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

Download details:

IP Address: 182.73.193.34

This content was downloaded on 15/06/2015 at 11:00

Please note that terms and conditions apply.

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

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

2015 Environ. Res. Lett. 10 064011

(http://iopscience.iop.org/1748-9326/10/6/064011)

Home Search Collections Journals About Contact us My IOPscience

Environ. Res. Lett. 10 (2015) 064011 doi:10.1088/1748-9326/10/6/064011

LETTER

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

IzuruTakayabu1, KenshiHibino2, Hidetaka Sasaki1, Hideo Shiogama3, NobuhitoMori4,Yoko Shibutani5 andTetsuya Takemi4

1 Meteorological Research Institute, Tsukuba, Ibaraki 305-0052, Japan2 University of Tsukuba, Tsukuba, Ibaraki 305-8577, Japan3 National Institute for Environmental Studies, Tsukuba, Ibaraki 305-8506, Japan4 Disaster Prevention Research Institute, KyotoUniversity, Uji, Kyoto 611-0011, Japan5 Tottori University, Tottori 680-8550, Japan

E-mail: takayabu@mri-jma.go.jp

Keywords: typhoon, haiyan, storm surge, worst case scenario, climate change

Supplementarymaterial for this article is available online

AbstractEffects of climate change on theworst case scenario of a storm surge induced by a super typhoon in thepresent climate are investigated through the case study of TyphoonHaiyan.We present the results ofour investigation on super-typhoonHaiyan by using a super high resolution (1 kmgrid) regionalmodel that explicitly handles cloudmicrophysical processes. As the parentmodel, we adopted theoperational weekly ensemble experiments (60 kmgrid) of the JapanMeteorological Agency, andcompared experiments using sea surface temperatures and atmospheric environmental parametersfrombefore the beginning of anthropogenic climate change (150 years ago)with those using observedvalues throughout the typhoon.Wewere able not only to represent the typhoon’s intensity but also toevaluate the influences of climate change onworst case storm surges in theGulf of Leyte due to atyphoonwith high robustness. In 15 of 16 ensemble experiments, the intensity of the simulatedworstcase storm in the actual conditions was stronger than that in a hypothetical natural conditionwithouthistorical anthropogenic forcing during the past 150 years. The intensity of the typhoon is translatedto a disastermetric by simulating the storm surge height by using a shallow-water long-wavemodel.The result indicates that theworst case scenario of a storm surge in theGulf of Leytemay beworse by20%, though changes in frequency of such events are not accounted for here.

1. Introduction

Human activities have changed the global climate andhave affected some extreme weather events (Bindoffet al 2013). A typhoon (i.e., a tropical cyclone or TC) isone of the most potentially destructive extremeweather events. Estimation of the power dissipationindex (cube of the maximum surface wind speedintegrated for the whole lifetime of the event), ameasure of the potential destructiveness of a TC overits lifetime, has already shown that TCs have becomestronger in the past thirty years (Emanuel 2005).Moreover, the increase in this index is highly corre-lated with the observed increasing trend in sea surfacetemperature (SST) during the same period. However,

studies disagree about whether changes in TC activitycan be attributed to human influence, owing toinsufficient observations and physical understanding(Bindoff et al 2013).

There is increasing interest in whether not onlylong-term trends of extreme events but also the char-acteristics of specific recently observed extreme eventscan be attributed to external drivers and natural cli-mate variability. Simulations to estimate how pre-dicted sea level increases up to 2050 or 2100 willchange the probability of future inundations in theeastern coast of the United States on the scale of thatcaused by Hurricane Sandy (Sweet et al 2013) haveindicated that the return period of a similar inunda-tion is likely to decrease.

OPEN ACCESS

RECEIVED

13November 2014

REVISED

2May 2015

ACCEPTED FOR PUBLICATION

27May 2015

PUBLISHED

11 June 2015

Content from this workmay be used under theterms of theCreativeCommonsAttribution 3.0licence.

