extreme sea surge responses to climate variability in ... · de valor extremo generalizado, y las...

Extreme Sea Surge Responses to Climate Variabilityin Coastal British Columbia, Canada

Dilumie S. Abeysirigunawardena,∗ Dan J. Smith,† and Bill Taylor‡

∗Neptune Canada and University of Victoria, Canada†Department of Geography, University of Victoria, Canada

‡Environment Canada

This article presents a statistical investigation of the spatial and temporal changes of extreme sea surgesin response to climate variability in coastal British Columbia. The study was based on an investigation ofin situ hourly tide gauge data spanning the interval from 1950 to 2007 at eleven tide gauges in the re-gion. The characteristics of the recorded extreme sea surges were parametrically modeled by the generalizedextreme value distribution while accounting for trends and dependence via climate covariate coefficients.The study confirms that decadal to interdecadal climatic variability is a fundamental element in explain-ing the changing frequency and intensity of sea surges in coastal British Columbia. For instance, the sea-surge response to climate variability impacts on sites along the British Columbia coast were found to befairly synchronous with an increase (decrease) in the magnitude of extreme sea surges in response to warm(cold) El Nino/La Nina-Southern Oscillation (ENSO) conditions. These trends make the flooding riskseven higher during warm ENSO conditions, especially if the global sea levels continue to rise as predictedby climate models. Key Words: climate variability, covariates, extreme sea surges, extreme value analysis, returnperiod.

Este artıculo presenta una investigacion estadıstica de los cambios espaciales y temporales de marejadas extremasen respuesta a la variabilidad del clima en la costa de Columbia Britanica. El estudio se baso en una investigacionin situ de datos horarios de mareografos que abarcan el intervalo de 1950 a 2007, en once mareografos de la region.Las caracterısticas de marejadas extremas registradas fueron modeladas de forma parametrica de la distribucionde valor extremo generalizado, y las tendencias y la dependencia a traves de coeficientes de covarianza delclima fueron tomados en cuenta. El estudio confirma que la variabilidad del clima decadal e interdecadal es unelemento fundamental en la explicacion del cambio de frecuencia y la intensidad de las marejadas en la costa deColumbia Britanica. Por ejemplo, la respuesta del mar a los efectos del aumento de la variabilidad del clima enlos sitios a lo largo de la costa de Columbia Britanica se encontraron a ser bastante sincronizado con un aumento(disminucion) en la magnitud de marejadas extremas en respuesta a oscilaciones de condiciones El Nino / LaNina (ENSO) tibios (frıos). Estas tendencias hacen que los riesgos de inundaciones sean aun mayores durantecondiciones calidas del ENSO, especialmente si el nivel del mar global sigue aumentando segun lo predichopor los modelos climaticos. Palabras clave: variabilidad del clima, covariables marejadas extremas, analisis de valoresextremos, perıodo de retorno.

Extreme sea surges are one of the greatest nat-ural threats to coastal communities in terms ofproperty devastation and lives lost (Murty 1988).

Typically, sea surges (also referred to as the residual sealevels) consist of two meteorological components—the

wind setup and barometric setup—and a thirdconcurrent factor resulting from the thermal expan-sion (contraction) of water due to warmer (colder)than normal climatic conditions (i.e., El Nino/La Nina-Southern Oscillation (ENSO) conditions). Once a

Annals of the Association of American Geographers, 101(5) 2011, pp. 992–1010 C© 2011 by Association of American GeographersInitial submission, July 2009; revised submission, February 2010; final acceptance, July 2010

Published by Taylor & Francis, LLC.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011

CV Impacts on Extreme Sea Surges in Coastal British Columbia 993

Figure 1. The geographical location of coastal of British Columbia, including the Queen Charlotte Islands in northern British Columbia andVancouver Island in southern British Columbia. Also shown are the Pacific region tide gauge stations and the location of Hacate Strait andthe Georgia Strait.

critical mean sea level threshold is exceeded, the eco-nomic, human, and ecological costs of even a smallincrease in extreme sea surges due to climate variabil-ity could be much higher in low-lying coastal commu-nities (Intergovernmental Panel on Climate Control2007). Given this, any evidence of possible links be-tween extreme sea surges and decadal to interdecadalscale climate variability modes is critical when it comesto defining acceptable levels of protection against ex-treme events.

Recent occurrences of extreme sea-surge eventsin conjunction with progressive mean sea-level risehave resulted in significant infrastructure damageto many coastal communities in British Columbia(Walker et al. 2007; Thomson, Bornhold, and Maz-zotti 2008). These events have raised concerns aboutour lack of understanding about the relations be-tween extreme sea surges and the changing climatein British Columbia. This study is primarily aimedat developing a better understanding of the tempo-ral and spatial distribution of extreme sea surges andthe response to various regional climate variabilitymodes by statistically exploring eleven historical long-

term water level records in coastal British Columbia(Figure 1).

The study methodology follows a development inthe extreme-value statistical theory, where the conven-tional generalized extreme value (GEV) distribution(Jenkinson 1969; Leadbetter, Lindgren, and Rootzen1983; Coles 2001) is applied to parametrically modelthe multiyear annual maximum sea surge events whileaccounting for trends due to climate variability via cli-mate covariate coefficients (Gilleland and Katz 2005).The sea surge magnitudes applied for the statistical anal-ysis are computed by taking the concurrent differencesbetween the observed in situ hourly tide gauge data andthe predicted astronomical tidal levels. This work be-comes the first of its kind in British Columbia, as nosimilar studies have been done previously in the region.

The article is organized as follows: The next sec-tion describes the study region and data. After that,we present the methodology, including some back-ground on extreme value statistics. The following sec-tion presents and compares the exceedance probabilityof extreme sea surges at each site and their sensitivityto various climate modes (i.e., warm, neutral and cold

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011

994 Abeysirigunawardena, Smith, and Taylor

ENSO states). Finally, we discuss and present conclu-sions of this study.

Data Preprocessing: Tide Gauge DataQuality Assurance

Hourly water level data from eleven tide gauge siteswith data covering at least twenty years or more wereconsidered for analysis (Figure 1). The data originatedfrom the Marine Environmental Data Service (MEDS)archives of Fisheries and Oceans Canada. The sta-tion name coordinates and acronyms are indicated inTable 1.

Prior to the statistical analysis, the observed waterlevel data were quality checked for inhomogeneities,abnormalities, and gaps resulting from changes in theinstrumentation, measurement techniques, or observa-tional practices. Potential problems due to large datagaps were eliminated by disregarding years with greaterthan 10 percent missing data. Because extremes gen-erally occur during fall and winter seasons (i.e., fromOctober to March of the following year), years that sat-isfied 100 percent monthly data coverage from Octoberto March were included in the analysis, even when theannual data coverage was less than 90 percent.

The presence of inhomogeneities and data abnor-malities was identified by comparing the mean sea level(MSL) anomaly plots against the gauge history reports.Appropriate corrections were applied when sufficientinformation was available to justify possible heterogene-ity resulting from gauge shifts. Some stations indicatedambiguous shifts in the annual MSL anomalies, suggest-ing possible temporal offsets associated with instrumentproblems or site relocations. Most of the shifts could

be linked to gauge movement or gauge corrections viathe gauge history information. Accordingly, the datafor Point Atkinson (7795), Tofino (8615), Vancouver(7735), and Bamfield (8545) indicated possible offsetsassociated with the MSL linked to gauge shifts, and Pa-tricia Bay (7277), Campbell River (8074), and BellaBella (8976) indicated significant offsets between esti-mated MSL (i.e., Z0 in the tidal constituents) and thecalculated MSL based on observed data.

