This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 46.18.87.93

This content was downloaded on 25/01/2017 at 14:42

Please note that terms and conditions apply.

Real-time extreme weather event attribution with forecast seasonal SSTs

View the table of contents for this issue, or go to the journal homepage for more

2016 Environ. Res. Lett. 11 064006

(http://iopscience.iop.org/1748-9326/11/6/064006)

Home Search Collections Journals About Contact us My IOPscience

You may also be interested in:

A comparison of model ensembles for attributing 2012 West African rainfall

Hannah R Parker, Fraser C Lott, Rosalind J Cornforth et al.

Climate change effects on the worst-case storm surge: a case study of Typhoon Haiyan

Izuru Takayabu, Kenshi Hibino, Hidetaka Sasaki et al.

Attributing human mortality during extreme heat waves to anthropogenic climate change

Daniel Mitchell, Clare Heaviside, Sotiris Vardoulakis et al.

Multi-method attribution analysis of extreme precipitation in Boulder, Colorado

Jonathan M Eden, Klaus Wolter, Friederike E L Otto et al.

Predicting uncertainty in forecasts of weather and climate

T N Palmer

Attribution of human-induced dynamical and thermodynamical contributions in extreme weather events

R Vautard, P Yiou, F Otto et al.

What is the current state of scientific knowledge with regard to seasonal and decadalforecasting?

Doug M Smith, Adam A Scaife and Ben P Kirtman

Relationship between North American winter temperature and large-scale atmospheric circulation

anomalies and its decadal variation

B Yu, H Lin, Z W Wu et al.

The changing influences of the AMO and PDO on the decadal variation of the Santa Ana winds

Andy K Li, Houk Paek and Jin-Yi Yu

Environ. Res. Lett. 11 (2016) 064006 doi:10.1088/1748-9326/11/6/064006

LETTER

Real-time extreme weather event attribution with forecastseasonal SSTs

KHaustein1, F E LOtto1, PUhe1,2, N Schaller3,MRAllen1,3, LHermanson4, NChristidis4, PMcLean4 andHCullen5

1 School of Geography and the Environment, University of Oxford,Oxford, UK2 Oxford e-ResearchCenter, University ofOxford,Oxford, UK3 Department of Atmospheric Oceanic and Planetary Physics, University ofOxford, Oxford, UK4 MetOfficeHadley Centre, Exeter, UK5 Climate Central,Washington, USA

E-mail: karsten.haustein@ouce.ox.ac.uk

Keywords: extreme event attribution, climate change, ensemblemodelling

AbstractWithin the last decade, extremeweather event attribution has emerged as a newfield of science andgarnered increasing attention from thewider scientific community and the public. Numerousmethods have been put forward to determine the contribution of anthropogenic climate change toindividual extremeweather events. So far nearly all such analyses were donemonths after an event hashappened.Here we present a newmethodwhich can assess the fraction of attributable risk of a severeweather event due to an external driver in real-time. Themethod builds on a large ensemble ofatmosphere-only general circulationmodel simulations forced by seasonal forecast sea surfacetemperatures (SSTs). Taking the England 2013/14winterfloods as an example, we demonstrate thatthe change in risk for heavy rainfall during the Englandfloods due to anthropogenic climate change, isof similarmagnitude using either observed or seasonal forecast SSTs. Testing the dynamic response ofthemodel to the anomalous ocean state for January 2014, wefind that observed SSTs are required toestablish a discernible link between a particular SST pattern and an atmospheric response such as ashift in the jetstream in themodel. For extreme events occurring under strongly anomalous SSTpatterns associatedwith known low-frequency climatemodes, however, forecast SSTs can providesufficient guidance to determine the dynamic contribution to the event.

1. Introduction

Warming of the climate system is unequivocal, andsince the 1950s, many of the observed changes in theEarthʼs climate system are unprecedented over decadesto millennia [13]. Since anthropogenic climate changeposes a variety of societal and socio-economic risksrelated to changes in the frequency of extreme weatherevents [2], the scientific community is challenged toprovide answers as to what these changes might be andwhether and to what extent such changes are alreadydetectable. As extreme events are rare by definition, it isoften futile to draw a confident conclusion regardingchanging risks from an analysis based on observationaldata alone. Using the power of large ensemble simula-tions, with climate models that have shown to

realistically reproduce key feedbacks and large-scaleprocesses in the climate system [18, 28], it is possible toobtain meaningful statistics on the magnitude andfrequency of occurrence of some extreme weatherevents and to assess changes in the statistics [22, 30].

Several of the current approaches of this so-calledprobabilistic event attribution utilize atmosphere-onlygeneral circulation models (AGCM) and hence rely onobserved sea surface temperatures (SSTs) [21, 22, 27];leading to publication months or years after the event.Given that public awareness for an extremeweather eventis limited to a short period after the event occurring [4],and links to climate change are discussed in the mediadespite the lack of scientific evidence, an attribution state-ment which is based on a very fast analysis frameworkproviding robust results would be desirable. If we are able

OPEN ACCESS

RECEIVED

18 January 2016

ACCEPTED FOR PUBLICATION

28April 2016

PUBLISHED

1 June 2016

Original content from thisworkmay be used underthe terms of the CreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2016 IOPPublishing Ltd

to present evidence for a causal link (or the absence of it)between a specific extreme event and climate change, thepublic will more likely understand the reality and theextent of the problem and it will be harder to dismiss cli-mate change or to blame it as the sole cause, respectively.Disaster risk management can be improved and the sci-enceon the topic of climate change canbe advanced.

Here we show that fast-track event attribution doesprovide robust assessments of the return time forextremeweather events that are comparablewith thewellestablished attribution of climate-related events frame-work [8]. Thenewapproach employs theweather@home(W@H) modelling environment for event attributionbased on the atmosphere-only model (HadAM3P) [17].Using seasonal forecast SST and sea ice concentration(SIC)data products insteadof observed data, ourmethodenables us to simulate thepossible ‘weather’ in advance ofan extreme weather event occurring and it allows us toprovide an evidence-based attribution statement imme-diately after the event orwhile it is still unfolding.