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

© 2015 IOPPublishing Ltd

Pall et al (2011) proposed a new approach forquantifying the roles of external drivers and naturalvariability on specific extreme events by performingtwo types of ensemble simulations with an atmo-spheric general circulation model (AGCM). One tookinto account anthropogenic changes in the well-mixedgreenhouse gas (GHG) concentration and the otherdid not. The first was forced by historical anthro-pogenic and natural forcing factors and observedSSTs, whereas in the second; historical changes inGHG concentrations and estimates of their effect onSST were omitted. Then, by comparing the results,they inferred how human activity influenced the like-lihood and intensities of events. This approach of Pallet al (2011) is called Probabilistic Event Attribution(PEA), and has been applied in some research ofrecent extreme weather and climate events, such asheat waves and river floods (Otto et al 2012, Christidisand Stott 2014, Shiogama et al 2014, Wolskiet al 2014).

The aim of this study was to estimate a robust sig-nal of climate change in increasing the severity of acoastal hazard by Typhoon Haiyan, as an example of aworst case scenario in the present climate (e.g. Moriet al 2014, Lin et al 2014). Here we focus on the ‘worstcase scenario’ in order to assess a physical upperbound that spawns disasters; we do not take intoaccount the frequency of its occurrence. For this pur-pose, we performed two types of ensemble simulationsof Typhoon Haiyan. The first used observed SSTs,atmospheric properties and GHG concentrations(ALL simulations), and the second ensemble usedcounterfactual natural external conditions (NATsimulations). Although the idea of our approach ispartly based on that of PEA, it should be noted that wedo not intend to investigate changes in the frequencyof typhoons. To evaluate possible amplification effects

due to climate change, we adopted ensemble predic-tion (Saito et al 2010) and applied dynamical down-scaling from a global model (WEP: Weekly EnsemblePrediction system of Japan Meteorological Agencywith the equivalent grid point resolution, 60 km)(Sakai 2009, Saito 2011) to a high-resolution WeatherResearch and Forecasting (WRF) model (Skamarocket al 2008) at the horizontal grid spacing of 1 km thatwas able to represent the strength of a category 5 TC(Gentry and Lackmann 2010, Kanada et al 2012).Withthe dynamical downscaled WRF 1 km models, weselected 16 experiments out of all 51 ensemblemembers.

When we discuss the activity of typhoons, fre-quency of genesis and tracks are important, as well asmaximum intensity. To discuss the disaster preven-tion, we should handle these three metrics jointly.However, for the purpose of disaster mitigation, it isworth discussing at least the change in maximumintensity of the worst typhoon. Because TyphoonHai-yan was the most powerful typhoon to make landfallto date (Mori et al 2014), it is very important to discusshow the worst case event would change under thewarming climate environment.

The paper has the following structure. The dyna-mical and ensemble downscaling procedure has beenintroduced in section 2. Section 3 describes the repre-sentativeness of the model typhoon, and also a signalof climate change appeared in the downscaling inte-grations. In section 4we present some discussions, andsection 5 is the conclusion.

2.Methods

The dynamical downscaling procedure adopted in thisstudy is shown schematically in figure 1. The parentmodel was the Japan Meteorological Agency (JMA)

Figure 1. Schematic diagramofmulti-time ensemble PGWDmethod.

2

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

weekly ensemble prediction model (WEP), which isthe operational model used for weekly forecasts(Sakai, 2009). WEP, which is a global spectral model(GSM) with an equivalent grid point resolution of60 km, was used to obtain the initial and boundaryconditions, but the initial perturbations were calcu-lated with a lower resolution GSM (equivalent hor-izontal grid resolution is 200 km). The ensemble had51 members, prepared using the singular vectormethod (Buizza and Palmer 1995). Time integrationwas from 12 UTC on 4 November 2013 until 12 UTCon 11November 2013.

To avoid the occurrence of a shock at the lateralboundary, we used a cascade of nested models, down-scaling finally to aWRFmodel with a 1 km grid resolu-tion. The first downscaling step was to a regionalmodel with a 20 km grid resolution (NHRCM20)(Sasaki et al 2011). To reproduce the TC track of theparent model in the downscaled model, we adoptedspectral nudging above 15 000 m. The next down-scaling step was to the regional model with a 5 km gridresolution (NHRCM05), and then two-way nestingfrom 3 kmWRF to 1 kmWRFmodels. In NHRCM05,the Kain-Fritsch convective parameterization scheme(Kain and Fritsch 1990, 1993) is used, and in theWRFmodels, cloud microphysical processes are explicitlyhandled.

