deo 2010 - on meteorological droughts in tropical pacific islands

Upload: dr-ravinesh-c-deo

Post on 05-Apr-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    1/10

    METEOROLOGICAL APPLICATIONSMeteorol. Appl. (2010)Published online in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/met.216

    On meteorological droughts in tropical Pacific Islands:time-series analysis of observed rainfall using Fiji

    as a case study

    Ravinesh C. Deo*Department of Mathematics, Faculty of Science, The University of Southern Queensland, Springfield QLD 4300, Australia

    ABSTRACT: Analysis of historical droughts was undertaken by converting observed monthly rainfall over the period19492008 to the Standardized Precipitation Index (SPI). The modified MannKendall test applied on monthly rainfallshowed statistically significant downward trends at significance level = 0.01. The Sens Slope, estimated from the time-series, revealed statistically significant decreases in annual rainfall ranging from 13 to 47 mm year1. Based on the SPI,drought duration and severity was non-uniform across stations over the 60 year period. The strongest impacted stations

    were located in western and northern Fiji where rainfall deficiency for the period 19691988 led to a dramatic increase inmoderate and severe droughts. The return periods of annual rainfall were much longer at these stations relative to those

    in outer-lying islands. While rainfall increased over the period 19892008, the actual amounts did not exceed 19491968totals, confirming a net shift towards drier conditions since the 1950s. This study has demonstrated that SPI can be auseful tool for diagnosis and monitoring meteorological droughts in tropical Pacific islands. Copyright 2010 RoyalMeteorological Society

    KEY WORDS droughts in Fiji; standardized precipitation index; return periods; rainfall trends

    Received 29 December 2009; Revised 10 May 2010; Accepted 24 May 2010

    1. Introduction

    Climate change due to elevated greenhouse gases, strato-spheric ozone depletion and atmospheric aerosols has aprofound impact on global, regional and local hydro-logical cycles (IPCC, 2007). As in other parts of thetropics, the climate of the tropical Pacific is fragileand responds rapidly to anthropogenic radiative forc-ings. This increases vulnerability to extreme climaticevents such as droughts and flash floods with very highconfidence (Mimura et al., 2007). In the foreseeablefuture these islands are not immune from impacts ofsea level rise, increasing severity of tropical cyclonesand frequent storm surge (IPCC, 2007). Hay et al.

    (2003) documented that the South Pacifics climatecompared to earlier records during the 20th centuryhas become drier and warmer by 15% and 0.8 C,respectively.

    Rainfall in small tropical Pacific islands is highly vari-able, both spatially and temporally, and fluctuates ondaily, monthly and annual time-scales (Mataki et al.,2006). These variations are mainly controlled by themovement of the Inter-tropical Convergence Zone, lying5N, and the South Pacific Convergence Zone (SPCZ),stretching east-southeast from Papua New Guinea to

    * Correspondence to: Ravinesh C. Deo, Department of Mathematics,Faculty of Science, The University of Southern Queensland, SpringfieldQLD 4300, Australia.E-mail: [email protected]; [email protected]

    Samoa (Basher and Zheng, 1998). The seasonal rainfall

    is largely dependent on north to south movement of theSPCZ and large-scale cloud bands associated with theposition of the ascending branch of the Walker Circula-tion along the Equatorial Pacific, movement of tropicalupper tropospheric troughs, and surface mid-latitude sys-tems such as cold fronts (Mataki et al., 2006). The IPCC(2007) demonstrated a dramatic decrease in summer rain-fall. While rainfall is projected to increase during winter,this may not be sufficient to compensate the shortfalls.The projections predict that a 10% reduction by 2050would lead to a 20% reduction in the size of the freshwa-ter lens in Kiribati (Christensen et al., 2007). The shifttowards extended periods of dry spells causes loss of

    soil fertility which could impact negatively on agricul-ture (IPCC, 2007). This could lead to economic losseson the order of US$2352 million by 2050, equivalentto 23% of Fijis GDP (World Bank, 2000).