In this case study, we use the example of the Eng-land winter floods 2013/14 to compare the attributionstatements obtained fromanalyzing an existing coupledGlobal-Regional Climate Model (GCM-RCM) ensem-ble forced with observed SSTs and SIC with a GCM-RCM ensemble forced with forecast SSTs and SIC.Since this case has been thoroughly investigated usingobserved SSTs [26], we can (1) built on the previousresults for January 2014 and (2) use the existing hugeensemble to estimate statistics of very rare events withmore confidence.We find that we are able to reproducethe attribution statement for a changing risk in pre-cipitation between natural (NAT; the world that mighthave been without anthropogenic climate change) andactual (ACT) conditions with high confidence. We alsolooked at near-surface temperature and found excellentagreement in the attribution result between ensemblesusing forecast and observed SSTs (section 3.1).

Since extreme events are a combination of an anom-alous state of the circulation (atmospheric noise) andthermodynamic and dynamic changes due to climatechange (anthropogenic signal), the identification of thesignal is the major challenge and notoriously difficultproblem when making attribution statements [22]. It isworth pointing out that the case-dependent (internal)dynamic contribution—detectable or not—will not con-siderably change the attributable human component,commonly expressed as the change in risk. Even themost extreme circulation regime could be largely unaf-fected by climate change [32]. Only the thermodynamicresponse and possible (forced) changes in the large-scaledynamics due to anthropogenicwarming alter the risk ina significant way. The return time estimate for anextreme event under current (ACT) or counterfactual(NAT) conditions will, however, change considerably inresponse to event-specific extreme dynamics. For exam-ple, a positive North Atlantic Oscillation (NAO+)increases the likelihood for a wet winter over Englandoccurring, regardless of climate change.

We use a suite of metrics to investigate the robust-ness of the combined dynamic and thermodynamicresponse in our two SST scenarios. We also test thedetectability of the dynamic response during the floodevent in the context of the emerging variability betweenthe different NAT SST patterns. Similar attemptsemploying a correlation method have been made fortheUKwinter floods [6]. In order to deepen our under-standing of the causes for extreme circulation regimes,we analyse global atmospheric teleconnections due toanomalous low-frequency ACT SST patterns and testthemagainstNCEPReanalysis [14].

We find that the success in reproducing the modelsthermodynamic response is associated with a less con-fident result in the models dynamic response. Withreference to the current transient climatological mean interms of SSTs, we find that the likelihood for a wetterwinter over England occurring is not changed whenusing forecast SSTs. Observed SSTs are required to iden-tify the models dynamical response. We show, however,that we can detect event-specific dynamic signals in caseswhere strongly anomalous low-frequency SST patternsare present (mainly tropical and Northern Pacificocean), exemplifiedby amodel analysis for the current ElNiño. Since climate modes such as El Niño can robustlypredicted ahead of the winter season, associated globalatmospheric teleconnections canbedetected aswell.

In section 2, we introduce the methodology, fol-lowed by the presentation and discussion of the resultsin section 3.We concludewith section 4.

2. Setup andmethodology

We utilize the W@H distributed volunteer computinginfrastructure to attribute changes in risk of extremeevents. The HadAM3P AGCM provides the boundaryconditions for the HadRM3P RCM with an Europeandomain on a rotated lat-lon grid [17].

The setup necessary to validate the real-time attribu-tion capability consists of four sets of ensemble simula-tions: two under current climate forcings (ACT) usingobserved SSTs/SICs (OSTIA [9]) and seasonal forecastSSTs/SICs (GloSea5) and two under natural conditions(NAT) for the two different sets of SSTs. For ACT, theIPCCAR5RCP4.5 forcing scenario is used and forNAT,the IPCC historical forcing for 1900 is used.While thereis only one realization of observed SSTs, the seasonalforecast SSTs are in itself an ensemble obtained by cou-pled climate models. The atmospheric initial conditionsare provided from previous model experiments toensure consistent spin-up. Variability is introducedusing initial condition perturbations. In order to quan-tify the dynamic model response, a AGCM climatologysimulation over the 1986–2014 period (CLIM) withcorresponding observed forcing conditions (historicalforcing until 2005, IPCC AR5 RCP4.5 forcing after-wards) and prescribed OSTIA SSTs and SIC was run.

2

Environ. Res. Lett. 11 (2016) 064006

We use 200 model ensemble member for each year inthis case.

GloSea5 is a seasonal forecast product derived fromthe UKMet Office HadGEM3GCM [33], coupled withnucleus for European modelling of the ocean (NEMO)ocean [16]. Climate forcings are set to observed valuesfor the hindcast period up to 2005 and follow the IPCCAR5 RCP4.5 scenario afterwards, consistent with ourHadAM3PACT and CLIM setup. It consists of 42 fore-castmembers, corresponding to the forecast runsmadein the three weeks (twice daily) before the latestmonthly seasonal prediction is issued publicly online.The forecast members are disseminated by the UKMetOffice as a monthly data package which also includes∼12×14 yr (1996–2009) of hindcast ensemble simu-lations for SST for the forthcoming sixmonths.

Being a forecast product froma free running coupledocean-atmosphere GCM ensemble prediction system, itis susceptible to deviations from the atmospheric meanstate after initialization, which, averaged over an ensem-ble of model runs, lead to systematic SST forecast biasesthat we have to account for. The UK Met Office hasdeveloped its own dedicated bias correction method,using the hindcast ensemble following the proceduredescribed in [1, 15]. End users then have to develop theirown bias correction method. We apply a slightly simpli-fied method (no noticeable difference when comparedwith the UK Met Office method), using the hindcastensemble average SST for each year and each month tothen calculate the difference with observed OSTIA SSTfor the same month. In the final step, the multi-yearaverage SSTdifference is applied to the forecast ensemblemember of each respective month. For our test case, theEngland winter floods 2013/14, the GloSea5 forecastissued in November 2013 for December 2013 to May2014 is used. In figures A1(a) and (c), the OSTIA andGloSea5 SST anomalies for January 2014, relative to the1996–2009 average, are plotted. ΔSST between OSTIAand GloSea5 (essentially the forecast error) is shown infigure A3(d). The SST anomalies and the forecast errorfor December and February are very similar in spatialextent andmagnitude (not shown). Sea ice extent used inour GloSea5 ACT simulations is directly derived fromSSTs. If the grid box average SST is<271.4 K, this box isassumed tobe coveredwith sea ice.