Sea Surge Computations

The observed water level at any given station hastwo site-specific additive components: the astronomicaltidal level and the residual or the sea surge component.Astronomical tidal predictions for average meteorolog-ical conditions at a given location are deterministic,provided that all of the amplitude and phase values ofthe gravitationally induced constituent tides are knownat the location (Foreman 1977). When meteorologicalconditions deviate from the average values, however,the actual times and heights of the tides can differ fromthose predicted. Such differences are termed sea surgesor residuals, and can be attributed to nontidal effectssuch as atmospheric pressure, winds, and ENSO effectson the total water levels.

For this work, the sea surge (residual) time series ateach site was computed by subtracting the predictedastronomical tidal level from the measured total waterlevels at the tide gauge. The station-specific astronomi-cal tides were based on the tidal constituents developedat the Department of Fisheries and Oceans (DFO; Fore-man 1977). Unrealistic sea surge events might appearor disappear suddenly, resulting from instrumentation

Table 1. Pacific region tide gauge stations adjacent to the coast of British Columbia with at least twenty years of data

Station coordinates

No. Station number Station name Latitude (Deg.) Longitude (Deg.) Duration (years)

1 7120 Victoria Harbour 48.42 123.37 682 7277 Patricia Bay 48.65 123.45 253 7735 Vancouver 49.29 123.11 514 7795 Point Atkinson 49.34 123.25 525 8074 Campbell River 50.04 125.25 336 8408 Port Hardy 50.72 127.49 397 8545 Bamfield 48.84 125.14 368 8615 Tofino 49.15 125.91 549 8976 Bella Bella 52.16 128.14 4610 9354 Prince Rupert 54.32 130.32 5711 9850 Queen Charlotte City 53.25 132.07 39

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


errors or seismic activity. Such events generally occurindependently of meteorological events with no similaroscillation observed at nearby tide gauges. To eliminatesuch events from the analysis, each annual maxima seasurge event was compared against the correspondingresidual water levels at the closest tide gauge stations,and with meteorological station data across two tidalcycles.

Climate Variability Indexes

Climate variability in the northeast Pacific is char-acterized by a number of climate variability indexes,such as the Multivariate ENSO Index (MEI), PacificDecadal Oscillation (PDO), Northern Oscillation In-dex (NOI), Aleutian Low Pressure Index (ALPI), andPacific North American teleconnection (PNA). Pre-viously, climate variability modes such as the ENSOand PDO have been directly linked to interannual sealevel variations along the eastern Pacific coastal mar-gin (Trenberth and Hurrell 1995; Crawford et al. 1999;Storlazzi, Willis, and, Griggs 2000; Subbotina, Thom-son, and Rabinovich 2001; Allan and Komar 2002;Abeysirigunawardena and Walker 2008). Given this,the extreme sea surge recurrences along coastal BritishColumbia are expected to shift in conjunction withthese climate variability modes as well.

This study investigates the relationship between ex-treme sea surges in coastal British Columbia and a num-ber of dominant climate variability modes in the north-east Pacific. The ALPI is a northeast Pacific regionalindex that indicates the relative intensity of the control-ling pressure system in the northeast Pacific and there-fore indirectly describes the relative strength of regionalwind patterns. The index is calculated as the mean area(km2) with sea-level pressure less than or equal to 100.5kPa expressed as an anomaly from the 1950 to 1997mean (Beamish, Neville, and Cass 1997). Positive ALPIindex values reflect a relatively strong or intense Aleu-tian low, whereas negative values indicate the opposite.ALPI values are published online by the CanadianDFO (http://www.pac.dfo-mpo.gc.ca/science/species-especes/climatology-ie/cori-irco/indices/alpi.txt).

The classification of ENSO events in this studyis based on the National Oceanic and Atmo-spheric Administration’s (NOAA) MEI (Wolter andTimlin 1993, 1998). The MEI is defined as theweighted average of a number of tropical Pacificenvironmental variables including sea surface tem-perature, east–west (i.e., zonal) and north–south(i.e., meridional) surface winds, sea-level pressure

and temperature, and cloudiness. Negative MEI val-ues represent the cold ENSO phase (La Nina) andpositive MEI values represent the warm ENSO phase(El Nino). Monthly MEI values are published online atNOAA-CIRES Climate Diagnostic Center’s Compre-hensive Ocean-Atmosphere Data Set (COADS; http://www.esrl.noaa.gov/psd/data/correlation/mei.data

The PDO index characterizes interannual variabilityin average north Pacific sea surface temperature andreflects northeast Pacific regional climatic variability(Mantua et al. 1997). PDO is well correlated withmany records of north Pacific and Pacific Northwestclimate and ecology, including sea-level pressure,winter land–surface temperature and precipita-tion, and stream flow (Mantua et al. 1997; Zhang,Wallace, and Battisti 1997; Hessl, McKenzie, andSchellhaas 2004; Schneider and Cornuelle 2005).Predominantly positive or negative sea–surface temper-ature anomalies along the Pacific Coast of North Amer-ica characterize the phase of the PDO pattern as beingin a “warm” or “cold” state, respectively. Monthly PDOindex values are published online by the Joint Institutefor Study of the Atmosphere and Ocean’s PDO site(ftp://ftp.atmos.washington.edu/mantua/pnw impacts/INDICES/PDO.latest; Mantua et al. 1997).

The NOI is the sea-level pressure anomaly betweenthe North Pacific High (NPH, 35◦N and 130◦W) regionof the northeast Pacific and the equatorial Pacific nearDarwin, Australia, a climatologically low-pressure re-gion (Schwing, Murphree, and Green 2002). The NOIis dominated by the interannual variations of ENSO,such that large positive (negative) NOI are associatedwith La Nina (El Nino; Schwing, Murphree, and Green2002). As NOI is partially based in the northeast Pacific,it provides a more direct connection between variousclimate processes in the northeast Pacific and remotelyteleconnected global climate events. The NOI timeseries is available online from NOAA’s Pacific Fish-eries Environmental Laboratory (http://www.pfeg.noaa.gov/products/PFEL/modeled/indices/NOIx/noix.html).

The PNA describes one of the most dominant modesof low-frequency variability in the Northern Hemi-sphere extratropics during the Northern Hemispherewinter (Van den Dool, Saha, and Johansson 2000).The PNA index is constructed through a linear combi-nation of normalized 700-mb height anomalies nearestto the nominal centers of the PNA pattern (Leathersand Palecki 1992). The PNA pattern occurs as a resultof an amplification or damping of the mean flow config-uration over the unique geography of the Pacific Basinand North America (Leathers and Palecki 1992). PNA

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


is strongly influenced by the mode of ENSO, where thepositive phase of PNA tends to be associated with warmENSO and the negative phase with cold ENSO.

Methodology

The methodology of this study is based on extremevalue theory, a widely applied approach for drawinginferences about the extremes of a stochastic processbased on relatively extreme observations of that pro-cess (Coles and Dixon 1999; Coles 2001; Butler et al.2007). Two main approaches have been proposed in theliterature: the block maxima models (e.g., annual max-ima) and the peaks-over-threshold models (POT). Asummary of these techniques can be found in Palutikof(1999). In this work, a GEV distribution is fitted to an-nual maxima sea surge data at each site using the block(i.e., annual) maxima models. The impacts of climatevariability on extreme sea surge recurrences are simu-lated by incorporating the dominant climate variabilityindexes as model parameter covariates. For the analysis,the R (R Development Core Team 2007) package “ex-tRemes” was used because of its ability to incorporatecovariate information into parameter estimates (Gille-land and Katz 2005).