    The El Nino Southern Oscillation (ENSO) plays asignificant role in determining the climate of smalltropical islands (Kumar et al., 2006). Being the hub of thePacific Ocean, Fiji enjoys tropical maritime conditions.The group lying between 15 and 18 S and 175 and177 W consists of over 300 islands. The limestoneislands are vulnerable to droughts and storm surges, whilelarger islands are volcanic and have well-established

    gullies and rivers. The southeast Trade Winds bringrain to heavily rain forested eastern zones, leaving thewestern (leeward) zone dry. The southeastern coasts and

    Copyright 2010 Royal Meteorological Society

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    2/10

    R. C. DEO

    high interiors experience persistent humid weather andorographic rain (Lal, 2004). Dry spells last 34 monthsand are associated with the ENSO cycle (Nicholls andWong, 1990) as the weakening of easterlies during an ElNino event due to the eastward shift of the SPCZ reducesconvection and cumulative rainfall.

    Only a few studies have focussed on modelling andanalysis of Fijis past and future climate. In reviewingoutput from a coupled gas-cycle/climate model (MAG-ICC; Wigley, 1994), Agrawala et al. (2003) noted a pro-

    jected increase in temperature by 0.5 C by 2025 andrainfall change by an appreciable magnitude but with nodefinitive direction of change. Mataki et al. (2006) anal-ysed climate trends from 1961 to 2006 for Nadi and Suva(Fiji Islands), and found increasing trend of temperatures,no definite trends in annual rainfall but a significant inter-annual variability in annual and summer rainfall. Studiesby Hayashi and Golder (1993) and Jones et al. (1998)have shown that austral summers are often characterized

    by relatively small sea-surface temperature anomalies inthe tropical Pacific compared to stronger warm and coldepisodes. During these summers there is a stronger link-age between the MaddenJulian Oscillation (MJO) andextreme precipitation events in the western South Pacificregion (Jones and Weare, 1996). Lal et al. (2002) inves-tigated the climatic response of Small Island States totransient increases in anthropogenic forcing using a rangeof coupled atmosphere-ocean global climate models. Anarea-averaged annual mean warming of 2 C by the2050s was evident. However, an increase in daily rain-fall intensity leading to more heavy rainfall events was

    also projected. Risbey et al. (2002) used results from anensemble of five GCMs to show a projected increase inrainfall of 3.3% by 2025 and 9.7% by 2100, whileFeresi et al. (2000) demonstrated a projected change inrainfall, but the direction of change was uncertain. Analy-sis of observed datasets by Manton et al. (2001) revealedsignificant decreases in the number of rain days since1961 throughout the western and central South Pacificbut increases in the north of French Polynesia and FijiIslands. Salinger et al. (2001) described the influence ofthe Inter-decadal Pacific Oscillation (IPO) on decadalclimate trends and inter-annual modulation of ENSO tele-connections throughout the southwest Pacific. Their anal-

    ysis for precipitation during June to July-August showedincreases in the far north Pacific, and decreases in theCoral Sea and Fiji regions. The precipitation decreasesappeared to be consistent with rises in mean sea levelpressure (MSLP) over the Coral Sea, but large increasesin the northeast did not appear to be directly relatedto MSLP.

    Despite the occurrence of sequential droughts in theFiji Islands over the last few decades, limited studieshave investigated Fijis drought history. This could bepartly because droughts have salient features, with aslow and periodic developmental phase, so timely and

    accurate prediction is often not easy. In the absenceof modelling expertize, rainfall indices based on in situmeasurements can provide quantitative estimates of the

    onset and withdrawal of meteorological droughts. Thisarticle reports time-series analysis of monthly rainfallto quantify severity and duration of droughts in thetropical Pacific islands, using Fiji as a case study. A long-term dataset (1949 2008) from geographically diversestations has been used to capture rainfall trends and

    drought history for different regions in the Fiji Islandsby computing the Standardized Precipitation Index (SPI)following McKee et al. (1993). The present study is thefirst to apply the SPI method to examine the droughthistory of the Fiji Islands.

    2. Data and analysis

    Historical monthly rainfall measured at seven mete-orological stations was extracted from the Compre-hensive Pacific Rainfall Data Base (PACRAIN) forthe period 1949 2008 (Figure 1(b)). These data have

    been quality checked for inhomogeneities and publishedwidely (e.g. He and Barnston, 1996; Griffiths et al.,2003). Supplementary data obtained from Fiji Mete-orological Services (FMS) were used to complementthe PACRAIN database for cross-checking and furthervalidation. It was verified that these data comprisedclimate records from Fijis Reference Climate Station(RCS) network or the Global Climate Observing System(GCOS) Surface Network (GSN) (Peterson et al., 1997).To ensure that rainfall records for the early 1950s didnot have significant calibration inaccuracies, data fromonly those GCOS stations which had less than 20% of

    daily missing values were analysed. The selected sta-tions had documented metadata consisting of a historyof site location, observing instruments and observingpractices.