The NAT SST patterns are calculated by removingthe modelled anthropogenic warming from theobserved and seasonal forecast SSTs. To achieve that,we use 11 GCM simulations from the Coupled ModelIntercomparison Project phase 5 (CMIP5) archive,averaging and subtracting the monthly climatologies(1996–2005) of the ACT (Historical) and the NAT (His-toricalNat) simulations. The resulting anomaly patternsrepresent 11 estimates of the possible impact of humanGHG emissions on SSTs. They are used to generate theNAT SSTs and to run the associated NAT experiments,alongwith pre-industrial atmosphericGHGconcentra-tions. In figures A1(d)–(l), the NAT patterns fromthe multi-model-mean (figure A1(d)–(f)), GISS-E2-H

model (figures A1(g)–(h)), and CCSM4 model (figuresA1(j)–(k)) are shown. See also figure S3 in [26]where all11 NAT patterns are provided. HadISST1 [24] linearSST trend (1870–2015) is shown for context as well(figureA1(b)).

3. Results

We focus on the comparison of five RCM ensemblesforced with OSTIA and GloSea5 SST/SIC, respec-tively. For the analysis, ∼17.000 ACT and 11 times∼10.000 NAT runs with OSTIA SSTs, ∼12.000 ACTand 11 times ∼8.000 NAT runs with GloSea5 SSTs,and∼6000ACTCLIM simulations forcedwithOSTIASSTs are used. We confine the attribution study toJanuary 2014 as it was the wettest such month in theobserved record then (5.2 mm d−1 on average over 8selected Southern England weather stations extractedfrom the UK Met Office digital archive), approxi-mately 10% above the previous record, whereas DJFwas wettest only by a small margin. For the evaluationof theCLIMexperiment, we use all wintermonths.

We investigate the changes in return times ofmonthly mean precipitation and near-surface temper-ature due to anthropogenic forcing in January 2014(figures 1(a) and (b)), followed by the analysis of thelarge-scale dynamic setup. We use a MSLP index overthe North Atlantic (NA) as well as a meridional MSLPgradient west of the UK in an attempt to disentangle thecontribution of changes in thermodynamic and large-scale dynamic processes (figure 1(c) and (d)). The analy-sis is supported by a model validation exercise withregard to known atmospheric teleconnections inresponse to anomalous low-frequency SSTpatterns.Ourstudy region over Southern England is defined from 50°to 52°Nand6.5°Wto2° E in theRCM.

3.1. Thermodynamic/anthropogenic responsePrecipitation and near-surface temperature are themost common metrics to estimate the change inatmospheric thermodynamics in response to any givenchange in external forcing. For January 2014, theobserved average rainfall of 5.2 mm d−1 has anestimated return time, based on all available observedNDJF data, of 85 yr (figure 5(c) in [26]). Using ACTand NAT simulations from the OSTIA RCM ensem-ble, our best estimate for the change in risk of a(natural) 1-in-a-100-year January rainfall event tooccur is an increase of 42% (now a 1-in-a-70-yearevent;figure 1(a)). IndividualNATensemblemembersshow a wide range between no change (GISS-E2-H)and 160% increase (CCSM4). Using CLIM results(relative to 1986–2011 climatology), [26] found thatthe modelled January 2014 precipitation and near-surface temperature anomalies are similar to observa-tions for the top 1% wettest (i.e. 1-in-a-100-yearreturn time) in the OSTIA ACT simulations. Despite anotable wet bias (∼0.4 mm d−1) in the RCM in all

3

Environ. Res. Lett. 11 (2016) 064006

winter months over southern England, this result isevidence for a reliable representation of the temporalprecipitation variability in the model in agreementwith previousmodel evaluation efforts [17].

Consistent with these results, for a 1-in-a-100-yearJanuary rainfall event, our GloSea5 results indicate anincreasing risk of 66% (nowa1-in-a-60-year event;figure1(a)). The return times for precipitation are very similarin magnitude, both in the OSTIA and the GloSea5 ACTensemble. While the pooled GloSea5 NAT ensemblegives a comparable result to OSTIA, individual GloSea5NAT experiments do different responses owing to differ-ent SST patterns. It is exemplified by CCSM4 and GISS-E2-Hwhich show the strongest andweakest signal in caseof OSTIA, respectively. We discuss the cause for the dif-ference in the signal associated with the SST patternsshown infigureA1(g) andA1(j) in section3.4.

For near-surface temperature (figure 1(b)), OSTIAand GloSea5 results provide a comparable and veryrobust attribution result. Albeit being a fairly moderatemonth in terms of observed near-surface temperatureanomaly, January 2014 was still warmer than average(relative to 1981–2010 in theCentral EnglandTimeseries)with a mean temperature of 279.4 K (+1.3 K anomaly).The return time for such an event would have changedfrom 1-in-a-14-year event under pre-industrial condi-tions to a 1-in-a-3-year event under current conditions,which correspondswith amore than four-fold increase inprobability of such a moderately warm January. Theresult for the change in risk is essentially the same usingGloSea5 forecast SST rather than OSTIA observed SST.IndividualNATsimulations show the samemagnitudeoftemperature response to a change in forcing for both, theOSTIAand theGloSea5 experiments.