Dynamical downscaling using 3 km/1 km WRFmodels has been done, but not for all ensemble mem-bers of WEP calculations because of the limitation ofthe computer resources. To cover the wide range ofensemble members, the following three cases areselected. They are, Casem02: where the track is nearestto the observed track (best track) around Leyte andSamar Island (124.8E). Case m11: where the mini-mum central pressure (MCP) of the typhoon calcu-lated in NHRCM05 is lowest among all. Case p18:where the MCP of NHRCM05 is the highest amongall. Because the timing of starting of the integrationalso affectsMCP, the following three times are selectedas the initial conditions. They are, #1000: 12UTC, 5,November, #1001: 18UTC 5, November, and #1002:00UTC 6,November. Thus we have 9 experiments fol-lowing the above conditions. Furthermore, to estimatethe storm surge around Tacloban caused by TyphoonHaiyan, we select all cases where the typhoon passedaround Leyte Island (125E), within the range of 50 kmto the best track data. Seven cases (05 m, 12 m, 15 m,21 m, 25 m, 06 p, 09 p) are found other than Casem02. For these seven cases, only #1001 (initial condi-tion at 18UTC 5, November) has been calculated. As aresult, we calculate 16 cases by using 3 km/1 kmWRFmodels.

For the NAT ensemble simulations, we appliedPseudo Global Warming Downscaling (PGWD), inwhich the boundary conditions were assumed to be alinear coupling of the WEP data and the differencecomponent of the climate change of air temperature

between the middle of the nineteenth centuryand 2013.

To estimate the hypothetical natural SST, weremoved linear trends in the monthly data from theHadISST (Hadley Centre Sea Ice and Sea Surface Tem-perature) data set (Rayner et al 2003) for 1870–2012from the original SST data used in the ALL runs(Christidis and Stott 2014, Shiogama et al 2014). Thedifference between ALL-SST and NAT-SST in the tar-get area (100°–180°E, 5°S–25°N) was between 0.2 Kand 0.8 K (as shown in figure S2). To remove anthro-pogenic atmospheric warming, we subtracted the dif-ferences of atmospheric temperature between two100-member ensemble simulations of the MIROC5AGCM (Shiogama et al 2014, Watanabe et al 2010),performed with and without human influence, fromthe WEP data. The hypothetical anthropogenic SSTsignals of theMIROC5 ensemble are the same as in ourregional climate model runs. We averaged the atmo-spheric temperature differences over the whole targetarea during November of 2010–2012 across 100ensemble members. The averaged tropospheric andstratospheric temperature differences were +0.5 K and–1.5 K, respectively. We call this effect ‘ATM’. For thedownscaling calculations, we used GHG (CO2, CH4

and N2O) levels in the 1850s reported by the Inter-governmental Panel on Climate Change Fifth Assess-ment Report (IPCC2013).

The Surge-Wave-Tide coupled model (SuWAT)(Kim et al 2008) was forced by the 1 km WRF modelcalculation result. SuWAT is a fully coupled model ofstorm surge and ocean waves based on the non-linearshallow-water equation and spectral wave model, andtakes into account atmospheric pressure, wind stress,and wave radiation stress (wave effects on current).Input data are surface wind and sea level pressure. Weapplied SuWAT to three domains with spatial resolu-tions ranging from 0.1° (D1) to 740 m (D3), with two-way nesting. The astronomical tide was excluded forcomputation and survey for simplicity, although themaximum astronomical tidal range is 0.7 m and wasclose tomeanwater level at the landfall ofHaiyan.

3. Results

3.1. Ensemble simulation of TyphoonHaiyanThe evolution and intensity of Haiyan were wellrepresented in all of the experiments. In WEP, thesimulated typhoon tracks around Leyte and SamarIsland (125°E), on which Tacloban is situated, weredistributed in a band about 500 km wide in thenorth–south direction with its centre north ofTacloban, and the tracks in the downscaled resultsfollowed the WEP result (figure 2). The westward-propagation speed of the typhoon was slower in themodel simulations, compared with the observedtrack (best-track data), and as a result, the TC landfallon Leyte and Samar Island was delayed about a half