GEV Distribution

Under weak assumptions, extreme value theory char-acterizes the GEV distribution as the limit distributionof maxima of a sequence of independent and identi-cally distributed observations from a continuous ran-dom variable. Thus, the GEV distribution can be used asan approximation to model the maxima of long (finite)sequences of random variables (Leadbetter, Lindgren,and Rootzen 1983; Coles 2001):

G (z; μ; σ ; ξ) = exp[− {1 + ξ (z − μ) /σ }−1/ξ+ ]

(1)where −∞ < μ < ∞, σ > 0, −∞ < ξ < ∞ are thelocation, scale, and shape parameters, respectively.

The sign of the shape parameter (ξ) determinesthe GEV distribution type (Leadbetter, Lindgren, andRootzen 1983; Coles 2001).

1. ξ > 0 corresponds to a Frechet type distributionhaving a heavy tail with a polynomial decay.

2. ξ < 0 corresponds to a Weibull type distributionwith a bounded upper tail, meaning that there isa finite value that the maximum cannot exceed.

3. ξ = 0 corresponds to a Gumbel type distributionwith an unbounded thin tail.

In this study, the parameter estimation for the GEVmodel was based on the maximum likelihood estimation(MLE) method, because of its ability to easily incorpo-rate covariate information into the model parameterestimates (Hosking and Wallis 1987). Another advan-tage of the MLE method is that approximate standarderrors for estimated parameters and return levels canbe automatically produced, either via the informationmatrix or through the profile likelihood method (Gille-land and Katz 2005). Nevertheless, the performance ofthe MLE method is known to be unstable and can giveunrealistic estimates for the shape parameter when thesample size is small (i.e., n ≤ 25; Hosking and Wal-lis 1987; Coles and Dixon 1999). When enough dataare available, however (n > 30), MLE is comparable inperformance to other alternative estimation techniques(Martins and Stedinger 2001; Gilleland and Katz 2005).

The magnitude of extremes is often expressed as thereturn levels of an extreme event (Zp), such that thereis a probability (p) that Zp is exceeded in any givenyear. The probability (p) is typically expressed as (1/p)years, where (1/p) is referred to as the return period.For example, the 100-year surge return level (Z100) ata given location will have a 1 percent (1/p = 1/100 =0.01) chance to occur in any given year. An alternativeexplanation would be that the extreme surge level cor-responding to Z100 will be exceeded at least once in 100years. Zp is obtained by inverting the fitted distributionfunction in Equation 1 and solving for the return levelsfollowing Equations 2.1 or 2.2.

Z p = μ − σ

ξ[1−{− log(1−p )}−ξ ], for ξ �= 0

(2)

Z p = μ − σ log{− log(1 − p)} for ξ = 0 (3)

Simulate the Effect of Climate Variability on SeaSurge Extremes Via Covariates

Because this study is the first of its kind that in-vestigates the relationship between extreme sea surgerecurrences and climate variability in costal BritishColumbia, the scope is limited to simple linear relation-ships between climate variability indexes and extreme

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


sea surges. As such, the climate-variability indexes (Y)are related to the GEV model coefficients (σ and ξ) aslinear covariates (Equation 3).

μ (X) = μ0 + μ1Y, ln σ (X)

= σ0 + σ1Y, ξ (X) = ξ0 + ξ1Y (4)

The Dependencies Between Climate Indexes.Initially the candidate climate indexes (i.e., MEI, PDO,PNA, ALPI, and NOI) are included as individual co-variates in the GEV model. The fit of each modelwas tested for significant improvement at the 95 per-cent confidence level against the “base model” (i.e.,the corresponding model without climate covariates).Any individual climate covariate that does not showstatistically significant improvements in the modifiedGEV model fit against the “base model” fit is con-sidered superfluous, whereas those with a significantlybetter fit are carried forward to the next level ofanalysis.

Assessing the Comparative Performance of theGEV Models with Climate Covariates. Evidently,many large-scale climate regimes described by variousclimate variability indexes are related. For example, thestrong climate regime shift that occurred in mid-1976was captured by all climate indexes considered in thisanalysis at approximately the same time, suggesting thatthe climatic phenomena that triggered the regime shifthas affected a wide range of climate indexes (Figure2). Due to these interdependencies, climate indexeswhen applied as individual covariates within a GEVmodel might show very similar responses. On the otherhand, when extreme surge climatology in a region is sig-nificantly affected by climate variability characteristicsunique to a climate index, they should indicate compar-atively better GEV model performance over others. Fol-lowing this, the likelihood ratio test was applied to testthe comparative performance of different GEV modelsin the presence of correlation between the explana-tory variables (i.e., climate variability indexes). In thisapproach, the GEV models were made more complexsystematically by increasing the number of explanatoryvariables in successive models.

This approach aids in choosing the GEV model thatbest describes the data without having more param-eters than necessary. Each time the fit diagnostics ofeach successive model is cross-compared with the for-mer via a likelihood ratio test to determine whetherthe new model is significantly better than the previous

one. The combination of climate indexes that best de-scribes the model fit are chosen for the final GEV modelto simulate the climate effects on extreme sea surgerecurrences. This approach avoids repetition of pat-terns that are common to many climate indexes withinthe GEV models and simplifies the final GEV model.Table 2 demonstrates the application of this test onextreme sea surge data at station 7120.

Inferences Under Different Climate States. Tomake meaningful extreme sea surge return level infer-ences from a GEV model with time-varying climatevariability covariates (see Equation 3), the models werereduced to a time-dependent function by setting theclimate variability indexes (i.e., covariates) to a set offixed average values conditional on three distinct cli-mate states. The climate states were chosen based on theEnvironment Canada climate classification scheme as(1) the extremely warm (strong El Nino), (2) extremelycold (strong La Nina), and (3) neutral. Table 3 summa-rizes the corresponding climate indexes averaged overthe range of years classified under these three climatestates.

Results

Sea Surge Dependencies in Coastal British Columbia

Table 4 demonstrates the overall average monthlymaximum and monthly mean positive sea surges (i.e.,residual water levels) along the British Columbia coastsince 1950. The statistics are comparable, with less than2-cm standard deviations among the eleven stations.None of the station records indicated extreme resid-uals exceeding the 2-m mark during this period. Theannual cycle of the mean and maximum sea surges atall stations are comparable, with the peak surge consis-tently coinciding with the fall and the winter seasons(i.e., from October–March). In general, the sea surgemagnitudes from the October to March period are 40to 50 percent higher than those that occur from Aprilto September, and December is the month with thestrongest sea surges.

Correlation analysis of annual maximum seasurges from 1950 to 2007 based on the Pearsonproduct–moment and the Spearman’s rank correlationcoefficients indicate two groups of tide gauges havingdistinctly different sea surge characteristics (Table 5).For instance, tide gauges located on Vancouver Islandand in the Strait of Georgia showed significant (at99 percent confidence level) correlations (correlation

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Table 2. Application of the likelihood ratio tests for the model selection of the generalized extreme value location parameteron annual maximum sea surge data at Station 7120

Significance compared with the chi-squarevalue for 0.05 significance level

NegativeModel Parameter log-likelihood 1 vs. 2 2 vs. 3 3 vs. 4

1 BASE 214.82.1 BASE + CV (μ 1 = MEI) 211.6 6.4 > 3.842.2 BASE + CV (μ 2 = NOI) 210.8 8 > 3.842.3 BASE + CV (μ 3 = PNA) 211.2 7.2 > 3.843.1 BASE + CV (μ 1 = NOI; μ 2 = PNA) 208.7 4.2 > 3.843.2 BASE + CV (μ 1 = NOI; μ 2 = MEI) 210.7 < 3.843.3 BASE + CV (μ 1 = PNA; μ 2 = MEI) 209.8 < 3.844.1 BASE + CV (μ 1 = NOI; μ 2 = PNA; μ 3 = MEI) 208.7 0 < 3.84

Selected final modelBASE + CV (μ 1 = NOI; μ 2 = PNA)

Note: CV = climate variability; MEI = Multivariate ENSO Index; NOI = Northern Oscillation Index; PNA = Pacific North American teleconnection.