Figure 1.Return time plot forOSTIA andGloSea5 ACT andNAT experiments for January 2014monthlymean precipitation (a),mean near-surface temperature (b),MSLP at 20°Wand 60° N (c), andMSLP gradient between 57° Nand 51° N (8° W) (d).Individual actual ensembles are represented by grey dots. CCSM4 (light green and light blue, resp.) andGISS-E2-H (dark green anddark blue, resp.) are highlighted as they show the strongest/weakest response in terms of change in return time formonthlymeanprecipitation (a). The pooled natural ensembles are highlighted in brown (OSTIA) and black (GloSea5). The 5%–95%uncertaintyrange is provided for the actual and the pooled natural ensembles.

4

Environ. Res. Lett. 11 (2016) 064006

3.2. Case specific dynamic responseHere, we first analyse the changing return times ofMSLP for 20° western longitude and 60° northernlatitude (figure 1(c)), the grid point coinciding with thecentre of the record low pressure as observed in January2014 (figure 1(a) in [26]) andhence chosen as an index.

In contrast to the OSTIA ensemble (ACT=red;pooled NAT=brown), in the GloSea5 simulations(ACT=yellow; pooled NAT=black)we find that therisk of such low pressure occurring under ACT condi-tions does not change for a 1-in-a-100 year event in aNAT world. The OSTIA forced simulations suggest anincreasing risk of 33%. We obtain more coherentresults when we look at a larger domain over the NAfrom 38° to 62°N and 8° to 32°W (not shown). Inboth, theOSTIA andGloSea5 experiment, the risk for a1-in-a-100-year low pressure situation occurring is notchanged significantly betweenNATandACT results.

Analysing the meridional MSLP gradient to the westof England and Scotland (which is considered less sensi-tive to the selection of the location and at least partlyrepresentative of potential changes in the atmosphericcirculation over the NA region), we find that a 1-in-a-100-year event in terms of pressure gradient strength inthe NAT scenario becomes a 1-in-a-60-year event underACT conditions using OSTIA SSTs (figure 1(d)). WithGloSea5 SSTs, a 1-in-a-100-year event (NAT) becomes a1-in-a-80-year event (ACT). The risk for a wetter winterunder ACT conditions occurring due to a stronger pres-sure gradient is higher in the OSTIA experiment, whichintuitively suggests a stronger dynamic component.Notethat the difference between the changing risk in OSTIAandGloSea5 is not significant though.

3.3. Atmospheric teleconnectionsIt can be shown, as discussed in the appendix, that ourW@H HadAM3P model is able to capture most of theobserved dynamic features in the troposphere inresponse to low-frequency SST changes. While theestablished teleconnections are not strong enough inJanuary 2014 todeterminewhether the rainfall anomalycan be related to case-specific anomalous circulationpatterns in the GloSea5 experiment, such a link appearsto exist when OSTIA SSTs are used (see figures A3 andA4 ). The SST pattern in the NE Pacific bears strongresemblance to the positive phase of the Pacific DecadalVariability/Oscillation (PDV/PDO+), associated withan increased likelihood for the North Atlantic Oscilla-tion (NAO) phase to be positive [23]. Together with anincreased chance for NAO+ conditions, we findpositive 200hPa zonal wind anomalies (more activejetstream) and an increased risk in precipitation overEngland, corroborating the results we found usingOSTIA SSTs in the previous section. Failure to forecastthe warm NE Pacific SST anomaly minimizes theattributable change in circulation patterns over the NA,which is what happened in the GloSea5 forecast forJanuary 2014. As a result, the odds for a winter over the

UK to be wetter than climatological average undercurrent conditions owing to anomalous atmosphericcirculation did not change in our GloSea5 ACTexperiment, which is consistentwithour results above.

While there is evidence thatAGCMs tend to producemore or less blocking as a function of Gulf Streamstrength [19], and hence the SST gradient in the NAregion. Our analysis of the teleconnection strength inHadAM3P suggests that the main cause for the missingNAO/jetstream signal in theGloSea5ACT experiment isthe difference in Pacific SSTs rather than the marginaldifferences between OSTIA and GloSea5 over the NA.That said, the slightly stronger NA SST gradient inOSTIA may well be caused by the Pacific SST anomalyvia a laggedPacificNorthAmericanpattern (PNA)-NAOlink. The MSLP signal over the NA is, however, almostcertainly caused by the direct thermodynamic feedbackto SSTs as warmer SSTs cause more buoyant atmo-spheric conditions,which, onaverage, lowerMSLP.

Noteworthy in this context is that, by construction,the AGCM results (forced SSTs) tend to be damped withregard to higher frequency internal atmospheric chaotic(white) noise due to the fact that thermal air-sea-fluxesare reduced by virtue of disabled air-to-sea heatexchange. In the real world, as well as in fully coupledGCMs such as GloSea5, the high-frequency atmosphericperturbations will interact with SSTs through surfaceturbulent heat fluxes (air-to-sea) and alter the SST pat-terns. This is particularly true over the NA region wherebaroclinicity leads to periods of high cyclonic activityassociated with anomalous atmospheric momentumflux and hence surface wind stress [3]. The highly sto-chastic nature of these processes will inevitably reducethe predictive skill of any set of forecast SSTs. Despitebeing equally affected by strong baroclinic activity, theNorthwest Pacific region ismore conducive to persistentSST patterns (as evident in form of ENSO and PDV) asthe Walker circulation and the anchoring effect of theRockyMountains can lead to constructive or destructiveinterferencebetween tropical and extra-tropical regions.

If, in contrast to DJF 2013/14, stronger SST anoma-lies exist, we can actually identify the case-specificdynamic component with our fast-track attributionmethod. For the ongoing El Niño event of 2015/16, infigure 2 we provide evidence that the dynamic comp-onent can be determined with confidence in response tothe anomalous Pacific SST pattern forOctober 2015.Weanalyse MSLP (figure 2, left column) and 200hPa zonalwind (figure 2, right column) in HadAM3P and NCEPReanalysis (figures 2(b) and (k)).

