WP4.3: Understanding Extreme Weather and Climate Events David B. Stephenson University of Reading Background on extremes Overview of WP4.3 The first task.

WP4.3: Understanding Extreme WP4.3: Understanding Extreme Weather and Climate EventsWeather and Climate Events

David B. Stephenson

University of Reading

• Background on extremes• Overview of WP4.3• The first task 4.3a some input needed

ENSEMBLES RT4/5 meeting, Paris, 10-11 February 2005

us!

Gare Montparnasse, 22 October 1895

The definition problem:The definition problem:Extreme events can

be defined by:

• Maxima/minima

• Magnitude

• Rarity

• Impact/losses

“Man can believe the impossible,

but man can never believe the

improbable.” - Oscar Wilde

IPCC 2001 definition of extreme event:IPCC 2001 definition of extreme event:“An extreme weather event is an event that is rare within its statistical reference distribution at a particular place. Definitions of "rare" vary, but an extreme weather event would normally be as rare or rarer than the 10th or 90th percentile.”

• No mention of magnitude• Why 10th or 90th percentile? • No mention of the tail (EVT)• Not the only definition

)(}Pr{ xFxXp

Other IPCC TAR 2001 definitions:Other IPCC TAR 2001 definitions:Extreme climate event:

“an average of a number of weather events over a certain period of time which is itself extreme (e.g. rainfall over a season)”

Simple extremes:“individual local weather variables exceeding critical levels on a continuous scale”

Complex extremes:“severe weather associated with particular climatic phenomena, often requiring a critical combination of variables”

A Risk Modelling PerspectiveA Risk Modelling Perspective

Severe events (extreme loss events) caused by:

• Rare weather events

• Extreme weather events (amenable to EVT)

• Clustered weather events (e.g. climate event)

Natural hazarde.g. windstorm

Damagee.g. building

Losse.g. claims ($)

Risk=p(loss)=p(hazard) X vulnerability X exposure

extreme loss is not always due to extreme weather!

The Generalised Extreme Value distributionThe Generalised Extreme Value distributionMaxima/minima and the extreme tail of distributionscan be modelled asymptotically by the 3-parameterGeneralised Extreme Value (GEV) distribution:

1

1exp)Pr(x

xX

Models the tail of the probability distribution

Valid only when sufficiently far into the tail

No universal, absolute criteria for how far is sufficiently far

What can we learn from the What can we learn from the study of tails (caudology?) …study of tails (caudology?) …

… … about the whole animal?about the whole animal?

PDF = Probable Dinosaur Function ??

Hypotheses about changing PDFsHypotheses about changing PDFsH0: No change (variation due to

sampling only) Sampling uncertainty can be tested

using tail model

HL: Change due to “mean effect” e.g. Mearns et al. (1984), Wigley

(1985), …

HS: Change due to “variance effect”

e.g. Katz and Brown (1992), Katz and coworkers …

HLS: Change due to mean and variance effects

e.g. Brown and Katz (1995), …

Ha: “Structural change” in shape etc

e.g. Kestin 2001, Antoniadou et al. 2001

Some key climate change questions …Some key climate change questions …• How can we best estimate future possible changes in tail

probabilities and return values for extreme events?

multi-model ensemble data multi-model ensemble data tail probabilities tail probabilities

• Do we understand the key processes that led to these changes?

• Which factors are most important for determining changes in extreme events?– Large-scale atmospheric circulation– Land surface conditions (soil moisture, snow, ice)– Sea surface conditions (N. Atlantic sea surface temperatures)– Model resolution

ENSEMBLES WP4.3: ENSEMBLES WP4.3: Understanding Extreme Weather and Climate EventsUnderstanding Extreme Weather and Climate Events   Provision of statistical methods for identifying

extreme events and the climate regimes with which they are associated. More robust assessments of the effects of climate change on the probability of extreme events and on the characteristics of natural modes of climate variability.

us!

The WP4.3 teamThe WP4.3 teamUREADMM: David Stephenson, Caio Coelho, Chris FerroKNMI: Frank Selten and Adri Buishand and postdoc???

CERFACS: Laurent Terray and Emilia Sanchez INGV: Silvio GualdiIFM: Mojib Latif, Ernst BedachtNERSC: Dag Steinskog, Helge Drange, Nils Gunnar Kvamsto

AUTH: Prof. Panagiotis Maheras, As. Prof. Helena Flocas, Anagnostopoulou, Konstantia Tolika, Maria Hatzaki

UEA: Jean Palutikof, Tom HoltUFR: Martin Beniston, Stephane Goyette

Work tasksWork tasksTask 4.3.a: Development and use of methodologies for the

estimation of extreme event probabilities. Which are the best methods for inferring probabilistic tail information from multi-model ensembles of climate model simulations?