3

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

day in the model simulations. This phase lag of thesimulated typhoon is not attributable to the WRFmodel itself, because when Typhoon Haiyan wassimulated with the same WRF 3 km/1 km modelsystem with reanalysis data as the parent model, theresult showed no phase lag (Mori et al 2014). Thisresult suggests that the phase lag in this ensembledownscaling experiment is attributable to the perfor-mance of WEP. Because of this phase lag, however,we compared the evolution of the typhoon in relationto its longitudinal position (figure 3). The compar-ison indicates that in the ALL simulations both

central surface pressure and maximum wind speedagree well with the best-track data, though theintensification of the TC that occurred before itslandfall on Leyte and Samar Island was slightlydelayed. The minimum central pressure (MCP) ofthe simulated typhoon was as high as 906 hPa (inexperimentm02 (#1001)), which corresponds well tothe intensity of the actual TC of 895 hPa as estimatedby the Dvorak method (Dvorak 1975). The rapiddecline of the typhoon strength after its landfall onLeyte and Samar Island was also reproduced by all ofthe ensemble simulations.

Figure 2.Predicted tracks of TyphoonHaiyan, of the down-scaling done byNHRCM05 fromWEP (in broken lines). The thick lineindicates the best track datawith the dots in the interval of six hours.

Figure 3.Time history of TyphoonHaiyan in ALL simulations (16 ensemble experiments) with the 1 kmWRFmodel. (a) Centralpressure; (b)maximumwind speed. Thewestward propagation speed of the typhoon is slightly slower in theWEP simulationscomparedwith the observed best track, so the changes in pressure andwind are shown in relation to longitude. The position ofTaclobane is around 125 E.

4

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

To evaluate how well the experiments couldreproduce Typhoon Haiyan, we used the results ofexperiment m02 (#1001), in which the simulatedtrack was nearest to the observed track (best track).The structure of the eye-wall clouds was representedby cascade downscaling to a 1 km grid scale. Becausethe inner core convective band just around the eye isthe engine of a typhoon, it is necessary to representthe structure of the eye-wall clouds for the sufficientintensification of a typhoon. Compared with weakerTCs, however, the radius of the eye of category 4 or 5typhoons becomes smaller (about 20 km) (Weath-erford and Gray 1988). Therefore, to represent thestructure of the eye-wall clouds, the model musthave a horizontal resolution of less than 10 km, and

use of a model with a 2 km or 1 km grid is betterbecause the simulation can be performed without acumulus convective scheme (Gentry and Lack-mann 2010, Kanada et al 2012). Supplementaryfigure S1 depicts the convective rain bands withinand around the TC core in the 20, 5 and 1 km mod-els. In the coarser resolution models, the eye-wallstructure appears smoothed. To simulate the inten-sity of the disturbance better, the structure withinthe TC’s core must be sufficiently resolved. Compar-ison of our simulation results (experiment m02(#1001)) with the radar reflectivity data for TyphoonHaiyan obtained by Guiuan station (figure 4)showed that WRF 1 km was able to reproduce thestructure around the eye-wall clouds well.

Figure 4. (a) Radar reflectivity at z = 2 km in them02 (1001) experimentwith the 1 kmWRFmodel (dBZ) (11UTCon 11August2014). (b) Radar reflectivity data for Guiuan station (dBZ), elevation angle 1° (0416UTCon 11August 2014).

5

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

3.2.Difference in the development of the TyphoonbetweenNAT andALL conditionsNext, we compared the ALL results with the NATresults obtained by the cascade downscaling methodand found that the influence of the present climatestate in comparison to the hypothetical natural condi-tion on the intensity of the worst case storm wasrobustly indicated. In 15 of the 16 WRF experiments,the MCP was lower in the ALL simulations (meandifference and standard deviation; −6.44 and4.98 hPa) than in the NAT simulations (figure 5(a)).In addition, the maximum surface wind speed wasstronger in 15 of 16 ALL simulations (mean differenceand standard deviation; 2.89 m s−1 and 2.06 m s−1)(figure 5(b)). This intensification has a 1% significantlevel.

To investigate whether the intensity of the worstcase storm was accounted for by the differences of cli-mate states between the real climate in November2013 and the hypothetical natural climate, we used themaximum potential intensity (MPI) following Ema-nuel (1986) to examine the differences in the potentialdevelopment of the typhoon between NAT and ALL.