The El Niño was well forecast by GloSea5 (August2015 forecast; figure 2(h); compare with OSTIA infigure 2(e)), which should be expected given the lead timeof sub-surface changes in the position of the oceanic ther-mocline due to anomalous West Wind Bursts monthsahead of the actual event. Since extreme events are oftenlinked to anomalous circulation patterns caused by ElNiño or La Niña, we are hence able to estimate theincreased risk of any given extreme event during such

5

Environ. Res. Lett. 11 (2016) 064006

anomalies. The results also suggest that the signal of anypotential change of dynamics due to climate change isvery small compared to the signal associated with ENSO.In fact, the change in risk for an extreme event occurringowing to case-specific dynamics is very similar in ourACTandNATexperiments (true for both SST scenarios).Notably, even the Reanalysis for October 2015 showsmost teleconnection features typical for El Niño, despitebeing a snapshot of only onemonth (figures 2(b) and (k)).

3.4. Implications for the risk assessmentBasedonourfindings so far, herewe summarizewhetherand towhat extent we can quantify changes in likelihoodfor an extreme event occurring as a results of aparticularly anomalousdynamic state of the atmosphere:

• Correlation coefficients between tropical or NorthPacific SSTs with MSLP and 200hPa zonal windanomalies (see figure A2) can provide a roughestimate of expected circulation changes as a func-tion of anomalous SST patterns. With OSTIA, theNorth Pacific SST anomalies in January 2014 arelarge enough to force a dynamic model response

over the NA region. It emerges both at surface level(MSLP) and jetstream level.

• In case of theGloSea5 experiment (ACT), the forecastSSTs do not link strongly with the low-frequencyclimate modes and the large-scale dynamics overEngland in January 2014, as the positive SST anomalyin thePacific is considerablyunderestimated. Inotherwords, the atmospheric dynamic features triggeredby the anomalous NE Pacific SSTs are absent in themodel in this case which hampers any attempt toquantify the attributable dynamic component.

• As far as the predictable component with forecastSSTs in more general terms is concerned [20], wefind that the associated attribution skill of GloSea5for PDV-related (Northern Pacific) climate modesis very limited, whereas for ENSO-related (Equator-ial Pacific) climatemodes it is notably better.

• A promising candidate to quantify the large-scaledynamics impact is the comparison of return times ofprecipitation from ACT and NAT experiments withthe model climatology of the most recent decade

Figure 2. Left column:W@HHadAM3PMSLP anomaly forOctober 2015 (reference period 1986–2014) for theOSTIAACT (a),OSTIANAT (d), GloSea5ACT (g) andGloSea5NAT (j) experiment.Middle column: correspondingNCEPReanalysis 200hPa zonalwind (b) andMSLP anomaly (k) forOctober 2015. In between the observedOSTIA (e) andGloSea5 forecast SST anomalies (h; forecastissued inAugust) that we used to force themodel experiments (naturalizedwithCMIP5 SST patterns inNAT cases). Right column:same as left columbut for 200hPa zonalmodel wind anomalies forOctober 2015.

6

Environ. Res. Lett. 11 (2016) 064006

(derived from the CLIM experiment). The dynamiccomponent is thedifference in return timeofprecipita-tion between CLIM and ACT, whereas the thermo-dynamic component is the difference between CLIMand NAT (expressed in terms of return times in eithercase). For the UK floods, 20%–30% of the increasedrisk can be attributed to case-specific atmosphericcirculation anomalies in theOSTIAACTexperiment.

• The duration of an extreme weather event is crucialwhen attributing a dynamic component. Dailyevents (<3 d) are much less affected by anomalouscirculation regimes, hence short extremes are lessattributable to atmospheric circulation changes. Asfor monthly or longer events, the comparison withsimilar circulation flow regimes can reveal robustattributable results regarding the dynamic comp-onent [6]. As demonstrated in [6], investigating DJFand 10 d maximum rainfall over the UK during thewinter 2013/14, the change in likelihood of an eventoccurring in response to anthropogenic/thermody-namic influences can vary by an order ofmagnitude.

• While the El Niño case of October 2015 suggests anegligible role for event-specific atmospheric circu-lation anomalies over the UK on the basis of ourresults (anomalous ACT and NAT MSLP and jet-stream patterns are very similar in all experiments;see figure 2), anthropogenic warming is believed tocause (forced) jetstream and circulation patternshifts to some extent [7, 11].

There is one last point to reiterate. As mentionedin section 3.1, the NAT results differ significantlybetween different SST patterns. Figures A1(d)–(l) sug-gest that these patterns vary in the same order as thedifference between OSTIA and GloSea5. They mayamplify or dampen the signal of a particular climatemode. They may also introduce a stronger or weaker(counterfactual) ocean cooling signal in general. Forexample, GISS-E2-H (figure A1(g)) amplifies the PDV+ signal in OSTIA in January 2014 (i.e. more pre-cipitation due to enhanced dynamic feedback), whileCCSM4 (figure A1(j)) represents the coldest NATexperiment (i.e. much less precipitation due to greatlyenhanced thermodynamic signal). The fact that we canidentify and disentangle internal dynamic and forcedthermodynamic causes when it comes to explainingthe difference in the attribution result of individualNAT experiments (due to their characteristic CMIP5-based SST pattern), further increases the confidence inour results, despite the apparent increase in uncer-tainty expressed in terms of ensemble spread.

4. Conclusions

With the new approach for event attribution intro-duced in this paper, we show that the coupled GCM-RCM ensemble forced with forecast SSTs and SIC

reproduces the attribution result for temperature andprecipitation for the England winter flood case, madewith the model forced with observed SSTs and SIC.This justifies the suitability of the approach in anoperational attribution framework.

When separating the case-specific contributiondue to anomalous large-scale dynamic weather pat-terns, we find that the experiments with forecast SSTsdo not give the same answer as those with observedSSTs if the Pacific is not forecast well. The additionalanalysis of the recent El Niño test case, however, doesdemonstrate that we can detect potential internaldynamic contributions to a extreme weather caseunder conditions where SSTs are better predicted. Wehave also shown that the model reproduces most rele-vant global teleconnection patterns in response to amajor Pacific ocean climate mode, re-emerging in theensemblemean even on amonthly basis.