Figure 2. Temporal distribution of Northern Oscillation Index (NOI), Pacific Decadal Oscillation (PDO), and El Nino/La Nina-SouthernOscillation (ENSO) indicating significant colinearities. Note the 1976 major climate regime shift being captured by all climate indexesapproximately at the same time.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Table 3. Average climate indexes for strong El Nino, strong La Nina, and neutral years

MAXSL(Chart, in cm) PDO MEI NOI PNA ALPI

Strong El Nino (warm ENSO) years: 1957, 1982, 1991, 1992, 1998536 0.219 0.928 –1.486 0.105 0.406

Strong La Nina (cold ENSO) years: 1973, 1974, 1988, 1989, 1999521 –0.370 –0.688 1.185 –0.081 –0.486

Neutral years: 1951, 1952, 1953, 1954, 1961, 1962, 1963, 1966, 1967, 1968, 1971, 1979, 1980, 1981, 1985, 1986, 1990, 1994, 1996, 2000,2001, 2003, 2004, 2005

527 –0.142 –0.050 0.175 –0.040 0.439

Note: The definitions of strong El Nino and La Nina years are based on the Environment Canada (Meteorological Services of Canada, The Green Lane)classification scheme. PDO = Pacific Decadal Oscillation; MEI = Multivariate ENSO Index; NOI = Northern Oscillation Index; PNA = Pacific NorthAmerican teleconnection; ALPI = Aleutian Low Pressure Index; ENSO = El Nino/La Nina-Southern Oscillation.

between 0.7 and 0.95). Such strong spatial dependen-cies can lead to flooding occurring simultaneously alongentire reaches. The correlation does, however, seemto decrease with increasing vertical distance from thesouthernmost stations. Stations 9850, 9354, and 8976located in proximity to Hacate Strait indicate only mod-erate to weak correlation (0.3–0.4) with the southernVancouver Island stations. The analysis suggests thepresence of two distinct spatial patterns of extreme seasurge conditions. Understanding these spatial depen-dencies would be beneficial toward improved realloca-tion of monitoring sites.

Understandably, correlation is a convenient but sim-ple approach to measure dependence among variables.Thus, it might not be the best measure of dependence

at extreme levels. Therefore, more sophisticated mul-tivariate extreme value methods are recommended asa rigorous basis for studying these spatial dependenciesfurther (Coles, Heffernan, and Tawn 1999).

Extreme Sea Surge Exceedances in Coastal BritishColumbia

GEV Model Approximations Without Climate Co-variates. The GEV approximations without climatecovariates (hereafter referred to as the base model) indi-cated a Weibull type distribution with a negative shapeparameter for all but one station (i.e., Station 9850),suggesting an upper limit for the extreme sea surge mag-nitudes (Table 6). Station 9850 has been identified by

Table 4. Overall averages of the (a) monthly maximum sea surges and (b) monthly mean sea surges recorded to date at theeleven tide gauge stations in coastal British Columbia, indicating a winter seasonal (October–March) preference for

high-magnitude sea surge occurrences

Overall average monthly extremesea surges (cm)

Overall average monthly meanpositive sea surges (cm)

Month M SD M SD

January 88.7 11.4 16.5 0.9February 82.5 10.7 16.1 0.9March 76.7 6.4 13.4 0.9April 63.4 9.1 10.4 0.7May 51.2 10.2 8.3 0.6June 52.6 14.6 7.8 0.7July 50.1 13.6 7.1 0.6August 43.8 10.1 6.9 0.7September 50.9 9.3 8.2 1.2October 73.3 14.0 11.9 1.5November 86.5 12.6 15.0 1.0December 95.1 16.7 17.1 0.9

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011

Tab

le5.

Pear

son

prod

uct–

mom

entc

orre

lati

onco

effic

ient

sofa

nnua

lmax

imum

sea

surg

esfr

om19

50to

2007

indi

cati

ngtw

odi

stin

ctly

diffe

rent

spat

ial

depe

nden

cies

(dem

arca

ted

byda

shed

rect

angl

es)

amon

gst

atio

nsin

coas

talB

riti

shC

olum

bia

1000

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Table 6. Estimated generalized extreme value (GEV) parameters for the base model at each station

GEV model coefficients for the base model

Location parameter (μ) Scale parameter (σ ) Shape parameter (ξ)Data coverage

Station (years) M SE M SE M SE

7120 68 56.01 1.49 10.99 1.07 –0.275 0.0907277 25 60.12 2.73 12.53 2.19 –0.568 0.1447735 51 66.5 1.97 12.69 1.41 –0.328 0.0997795 52 67.3 1.86 12.42 1.25 –0.274 0.0628074 33 71.5 2.73 14.18 2.01 –0.372 0.1238408 39 54.8 2.12 11.99 1.72 –0.536 0.1308545 36 57.9 2.80 14.57 2.12 –0.342 0.1498615 54 61.6 2.19 14.71 1.51 –0.097 0.0708976 46 59.6 2.00 14.47 1.39 –0.270 0.0809354 57 64.2 1.40 10.27 1.04 –0.134 0.1119850 39 69.7 2.47 13.29 1.84 –0.017 0.146

Figure 3. The return level plots at the eleven tide gauge stations in coastal British Columbia, developed based on generalized extreme value(GEV) distribution. The 5 percent and 95 percent confidence bands together with the annual maxima for each station are also shown. Thefitted density following the data indicates that the selected GEV models for the data are satisfactory choices.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


the DFO as one with significant shallow water effects.Although observed annual maxima sea surges fall wellwithin the 95 percent confidence limits in all the GEVapproximations, there is an apparent underestimationof the GEV approximations at the tail end. The ex-treme sea surge return level projections show just threestations (i.e., Stations 8615, 8074, and 9850) exceedingthe 1-m mark for the 100-year return level (Figure 3).Stations located in close proximity (i.e., 7120 and 7277,7735 and 7795) showed return levels with comparablemagnitudes (Table 7).

GEV Model Approximations with Climate Co-variates. Certain stations indicated significantimprovement in the GEV approximations with indi-vidual climate covariates, compared to their base fits(Table 8). Over 55 percent of the stations showed sta-tistically significant model improvement when PNAand NOI were applied as location parameter covariates.Both of the indexes are known to describe large-scalepressure patterns directly affecting the northeast Pa-cific sea surface temperature and storm frequency. ThePNA has a strong influence on the strength and lo-cation of the jet stream that influences storminess incoastal British Columbia. The MEI, which describesthe remotely teleconnected oceanic and atmosphericpatterns, appears to have significant influences on scaleand shape parameter, and PDO shows the least amountof impact on extreme sea surges in coastal BritishColumbia.

Generally speaking, the location parameter that de-fines the mean of the extreme residuals within the GEVmodel is the most affected by climate variability (i.e.,over 60 percent of the stations). The least affected is the

scale parameter of the models. As far as the sign of theclimate covariate coefficients are concerned, there is anotable similarity in the location parameter at all sta-tions, suggesting spatially consistent climate variabil-ity impacts on extreme sea surges in coastal BritishColumbia. A strong presence of ALPI, NOI, PNA,and MEI is seen among the final extreme value sta-tistical models (GEV) at each station (Table 9). Thefact that certain climate covariates became redundantin the presence of others might be due to either the con-sideration of multiple climate indexes that describe thesame climatic phenomenon or some climate covariatesjust not being related to the levels of extreme surgesconsidered.

