoSwiss model output of
COSMO-1E (11 members, resolution 1.1 km, forecast period 33-45 hours)
COSMO-2E (21 members, resolution 2.2 km, forecast period 5 days)
The frequency of each event is informed by their probability in the ensemble
forecast and is equal to 1/11 or 1/21 for COSMO-1E or COSMO-2E, respectively.
Parameters
----------
fp_file : str
string directing to one netcdf file
run_datetime : datetime
The starting timepoint of the forecast run
of the cosmo model
event_date : datetime, optional
one day within the forecast
period, only this day (00H-24H) will be included in the hazard
model_name : str,optional
provide the name of the COSMO model,
for the description (e.g., 'COSMO-1E', 'COSMO-2E')
description : str, optional
description of the events, defaults
to a combination of model_name and run_datetime
intensity_thres : float, optional
Intensity threshold for storage in m/s. Default: class attribute
StormEurope.intensity_thres (same as used by WISC SSI calculations)
Returns
-------
haz : StormEurope
StormEurope object with data from COSMO ensemble file.
"""
intensity_thres = cls.intensity_thres if intensity_thres is None else intensity_thres
# read intensity from file
ncdf = xr.open_dataset(fp_file)
ncdf = ncdf.assign_coords(date=('time',ncdf["time"].dt.floor("D").values))
if event_date:
try:
stacked = ncdf.sel(
time=event_date.strftime('%Y-%m-%d')
).groupby('date').max().stack(intensity=('y_1', 'x_1'))
except KeyError as ker:
raise ValueError('Extraction of date and coordinates failed. This is most likely '
'because the selected event_date '
f'{event_date.strftime("%Y-%m-%d")} is not contained in the '
'weather forecast selected by fp_file {fp_file}. Please adjust '
f'event_date or fp_file.') from ker
considered_dates = np.datetime64(event_date)
else:
time_covered_step = ncdf['time'].diff('time')
time_covered_day = time_covered_step.groupby('date').sum()
# forecast run should cover at least 18 hours of a day
considered_dates_bool = time_covered_day >= np.timedelta64(18,'h')
stacked = ncdf.groupby('date').max()\
.sel(date=considered_dates_bool)\
.stack(intensity=('y_1', 'x_1'))
considered_dates = stacked['date'].values
stacked = stacked.stack(date_ensemble=('date', 'epsd_1'))
stacked = stacked.where(stacked.VMAX_10M > intensity_thres)
stacked = stacked.fillna(0)
# fill in values from netCDF
intensity = sparse.csr_matrix(stacked.VMAX_10M.T)
event_id = np.arange(stacked.date_ensemble.size) + 1
date = np.repeat(
np.array(datetime64_to_ordinal(considered_dates)),
np.unique(ncdf.epsd_1).size
)
orig = np.full_like(event_id, False)
orig[(stacked.epsd_1 == 0).values] = True
if description is None:
description = (model_name +
' weather forecast windfield ' +
'for run startet at ' +
run_datetime.strftime('%Y%m%d%H'))
# Create Hazard
haz = cls(
intensity=intensity,
event_id=event_id,
centroids = cls._centroids_from_nc(fp_file),
# fill in default values
orig=orig,
date=date,
event_name=[date_i + '_ens' + str(ens_i)
for date_i, ens_i
in zip(date_to_str(date), stacked.epsd_1.values + 1)],
frequency=np.divide(
np.ones_like(event_id),
np.unique(ncdf.epsd_1).size),
)
# close netcdf file
ncdf.close()
haz.check()
return haz
[docs]
def read_icon_grib(self, *args, **kwargs):
"""This function is deprecated, use StormEurope.from_icon_grib instead."""
LOGGER.warning("The use of StormEurope.read_icon_grib is deprecated."
"Use StormEurope.from_icon_grib instead.")
self.__dict__ = StormEurope.from_icon_grib(*args, **kwargs).__dict__
[docs]
@classmethod
def from_icon_grib(cls, run_datetime, event_date=None, model_name='icon-eu-eps',
description=None, grib_dir=None, delete_raw_data=True,
intensity_thres=None):
"""Create new StormEurope object from DWD icon weather forecast footprints.