Guidance regarding the reliability of the attributionresult, particularly over theNA region, is provided by thespread of the individual NAT ensemble experiments.Their impact on the results was demonstrated and couldbe attributed either to thermodynamic or dynamic cau-ses based on simple physical considerations and the ana-lysis of the teleconnections. Changing SST patterns causechanges in the large-scale dynamics, whereas larger SSTanomalies cause a stronger response in formof increasedor reduced rainfall.

For the main analysis of the UKwinter floods 2013/14, we found that the contribution to the change in riskof the event by changes in the internal case-specificatmospheric dynamics are not reliably detected usingforecast SSTs. However, this should not invalidate ourrapid attribution result as the thermodynamic changesare themain cause of the increase in risk. This is also truefor any forced dynamic contribution due to climatechange. Observed SSTs continue to be required in orderto gain a comprehensive understanding of the drivers ofthis particular extremeprecipitation event.

We also evaluated the model fidelity with regard toreproduction of atmospheric teleconnection patterns.We found that the dynamic contribution to an extremeevent can be detectedwhen particular SSTmodes exceeda certain thresholds. These findings provide very usefuladditional guidance regarding the appropriate use of SSTdata tomake predictions about changing risks for certainextreme events, namely those that are related to changesin the atmospheric circulation.

Apart from the standard model validation whichconcerns the representation of precipitation variabilityand magnitude [17, 18], the fact that HadRM3P simu-lates scaling factors for rainfall with temperature overthe UK that are in agreement with theoretical con-siderations (Clausius–Clapeyron scaling in absence ofchanging atmospheric dynamics), observations [10],and high-resolutionmodel simulations [5] is reassuringwith regard to our attribution statement. Nonetheless,biases are still present in themodel. With a catalogue ofpre-generated global bias data for relevant metrics, we

7

Environ. Res. Lett. 11 (2016) 064006

will make sure that we are realistic in what is achievablewithin the realm of our analysis framework. For exam-ple, there may be structural model uncertainty relatedto underestimated dynamical responses due to missingtroposphere–stratosphere interaction as observed inlow-top models [25]. We may also discover that forcedchanges owing to shifts in atmospheric circulation pat-terns in response to anthropogenic forcing are alreadycontributing significantly to any risk assessment.

Given that there are often conflicting messages andspeculations in the immediate aftermath of damagingweather events about whether there is a link to climatechange or not, we have demonstrated here that our newmethod can deliver robust attribution statements as havebeen frequently requested [29]. It provides a uniqueopportunity to assess a multitude of extreme weatherevents in real-time, which, if carefully communicated,has the potential to provide scientific evidence in adebate that otherwisewould bedominated byopinion.

Acknowledgments

We thank A Scaife and three anonymous referees fortheir valuable comments which helped to improve themanuscript.Wewould like to thank our colleagues at theOxford eResearch Centre: A Bowery, J Miller and DWallom for their technical expertise. We would also liketo thank all of the volunteers who have donated theircomputing time to climateprediction.net and W@H.The data for this paper are available by request from thecorresponding authors.

Appendix

Given the non-significant nature of changes in MSLPin both, the GloSea5 and OSTIA experiment, here weanalyse themodel response to anomalous SST patternsassociated with major climate modes. In figure A1, theresulting SST anomalies after ‘naturalizing’ the

Figure A1. Left column: AnalysedOSTIA SST for January 2014 as used for theACT experiment (a). OSTIAminus the differencebetweenACT andNATCMIP5Multi-Model-Mean SSTs for January 2014 as used in theNAT experiments (d). Same as forMMMin(b) but withGISS-E2-H (g) andCCSM4 SSTs (j).Middle column:HadISST1 observed linear SST trend from1870-2015 (b).MMM(e), GISS-E2-H (h) andCCSM4 (k)ACT-NAT SSTpatterns before subtraction from the observed/forecast SSTs. Right column:Forecast GloSea5 SST for January 2014 as used in the corresponding ACT experiment (c). GloSea5minus difference betweenACT andNATCMIP5MMM (f), GISS-E2-H (i) andCCSM4SSTs (l) for January 2014 as used in theGloSea5NAT simulations. GISS-E2-H andCCSM4 are at the high and the low end of theOSTIANAT simulations in termof change in return time of (extreme) precipitationevents in comparison to theOSTIAACT experiment.

8

Environ. Res. Lett. 11 (2016) 064006

observed SSTs are shown to put the variability incontext of the climate mode variability. This analysisenables us to put the results over the NA region into aglobal perspective and will also be able to determinethemodels fidelity to reproduce observed atmosphericvariability. NCEP Reanalysis is used as a surrogate forobservations. It should be noted that all modes of NHatmospheric variability (e.g. maximum of Jet position,blocking frequency and duration, NAO power spec-trum) are well represented [18]. There are regionswhere the model does have a bias with respect toobservation, [17] but these biases tend to have thesame magnitude in ACT and NAT model simulationswhich is why we are confident that existing biases donot compromise our attribution statements.

We focus on the anomalies of MSLP, 500hPa Geo-potential and 200hPa zonal wind to the two modes ofthe PDV/PDO. PDV is thereby defined as the leadingPrincipal Component (PC1) of North Pacific monthlySST variability. We have sub-divided the climatology

period into years of the positive (PDV+; 10 yr) and thenegative PDV phase (PDV–; 12 yr), respectively. Foreach year and month, we use an ensemble of 200HadAM3P global model members. The strategyenables us (1) to identify regions where we can attri-bute extreme events more reliably and (2) to deter-mine the magnitude of the atmospheric modelresponse in comparisonwith observations.