We estimated the effect of changes in SST and atmo-spheric temperatures (ATM) on the MPI separately(supplementary table S1). The total contribution ofSST and ATM to the MPI difference was −12 hPa(supplementary table S2). The contribution from SSTchanges alone ranged from −17 to −16 hPa, and thatfrom ATM alone ranged from +3 to +4 hPa. The dif-ference in the sign of the contributions from SST andATM is consistent with the results of a previous study(Knutson and Tuleya 2004). Our result suggests thatthe SST change is themain contributor to theMPI dif-ference between NAT and ALL. In the oceanic regionaround the Philippines, the SST was from +0.2 to+0.8 K higher in the ALL simulations compared to theNAT simulations (supplementary figure S2), and thesehigher SSTs can account for theMPI difference.

3.3. Storm surge estimationFinally, we used the nonlinear shallow-water long-wave model SuWAT (Kim et al 2008) to estimateincreases in the water surface elevation due to theworst case storm. We focused on the maximum waterelevation, the storm surge height at Tacloban, whereTyphoon Haiyan caused catastrophic damage. Toestimate a disaster at such a specific point, the largestpermissible bias in the track compared to the best trackis 50 km (empirically estimated from the maximumwind speed radius of the typhoon). Because only tenensemble members satisfied this condition for each ofALL and NAT experiments, we applied SuWAT toestimate the surge at Tacloban to only 2 × 10 experi-ments. The average maximum surge height in the tenD3 experiments was 2.60 m (standard deviation1.36 m) and 2.19 m (standard deviation 1.00 m) inALL and NAT, respectively, an amplification ofaround 20% in ALL (supplementary table S3). Themaximum surge height along the coast, simulated inexperiment m02 (#1001), was 4.27 m in ALL, whereasin NAT it decreased to 3.80 m (figure 6, shown withland inundation survey data (Mori et al 2014, Tajimaet al 2014) which influenced local bathymetry andwind waves). Although there is not real surge data, thereanalysis of maximum surge height shows 5.15 m(Mori et al 2014). While the surge height in ALL isunderestimated in comparison with the reanalysis, thespatial pattern of maximum surge height is similar toreanalysis byMori et al (2014). As the first seichemodein Leyte Gulf increased the amplitude of the stormsurge at Tacloban, the surge height at any specificpoint is very sensitive to the position of the TC track.We estimated the mean maximum surge height usingthe ALL surface wind amplification data (5.82%; thedifference between NAT and ALL) with the NAT trackdata (ALL_Pseudo). The results showed that thedifference in the track between the ALL and NATsimulations had limited influence on the mean max-imum surge height. Thus, the difference in maximumsurge height between them is explained mainly by the

Figure 5.Comparison of (a)MCP and (b)maximum surfacewind speed between theNAT andALL experiments.

6

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

difference in the cyclone intensity. The difference ofmaximum surge height between NAT and ALL is0.47 m and amplification of surge height is muchlarger than surface wind amplification due to non-linear characteristics of momentum transfer betweenatmospheric and ocean interface.

4. Influence of extreme sea level changewithin these 150 years

There are many studies of historical sea level rise(IPCC-AR5, WGI, chapter 13). For example, ChurchandWhite (2011) investigate sea level change from thelate 19th to the early 21st century. Although there wereonly a few tide gauges that operated in themiddle 19thcentury, their papers figure 7 shows some sea levelchange records estimation from several data sources.

From their figure, c.a. 0.25–0.30 m sea level rise isfound and other studies also show similar valuearound this area if we exclude local land subsidence.Wind driven circulation or ENSO signal may alsoproduce sea level change locally, but as shown inWoodworth et al (2008), the time scale caused by suchocean circulation is much shorter than 150 years. TheIPCC Fifth Assessment Report also discusses thevulnerability of coastal regions to other physicalprocesses, such as storm surges at extreme sea level.Changes in the severity of storm surges are some of theexamples of how climate change affects coastalregions, but it is difficult tomake an impact assessmentat particular regions quantitatively. In our paper, thestorm surge height increases at most 0.50 m betweenNAT and ALL experiments, caused by the intensifica-tion of the typhoon which is larger than historical sealevel rise (estimated as 0.25–0.30 m).

Figure 6.Maximumwater surface elevation predicted by using the SuWATmodel (domainD3). Results of experimentm02 (1001) in(a) ALL and (b)NAT. The coloured dots around the coastline show inundation data obtained by a field survey.