New files are available for 24 hours on
https://opendata.dwd.de, old files can be processed if they are
already stored in grib_dir.
One event is one full day in UTC. Current setup works for runs
starting at 00H and 12H. Otherwise the aggregation is inaccurate,
because of the given file structure with 1-hour, 3-hour and
6-hour maxima provided.
The frequency for one event is 1/(number of ensemble members)
Parameters
----------
run_datetime : datetime
The starting timepoint of the forecast run
of the icon model
event_date : datetime, optional
one day within the forecast
period, only this day (00H-24H) will be included in the hazard
model_name : str,optional
select the name of the icon model to
be downloaded. Must match the url on https://opendata.dwd.de
(see download_icon_grib for further info)
description : str, optional
description of the events, defaults
to a combination of model_name and run_datetime
grib_dir : str, optional
path to folder, where grib files are
or should be stored
delete_raw_data : bool,optional
select if downloaded raw data in
.grib.bz2 file format should be stored on the computer or
removed
intensity_thres : float, optional
Intensity threshold for storage in m/s. Default: class attribute
StormEurope.intensity_thres (same as used by WISC SSI calculations)
Returns
-------
haz : StormEurope
StormEurope object with data from DWD icon weather forecast footprints.
"""
# pylint: disable=protected-access
intensity_thres = cls.intensity_thres if intensity_thres is None else intensity_thres
if not (run_datetime.hour == 0 or run_datetime.hour == 12):
LOGGER.warning('The event definition is inaccuratly implemented '
'for starting times, which are not 00H or 12H.')
# download files, if they don't already exist
file_names = download_icon_grib(
run_datetime, model_name=model_name, download_dir=grib_dir)
# create centroids
nc_centroids_file = download_icon_centroids_file(model_name, grib_dir)
# read intensity from files
for ind_i, file_i in enumerate(file_names):
gripfile_path_i = Path(file_i[:-4])
with open(file_i, 'rb') as source, open(gripfile_path_i, 'wb') as dest:
dest.write(bz2.decompress(source.read()))
ds_i = xr.open_dataset(gripfile_path_i, engine='cfgrib')
if ind_i == 0:
stacked = ds_i
else:
stacked = xr.concat([stacked,ds_i], 'valid_time')
# create intensity matrix with max for each full day
stacked = stacked.assign_coords(
date=('valid_time', stacked["valid_time"].dt.floor("D").values))
if event_date:
try:
stacked = stacked.sel(
valid_time=event_date.strftime('%Y-%m-%d')).groupby('date').max()
except KeyError as ker:
raise ValueError('Extraction of date and coordinates failed. This is most likely '
'because the selected event_date '
f'{event_date.strftime("%Y-%m-%d")} is not contained in the '
'weather forecast selected by run_datetime'
f'{run_datetime.strftime("%Y-%m-%d %H:%M")}. Please adjust '
'event_date or run_datetime.') from ker
considered_dates = np.datetime64(event_date)
else:
time_covered_step = stacked['valid_time'].diff('valid_time')
time_covered_day = time_covered_step.groupby('date').sum()
# forecast run should cover at least 18 hours of a day
considered_dates_bool = time_covered_day >= np.timedelta64(18,'h')
stacked = stacked.groupby('date').max().sel(date=considered_dates_bool)
considered_dates = stacked['date'].values
stacked = stacked.stack(date_ensemble=('date', 'number'))
stacked = stacked.where(stacked > intensity_thres)
stacked = stacked.fillna(0)
event_id = np.arange(stacked.date_ensemble.size) + 1
date = np.repeat(
np.array(datetime64_to_ordinal(considered_dates)),
np.unique(stacked.number).size
)
orig = np.full_like(event_id, False)
orig[(stacked.number == 1).values] = True
if description is None:
description = ('icon weather forecast windfield for run started at ' +