    This paper reports the cumulative monthly rain-fall which is expected to be less affected by out-liers compared to daily rainfall datasets. Accordingto Manton et al. (2001) these stations have a doc-umented history of changes such as those involv-ing instrumentation, observation practices and the sta-tions immediate environment (metadata). As such,the stations have a long, continuous and homoge-neous record with minimum influence from urbaniza-tion, calibration inaccuracies and are generally consideredhigh-quality.

    Rainfall percentiles were computed by sorting values inascending order with percentile order (p) for each valuebeing, p = (n/(N + 1)) 100 where n = the order ofrainfall value. The standardized rainfall index (SPI)was computed following McKee et al. (1993). SinceSPI is the standardized value of rainfall it allowedcomparison of droughts in climatically diverse regions.Computing the SPI involved fitting gamma probabilitydensity functions to given distribution of rainfall. Thegamma distribution was defined by its probability density

    function:g(P) = 1

    ()P1ex/ (1)

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    3/10

    DROUGHTS IN TROPICAL PACIFIC ISLANDS A CASE STUDY OF FIJI

    (a)

    (b)

    Figure 1. (a) The South Pacific region showing the Fiji Islands, and (b) meteorological stations for Fiji Islands. Source of (a): http://www.

    nationalgeographic.com/xpeditions/

    The parameters, and were estimated using maxi-mum likelihood solutions, where:

    = 14A

    1 +

    1 + 4A

    3

    (2)

    and,

    = P

    (3)

    where A = ln(P ) ln(P ) /N, n = number of rain-fall observations.

    The cumulative probability was given by:

    G(P) =P

    0

    g(P)dP = 1()

    P0

    x1eP / dP (4)

    Letting t = P / the equation becomes an incompletegamma function:

    G(P) = 1()

    t0

    t1etdt (5)

    Since the gamma function is undefined for P = 0, thecumulative probability becomes:

    H( P ) = q + (1 q) G(P ) (6)where q is the probability of a zero. The cumulativeprobability H( P ) was transformed to a standard normalrandom variable with mean zero and variance of one,which gave the SPI:

    SP I = +[t c0

    +c1t

    +c2t

    2

    1 + d1t+ d2t2 + d3t2 ], (7)for 0.5 < H(P) < 1.0

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    4/10

    R. C. DEO

    and

    SP I = [t c0 + c1t + c2t2

    1 + d1t + d2t2 + d3t2], (8)

    for 0 < H(P) < 0.5

    for,

    t =

    ln

    1

    (H(P))2

    , for 0 < H( P ) 0.5 (9)

    and

    t =

    ln

    1

    (1.0 H(P))2

    , for 0.5 H(P) < 1.0 (10)

    and c0 = 2.515517, c1 = 0.802853, c2 = 0.010328,d1 = 1.432788, d2 = 0.189269 and d3 = 0.001308.

    The drought part of the SPI range is split into

    moderately dry (1.5 < SPI 1.0), severely dry(1.5 S P I < 2.0) and extremely dry (SP I 2.0)conditions (McKee et al., 1993). The categories ofdrought months were defined by SP I 2.0 as extremedrought month, 2.0 < SPI 1.5 as a severe droughtmonth and 1.5 < S P I 1.0 as a moderate droughtmonth.

    The return period of total annual rainfall at each stationwas computed to estimate the interval of time betweenevents of a given rainfall amount. This provided anaverage estimate of time elapsed between two successiverealizations of a rainfall event in that year. The return

    period, R, of any given rainfall amount was computed as:

    R = n + 1m

    (11)

    where n is the number of years on record, and m is therank of the rainfall event in consideration.

