1
MTMG50: CLIMATE
SERVICES AND IMPACT
MODELLING

TECHNICAL ASSIGNMENT:
ESTIMATING WIND RISK

JOHN METHVEN & DAVID BRAYSHAW

DEPARTMENT OF METEOROLOGY
UNIVERSITY OF READING









2
Climate Service Technical Assignment: A
Estimating risk associated with wind damage over Europe
Setting the scene
One of the most costly natural hazards for Northern Europe arises from extreme winds and
the resulting injury to people and damage to infrastructure, such as buildings, power cables
and so on. Even globally, wind hazard is among the most damaging. In Northern Europe the
strongest winds are associated almost exclusively with the passage of extratropical cyclones
and the mesoscale structures embedded within them, such as fronts and sting jets.

The insurance and re-insurance industry need to estimate the risk of damaging winds across
the regions of interest. Broadly, this includes two components: firstly changes in the hazard
(and whether or not wind extremes might change with changing climate), and, secondly,
changes in the exposure and vulnerability associated with human activity (such as new
buildings, land use change, changes in the distribution and value of insured properties etc).

Defining the problem
In recent years, insurance companies have commissioned climate services to estimate the
risk of extreme winds and associated damage. Direct observational records of wind-storm
damage and insurance loss are relatively short, so re-analyses (and long historical climate
simulations) have been used to estimate the distributions of wind and damage functions
(e.g., the XWS Wind Catalogue, link below). In addition, some companies have started using
seasonal forecasts to estimate risk for the months ahead.

The typical approach to approximating wind damage potential, following Klawa and Pinto
(2003), is to calculate a loss potential or Storm Severity Index (SSI):
= (

98
− 1)
3
for > 98

where loss is only assumed to occur if, V, the local wind speed (at a grid point at a given
time) exceeds the 98th percentile of wind speed at that location, calculated from a long
historical dataset. Note the dependence on “wind excess” cubed. Variable p refers to
population density as a measure of exposure to the hazard, but in this assignment we will
neglect this factor and set p=1.

References
The XWS (extreme wind storms) Wind Catalogue: http://www.europeanwindstorms.org/

Klawa and Ulbrich (2003). A model for the estimation of storm losses and the identification
of severe weather storms in Germany, Nat Haz Earth Sys Sci, 3, 725-732.

Leckebusch, G. C., U. Ulbrich, L. Fröhlich, and J. G. Pinto, 2007: Property loss potentials for
European midlatitude storms in a changing climate. Geophys. Res. Lett., 34, L05703,
doi:10.1029/2006GL027663.
3
Questions for this technical assignment
Q1. What does the statistical distribution of observed wind strength look like calculated
from a multi-decadal observational record at three locations (near London, Paris and
Hamburg)? What are the wind speeds corresponding to the 50th, 75th and 98th percentiles of
the wind speed distribution at each location?
Q2. How does the 98th percentile of observed wind speed vary geographically (viewed on a
map) across Northwest Europe?
Q3. How does accumulated loss potential (estimated from the sum of SSI over the historical
record, using daily data) vary geographically across Northwest Europe?
Q4. Using the raw sub-seasonal-to-seasonal model S2S hindcasts, what are the 98th
percentiles of wind speed at the locations near London, Paris and Hamburg? How do they
differ from the 98th percentiles from observations? Do the values of the 98th percentile vary
with forecast lead time?
Q5. Consider the re-analysis data for one high impact cyclone crossing Northwest Europe in
the years where you have S2S hindcast data. What is the loss potential of the storm
estimated from observations?
Q6. How suitable are the S2S forecasts for:
i) Forecast and warning context: detecting the possibility of a high severity wind
storm in this case study time window. How many days ahead is the signature of
this storm detected?
ii) Climate risk context: estimating the probability of a storm at least as strong (in
terms of SSI) as the one observed? Restrict your analysis to the 3 locations in Q4.
Note that if you calculate model SSI using the 98th percentile of the model wind
speed distribution, this makes some allowance for model bias.
Data available to use in the assignment
data/ERA5/europe/europe_1halfx1half_ERA5_winds.zip
- Contains ERA5 re-analysis wind data covering the years 1979-2018.
data/S2S/europe/ecmwf - S2S hindcast data from ECMWF, winters 1999/2000 – 2009/10
data/S2S/europe/ncep – S2S hindcast data from NCEP, winters 1999/2000 – 2009/10
Note that you do not need to use everything. It is up to you to select. All the data is on a
common grid with 1.5 degree spacing in longitude and latitude. The hindcasts were
calculated with a consistent version of the forecasting system (model and data assimilation).
Each file contains one ensemble forecast from a particular date. Note that the S2S forecast
start dates are not regular for ECMWF (only twice per week) but are daily for NCEP. For both
4
centres, the data is recorded every day into the forecast (the lead time) and each file
contains the data for all ensemble members.
Template Python routines to get you started on the assignment
readera_ground_cartopy.py – read ground data NetCDF file and plot on a map using cartopy
readera_wind_series.py – read a sequence, at daily intervals, of NetCDF files containing
ERA5 wind data, extract wind components from a particular location and store a time series.
Instructions for the assignment
Write a technical report which addresses the 6 questions posed above in turn, using a new
section for each question. In each section,
• Explain what data you used to tackle the question, giving reasons why it is suitable.
• Describe the calculations you performed with the data (mathematically and in words
– not the code) and important considerations for your calculations, referring to the
Lecture Notes for guidance on how to approach such problems and best practice.
• Summarise the results and your answer to the questions.
• Present your results using key statistics and figures (with captions).
• Explain the limitations of your approach and the calculations.
Begin the report with a short introduction (less than half a page) setting the scene on the
purpose of this climate service assignment and end with a summary of findings.
Submit the report as a PDF file with a maximum of 10 sides A4 in 11-point font including all
figures and references. You can use a slightly smaller font for figure captions and references.
Submit the Python files for the code you used to answer these questions (upload the .py
files at the assignment submission point on Blackboard).








Please turn over.
5
Assessment – breakdown of marks
The marks on this assessment (worth 70% of the module marks) will be allocated as follows:
20% Python code structure and quality. Use the guidelines in the Lab Course Notes on
good programming practice, including PEP8, when you write your code.
40% Your approach to the tasks, linking to relevant literature tackling similar problems,
and the results obtained (Q. 1, 2, 3, 5). The scientific basis and originality of your
approach and quality of the results will be considered. Make good use of the Lecture
Notes, and references therein, to approach the task in a rigorous way.
20% Synthesis and discussion of the suitability of S2S forecasts for the climate service
addressed in this technical assignment (including Q. 4 and 6). Can useful forecasts of
the impact variable (i.e., wind damage or hydroelectric power station flow rate) be
generated by using an impact model driven by the S2S forecast data? Outline the
pros and cons of your approach as a climate service.
20% Presentation of the report, including quality of the writing and data presentation in
terms of graphs and statistics.