run_datetime.strftime('%Y%m%d%H'))
# Create Hazard
haz = cls(
intensity=sparse.csr_matrix(stacked.gust.T),
centroids=cls._centroids_from_nc(nc_centroids_file),
event_id=event_id,
date=date,
orig=orig,
event_name=[date_i + '_ens' + str(ens_i)
for date_i, ens_i
in zip(date_to_str(date), stacked.number.values)],
frequency=np.divide(
np.ones_like(event_id),
np.unique(stacked.number).size),
)
haz.check()
# delete generated .grib2 and .4cc40.idx files
for ind_i, file_i in enumerate(file_names):
gripfile_path_i = Path(file_i[:-4])
idxfile_path_i = next(gripfile_path_i.parent.glob(
str(gripfile_path_i.name) + '.*.idx'))
gripfile_path_i.unlink()
idxfile_path_i.unlink()
if delete_raw_data:
#delete downloaded .bz2 files
delete_icon_grib(run_datetime, model_name=model_name, download_dir=grib_dir)
return haz
@staticmethod
def _centroids_from_nc(file_name):
"""Construct Centroids from the grid described by 'latitude' and
'longitude' variables in a netCDF file.
"""
LOGGER.info('Constructing centroids from %s', file_name)
ncdf = xr.open_dataset(file_name)
create_meshgrid = True
if hasattr(ncdf, 'latitude'):
lats = ncdf.latitude.data
lons = ncdf.longitude.data
elif hasattr(ncdf, 'lat'):
lats = ncdf.lat.data
lons = ncdf.lon.data
elif hasattr(ncdf, 'lat_1'):
if len(ncdf.lon_1.shape)>1 & \
(ncdf.lon_1.shape == ncdf.lat_1.shape) \
:
lats = ncdf.lat_1.data.flatten()
lons = ncdf.lon_1.data.flatten()
create_meshgrid = False
else:
lats = ncdf.lat_1.data
lons = ncdf.lon_1.data
elif hasattr(ncdf, 'clat'):
lats = ncdf.clat.data
lons = ncdf.clon.data
if ncdf.clat.attrs['units']=='radian':
lats = np.rad2deg(lats)
lons = np.rad2deg(lons)
create_meshgrid = False
else:
raise AttributeError('netcdf file has no field named latitude or '
'other know abrivation for coordinates.')
ncdf.close()
if create_meshgrid:
lats, lons = np.array([np.repeat(lats, len(lons)),
np.tile(lons, len(lats))])
cent = Centroids(lat=lats, lon=lons, on_land='natural_earth')
return cent
def _combine_events(self, event_ids):
"""combine the intensities of two events using max and adjust event_id, event_name,
date etc of the hazard
the event_ids must be consecutive for the event_name field to behave correctly
Parameters
----------
event_ids : array
two consecutive event ids
"""
select_event_ids = np.isin(self.event_id, event_ids)
select_other_events = np.invert(select_event_ids)
intensity_tmp = self.intensity[select_event_ids, :].max(axis=0)
self.intensity = self.intensity[select_other_events, :]
self.intensity = sparse.vstack([self.intensity, sparse.csr_matrix(intensity_tmp)])
self.event_id = np.append(self.event_id[select_other_events], self.event_id.max() + 1)
self.date = np.append(self.date[select_other_events],
np.round(self.date[select_event_ids].mean()))
name_2 = self.event_name.pop(np.where(select_event_ids)[0][1])
name_1 = self.event_name.pop(np.where(select_event_ids)[0][0])
self.event_name.append(name_1 + '_' + name_2)
fraction_tmp = self.fraction[select_event_ids, :].max(axis=0)
self.fraction = self.fraction[select_other_events, :]
self.fraction = sparse.vstack([self.fraction, sparse.csr_matrix(fraction_tmp)])
self.frequency = np.append(self.frequency[select_other_events],
self.frequency[select_event_ids].mean())
self.orig = np.append(self.orig[select_other_events],
self.orig[select_event_ids].max())
if self.ssi_wisc.size > 0:
self.ssi_wisc = np.append(self.ssi_wisc[select_other_events],
np.nan)
if self.ssi.size > 0:
self.ssi = np.append(self.ssi[select_other_events],
np.nan)
if self.ssi_full_area.size > 0:
self.ssi_full_area = np.append(self.ssi_full_area[select_other_events],
np.nan)
self.check()
[docs]
def calc_ssi(self, method='dawkins', intensity=None, on_land=True,
threshold=None, sel_cen=None):
"""Calculate the SSI, method must either be 'dawkins' or 'wisc_gust'.