Task 4.3.b: Exploring the relationships between extreme events, weather systems and the large-scale atmospheric circulation/climate regimes. How do different large-scale factors influence weather extremes?

Task 4.3.c: The influence of anthropogenic forcings on the statistics of extreme events. How are extreme events likely to behave in the future?

Workpackage deliverablesWorkpackage deliverablesWP4.3a: Statistical methods for identifying regimes and estimating extreme-value tail probabilities using multi-model gridded data. Reports will be written up on this and disseminated to all partners and software in MATLAB (and possibly IDL) will be made freely available.

WP4.3b: An analysis of which factors are the most important in determining extreme events in Europe obtained by applying the techniques developed in WP4.3a to the multi-model ensemble of coupled and time-slice simulations. Oceanic (e.g. SST and THC) and atmospheric (large-scale flow) factors will be investigated. Extremes in both Northern and Southern Europe will be addressed.

WP4.3c: The behaviour of extremes in ENSEMBLES coupled scenarios will be analysed and interpreted using techniques and ideas developed for deliverables WP4.3a and WP4.3b.

Kilometre-stonesKilometre-stonesM4.3.1: Spatial Extreme Value (SEV) model developed and coded

up in MATLAB. (month 18)

M4.3.2: Techniques for extracting large-scale regimes developed. (month 18)

M4.3.3: Preliminary analysis of extremes and regimes in coupled runs completed. (month 24)

M4.3.4: Key large-scale factors for extremes identified. (month 36)

M4.3.5: Extremes in scenario runs summarised. (month 48)

What key scientific questions would you like to address?What key scientific questions would you like to address?UREADMM What are the key processes that create extreme events and what are the

best ways to diagnose these processes?

KNMI Influence of the large scale circulation on extreme temperatures and precipitation

CERFACS Understanding the links between large-scale circulation (LSC) patterns and extremes

Predict changes in EE distribution due to anthropogenic forcing and the links with changes of the mean hydrological cycle

INGV The relationship between the extreme events in the Euro-Mediterraneanregion and the large scale circulation

IFM Will there be more extreme events in future climate (2xCO2) ?

NERSC How do different large-scale factors influence weather extremes?

How are extreme events likely to behave in the future and what is the uncertainty in the prediction (e.g. due to model horizontal resolution, etc.)?

AUTH How are extreme events likely to behave in the future in Mediterranean?

Sensitivity of changes in extremes to horizontal resolution in Mediterranean

UEA What are the linkages between circulation types and the occurrence ofextremes at present and in the future under conditions of climate change?How will changes in weather types affect the occurrence of extremes?What other factors are responsible for the changing occurrence of

extremes?

What types of extreme event are you most interested in?What types of extreme event are you most interested in?

UREADMM Heat waves, extratropical storm-related extremes

KNMI Occurences of heat waves and multi-day rainfall extremes

CERFACS Precipitation, temperature and storms

INGV Temperature (heat waves and cold air outbreaks) and Precip (floods and droughts) in the Euro-Mediterranean region.

IFM Extreme windspeed, precipitation, temperature

NERSC Extreme events associated with: extra-tropical cyclones, polar lows, Arctic fronts and (high latitude) topography

AUTH Precipitation and Temperature extremes in the Mediterranean

UEA Floods (heavy rainfall events); heat waves (high temp events

UFR ???

Which factors controlling these extremes are you interested Which factors controlling these extremes are you interested in investigating?in investigating?

UREADMM Large-scale flow patterns

KNMI Large scale flow regimes and soil moisture conditions

CERFACS Tropical and extra-tropical LSC patterns, surface conditions

INGV Large scale regimes in the North Atlantic (e.g., NAO), SSTs in the tropical Oceans

IFM Changes/shift in general circulation, upper atmosphere

NERSC The large-scale atmospheric circulation

Low frequency variability, ocean and sea-ice state

Model uncertainty (sensitivity to resolution)

AUTH Regional circulation (circulation types) surface and 500hPa conditions

UEA Changing frequencies of weather types; changes in the relationship between extremes and weather types

UFR ???

What software do you normally use to analyse data?What software do you normally use to analyse data?UREADMM MATLAB, R

KNMI Fortran routines and GrADS

CERFACS IDL, ncl or fortran routines for the analysis ferret or ncl for the graphics. netcdf for the data.