Extreme Sea Surge Exceedances Under ENSOConditions. Using the finalized GEV approximationswith climate covariates, the changes in the exceedanceprobability of extreme sea surges under warm, neu-tral, and cold ENSO conditions were examined. Theaverage climate index values for the three distinctclimatic states were calculated (Table 3). For illus-tration the final results for Station 7120 are shown(Figure 4).

According to the results, for a given return period,nine out of the eleven stations (>80 percent) indi-cated higher (lower) residual water level occurrencesunder warm (cold) ENSO conditions, compared to thebase case (i.e., projections without climate considera-tions). Table 10 summarizes the station-specific extremesea surge magnitudes with 1 percent probability of ex-ceedance in each year under different climate state. Ac-cordingly, for a given return period, all stations indicateconsistently decreasing return levels from warm ENSOtoward cold ENSO conditions. For instance, number

Table 7. Estimated sea surge return levels in coastal British Columbia with no climate considerations (base model)

Return period (years)Data coverage

Station (years) 2 5 10 20 25 50 100

7120 68 59.8 69.5 74.5 78.3 79.4 82.3 84.77277 25 64.3 72.8 76.0 78.1 78.6 79.8 80.67735 51 70.9 81.6 86.7 90.6 91.7 94.5 96.77795 52 71.7 82.6 88.2 92.6 93.8 97.1 99.88074 33 76.3 87.8 93.1 97.0 98.0 100.7 102.78408 39 58.8 67.2 70.5 72.6 73.2 74.4 75.38545 36 62.9 75.0 80.8 85.1 86.2 89.3 91.68615 54 66.9 82.2 91.4 99.6 102.1 109.4 116.38976 46 65.9 77.0 82.6 87.1 88.3 91.7 94.49354 57 67.8 78.1 84.1 89.3 90.9 95.4 99.49850 39 74.9 90.0 99.7 108.8 111.7 120.4 129.0

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011

Tab

le8.

Clim

ate

cova

riat

eco

effic

ient

sexp

ress

edto

the

first

deci

mal

whe

nap

plie

das

indi

vidu

alco

vari

ates

Loca

tion

para

met

er(μ

)Sc

ale

para

met

er(σ

)Sh

ape

para

met

er(ξ

)

Dat

aco

vera

geLi

near

Stat

ion

(yea

r)tr

end

ALP

IM

EIN

OI

PDO

PNA

ALP

IM

EIN

OI

PDO

PNA

ALP

IM

EIN

OI

PDO

PNA

7120

680.

10.

74.

8–2

.53.

110

.6–0

.11.

7–1

.00.

81.

1–0

.00.

2–0

.10.

10.

272

7725

0.3

1.8

1.5

–1.3

2.2

12.8

–0.3

–2.9

–1.1

–1.3

–5.4

–0.0

–0.3

–0.1

0.2

0.2

7735

510.

21.

42.

5–1

.62.

39.

00.

00.

4–0

.11.

3–0

.50.

00.

3–0

.10.

20.

677

9552

–0.1

0.3

2.5

–1.7

2.4

8.8

–0.5

3.4

–0.8

0.1

–6.1

–0.0

0.3

–0.2

–0.2

–0.9

8074

33–0

.11.

04.

2–2

.84.

37.

2–0

.02.

11.

32.

2–1

.1–0

.4–0

.6–0

.1–0

.20.

584

0839

–0.2

0.6

1.4

–1.0

1.7

5.9

0.1

0.4

–0.5

0.5

2.5

0.0

0.1

–0.0

0.0

0.2

8545

360.

21.

97.

9–4

.36.

813

.50.

45.

6–2

.34.

63.

20.

00.

4–0

.20.

50.

386

1554

–0.2

1.2

4.3

–4.0

1.3

10.8

–0.1

3.1

–1.2

2.9

0.3

–0.0

0.7

–0.3

–0.0

1.9

8976

46–0

.21.

23.

5–2

.22.

66.

5–0

.4–1

.50.

9–1

.2–3

.3–0

.1–0

.10.

1–0

.1–0

.293

5457

0.2

1.5

4.0

–1.9

3.9

7.5

0.5

–0.2

0.1

–0.5

–2.3

0.1

–0.2

0.0

–0.0

0.1

9850

39–0

.00.

83.

6–1

.22.

35.

90.

60.

1–0

.60.

23.

30.

0–0

.00.

1–0

.1–0

.4

Not

e:T

hepo

siti

ve/n

egat

ive

(+/-

)sig

nin

each

cell

indi

cate

sthe

rela

tion

ship

(i.e

.,tr

end)

betw

een

the

mod

elpa

ram

eter

and

the

clim

ate

cova

riat

e.T

hesh

aded

cells

indi

cate

the

coef

ficie

ntst

hata

resi

gnifi

cant

atth

e95

perc

ent

leve

l.N

ote

that

the

resu

lts

are

appr

oxim

ates

toth

efir

stde

cim

al.A

LPI=

Ale

utia

nLo

wPr

essu

reIn

dex;

MEI

=M

ulti

vari

ate

ENSO

Inde

x;N

OI=

Nor

ther

nO

scill

atio

nIn

dex;

PDO

=Pa

cific

Dec

adal

Osc

illat

ion;

PNA

=Pa

cific

Nor

thA

mer

ican

tele

conn

ecti

on.

1003

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Table 9. Climate covariates coefficients for the final models at each tide-gauge station, selected from the likelihood ratio tests

Station Location parameter (μ) Scale parameter (σ ) Shape parameter (ξ)

7120 μ = 57.5 – 2.28(NOI) + 7.04(PNA) σ = 10.5 ξ = –0.3687277 μ = 60.1 σ = 12.5 ξ = –0.5687735 μ = 68.6 + 1.37(ALPI) σ = 11.8 ξ = –0.2657795 μ = 68.5 σ = 13.1 ξ = –0.511 + 0.331(MEI)8074 μ = 71.5 – 2.8(NOI) σ = 13.5 ξ = –0.4218408 μ = 57.8 + 5.9(PNA) σ = 11.5 ξ = –0.5558545 μ = 59.1 – 4.3(NOI) σ = 13.1 ξ = –0.4418615 μ = 62.4 σ = 13.4 ξ = –0.128 + 1.94(PNA)8976 μ = 59.4 + 1.2(ALPI) – 2.1(NOI) σ = 11.5 ξ = –0.272 –0.109(ALPI)9354 μ = 65.3 + 1.0(ALPI) + 2.9(MEI) σ = 9.3 ξ = –0.1149850 μ = 69.7 σ = 13.3 ξ = –0.017

Note: NOI = Northern Oscillation Index; PNA = Pacific North American teleconnection; ALPI = Aleutian Low Pressure Index; MEI = Multivariate ENSOIndex; ENSO = El Nino/La Nina-Southern Oscillation.

of stations with sea surge magnitudes corresponding to1 percent exceedance reaching the 1-m mark would in-crease from 33 percent in the base model to 56 percentunder warm ENSO conditions.

Figure 4. Estimated return levels for residual water levels (seasurges) under warm El Nino/La Nina-Southern Oscillation (ENSO;red dashed line), neutral (blue dashed line), and cold ENSO (greendashed line) conditions, from having fit maximum annual residualwater levels at Station 7120 to a generalized extreme value (GEV)distribution with climate variability effects accounted as covariates.Results for no climate considerations (black continuous line) andthe 95 percent confidence limits (blue continuous line) are includedfor comparison purposes. Similar probability curves have been con-structed for each tidal station of the British Columbia coast (resultsnot shown). (Color figure available online.)