'dawkins', after Dawkins et al. (2016),
doi:10.5194/nhess-16-1999-2016, matches the MATLAB version.
ssi = sum_i(area_cell_i * intensity_cell_i^3)
'wisc_gust', according to the WISC Tier 1 definition found at
https://wisc.climate.copernicus.eu/wisc/#/help/products#tier1_section
ssi = sum(area_on_land) * mean(intensity)^3
In both definitions, only raster cells that are above the threshold are
used in the computation.
Note that this method does not reproduce self.ssi_wisc, presumably
because the footprint only contains the maximum wind gusts instead of
the sustained wind speeds over the 72 hour window. The deviation may
also be due to differing definitions of what lies on land (i.e. Syria,
Russia, Northern Africa and Greenland are exempt).
Parameters
----------
method : str
Either 'dawkins' or 'wisc_gust'
intensity : scipy.sparse.csr
Intensity matrix; defaults to
self.intensity
on_land : bool
Only calculate the SSI for areas on land,
ignoring the intensities at sea. Defaults to true, whereas
the MATLAB version did not.
threshold : float, optional
Intensity threshold used in index
definition. Cannot be lower than the read-in value.
sel_cen : np.array, bool
A boolean vector selecting centroids.
Takes precendence over on_land.
"""
if intensity is not None:
if not isinstance(intensity, sparse.csr_matrix):
intensity = sparse.csr_matrix(intensity)
else:
pass
else:
intensity = self.intensity
if threshold is not None:
assert threshold >= self.intensity_thres, \
'threshold cannot be below threshold upon read_footprint'
intensity = intensity.multiply(intensity > threshold)
else:
intensity = intensity.multiply(intensity > self.intensity_thres)
cent = self.centroids
area_pixel = cent.get_area_pixel()
if sel_cen is not None:
pass
elif on_land is True:
sel_cen = cent.on_land.astype(bool)
else: # select all centroids
sel_cen = np.ones_like(area_pixel, dtype=bool)
ssi = np.zeros(intensity.shape[0])
if method == 'dawkins':
area_c = area_pixel / 1000 / 1000 * sel_cen
for i, inten_i in enumerate(intensity):
ssi_i = inten_i.power(3).dot(area_c)
# matrix crossproduct (row x column vector)
ssi[i] = ssi_i.item(0)
elif method == 'wisc_gust':
for i, inten_i in enumerate(intensity[:, sel_cen]):
area = np.sum(area_pixel[inten_i.indices]) / 1000 / 1000
inten_mean = np.mean(inten_i)
ssi[i] = area * np.power(inten_mean, 3)
return ssi
[docs]
def set_ssi(self, **kwargs):
"""Wrapper around calc_ssi for setting the self.ssi attribute.
Parameters
----------
kwargs :
passed on to calc_ssi
"""
self.ssi = self.calc_ssi(**kwargs)
[docs]
def plot_ssi(self, full_area=False):
"""Plot the distribution of SSIs versus their cumulative exceedance
frequencies, highlighting historical storms in red.