In figure A2 (left and middle column), the correla-tion of OSTIA (figure A2(a)) and ERSSTv4 SSTs(figure A2(b)) with detrended NE Pacific SSTs (180°–240° E and 30°–60° N) is shown for the winter season(DJF). The correlation of global with NE PacificMSLPand global with NE Pacific 200hPa zonal wind speed isshown in figure A2(c) (NCEP)/figure A2(d) (W@H)and figure A2(e)/(f), respectively. Most of the correla-tion patterns found in the Reanalysis are also presentin the model. Differences are most pronounced in theequatorial Atlantic (model shows stronger couplingwith the NE Pacific SSTs) for MSLP and south of 30° S

Figure A2. Left column: correlation coefficient between SSTs inNEPacific region (180°–240° Nand 30°–60° N)withOSTIA SSTs (a).Correlation between PNE SSTs andNCEPReanalysisMSLP (c) and zonal wind speed at 200hPa (e). Right column: same as (a) but forERSSTv4 SSTs (b). Same as (c) and (d) but forMSLP (e) and zonal wind speed speed (f) fromW@HHadAM3Pmodel. Referenceperiod is 1986—2014 andDJFmean is shown in all plots.

9

Environ. Res. Lett. 11 (2016) 064006

for both, MSLP and 200hPa zonal wind. The correla-tion is stronger in Reanalysis over North America andthe NA region. Dynamically, the link is established bybaroclinic waves forming the NA storm tracks inresponse to geopotential height anomalies associatedwith negative PNA [23], which projects on the PDV+pattern. As a result, the MSLP response to PDVbetween model and Reanalysis is very different overthe NA as evident from figure A3(k) (SST pattern),A3(j) (NCEP), and A3(l) (W@H) for PDV–. For PDV+, the problem is less distinct (figures A3(g)–(i)).NCEP clearly shows an increased MSLP gradient(figure A3(g))which is indicative of a downstream linkbetween PNA (negative) and the (NAO; positive)indeed. The model on the other hand shows a muchweaker response and a displacement of the MSLPanomalies tomore equatorial latitudes (figure A3(i)).

Since the NA is the very region where we focuswith our test case, we explore the large-scale dynamicsin more detail. Per definition, the strength of the

MSLP gradient determines the NAO phase, which inturn is dynamically controlled by upper level winds,mainly through large-scale modulations of the normalpatterns of zonal and meridional heat and moisturetransport [12]. More precise, the MSLP anomalies arethe surface manifestation of jetstream anomalies thatcontrol the cyclonic activity (formation of baroclinicEddies) over the NA. Therefore, on average the jet-stream anomalies should coincide with the MSLP.Since low-frequency atmospheric circulation anoma-lies are barotropic, mid-troposphere Geopotential(500hPa) and MSPL anomalies are largely propor-tional [23], we focus on MSLP for the remainder ofthis analysis. One region where 500hPa Geopotentialanomalies do have amuch stronger signal during PDV+ is the northeasternmost corner of the Pacific andWest Canada, associated with a sometimes ‘ridicu-lously resilient ridge’which can cause harmnot only inCalifornia [31].

Figure A3.Top row:mean sea level pressure (MSLP) anomaly for January 2014 (reference period 1986–2012) fromNCEPReanalysis(a) andW@HHadAM3Pmodel forcedwithOSTIA SSTs (c). It is contrastedwith the analysed SST anomalies for January 2014 fromOSTIA (b). The reference period is 1996–2009, which corresponds with the hindcast period ofGloSea5.Middle row:ΔSST betweenOSTIA andGloSea5 forecast SSTs for January 2014 (d). Note the underestimation of positive SSTs inGloSea5 in the PNE region.GloSea5 forecast SST anomalies for January 2014 (e). On the right,MSLP anomaly fromW@HHadAM3Pmodel forcedwithGloSea5SSTs (f). Bottom two rows:DJFMSLP anomaly for all PDVpositive/negative years (1986–2014 reference period) fromNCEPReanalysis (g/j) andW@HHadAM3Pmodel forcedwithOSTIA SSTs (i/l). Between the two the corresponding detrended/non-detrendedOSTIA SST anomaly forDJF during PDVpositive/negative years (h/k).Model experiments are forcedwith actual GHGconcentrations (ACT).

10

Environ. Res. Lett. 11 (2016) 064006

Comparing figures A3(g) and A4(g), the strongestMSLP gradient anomalies are firmly linked to thehighest positive jetstream anomalies. Both Reanalysisand model agree very well over the North Pacific,which is particularly remarkable as we are comparing200 model simulations with one pseudo-observationfor PDV+ years. Yet, the magnitude of the anomaliesis almost identical. In fact, the model ensemble meanconverges very quickly (10 randomly picked ensemblemember are already sufficient). This is also the reasonwhy the magnitude of the anomalies in NCEP andHadAM3P is so similar. 10 yr are enough to filter outalmost all higher-frequency noise. Over the NAregion, the jetstream anomalies show the same generalpattern with preponderance of NAO+ conditionsduring PDV+ winters (PNA–). Lacking a strong NA–MSLP gradient, the 200hPa zonal wind speed in themodel is weaker over the NA as well. Interestingly,model and Reanalysis agree better with regard to jet-stream anomaly strength over the NA for PDV–yearsduring which baroclinicity is reduced and hence theNAO more negative on average (not shown). It is lesssurprising given that the MSLP gradient is similar inmodel and Reanalysis (despite the absolute MSLPvalues being different; see figures A3(j) and (l)).

References

[1] Arribas A et al 2011TheGloSea4 ensemble prediction systemfor seasonal forecastingMon.Weather Rev. 139 1891–910

[2] BindoffNL et al 2013Detection and attribution of climatechange: from global to regionalClimate Change 2013: ThePhysical Science Basis Contribution ofWorkingGroup I to the

Fifth Assessment Report of the IPCC (Cambridge: CambridgeUniversity Press)

[3] BrethertonC S andBattisti D S 2000An interpretation of theresults from atmospheric general circulationmodels forced bythe time history of the observed sea surface temperaturedistributionGeophys. Res. Lett. 27 767–70

[4] Capstick S,Whitmarsh L, PoortingaW, PidgeonN andUphamP2015 International trends in public perceptions ofclimate change over the past quarter centuryWIREs Clim.Change 6 35–61