7

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

We also examined the influence of SLR of 0.30 m,on the storm surge level with higher mean sea levelcondition. However, the maximum sea surface levelconsidering a higher mean sea level condition didn’tgive significant changes in the storm surge height.Theoretically, the storm surge height is inversely pro-portional to the depth of the bay. Since the water depthnear Tacloban in Leyte Gulf has depth of approxi-mately 10 m, 0.30 mmean sea level rise almost cancelsout the change of total surge height, although thereremain some dependencies on characteristics of thebay and cyclones. The combination of a storm surgeand sea level rise is nonlinear, thus it is important toknow how sea level rise and dynamic effects, wind andpressure surges, will change in the future.

5. Conclusion

Typhoon Haiyan (local name Yolanda), the mostcatastrophic tropical cyclone ever to land in thewestern North Pacific Ocean, struck the Philippineson 8 November 2013. The typhoon and especially thestorm surge in the Leyte Gulf that accompanied itkilled more than 6000 people in Tacloban (Schierme-ier 2013). We conducted ensemble simulations withvery high resolution regional climate models and asurgemodel, and reproducedwell the pressure depres-sion, wind speed and surge level of Typhoon Haiyan,as an example of a worst case scenario. Furthermore,we compared these results with the results of ensemblesimulations of a hypothetical natural event, one with-out human influences, and found that the simulatedworst case typhoon and the accompanying stormsurge in the real condition became worse than those inthe hypothetical natural climate without anthropo-genic forcing. In 15 of 16 ensemble simulations, thetyphoon became stronger than it did in the hypothe-tical natural cases, and the height of the storm surgearoundTacloban increased by around 20%.

When we made the NAT external condition, weomitted anthropogenic warming in air temperaturedifference as well as that in SST. It is well known thatvertical wind shear and mid-level entropy deficit areimportant in determining the intensity of tropicalcyclone (Gray (1975), Tang and Emanuel(2010, 2012)). Although previous studies have sug-gested the weakening of Pacific Walker circulation(reductions of wind shear in the western Pacificregion), and anthropogenic contributions to thisobserved trend, there are substantial uncertainties inobservations and biases in model simulations (Toki-naga et al 2012, DiNezio et al 2013). Supplementaryfigure S3 shows the histograms of vertical wind shearin the 100-member ensembles of ALL-gcm and NAT-gcm runs of MIROC5-AGCM (Shiogama et al 2014).Although anthropogenic factors weaken the verticalwind shear in the ensemble mean (which would

increase the Typhoon intensity (Tang and Ema-nuel, 2010, 2012)), the variances due to internal varia-bility are large. Because of the poor signal-to-noiseratio, we have not subtracted these differences in windshear from the NAT conditions of RCM runs. It mightcause our estimates of anthropogenic influence on theTyphoon intensity to becomemore conservative.

We should also note here, that upper-ocean mix-ing is not accounted for because we do not have asound assumption for prescribing the upper-oceanstructure in natural climate conditions. Vincent et al(2014) insisted that themixing has a negative feedbackon TC intensity for very strong storms, but the influ-ence of such a mechanism is not accounted in ourstudy.

As we introduced in section 1, to assess the disasterrisk reduction caused by the climate change, we hadquite a different approach, rather than considering thedisastermitigation as we discussed in the paper. This isto consider the disaster prevention. For this purpose,we should estimate changes in the frequency of eventoccurrence. However, Hansen et al (2014) suggeststhat it is difficult to estimate the change of likelihoodwith a high confidence level when the sample numbersare not high enough, especially for extreme events. Toincrease the sample number efficiently in cases wherewe are not able to conduct sufficient ensemble experi-ments, we have another option of introducing a sto-chastic hurricane model of the type used in Lin et al(2012). This method imposes a stronger limitation ofthe assumed hurricane structure and intensity than adynamically downscaling approach. Thus, it is indis-pensable to carry out both types of approach jointly, toclarify the disaster risk of extreme events as a whole.

There are uncertainties associated with estima-tions of the effects of anthropogenic signals on SST(Christidis and Stott 2014, Shiogama et al 2014) andattribution analysis results may also be sensitive tomodels used. We showed here that very high resolu-tion regional climate models are necessary to simulatea category 5 typhoon. It should be kept in mind thatmore accurate representation of the typhoon withhigher spatial resolution does not necessary ensuremore reliable estimates of the anthropogenic con-tribution. Inter-comparison studies of very high reso-lution regional climate models should be carried outin which different estimates of anthropogenic signalsare used.