    2.1. Statistical testing

    The time-series data were subjected to the modifiedMannKendall (MK) test to detect trends (Hirsch andSlack, 1984). This procedure has been widely adopted

    since Mann (1945) and Kendall (1975). The MK testis a rank-based nonparametric procedure, capable ofaccounting for missing values, serial correlation andnumbers below detection limits (Hirsch and Slack, 1984).It is less affected by outliers, because its statistic is basedon sign of differences and not directly on the values(Onnz and Bayazit, 2003). The probability (p) of MKstatistic (z) is estimated by:

    p = 0.5 (|Z|)

    (|Z|) = 1

    2 |Z|

    0

    et2/2dt

    (12)

    Trends are indicated by the sign of the Z-value, witha positive showing an upward trend (Yue and Hashino,

    2003). It is noteworthy that the MK test requires at leastfour values in the time series, so the monthly datasetspanning 60 years was sufficiently large for significancetesting. The Sens Slope was estimated to check mag-nitudes of trends (Theil, 1950; Sen, 1968) by assumingtrends as linear such that f(t) = Q(t) + B. To compute

    the Sens Slope, all pair-wise slopes for the particulartime-scales were calculated (Yue and Hashino, 2003).Since Sens Slope is insensitive to outliers it was morerobust than the usual regression slopes, thus providing arealistic measure of the rainfall trends.

    3. Results and discussion

    3.1. Total rainfall accumulation

    Figure 2 presents the monthly averaged rainfall. It isimmediately evident that rain accumulation per monthhas a high degree of spatial and temporal variability. On

    average, March appears to be the wettest and July thedriest month for all stations. The rainfall amounts for thewettest to driest month are discernibly different amongstations, suggesting widespread variability in climaticconditions throughout Fiji. A closer scrutiny shows thatduring July, Rotuma experienced 63% of total rainfallof March, whereas Ono-I-Lau 35%, Nacocolevu andUdu Point 27%, and Nadi and Labasa only 1214%.The significant reduction in rain accumulation during Julyfor Nadi and Labasa indicates a marked shift to drierconditions in western and northern Fiji, the regions ofmajor economic activity.

    Figure 3 compares rainfall totals against percentiles forMarch and July. For both months, there was a distinctpattern in rain accumulation among stations. The wettestmonth received almost twice as much rain compared tothe driest month. During the driest month the highestrainfall total was recorded for Rotuma, followed byNausori, and the lowest for Nadi and Labasa over theentire range of percentiles. The rainfall values recordedat other stations had similarly low values, but higherthan those at Nadi. Interestingly, rainfall totals for Nadiamounted to less than 50% of totals at Nausori for allpercentiles. Taken together, Figures 2 and 3 suggest thatdrought severity in northern and western Fiji (i.e. Nadi

    and Labasa) is probably the greatest compared to otherstations.

    The monthly analysis has established distinctions inrainfall variability across different percentiles. To exam-ine historical changes, Figure 4 displays time series ofannual rainfall, together with the significance of trendsin the data. As demonstrated by negative Z scores, allstations experienced a generally decreasing trend. Notethat the strength of trend is depicted by the magnitudeof the Z score. The most rapidly decreasing trend is evi-dent at Labasa with Z score 3.3, followed by Nadiwith Z score 3.1, while the weakest trend was foundat Rotuma with Z score 2.2. Identically downwardtrends are found at Nacocolevu and Nausori, with Z score2.9. The magnitude of mean change as demonstrated

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    5/10

    DROUGHTS IN TROPICAL PACIFIC ISLANDS A CASE STUDY OF FIJI

    50

    100

    150

    200

    250

    300

    350

    400

    November

    December

    month of the year

    October

    September

    July

    August

    June

    May

    April

    March

    February

    January

    monthlyaveraged

    rainfall(mm)

    Figure 2. The monthly rainfall averaged over period 1949 2008. Symbols: filled triangle () Rotuma; empty circle ( ) Nadi; filled

    circle ( ) Labasa; filled square ( ) Nausori; empty square ( ) Nacocolevu; solid line with no symbol ( ) Udu Point; cross ( )

    Ono-I-Lau.