Even though climate variability modes are known tobe short lived, they generally prevail long enough (i.e.,decadal to interdecadal scale) to impose significant im-pacts on vulnerable coastal margins and infrastructure.This study clearly demonstrates how by including in-dexes of climate variability as covariates, the probabilityof an extreme sea surge will be higher (lower) in someyears under warm (cold) ENSO than it would have beenif climate variability had not been accounted for. Thus,the proposed approach might provide a basis for iden-tifying periods in which the probability of flooding isparticularly low or high and targeting resources accord-ingly during the planning stages. In addition, consider-ing climate variability in the analysis would ensure thatthe standard errors associated with the estimated re-turn levels would be more realistic, whereas they might

Table 10. Station-specific extreme sea surge occurrenceshaving 1 percent exceedance in a given year with climate

considerations

Extreme sea surge projections with 1 percentexceedance probability in each year (in cm)

Warm ColdStation Base ENSO Neutral ENSO

7120 84.9 84.9 80.1 77.57735 96.7 100.4 99.3 99.27795 99.8 107.7 90.0 84.98074 102.7 103.1 98.5 95.78408 75.3 77.5 76.7 76.48545 95.1 91.5 84.2 79.98615 116.3 136.1 102.2 96.78976 94.4 92.9 89.3 91.89354 100.6 101.7 98.9 95.7

Note: ENSO = El Nino/La Nina-Southern Oscillation.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Figure 5. Surge–tide relationship and the development of the extreme surge event of 24 December 2003 (top panel). Impacts of the stormsurge event on eastern Graham Island, British Columbia are shown (bottom panel). Approximately 2.5 m of shoreline was lost at this locationalong Highway 16 (upper) compromising the road shoulder and bed. Extensive coastal flooding also occurred, damaging buildings and sendingtons of drift logs onto nearby roads and properties.

become unrealistically small (or large) if changes overtime are ignored.

Spatial Dependencies of Extreme Sea Surgesin Coastal British Columbia

A strong spatial dependency of extreme sea surgescan lead to flooding occurring simultaneously along en-tire reaches of coastline. Two surge events were chosento demonstrate the risk of concurrent occurrence of ex-treme sea surge events in coastal British Columbia. Thespatial dispersion of the two events was assessed via acomparison of the return levels of the maximum seasurges and the duration of the events.

The first case study is a high-magnitude sea surgeevent that occurred on 24 December 2003. This eventresulted in extensive damage to the eastern coast of

Graham Island, Queen Charlotte Island. During thisevent, an intense storm system resulted in strong south-easterly winds reaching 111 km/hour−1, generating amaximum surge of 0.73 m above the predicted tide attide gauge Station 9850. The peak surge occurred atlow tide and the maximum total water level (TWL) of8.06 m (CD) happened five hours later (Figure 5). Thisevent caused extensive coastal flooding, tens of metersof coastal erosion, and damage to roads and criticalinfrastructure.

The second case study is based on an event that re-sulted in the highest extreme TWL event on the innersouth coast of British Columbia. The event occurredon 16 December 1982, when a powerful southeasterwith average winds of 50 km/hour and high tides gener-ated a high water mark of 5.6 m (CD) at Station 7795(Figure 6).

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Table 11. The sea surge characteristics and the projected annual percentage exceedance of surges (bolded columns) observedat each tide gauge station at the time of the strongest ever extreme sea surge event occurrence in southern

(16 December 1982) and northern (24 December 2003) British Columbia coasts

16 December 1982 event 24 December 2003 event

Station TWL (m) Tide (m) Surge (cm)

Annualpercentage

exceedance (%) TWL (m) Tide (m) Surge (cm)Annual percentage

exceedance (%)

7120 3.2 2.3 85.0 0.9 2.8 2.4 43.7 93.07735 5.6 4.7 87.8 8.3 3.6 3.2 46.0 90.97795 5.6 4.6 92.2 5.3 4.9 4.4 46.4 97.18074 5.2 4.3 88.3 19.1 4.9 4.4 43.9 99.08408 — — — — 4.4 3.9 51.5 95.68545 3.4 2.6 82.3 8.1 2.8 2.3 54.7 63.78615 3.3 2.5 80.6 22.2 3.2 2.7 53.4 73.98976 4.4 3.9 54.9 74.6 3.8 3.3 48.0 91.19354 6.2 5.7 48.0 98.5 2.8 2.2 59.6 63.39850 6.5 6.0 43.6 99.9 4.6 3.9 66.4 55.5

Note: TWL = total water level.

Figure 6. Surge–tide relationship and the development of the extreme surge event of 16 December 1982 (top panel). Impacts of the stormsurge event in southern British Columbia are shown (bottom panels). Extensive coastal flooding occurred in Boundary Bay, Mud Bay, andWestham Island and resulted in the highest ever water levels in southern British Columbia. TWL = total water level.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Table 12. The sea surge count distribution versus the percentile astronomical tide levels for the fifty highest surges on recordat the eleven tide gauges

Fifty highest recorded sea surge distribution versus percentile astronomical tide (AT) level (count)

Tide-gauge station 95th < AT95th > AT

> 90th90th > AT

> 75th75th > AT

> 50th50th > AT

> 10th10th > AT

> 5th 5th > AT

7120 2 1 3 15 21 4 47277 3 4 1 18 19 4 17735 1 1 1 10 11 3 237795 3 1 9 16 18 1 28074 6 4 4 6 26 4 08408 5 3 6 10 18 5 38545 2 1 13 12 20 1 18615 2 1 6 12 24 3 28976 2 3 11 11 18 2 39354 2 2 6 21 14 3 29850 0 2 9 12 24 0 3

Average events 3 2 6 13 19 3 4% with respect to the total

events considered (50)5 4 12 26 38 6 8

The surge developed at each of these stations was ex-amined with respect to the day prior to the event, theevent day, and the day after the event (i.e., seventy-two hours). Preliminary correlation analysis suggestedthat sea surges in Georgia and Hecate Straits are weaklycorrelated. Table 11 summarizes the extreme sea surgesmeasured at each station during the two events andthe projected annual percentage exceedances. The an-nual percentage exceedance of the maximum sea surgeelevations measured at each station during the 16 De-

cember event indicate a gradual transition from low(i.e., rare intense events) and moderate to high (i.e.,weak frequent events) toward the north. The low an-nual probability of exceedance (<10 percent at Sta-tions 7120, 7735, and 7795) indicates that an eventtypical of 16 December is capable of generating excep-tionally high surges over a wide region of the south-ern British Columbia coastline. The fact that the sameevent generated low-magnitude surge events furthernorth with annual probability of exceedance above 70

Table 13. The surge tide levels of the highest sea surge and highest total water level event on record at each tide gauge stationin coastal British Columbia

Surge-tide levels of the highest surge events Surge-tide levels of the highest total water-level event

Tide-gauge station Surge (cm) AT (cm) AT percentile Surge (cm) AT (cm) AT percentile

7120 85 231 50th–75th 66 305 >95th7277 84 240 10th–50th 71 359 >95th7735 124 91 < 5th 69 488 >95th7795 104 336 50th–75th 90 470 >95th8074 99 145 10th–50th 88 439 >95th8408 75 109 5th–10th 54 540 >95th8545 94 222 50th–75th 67 383 >95th8615 128 191 10th–50th 74 399 >95th8976 97 207 10th–50th 56 538 >95th9354 98 343 10th–50th 53 738 >95th9850 124 272 10th–50th 65 741 >95thAverage 101 217 10th–50th 68 491 >95th

Note: AT = astronomical tide.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


percent (i.e., Stations 8976, 9850, and 9354) suggeststhat an event typical of 16 December might not causewidespread flooding in the northern coasts of BritishColumbia.