Acknowledgments

This work was supported by the SOUSEI programmesof the Ministry of Education, Culture, Sports, Scienceand Technology (MEXT) of Japan. The Guiuan radardata were supplied by PAGASA, the Philippines, andDr Kubota of Japan Agency for Marine-Earth Scienceand Technology (JAMSTEC) helped us to read the

8

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al

original data. We also acknowledge Dr Fujita of theNumerical Prediction Division of the Japan Meteor-ological Agency (JMA) and Dr Yamaguchi of theMeteorological Research Institute for providing uswith information about JMA's operational weeklyensemble prediction system.

References

BindoffNL et al 2013Detection and attribution of climate change:from global to regionalClimate Change 2013: The PhysicalScience Basis. Contribution ofWorkingGroup I to the FifthAssessment Report of the Intergovernmental Panel on ClimateChange ed T F Stocker, DQin, G-KPlattner,MTignor,SKAllen, J Boschung, ANauels, YXia, VBex andPMMidgley (Cambridge, UnitedKingdom: CambridgeUniversity Press)

Buizza R and Palmer TN1995The singular-vector structure of theatmospheric global circulation J. Atmos. Sci. 52 1434–56

Christidis N and Stott P A 2014Change in the odds of warmyearsand seasons due to anthropogenic influence on the climateJ. Climate 27 2607–21

Church J andWhiteN 2011 Sea-level rise from the late 19th to theearly 21st century Surv. Geophys. 32 1–18

DiNezio PN,Vecchi GA andClement AC 2013Detectability ofchanges in theWalker circulation in response to globalwarming J. Clim. 26 4038–48

DvorakVF 1975Tropical cyclone intensity analysis and forecastingfrom satellite imagery from satellite imageryMon.Wea. Rev.103 420–30

Emanuel KA 1986An air-sea interaction theory for tropicalcyclones: I. Steady-statemaintenance J. Atmos. Sci. 43585–604

Emanuel KA 2005 Increasing destructiveness of tropical cyclonesover the past 30 yearsNature 436 686–8

GentryMS and LackmannGM2010 Sensitivity of simulatedtropical cyclone structure and intensity to horizontalresolutionMon.Wea. Rev. 138 688–704

GrayWM1975Tropical cyclone genesis in theWesternNorthPacificEnvpredrschfac Technical PaperNo 16–75 p 66

HansenG, AuffhammerMand SolowAR2014On the attributionof a single event to climate change J. Climate 27 8297–301

IPCC,Climate Change 2013The physical science basisContributionofWorkingGroup I to the Fifth Assessment Report of theIntergovernmental Panel on Climate Change (Cambridge,UnitedKingdom: CambridgeUniversity Press) p 1535

Kain J S and Fritsch JM1990A one-dimensional entraining/detraining plumemodel and its application in convectiveparameterization J. Atmos. Sci. 47 2784–802

Kain J S and Fritsch JM1993Convective parameterization formesoscalemodels: theKain-Fritcsh schemeTheRepresentation of Cumulus Convection inNumericalModelsedKAEmanuel andD JRaymond (Boston,MA: AmericanMeteorological Society) p 246

Kanada S,NakanoM,WadaA andKato T 2012 Effect of PBLschemes on the development of intense tropical cyclonesusing a cloud resolvingmodel J. Gophys. Res. 117D3

KimS-Y, YasudaT andMaseH2008Numerical analysis of effects oftidal variations on storm surges andwavesAppl. OceanRes.30 311–22

KnutsonTR andTuleya R E 2004 Impact of CO2-inducedwarmingon simulatedHurricane intensity and precipitation:Sensitivity to the choice of climatemodel and convectiveparameterization J. Clim. 17 3477–95

Lin I-I, Pun I-F and LienC-C 2014 ‘Category-6’ supertyphoonHaiyan in global warming hiatus: contribution fromsubsurface oceanwarmingGeophys. Res. Lett. 41 8547–53

LinN, Emanuel K,OppenheimerMandVanmarcke E 2012Physically based assessment of hurricane surge threat underclimate changeNat. Clim. Change 2 462–7