INGV GrADS and Matlab

IFM Matlab, Fortran, Grads

NERSC Matlab

AUTH Fortran and Visual Basic routines

UEA Matlab, R

UFR ???

R software: www.r-project.org (free, unix/microsoft/mac, statistical language)

What statistical tools might be useful for exploring extreme What statistical tools might be useful for exploring extreme events in the simulations?events in the simulations?

UREADMM Spatio-temporal exploratory methods and probability models  KNMI Analysis of extremes with covariates fi indices of large scale flow

regimes. These methods can be developed using the 62 member ensemble of 140 year long transient coupled simulations that we have done.

CERFACS characterize completely extremes propertiesimprove the methods to compare observed extremes to simulated ones. address the questions of stationarity and homogenization

of obs daily dataINGV Clustering defined using quantiles and thresholds exceedancesIFM percentile analysis, use of extrem indices (e.g. number of frost

days, number wet days,....), ARMA model to find out, if a trend is a product of quasistationary climate

NERSC Multivariate extreme analysis

AUTH Statistica, R, SPSS, Stardex Diagnostic toolsUEA GEV; r largest; covariates/joint probabilitiesUFR ???

Original person months in WP4.3Original person months in WP4.3

Partner P.I. 4.3a 4.3b 4.3c Total

UREADMM Stephenson 12 0 0 12

KNMI Selten 12 0 0 12

CERFACS Terray 0 6 0 6

INGV Gualdi 0 6 0 6

KIEL Latif 0 6 0 6

NERSC Kvamsto 0 6 0 6

AUTH Maheras 0 0 12 12

UEA Palutikof 0 0 9 9

UNIFR Beniston 0 0 3 No cost

Total 24 24 24 72

DoW person months in WP4.3DoW person months in WP4.3

Partner P.I. 4.3a 4.3b 4.3c Total

UREADMM Stephenson 12 0? 0? 12

KNMI Selten 12 0? 0? 12

CERFACS Terray 2 4 0 6

INGV Gualdi 2 10 0 12

KIEL Latif 0 6 0 6

NERSC Kvamsto 1 4 0 5

AUTH Maheras 4? 4? 4? 12

UEA Palutikof 3? 3? 3? 9

UNIFR Beniston 0 0 3 No cost

Total 36 31 7 74

First 18 monthsFirst 18 monthsD4.3.1: Statistical methods for identifying regimes and estimating extreme-value tail probabilities using multi-model gridded data. Reports will be written up on this and disseminated to all partners and software in MATLAB (and possibly IDL) will be made freely available (Month 18 – February 2007)

Milestones M4.3.1: Development of methodologies to explore climate variability and extreme events, tested initially on existing simulations, for use with the ENSEMBLES multi-model system (Month 18 – February 2007)

What statistical tools need developing?What statistical tools need developing?• What are the scientific questions/hypotheses?

questions design of experiments & analysis

• Which type of extremes are to be investigated?

• Which type of processes are important?

Some ideas:

• Toolkit of exploratory tools for extreme events

• Spatial extreme value model

multi-model ensemble data multi-model ensemble data tail probabilities tail probabilities

• Extremes statistical model workshops

Extremes workshops in Switzerland• PRUDENCE project: small workshops in 2003

and 2004 in Chateau D’Oex

• ENSEMBLES – workshop 5-8 March 2005 Chateau D’Oex to discuss the development of statistical methods

• 2006,2007,2008,… - would like to continue these workshops

Martin Beniston’s website: http://www.unifr.ch/geoscience/geographie/ENSEMBLES/rt8

Some statistical issues Some statistical issues • Exploratory methods (curse of dimensionality)• Dependency

– Spatial (between variables at different places)– Temporal

• Short-term (e.g. between daily values)• Long-term (e.g. between winters)

• Significance testing– Pointwise or simultaneous?– Local or global?– Power (i.e. failure to detect a change – type 2 error)

• Large data sets (many grid point variables)• Spatial pooling/borrowing

What I would like to see in WP4.3What I would like to see in WP4.3• Good interaction between all partners rather than

unilateral work

• Good communication between all partners (e.g. via messages sent to all on WP4.3 email list)

• Some exciting scientific questions rather than just minimal achievement of deliverables

• Joint WP4.3 papers and talks

• Shared knowledge, software, and data

What colours do you see?What colours do you see?92% 99.6%

6% 0.25%deuteranopic

2% 0.10%protanopic

http://www.vischeck.com

Can you see a difference?Can you see a difference?

I can’t! Nor can 6% of the male population

Can you see a difference?Can you see a difference?

I can’t! Nor can 6% of the male population

That’s all FolksThat’s all Folks