6.3.2 CLIMADA model demo

Model Demonstration of CLIMADA: A probabilistic risk
quantiﬁcation and adaptation economics platform
Hands-on demonstrations during the NAP Expo 2019
Incheon, South Korea
Evelyn Mühlhofer, Weather and Climate Risks Group, Federal Institute of Technology
Zurich, Switzerland (evelyn.muehlhofer@usys.ethz.ch)
April 10, 2019
The inputs are mainly based on the tutorials available on GitHub (https://github.com/
CLIMADA-project/climada_python/tree/master/script/tutorial), elaborated by Gabriela Az-
nar Siguan (gabriela.aznar@usys.ethz.ch)
Overview
The functionality of CLIMADA is gathered in the following classes:
• Entity: socio-economic models
– Exposures: exposed values
* BlackMarble: regional economic model from nightlight intensities and economic
indicators
* LitPop: data model of economic exposure from nightlight intensities and popula-
tion count
– ImpactFuncSet: collection of impact functions per hazard
* ImpactFunc: one adjustable impact function
* IFTropCyclone: deﬁnition of impact functions for tropical cyclones
– DiscRates: discount rates per year
– MeasureSet: collection of measures for adaptation
• Hazard: meteorological models
– TropCyclone: tropicalcyclone events
1

Step 1: Generating a Hazard Set
The most operational hazard module is currently the one for Tropical Cyclones (TC), which is why
we will focus on this for demonstrative purposes.
As always, there are different ways of generating or reading in a hazard set from already
existing sources (e.g. MATLAB or Excel). In this demonstration, we will generate a hazard set from
scratch using historical data from IBTrACS, and processing them with the TropCyclone module.
In [135]: # Import required packages:
from climada.hazard import Hazard
from climada.hazard import TCTracks
from climada.hazard import Centroids, TropCyclone
%matplotlib inline
In [136]: # First, we generate a "container" for our hazard, the Hazard class.
hazard_tc = Hazard('TC')
Next, we download TC track data from IBTrACS For this, it is easies to manually download
the data from IBTrACS:
• Go to ftp://eclipse.ncdc.noaa.gov/pub/ibtracs//v04r00/provisional/netcdf/
• Download the data from the folder IBTrACS.ALL.v04r00.nc
• Save the downloaded ﬁle to climada_python/data/system.
In [137]: # Store tc tracks in a sub-class of the hazard class: TCTracks.
selected_ibtracs = TCTracks()
# Of the downloaded data, select Northern Indian basin between 1980 and 2018
selected_ibtracs.read_ibtracs_netcdf(year_range=(1980, 2018), basin='NI')
# Generate interpolated track values to 1h time steps
selected_ibtracs.equal_timestep(time_step_h=1, land_params=False)
#set equal time steps here to shorter!! min 1h
print('Number of tracks:', selected_ibtracs.size)
fig, ax = selected_ibtracs.plot()
2019-04-05 11:50:33,709 - climada.hazard.tc_tracks - INFO - Interpolating 92 tracks to 1h time s
Number of tracks: 92
2

In [142]: # We could generate more tracks, e.g. 5 probabilistic events for each historic one
selcted_ibtracs_prob = TCTracks()
selcted_ibtracs_prob.read_ibtracs_netcdf(year_range=(1980, 2018), basin='NI')
selcted_ibtracs_prob.equal_timestep(time_step_h=1, land_params=False)
selcted_ibtracs_prob.calc_random_walk(ens_size=5) # generate 5 prob. per 1 historic tr
fig, ax = selcted_ibtracs_prob.plot()
2019-04-05 11:57:51,963 - climada.hazard.tc_tracks - INFO - Interpolating 92 tracks to 1h time s
2019-04-05 11:57:56,482 - climada.hazard.tc_tracks - INFO - Computing 460 synthetic tracks.
2019-04-05 11:58:02,342 - climada.hazard.tc_tracks - DEBUG - No historical track of category Hur
3

From the tracks we now want to calculate actual windﬁelds and map them to positions on the
ground ("centroids"). This is done as follows (for now, only with the historic tracks)
In [144]: # Construct centroids (we will narrow it down to the area of Bangladesh)
min_lat, max_lat, min_lon, max_lon = 20.766380, 26.536520, 92.666935, 88.121288
resol = 120
cent = Centroids()
cent.coord = (np.mgrid[min_lat : max_lat : complex(0, resol), min_lon : max_lon : comp
reshape(2, resol*resol).transpose()
cent.id = np.arange(cent.lat.size)
cent.check()
cent.plot()
2019-04-05 12:00:50,345 - climada.util.checker - DEBUG - Centroids.region_id not set.
Out[144]: (<Figure size 648x936 with 1 Axes>,
<cartopy.mpl.geoaxes.GeoAxesSubplot at 0x1a34692a58>)
4