MoriN, KatoM,KimS,MaseH, Shibutani Y, Takemi T,Tsuboki K andYasuda T 2014 Local amplification of stormsurge by Super TyphoonHaiyan in Leyte Gulf, GeoResearchLetters 41 5106–13

Otto F E L,MasseyN, vanOldenborghG J, Jones RG andAllenMR2012 Reconciling two approaches to attribution of the 2010Russian heat waveGeophys. Res. Lett. 39 L04702

Pall P, Aina T, StoneDA, Stott PA,NozawaT,Hilberts AG J,LohmanDandAllenMR2011Anthropogenic greenhousegas contribution toflood risk in England andWales inautumn 2000Nature 470 380–4

RaynerNA, ParkerDE,Holton EB, FollandCK,Alexander LV,Rowell D P, Kent EC andKaplan A 2003Global analysis ofsea surface temperature, sea ice, and nightmarine airtemperature since the late nineteenth century J. Geophys. Res.108 4407

SaitoK, Kuroda T, KuniiM andKohnoN2010Numericalsimulation ofMyanmar cycloneNargis and the associatedstorm surge: II. Ensemble prediction J.Met. Soc. Japan 88547–70

SaitoK 2011Comparison of initial perturbationmethods for themesoscale ensemble prediction systemof themeteorologicalresearch institute for theWWRPBsijing 2008Olympicsresearch and development project (B08RDB)Tellus 63A445–67

Sakai R 2009The one-week ensemble prediction system at JMAAnnual Report of Numerical PredictionDivision, JMA 55117–25 (in Japanese)

SasakiH,Murata A,HanafusaM,Oh’izumiMandKuriharaK 2011Reproducibility of present climate in a non-hydrostaticregional climatemodel nestedwithin an atmosphere generalcirculationmodel SOLA 7 173–6

SchiermeierQ 2013Did climate change cause TyphoonHaiyan?Nature doi:10.1038/nature.2013.14139

ShiogamaH,WatanabeM, ImadaY,MoriM,Kamae Y, IshiiM andKimotoM2014Attribution of the June-July 2013 heat wavein the southwesternUnited States SOLA 10 122–6SweetW, SweetW., Zerbas C., Gill S. and Park J. 2013Hurricane Sandy inundation probabilities today andtomorrow [in Explaining extreme events of 2012 from aclimate perspective]Bull. Amer.Meteor. Soc. 94 S17–20

SkamarockWC,Klemp J B,Dudhia J, Gill DO, BarkerDM,DudaMG,HuangX-Y,WangWandPowers J G 2 2008Adescription of the advaned researchWRF version 3NCARTeck. Note 475 113National Center for AtmosphericResearch, Boulder, Colo. (2008)

TajimaT et al 2014 Initial report of JSCE joint survey on the stormsurge disaster caused by TyphoonHaiyanCoastal Eng. J. 561450006

TangB and Emanuel K 2010Midlevel ventilation constraint ontropical cyclone intensity J. Atmos. Sci. 67 1817–30

TangB and Emanuel K 2012A ventilation index for tropicalcyclonesBull. Atmes.Met. Soc. 92 1901–12

TokinagaH, Xie S-P, Desar C, Kosaka Y andOkumura YM2012Slowdown of theWalker circulation driven by tropical Indo-PacificwarmingNature 491 439–43

Vincent EM, Emanuel KA, LengaigneM,Vialard J andMadecG2014 Influence of upper ocean stratification interannualvariability on tropical cyclones J. Adv.Model. Earth Syst. 6680–99

WatanabeM et al 2010 Improved climate simulation byMIROC5:mean states, variability, and climate sensitivity J. Climate 236312–35

WeatherfordC L andGrayWM1988Typhoon structure as revealedby aircraft reconnaissance: II. Structural variabilityMon.Wea. Rev. 116 1044–56

Wolski P, StoneD, TadrossM,WehnerMandHewitson B 2014Attribution offloods in theOkavango basin, SouthernAfricaJ. Hydrology 511 350–8

WoodworthPL,WhiteN J, Jevrejeva S,Holgate S J, Church JA andGehrelsWR2008Evidence for the accelerations of sea level onmulti-decade and century timescales Int. J.Climatol.29777–89

9

Environ. Res. Lett. 10 (2015) 064011 I Takayabu et al