Returns
-------
fig : matplotlib.figure.Figure
ax : matplotlib.axes._subplots.AxesSubplot
"""
if full_area:
ssi = self.ssi_full_area
else:
ssi = self.ssi
# data wrangling
ssi_freq = pd.DataFrame({
'ssi': ssi,
'freq': self.frequency,
'orig': self.orig,
})
ssi_freq = ssi_freq.sort_values('ssi', ascending=False)
ssi_freq['freq_cum'] = np.cumsum(ssi_freq.freq)
ssi_hist = ssi_freq.loc[ssi_freq.orig].copy()
ssi_hist.freq = ssi_hist.freq * self.orig.size / self.orig.sum()
ssi_hist['freq_cum'] = np.cumsum(ssi_hist.freq)
# plotting
fig, axs = plt.subplots()
axs.plot(ssi_freq.freq_cum, ssi_freq.ssi, label='All Events')
axs.scatter(ssi_hist.freq_cum, ssi_hist.ssi,
color='red', label='Historic Events')
axs.legend()
axs.set_xlabel('Exceedance Frequency [1/a]')
axs.set_ylabel('Storm Severity Index')
axs.ticklabel_format(axis='y', style='sci', scilimits=(0, 0))
plt.show()
return fig, axs
[docs]
def generate_prob_storms(self, reg_id=528, spatial_shift=4, ssi_args=None,
**kwargs):
"""Generates a new hazard set with one original and 29 probabilistic
storms per historic storm. This represents a partial implementation of
the Monte-Carlo method described in section 2.2 of Schwierz et al.
(2010), doi:10.1007/s10584-009-9712-1.
It omits the rotation of the storm footprints, as well as the pseudo-
random alterations to the intensity.
In a first step, the original intensity and five additional intensities
are saved to an array. In a second step, those 6 possible intensity
levels are shifted by n raster pixels into each direction (N/S/E/W).
Caveats:
- Memory safety is an issue; trial with the entire dataset resulted
in 60GB of swap memory being used...
- Can only use numeric region_id for country selection
- Drops event names as provided by WISC
Parameters
----------
region_id : int, list of ints, or None
iso_n3 code of the
countries we want the generated hazard set to be returned for.
spatial_shift : int
amount of raster pixels to shift by
ssi_args : dict
A dictionary of arguments passed to calc_ssi
kwargs :
keyword arguments passed on to self._hist2prob()
Returns
-------
new_haz : StormEurope
A new hazard set for the given country.
Centroid attributes are preserved. self.orig attribute is set
to True for original storms (event_id ending in 00). Also
contains a ssi_prob attribute,
"""
# bool vector selecting the targeted centroids
if reg_id is not None:
self.centroids.set_region_id(overwrite=False)
if not isinstance(reg_id, list):
reg_id = [reg_id]
sel_cen = np.isin(self.centroids.region_id, reg_id)
else: # shifting truncates valid centroids
sel_cen = np.zeros(self.centroids.shape, bool)
sel_cen[
spatial_shift:-spatial_shift,
spatial_shift:-spatial_shift
] = True
sel_cen = sel_cen.reshape(self.centroids.size)
# init probabilistic array
n_out = N_PROB_EVENTS * self.size
intensity_prob = sparse.lil_matrix((n_out, np.count_nonzero(sel_cen)))
ssi = np.zeros(n_out)
LOGGER.info('Commencing probabilistic calculations')
for index, intensity1d in enumerate(self.intensity):
# indices for return matrix
start = index * N_PROB_EVENTS
end = (index + 1) * N_PROB_EVENTS
intensity_prob[start:end, :], ssi[start:end] =\
self._hist2prob(
intensity1d,
sel_cen,
spatial_shift,
ssi_args,
**kwargs)
LOGGER.info('Generating new StormEurope instance')
base = np.repeat((self.event_id * 100), N_PROB_EVENTS)
synth_id = np.tile(np.arange(N_PROB_EVENTS), self.size)
event_id = base + synth_id
new_haz = StormEurope(
ssi_full_area=ssi,
intensity=sparse.csr_matrix(intensity_prob),
# don't use synthetic dates; just repeat the
# historic dates
date=np.repeat(self.date, N_PROB_EVENTS),
# subsetting centroids
centroids=self.centroids.select(sel_cen=sel_cen),
# construct new event ids
event_id=event_id,
# frequency still based on the historic number of
# years
frequency=np.divide(np.repeat(self.frequency, N_PROB_EVENTS),
N_PROB_EVENTS),
orig=(event_id % 100 == 0),
)
new_haz.check()
return new_haz
def _hist2prob(self, intensity1d, sel_cen, spatial_shift, ssi_args=None,
power=1.15, scale=0.0225):
"""Internal function, intended to be called from generate_prob_storms.