Even though the winds were blowing from the south-east sector during both events, the extreme surge ex-ceedance corresponding to the 24 December eventindicated a different spatial distribution of surges com-pared to the 16 December event. The two events signifydistinctly different spatial dependencies of storm surgeoccurrences in southern and northern British Columbiacoastlines. This, in combination with the existing weakcorrelation between the surges in northern and south-ern coastal stretches, suggests that the possibilities of si-multaneous flooding occurring along the entire BritishColumbia coastal reach from a single sea surge eventmight be rare. The location of the jet stream might be adetermining factor of the location and the spatial extentof damage due to surges in coastal British Columbia. Asdemonstrated earlier, a combination of maximum seasurge return levels corresponding to the same event ata number of tide gauge stations that are spatially apartprovides excellent indications as to the spatial disper-sion and dependencies of these events. Similar knowl-edge on many events might facilitate improvements inspatial predictions and the planning and designing ofsea level monitoring networks.

Coastal Flooding and Surge–Tide Interactions

Positive surges occurring at high tide often producecoastal flooding. In macrotidal areas, the impacts of asurge on a coast will depend on the tide level at thetime of the surge peak. Impacts will be at a maximumfor positive surges occurring at spring high tide, whereaseven exceptional surges might escape unnoticed whenthey occur at low tide levels. Pugh and Vassie (1979,1980) have shown that surges and tides are statisticallyindependent, but in shallow waters nonlineartide–surge interactions could occur due to the effectsof friction and advection. Such interactions could de-crease the surge levels at high water and increase surgelevels on the rising tide (Prandle and Wolf 1978).

An analysis of the astronomical tide levels corre-sponding to the fifty highest sea surge events on recordat the eleven tidal stations (550 events in total) indi-cated that, on average, just 5 percent of the highest fiftysurges coincided with an astronomical tide level equalto or higher than the ninety-fifth percentile level. Onthe other hand, about 75 to 80 percent of extreme surgescoincided with astronomical tides below the seventy-

fifth percentile level. This suggests a scarcity of highsurges occurring at the time of high astronomical tidesin coastal British Columbia (Table 12).

Although the threat of an extreme surge coincidingwith a spring high tide is ever present (5 percent proba-bility), the fact that almost 80 percent of the 550 high-est surge events analyzed coincided with astronomicaltides below the seventy-fifth percentile level suggeststhat serious coastal flooding in coastal British Columbiais governed by extreme astronomical tide levels ratherthan by extreme sea surges. For instance, the astronom-ical tide corresponding to the highest ever sea surgeevent at each tide gauge in coastal British Columbia(eleven events) indicated tidal levels below the fifti-eth percentile, whereas the highest ever TWL event(eleven events) coincided with astronomical tidal lev-els exceeding the ninety-fifth percentile level. Thus itis likely that most of the exceptionally high historicalsurges in coastal British Columbia have escaped unno-ticed due to their occurrence at lower tide levels (Table13). This might be mainly due to the presence of sub-stantially higher overall magnitudes of tides than surgelevels in macrotidal settings.

Discussion and Conclusions

The key focus of this article was to establish the ex-treme sea surge climatology in coastal British Columbiaby properly accounting for climate variability andchange signals embedded in the observational data. Themethodology provided a flexible and statistically rigor-ous approach for quantifying changes in the extremeproperties of sea surge processes by incorporating morephysical information into the extreme value analysis. Inparticular, we have attempted to improve the parame-ter estimates by explicitly incorporating knowledge oflinear dependencies to climate variability and change.The results of this study suggest that the benefits ofan analysis of this nature could be substantial withinan oceanographic context. One key practical issue re-volves around the fact that climate patterns are shortlived (i.e., decadal to interdecadal scale) and, as a re-sult, it is unclear how the climate excursions that triggerchanges in extreme sea surges will continue. Becauseof this, projections cannot be extrapolated indefinitelyinto the future. Thus, in practical applications, the out-comes should be expressed as percentage exceedancein each year linked to a prevailing climate state (i.e.,warm, cold, or neutral ENSOs).

The overall strategy for the final model selectionwith climate covariates might seem simple, due to its

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


initial model selection being performed separately foreach of the three GEV parameters (location, scale, andshape). Admittedly, these parameters are quite highlycorrelated, so that a certain covariate included in amodel parameter is likely to impact which covariatesare included in the other model parameters. Acknowl-edging this limitation, we intend to compare the perfor-mance of the simple approach that we have proposedin this article against the results of a more advancedapproach where we will simultaneously incorporatedifferent combinations of covariates into differentGEV parameters using the likelihood ratio tests. In thelatter approach the number of possible models would be-come large and the process of testing would need to beautomated.

The historical data show a fairly synchronous in-crease in extreme sea surge exceedances under warmENSO conditions. Sea surge signals in contrast, appearto be weak during cold ENSOs. Significant relationshipswere evident between the sea surge climatology acrosscoastal British Columbia and the ALPI, NOI, PNA, andMEI climate excursions. These observed changes couldbe largely attributed to changes in residual sea levelsresulting from warmer oceans and changes in storm ac-tivity due to shifts in the Pacific storm track in responseto the North Pacific climate variability patterns (Tren-berth and Hurrell 1995; Crawford et al. 1999; Stor-lazzi, Willis, and Griggs 2000; Subbotina, Thomson, andRabinovich 2001; Allan and Komar 2002; Abeysirigu-nawardena and Walker 2008). No significant and con-sistent long-term sea surge trends were apparent.

Sea surge occurrences in coastal British Columbiashow two distinct spatial dependencies among tidegauges located in the Hacate Strait and Georgia Straitregions of British Columbia. The results suggest possi-ble coastal flooding occurring simultaneously along awider coastal reach within the two regions, and theobserved weak dependencies between the two regionssuggest rare possibilities of simultaneous flooding oc-curring along the entire British Columbia coastal reachfrom a single sea surge event. Knowledge of such spatialdependencies helps determine to what extent differentstations could be used to describe extreme sea surge con-ditions and facilitates reallocation of monitoring sitesto improve spatial representation of sea level measure-ments in the region.

Apparently, serious coastal flooding in BritishColumbia is governed by extreme astronomical tide lev-els and not by extreme surges. It is therefore highly likelythat most of the exceptionally high historical surges in

coastal British Columbia have escaped unnoticed dueto their occurrence at lower tide levels. Nevertheless, itis argued that the extreme surge coincidence with lowto mid tides in coastal British Columbia could be par-tially due to chance. This diminishes the risk of gettingvery high sea levels as a result of extreme sea surges, butthe fact that extreme sea surges do still, infrequently,occur in conjunction with high tides means simply thatthe risk has not been eliminated completely. Thus, thethreat of extreme flooding due to such events is everpresent. For instance, if the greatest storm surge mea-sured at Station 9850 (124 cm) had occurred at thetime of the stronger astronomical tide of the 24 De-cember event (741 cm), coastal flooding could havebeen as much as 60 cm higher than any recorded levelin northern British Columbia. Similarly, if the greateststorm surge measured at Station 7795 (104 cm) hadoccurred at the time of the stronger astronomical tideof the 16 December event (470 cm), coastal floodingcould have been as much as 14 cm higher than anyrecorded level in southern British Columbia. Thus, therisk of occurrence of such events in the future shouldbe of concern.

ReferencesAbeysirigunawardena, D. S., and I. J. Walker. 2008. Sea-level

responses to climatic variability and change in north-ern British Columbia. Atmosphere-Ocean 46 (3): 277–96.