[5] Chan SC,Kendon E J, Roberts NM, FowlerH J andBlenkinsop S 2015Downturn in scaling ofUK extreme rainfallwith temperature for future hottest daysNat. Geosci. 9 24–8

[6] ChristidisN and Stott P A 2014 Extreme rainfall in theUnitedKingdomduringwinter 2013/14: the role of atmosphericcirculation and climate changeBull. Am.Meteorol. Soc. 96S1–S172

[7] ChristidisN and Stott P A 2016Changes in the geopotentialheight at 500 hPa under the influence of external climaticforcingsGeophys. Res. Lett. 42 10798–806

[8] ChristidisN, Stott PA, Scaife A, Arribas A, JonesG S,CopseyD, Knight J R andTennantW J 2013AnewHadGEM3—a based system for attribution ofweather and climate-related extreme events J. Clim. 26 2756–83

[9] DonlonC J,MartinM, Stark J, Roberts-Jones J, Fiedler E andWimmerW2012Theoperational sea surface temperatureand sea ice analysis (OSTIA) systemRemote Sens. Environ.116140–58

[10] Durack P J,Wijffels S E andMatear R J 2012Ocean salinitiesreveal strong global water cycle intensification during 1950 to2000 Science 336 455–8

[11] HortonDE, JohnsonNC, SinghD, SwainDL,RajaratnamB andDiffenbaughNS 2015Contribution ofchanges in atmospheric circulation patterns to extremetemperature trendsNature 522 465–9

[12] Hurrell JW1995Decadal trends in theNorth AtlanticOscillation: regional temperatures and precipitation Science269 676–9

[13] IPCC2013 Summary for policymakersClimate Change 2013:The Physical Science Basis Contribution ofWorkingGroup I to

Figure A4. Same asfigure A3 but for 200hPa zonal wind speed anomalies. In the bottom row, only the PDVpositive case is shown (g)–(i). JFM rather thanDJF period is used.

11

Environ. Res. Lett. 11 (2016) 064006

the Fifth Assessment Report of the IPCC (Cambridge: CambridgeUniversity Press)

[14] Kalnay E et al 1996TheNCEP/NCAR40-year reanalysisprojectBull. Am.Meteorol. Soc. 77 437–70

[15] MacLachlanC et al 2014Global Seasonal forecast systemversion 5 (GloSea5): a high-resolution seasonal forecast systemQ. J. R.Meteorol. Soc. 141 1072–84

[16] MadecG 2008NEMOocean EngineNote du Pole deModelisation (Paris: Institut Pierre-Simon Laplace (IPSL))

[17] MasseyN, Jones R,Otto F E L, Aina T,Wilson S,Murphy JM,Hassell D, Yamazakif YH andAllenMR2014Weather@home—development and validation of a very large ensemblemodelling system for probabilistic event attributionQ. J. R.Meteorol. Soc. 141 1528–45

[18] Mitchell DM,MasseyN,Haustein K,Harvey B,Davini P,Otto F E L andAllenMR2016The dynamics of extremes:implications for event attributionClim. Dyn. submitted

[19] O’Reilly CH,Minobe S andYoshida AK 2016The influence ofthe gulf streamonwintertime european blockingClim.Dyn.(doi:10.1007/s00382-015-2919-0)

[20] Otto F E L, Boyd E, Jones RG, Cornforth R J, James R,ParkerHR andAllenMR2015Attribution of extremeweatherevents inAfrica: a preliminary exploration of the science andpolicy implicationsClim. Change 132 531–43

[21] Otto F E L,MasseyN, vanOldenborghG J, Jones RG andAllenMR2012Reconciling two approaches to attribution ofthe 2010Russian heat wave J. Geophys. Res. 39 L04702

[22] Pall P, Aina T, StoneD, Stott PA,NozawaT,Hilberts AG J,LohmannDandAllenMR2011Anthropogenic greenhousegas contribution toflood risk in England andwales in autumn2000Nature 470 382–6

[23] Pinto JG, ReyersM andUlbrichU 2011The variable linkbetween PNAandNAO in observations and inmulti-centuryCGCMsimulationsClim.Dyn. 36 337–54

[24] RaynerNA, ParkerDE,Horton EB, FollandCK,Alexander LV, Rowell DP, Kent EC andKaplanA 2003Globalanalyses of sea surface temperature, sea ice, and nightmarineair temperature since the late nineteenth century J. Geophys.Res. 108 4407

[25] Scaife AA et al 2012Clim.Dyn. 38 2089–97[26] SchallerN et al 2016Human influence on climate in the 2014

Southern Englandwinterfloods and their impactsNat. Clim.Change in press (doi:10.1038/nclimate2927)

[27] SchallerN,Otto F E L, vanOldenborghG J,MasseyN andSparrow S 2014The heavy precipitation event ofMay–June2013 in the upper danube and elbe basins (in ‘Explainingextremes of 2013 from a climate perspective’)Bull. Am.Meteorol. Soc. 95 69–72

[28] StainforthDA et al 2005Uncertainty in predictions of theclimate response to rising levels of greenhouse gasesNature(doi:10.1038/nature03301)

[29] Stott PA, AllenM,Christidis N,Dole RM,HoerlingM,HuntingfordC, Pall P, Perlwitz J and StoneD2013Attribution ofweather and climate-related eventsClimateScience for Serving Society (TheNetherlands: Springer)pp 307–37

[30] Stott PA, StoneDA andAllenMR2004Humancontribution to the European heatwave of 2003Nature 432610–4

[31] SwainDL 2015A tale of twoCalifornia droughts: lessonsamidst recordwarmth and dryness in a region of complexphysical and human geographyGeophys. Res. Lett. 429999–10003

[32] TrenberthKE, Fasullo J T and Shepherd TG2015Attributionof climate extreme eventsNat. Clim. Change 5 725–30

[33] WaltersDN et al 2011Themet office unifiedmodel globalatmosphere 3.0/3.1 and JULES global land 3.0/3.1configurationsGeoscientificModel Dev.s 4 919–41

12

Environ. Res. Lett. 11 (2016) 064006