Generates six permutations based on one historical storm event, which
it then moves around by spatial_shift gridpoints to the east, west, and
north.
Parameters
----------
intensity1d : scipy.sparse.csr_matrix, 1 by n
One historic event
sel_cen : np.ndarray(dtype=bool)
which centroids to return
spatial_shift : int
amount of raster cells to shift by
power : float
power to be applied elementwise
scale : float
weight of probabilistic component
ssi_args : dict
named arguments passed on to calc_ssi
Returns
-------
intensity : np.array
Synthetic intensities of shape (N_PROB_EVENTS, length(sel_cen))
ssi : np.array
SSI per synthetic event according to provided method.
"""
if not ssi_args:
ssi_args = {}
shape_ndarray = tuple([N_PROB_EVENTS]) + self.centroids.shape
# reshape to the raster that the data represents
intensity2d = intensity1d.reshape(self.centroids.shape)
# scipy.sparse.csr_matrix elementwise methods (to avoid this:
# https://github.com/ContinuumIO/anaconda-issues/issues/9129 )
intensity2d_sqrt = intensity2d.power(1.0 / power).todense()
intensity2d_pwr = intensity2d.power(power).todense()
intensity2d = intensity2d.todense()
# intermediary 3d array: (lat, lon, events)
intensity3d_prob = np.ndarray(shape_ndarray)
# the six variants of intensity transformation
# 1. translation only
intensity3d_prob[0] = intensity2d
# 2. and 3. plusminus scaled sqrt
intensity3d_prob[1] = intensity2d - (scale * intensity2d_sqrt)
intensity3d_prob[2] = intensity2d + (scale * intensity2d_sqrt)
# 4. and 5. plusminus scaled power
intensity3d_prob[3] = intensity2d - (scale * intensity2d_pwr)
intensity3d_prob[4] = intensity2d + (scale * intensity2d_pwr)
# 6. minus scaled sqrt and pwr
intensity3d_prob[5] = (intensity2d
- (0.5 * scale * intensity2d_pwr)
- (0.5 * scale * intensity2d_sqrt))
# spatial shifts
# northward
intensity3d_prob[6:12, :-spatial_shift, :] = \
intensity3d_prob[0:6, spatial_shift:, :]
# southward
intensity3d_prob[12:18, spatial_shift:, :] = \
intensity3d_prob[0:6, :-spatial_shift, :]
# eastward
intensity3d_prob[18:24, :, spatial_shift:] = \
intensity3d_prob[0:6, :, :-spatial_shift]
# westward
intensity3d_prob[24:30, :, :-spatial_shift] = \
intensity3d_prob[0:6, :, spatial_shift:]
intensity_out = intensity3d_prob.reshape(
N_PROB_EVENTS,
np.prod(self.centroids.shape)
)
ssi = self.calc_ssi(intensity=intensity_out, **ssi_args)
return intensity_out[:, sel_cen], ssi
# pylint: disable=invalid-name
def generate_WS_forecast_hazard(run_datetime=None,
event_date=None,
haz_model='icon-eu-eps',
haz_raw_storage=None,
save_haz=True):
"""use the initialization time (run_datetime), the date of the event and
specify the forecast model (haz_model) to generate a Hazard from forecast
data either by download or through reading from existing file.