In [145]: # construct tropical cyclones from historic tracks, with adequate windfields
tc_Bangladesh = TropCyclone()
tc_Bangladesh.set_from_tracks(selected_ibtracs, centroids=cent)
tc_Bangladesh.check()
# Version with probabilistic paths:
5

# tc_Bangladesh_prob = TropCyclone()
# tc_Bangladesh_prob.set_from_tracks(selected_ibtracs_prob, centroids=cent)
# tc_Bangladesh_prob.check()
2019-04-05 12:02:30,068 - climada.hazard.trop_cyclone - INFO - Mapping 552 tracks to 14400 centr
2019-04-05 12:03:05,949 - climada.hazard.trop_cyclone - DEBUG - Append events.
2019-04-05 12:03:06,977 - climada.hazard.trop_cyclone - DEBUG - Compute frequency.
In [146]: # plot most intense event
tc_Bangladesh.plot_intensity(-1)
# plot highest hazard intensity at each point
tc_Bangladesh.plot_intensity(0)
# plot highest hazard intensity for different return periods (10, 50, 75, 100 years)
_, _, res = tc_Bangladesh.plot_rp_intensity([10, 50, 75, 100])
2019-04-05 12:03:47,975 - climada.hazard.base - INFO - Computing exceedance intenstiy map for re
6

In [147]: # Now that we generated the hazard from the initial data,
# we just need to fill it into our intially created hazard container:
hazard_tc.append(tc_Bangladesh)
hazard_tc.check()
Step 2: Generating Exposure
Exposure can describe the geographical distribution of people, livelihoods and assets or infras-
tructure; all items potentially exposed to hazards. As shown in the overview, an "entity" class
exists, into which the exposed values ("exposure") are loaded as a sub-class.
Two models currently exist for generating such exposures (BlackMarble and LitPop) by de-
fault. Both contain estimates of the distribution of monetary asset values and population on a
global scale at 1x1km resolution.
9

There are other ways of generating exposures, e.g. from a GeoDataFrame of Python’s library
geopandas or from loading respective information from an excel-ﬁle.
In [81]: # Import required packages:
import numpy as np
import pandas as pd
from matplotlib import colors
from iso3166 import countries as iso_cntry
%matplotlib inline
from climada.entity import Entity
from climada.entity import Exposures
from climada.entity import BlackMarble
from climada.entity.exposures.litpop import LitPop
In [109]: # Initiate an entity class as "container" for exposures:
entity_trial = Entity()
# Generated a Black Marble exposure
exposure_trial = BlackMarble()
# Fill the generated exposure with actual values for a country (Bangladesh)
exposure_trial.set_countries(['Bangladesh'], 2013, res_km=1.0)
exposure_trial.set_geometry_points()
# Add the exposure as a sub-class to entity
entity_trial.exposures = Exposures(exposure_trial)
entity_trial.exposures['if_TC'] = entity_trial.exposures['if_']
entity_trial.exposures.assign_centroids(hazard_tc)
entity_trial.exposures.check()
In [110]: # plot exposure with major towns (pop_name=True):
entity_trial.exposures.plot_hexbin(pop_name=True)
# plot exposure with log-normal colormap for better contrasts:
norm=colors.LogNorm(vmin=500, vmax=4.0e9)
entity_trial.exposures.plot_hexbin(norm=norm)
array([[<cartopy.mpl.geoaxes.GeoAxesSubplot object at 0x1a419f7eb8>]],
dtype=object))
10