Allan, J. C., and P. D. Komar. 2002. Extreme storms on thePacific Northwest coast during the 97–98 El Nino and1998–99 La Nina. Journal of Coastal Research 18:175–93.

Beamish, R. J., C. E. Neville, and A. J. Cass. 1997. Productionof Fraser River sockeye salmon (Oncorhynchus nerka) inrelation to decadal-scale changes in the climate and theocean. Canadian Journal of Fisheries and Aquatic Sciences54:543–54.

Butler, A., J. E. Heffernan, J. A. Tawn, R. A. Flather, andK. J. Horsburgh. 2007. Extreme value analysis of decadalvariation in storm surge elevations. Journal of MarineSystems 67:189–200.

Coles, S. G. 2001. An introduction to statistical modeling ofextreme-values. London: Springer Verlag.

Coles, S. G., and M. J. Dixon. 1999. Likelihood-basedinference for extreme-value models. Extremes 2:5–23.

Coles, S. G., J. E. Heffernan, and J. A. Tawn. 1999. De-pendence measures for extreme value analyses. Extremes2:339–65.

Crawford, W. R., J. Cherniawkey, M. Foreman, and P. Chan-dler. 1999. El Nino sea-level signal along the westcoast of Canada. Scientific Report No. 10, North PacificMarine Science Organization (PICES), Sidney, BritishColumbia, Canada.

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011


Foreman, M. G. G. 1977. Manual for tidal heights analysisand prediction. PMS Report 77–10, Institute of OceanSciences, Patricia Bay, Victoria, BC, Canada.

Gilleland, E., and R. W. Katz. 2005. Tutorial for the Extremestoolkit: Weather and Climate Applications of Extreme-value Statistics. http://www.assessment.ucar.edu/toolkit/(last accessed 10 June 2011).

Hessl, A. M., D. McKenzie, and R. Schellhaas. 2004. Droughtand Pacific decadal oscillation linked to fire occurrencein the inland Pacific Northwest. Ecological Applications12 (2): 425–42.

Hosking, J., and J. Wallis. 1987. Parameter and quantile es-timation for the generalized Pareto distribution. Techno-metrics 29 (3): 339–49.

Intergovernmental Panel on Climate Change. 2007: Inter-governmental Panel on Climate Change (IPCC), Climatechange 2007: The physical basis. Summary for policymakers.Geneva, Switzerland: World Meteorological Organiza-tion (WMO) and UN Environment Program (UNEP).http://www.ipcc.ch/publications and data/ar4/syr/en/contents.html (last accessed 10 June 2011).

Jenkinson, A. F. 1969. Statistics of extremes, estimation ofmaximum floods. World Meteorological OrganizationTechnical Notes 98, Chapter 5. Geneva: World Me-teorological Organization.

Leadbetter, M. R., G. Lindgren, and H. Rootzen. 1983. Ex-tremes and related properties of random sequences and series.New York: Springer Verlag.

Leathers, D. J., and M. A. Palecki. 1992. The Pacific/NorthAmerican teleconnection pattern and the United Statesclimate: Part II. Temporal characteristics and index spec-ification. Journal of Climate 5:707–16.

Mantua, N. J., S. R. Hare, Y. Zhang, J. M. Wallace, and R.C. Francis. 1997. A Pacific interdecadal climate oscilla-tion with impacts on salmon production. Bulletin of theAmerican Meteorological Society 78:1069–79.

Martins, E. S., and J. R. Stedinger. 2001. General-ized maximum-likelihood Pareto-Poisson estimators forpartial duration series. Water Resources Research 37 (10):2551–57.

Murty, T. S. 1988. List of major natural disasters, 1960–1987.Natural Hazards 1:303–04.

Palutikof, J. P. 1999. A review of methods to calculateextreme wind speeds. Journal of Applied Meteorology6:119–32.

Prandle, D., and J. Wolf. 1978. The interaction of surge andtide in the North Sea and River Thames. GeophysicalJournal of the Royal Astronomical Society 55:203–16.

Pugh, D. T., and J. M. Vassie. 1979. Extreme sea-levels fromtide and surge probability. In Proceedings of the 16thCoastal Engineering Conference, 1978. Vol. 1, 911–30.New York: American Society of Civil Engineers.

———. 1980. Application of the joint probability methodfor extreme sea level computations. In Proceedings of theInstitute of Civil Engineering 69: 959–75.

R Development Core Team. 2007. R: A language and environ-ment for statistical computing. Vienna, Austria: R Founda-tion for Statistical Computing. http://www.r-project.org(last accessed 10 June 2011).

Schneider, N., and B. D. Cornuelle. 2005. The forcing of thePacific decadal oscillation. Journal of Climate 18 (21):4355–73.

Schwing, F. B., T. Murphree, and P. M. Green. 2002. A cli-mate index for the northeast Pacific. Progress in Oceanog-raphy 53:115–39.

Storlazzi, C. D., C. M. Willis, and G. B. Griggs. 2000. Com-parative impacts of the 1982–83 and 1997–98 El Ninowinters on the central California coast. Journal of CoastalResearch 16:1022–36.

Subbotina, M. M., R. E. Thomson,, and A. B. Rabi-novich. 2001. Spectral characteristics of sea-level vari-ability along the west coast of North America duringthe 1982–83 and 1997–98 El Nino events. Progress inOceanography 49:353–73.

Thomson, R. E., B. D. Bornhold, and S. Mazzotti. 2008.An examination of the factors affecting relative and absolutesea level in British Columbia. Canadian Technical Reportof Hydrography and Ocean Sciences 260, Fisheries andOceans Canada, Ottawa, Ontario, Canada.

Trenberth, K. E., and J. W. Hurrell. 1995. Decadal cou-pled atmosphere-ocean variations in the north PacificOcean. In Climate change and northern fish populations, ed.R. J. Beamish, 14–24. Ottawa: Canadian GovernmentPublishing.

Van den Dool, H. M., S. Saha, and A. Johansson. 2000.Empirical orthogonal teleconnections. Journal of Climate13:1421–35.

Walker, I. J., J. V. Barrie, A. H. Dolan, Z. Gedalof, G. Man-son, D. Smith, and S. Wolfe. 2007. Coastal vulnerabilityto climate change and sea-level rise, Northeast GrahamIsland, Haida Gwaii (Queen Charlotte Islands), BritishColumbia. Final Report Project A580, Canada ClimateImpacts and Adaptations Program, Natural ResourcesCanada.

Wolter, K., and M. S. Timlin. 1993. Monitoring ENSO inCOADS with a seasonally adjusted principal compo-nent index. In Proceedings of the 17th Climate DiagnosticsWorkshop. Oklahoma Climate Survey, CIMMS and theSchool of Meteorology, University of Oklahoma, 52–57.

———. 1998. Measuring the strength of ENSO—How does1997/98 rank? Weather 53:315–24.

Zhang, Y., J. M. Wallace, and D. S. Battisti. 1997. ENSOlike inter-decadal variability: 1900–93. Journal of Climate10:1004–20.

Correspondence: Neptune Canada, 2300 McKenzie Ave, Victoria, BC V8P 5C2, Canada, and University of Victoria, Victoria, BC V8W2Y2, Canada, e-mail: [email protected] (Abeysirigunawardena); Department of Geography, University of Victoria, P.O. Box 3060 STN CSC,Victoria, BC V8W 3R4, Canada, e-mail: [email protected] (Smith); Environment Canada, Environment Canada—Pacific & Yukon Region,Vancouver, BC V6C 3S5, Canada, e-mail: [email protected] (Taylor).

Dow

nloa

ded

by [D

an S

mith

] at 0

8:04

02

Aug

ust 2

011

extreme sea surge responses to climate variability in ... · de valor extremo generalizado, y las...

Documents