Parameters
----------
run_datetime: datetime.datetime, optional
The starting timepoint of the forecast run
of the icon model, defaults to today 00:00 hours
event_date: datetime.datetime, optional
one day within the forecast
period, only this day (00H-24H) will be included in the hazard,
defaults to the day after tomorrow
haz_model: str, optional
select the name of the model to
be used: one of ['icon-eu-eps', 'icon-d2-eps', 'cosmo1e_file',
'cosmo2e_file'],
default is 'icon-eu-eps'.
haz_raw_storage: str, optional
path to folder, where netcdf files
are stored, mendatory when haz_model is 'cosmo1e_file' or
'cosmo2e_file'. the string must contain one set of curly brakets,
used by the function str.format to include
run_datetime.strftime('%y%m%d%H').
Default None resolves to "cosmoe_forecast_{}_vmax.nc" in
CONFIG.hazard.storm_europe.forecast_dir
save_haz: bool, optional
flag if resulting hazard should be saved in
CONFIG.hazard.storm_europe.forecast_dir, default is True.
Returns
-------
Hazard
hazard
str
haz_model
datetime.datetime
run_datetime
datetime.datetime
event_date
"""
FORECAST_DIR = CONFIG.hazard.storm_europe.forecast_dir.dir()
if run_datetime is None:
run_datetime = dt.datetime.today().replace(hour=0, minute=0, second=0, microsecond=0)
if event_date is None:
event_date = dt.datetime.today().replace(hour=0, minute=0, second=0, microsecond=0) \
+ dt.timedelta(days=2)
if haz_model in ['cosmo1e_file', 'cosmo2e_file']:
if haz_model == 'cosmo1e_file':
haz_model='C1E'
full_model_name_temp = 'COSMO-1E'
if haz_model == 'cosmo2e_file':
haz_model='C2E'
full_model_name_temp = 'COSMO-2E'
haz_file_name = (f'{HAZ_TYPE}_{haz_model}_run{run_datetime.strftime("%Y%m%d%H")}'
f'_event{event_date.strftime("%Y%m%d")}.hdf5')
haz_file = FORECAST_DIR / haz_file_name
if haz_file.exists():
LOGGER.info('Loading hazard from %s.', haz_file)
hazard = StormEurope.from_hdf5(haz_file)
else:
LOGGER.info('Generating %s hazard.', haz_model)
if not haz_raw_storage:
haz_raw_storage = FORECAST_DIR / "cosmoe_forecast_{}_vmax.nc"
fp_file = Path(str(haz_raw_storage).format(run_datetime.strftime('%y%m%d%H')))
hazard = StormEurope.from_cosmoe_file(
fp_file,
event_date=event_date,
run_datetime=run_datetime,
model_name=full_model_name_temp
)
if save_haz:
hazard.write_hdf5(haz_file)
elif haz_model in ['icon-eu-eps', 'icon-d2-eps']:
if haz_model == 'icon-eu-eps':
full_model_name_temp = haz_model
haz_model='IEE'
if haz_model == 'icon-d2-eps':
full_model_name_temp = haz_model
haz_model='IDE'
haz_file_name = (f'{HAZ_TYPE}_{haz_model}_run{run_datetime.strftime("%Y%m%d%H")}'
f'_event{event_date.strftime("%Y%m%d")}.hdf5')
haz_file = FORECAST_DIR / haz_file_name
if haz_file.exists():
LOGGER.info('Loading hazard from %s.', haz_file)
hazard = StormEurope.from_hdf5(haz_file)
else:
LOGGER.info('Generating %s hazard.', haz_model)
hazard = StormEurope.from_icon_grib(
run_datetime,
event_date=event_date,
delete_raw_data=False,
model_name=full_model_name_temp
)
if save_haz:
hazard.write_hdf5(haz_file)
else:
raise NotImplementedError("specific 'WS' hazard not implemented yet. " +
"Please specify a valid value for haz_model.")
# check if hazard is successfully generated for Forecast
if not isinstance(hazard, Hazard):
LOGGER.warning('Hazard generation unsuccessful.')
return hazard, haz_model, run_datetime, event_date