    0

    50

    100

    150

    200

    250

    300

    350

    400

    450

    500

    0

    100

    200

    300

    400

    500

    600

    700

    800

    900

    1000

    0 20 40 60 80 100

    percentiles

    0 20 40 60 80 100

    percentiles

    cumulativerainfall(

    mm)

    (a) (b)

    Figure 3. The cumulative monthly rainfall (mm) over the period 19492008 against percentiles: (a) wettest month (March), (b) driest month

    (July). Symbols: filled triangle () Rotuma; empty circle ( ) Nadi; filled circle ( ) Labasa; filled square ( ) Nausori; empty square ( )

    Nacocolevu; solid line with no symbol ( ) Udu Point; cross ( ) Ono-I-Lau.

    by a Sens Slope 47.1 mm year1 was the highestfor Labasa Mill, followed by 34.9 mm year1 forNadi and 31.3 mm year1 for Udu Point. Consistentwith weakest downward trend, Rotuma demonstrated thesmallest Sens Slope of 12.9 mm year1. In terms ofsignificance, decreases at all stations except Rotuma werestatistically significant at the 95% level of confidence.

    Figure 4 reveals that there was a marked de-cline in rainfall from 1960 onwards. This suggests aprevalence of extreme conditions, particularly in Nadi,

    Labasa and Nacocolevu where changes were mostpronounced. For a closer examination, mean annualrainfall is plotted in intervals of 19491968, 19691988and 19892008 (Figure 5). Clearly, the data show muchhigher rainfall during the first 20 year period comparedto other periods. The rainfall at Nadi and Labasa oscil-lated around 35004000 mm year1 over 19491968.Beyond 1968, the dramatic decrease resulted in aver-age value of10002000 mm year1. Although similartrends were found at Nausori and Rotuma, decreases

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    6/10

    R. C. DEO

    0

    2000

    4000

    6000

    0

    2000

    4000

    6000

    0

    2000

    4000

    6000

    0

    2000

    4000

    6000

    0

    2000

    4000

    6000

    (a)

    ye

    arlyraintall(mm)

    yearlyraintall(mm)

    yearlyraintall(mm

    )

    (b) (c)

    0

    2000

    4000

    6000

    (d)

    0

    2000

    4000

    6000

    (g)1950

    1960

    1970

    1980

    1990

    2000

    2010

    1950

    1960

    1970

    1980

    1990

    2000

    2010

    1950

    1960

    1970

    1980

    1990

    2000

    2010

    (e) (f)

    Figure 4. The time series of annual rainfall for the period 19492008. (a) Nadi (Sens Slope 34.9 mm year1, Z-value 3.1, p-value 0.002);(b) Nausori (Sens Slope 27.6 mm year1, Z-value 2.9, p-value 0.004); (c) Rotuma (Sens Slope 12.9 mm year1, Z-value 2.2, p-value0.026); (d) Labasa (Sens Slope 47.1 mm year1, Z-value 3.3, p-value 0.000); (e) Udu Point (Sens Slope 31.1 mm year1, Z-value 2.9,p-value 0.003); (f) Ono-I-Lau (Sens Slope

    20.9 mm year1, Z-value

    4.4, p-value 0.000); (g) Nacocolevu (Sens Slope

    27.1 mm year1,

    Z-value 2.9, p-value 0.004). Note: dashed line denotes Sens Slope, negative Z-value shows a downward trend and p-value shows statisticalsignificance of trend using the non-parametric MannKendall test.

    were much smaller than those at Nadi and Labasa.From 19691988, Nadi and Labasa experienced lowestrainfall in history of 5001000 mm year1. Althougha decreasing trend was not so noticeable over the period1949 1968, there was a gradual decrease in rainfall overthe period 1969 1988 for Rotuma and Nausori. Overthe last 20 years, rainfall amounts increased monotoni-cally. However, the recorded values remained below the19491968 totals.

    3.2. Drought severity and duration

    The cumulative yearly rainfall, represented as the droughtindex, SPI, is shown in Figure 6. The index is normalizedby its mean and standard deviation so comparisons ofSPI in geographically different regions are possible.Consistent with Figures 4 and 5, SPI shows a generalshift towards increasingly negative values at all stations,with statistically significant negative increases at Nadi,Labasa, Nacocolevu and Ono-I-Lau at the 95% level ofsignificance.