Step 3: Deﬁning Vulnerability (Impact Functions)
Impact Functions relate hazard intensity to the ensuing damage of the exposed assets or people.
Again, default functions exist, functions can be read in from excel or be manually created
(which is demonstrated here).
First, an Impact Function Set needs to be created (a "container" for the Impact Functions). This
set is then ﬁlled with an impact function, which we in turn create manually. Lastly, the impact
function set needs to be added to the "overall container", the entity.
In [111]: # loading necessary package:
from climada.entity import ImpactFuncSet
12

from climada.entity import ImpactFunc
import numpy as np
import matplotlib.pyplot as plt
In [112]: # Create Set:
imp_fun_set = ImpactFuncSet()
In [113]: # Create Impact Function from scratch
imp_fun = ImpactFunc()
imp_fun.haz_type = 'TC'
imp_fun.id = 1
imp_fun.name = 'TC random function'
imp_fun.intensity_unit = 'm/s'
imp_fun.intensity = np.linspace(0, 100, num=20)
imp_fun.mdd = np.sort(np.random.rand(20))
imp_fun.paa = np.sort(np.random.rand(20))
imp_fun.check()
In [114]: # Add Impact function to Set
imp_fun_set.add_func(imp_fun)
In [115]: # Add set to entity and plot
entity_trial.impact_funcs = imp_fun_set
entity_trial.impact_funcs.plot('TC')
[<matplotlib.axes._subplots.AxesSubplot at 0x1a1c24ccc0>])
13

Risk Quantiﬁcation, Damage Calculations ("Risk Today")
In [148]: # load necessary package
from climada.engine import Impact
In [149]: # Calculate impact (damage) by combining exposure, hazard and impact function
impact_trial = Impact()
impact_trial.calc(entity_trial.exposures, entity_trial.impact_funcs, hazard_tc)
2019-04-05 12:04:48,480 - climada.engine.impact - INFO - Exposures matching centroids found in c
2019-04-05 12:04:48,550 - climada.engine.impact - INFO - Calculating damage for 49270 assets (>0
In [150]: # Calculate impact exceedence frequency curve
freq_curve_fl = impact_trial.calc_freq_curve()
freq_curve_fl.plot();
print('Expected average annual impact: {:.3e} USD'.format(impact_trial.aai_agg))
# Plot average annual impact (damage) at each exposure
impact_trial.plot_eai_exposure(buffer_deg=1.0);
Expected average annual impact: 6.887e+08 USD
14

Towards Total Future Risk: Additional Risk from Socio-Economic Development & Cli-
mate Change
Steps 1-3 would now be repeated, but the exposure and hazard set would be modiﬁed, to capture
future scenarios:
• Population and Economic Growth --> Multiply asset and population values in exposure with
adequate factor
• Climate Change --> Increase frequency or intensity of event in hazard set
Step 4: Adaptation Measures Parametrization
Again, steps 1-3 are repeated. This time, hazard set and impact function would be modiﬁed, to
capture adaptation efforts:
15

• Impact at same intensity is lowered
• Frequency or intensity of hazard are lowered
Let’s create a new impact function, derived from the one above but representing the effects of an
adaptation measure on the impact e.g. building code (BC) set to resist winds up to 60 m/s:
In [151]: # effect of mdd_impact, paa_impact, hazard_inten_imp
%matplotlib inline
import numpy as np
from climada.entity import Measure
In [161]: # impact function which is 0 up to 62.5 m/s
imp_fun_BC = ImpactFunc()
imp_fun_BC.haz_type = 'TC'
imp_fun_BC.id = 2
imp_fun_BC.name = 'TC function with Building Code'
imp_fun_BC.intensity_unit = 'm/s'
imp_fun_BC.intensity = np.linspace(0, 100, num=20)
imp_fun_BC.mdd = np.append(np.zeros(13),np.sort(np.random.rand(7)))
imp_fun_BC.paa = np.append(np.zeros(13),np.sort(np.random.rand(7)))
imp_fun_BC.check()
In [158]: # Add Impact function to Set
imp_fun_set.add_func(imp_fun_BC)
In [159]: # Add set to entity and plot the old & new one
entity_trial.impact_funcs = imp_fun_set
entity_trial.impact_funcs.plot('TC')
[<matplotlib.axes._subplots.AxesSubplot at 0x1a3a482518>,
<matplotlib.axes._subplots.AxesSubplot at 0x1a31c75128>])
16

Result: Adaptation Options Appraisal
The total risks with and without adaptation measures can now be compared (incl. under different
socio-economic and climate change scenarios). The difference is total future risk is hence the "re-
duced damage" or "beneﬁt". Given that implementation costs are known for each of the measures,
cost-beneﬁt measures may now be calculated.
17

6.3.2 CLIMADA model demo

More Related Content

Similar to 6.3.2 CLIMADA model demo (20)

More from NAP Events (20)

Recently uploaded (20)

6.3.2 CLIMADA model demo