    The distribution of SPI derived from moving accumu-lations of monthly rainfall showed different probabilitiesfor all stations (Figure 7). It is clearly evident that the

    PDF is most symmetric for Rotuma. This agrees withRotuma having more consistent rainfall over the studyperiod and perhaps less dynamic drought conditions.Interestingly, the PDF of Nadi and Labasa shows multi-modal behaviour, with substantial fluctuations in PDFrelative to other stations. A calculation of total areaunder the curve for all SPI 1.0 for Labasa and Nadishowed 12.5% increase relative to Rotuma and 1.24%increase relative to Nausori. Based on the distribution of

    SPI, drought conditions in Nadi and Labasa appear to bemuch stronger than those at other stations.

    The severity and duration of meteorological droughts,

    in the number of drought months, was characterized byconsidering all months with SPI 2.0 as an extremedrought month, all months with 2.0 < SPI 1.5as a severe drought month and all months with1.5 < SPI 1.0 as a moderate drought month(McKee et al., 1993). The choice of time slice was basedon Figure 5 which showed categorically different rainfallregimes prevalent every 20 years.

    All stations except Rotuma and Nausori experi-

    enced highest number of extreme drought monthsduring the period 19491968 (Figure 8(a)). The longestduration was for Labasa (6 months decade1), followed

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    7/10

    DROUGHTS IN TROPICAL PACIFIC ISLANDS A CASE STUDY OF FIJI

    0

    1000

    2000

    3000

    4000

    5000

    6000

    1950 1955 1960 1965 1968

    yearlyrainfall(mm)

    0

    1000

    2000

    3000

    4000

    5000

    6000

    1970 1975 1980 1985 1988

    yearlyrainfall(mm)

    0

    1000

    2000

    3000

    4000

    5000

    6000

    1990 1995 2000 2005

    year

    2008

    yearlyrainfall(mm)

    (a)

    (b)

    (c)

    Figure 5. The 20-year time-slice analysis of total yearly rainfall over

    the periods 19491968, 19691988 and 19892008. Symbols: filled

    triangle () Rotuma; empty circle ( ) Nadi; filled circle ( ) Labasa;

    filled square ( ) Nausori.

    by Ono-I-Lau (5 months decade1) and Nacocolevuand Nadi (4 months decade1). During the period19691988, there was a marked increase of2.5 monthdecade1 for Nausori and 4 months decade1 forRotuma and a relative decrease at other stations. Itis noteworthy that the last 20 year time slice showed

    increases in duration of extreme droughts for Nadi,Labasa, Udu Point, Ono-I-Lau and Nacocolevu but totaldurations for the period 19892008 remained below thedurations for the period 19491968. Together with rain-fall trends (Figure 5), this analysis demonstrates shorterduration of extreme droughts during the first 20 years,an increase during the mid period and a further reductionduring the last two decades.

    For severe droughts, the highest number of droughtmonths registered for the period 1969 1988 was forNadi (9 months decade1) followed by Labasa andUdu Point (7.5 months decade1) (Figure 8(b)). Overthe last 20 years there was a significant reduction forNadi, followed by Labasa, Nausori, Udu Point andNacocolevu relative to total number of severe drought

    months for the periods 19491968 and 19691988.However, for Rotuma, the number of severe droughtmonths increased to 8 months decade1 for the period1988 2008, indicating that Rotuma is becoming lessimmune to dry conditions.

    In terms of moderate droughts, successive increases

    for Nadi, Nausori and Ono-I-Lau during all three peri-ods are suggesting gradual shifts towards a drier cli-mate (Figure 8(c)). For Labasa, there was an increasein moderate drought months during the mid-period rel-ative to first 20 years, but a decline for the period1989 2008. Despite this decrease, the total numberof all drought months (SP I 1.0) over the period19892008 remained significantly higher than the1949 1968 totals (Figure 8(d)). Identical trends in thetotal number of drought months were evident for Nadiand Labasa during all three periods. That is, an increasein the total number of drought months for the period19681988 and a decrease for the period 19892008.

    For Nausori, the total number of drought months hasincreased for the period 19892008 compared to otherperiods, suggesting that the present climate is probablybecoming drier, even on the windward side.

    3.3. Return periods

    The results presented so far have shown increases in theduration and severity of droughts at all stations, albeitwith varying impacts across Fijis islands. In Figure 9,rainfall return periods at each station are shown. For anygiven rainfall amount the return period for Rotuma wasthe smallest. This is consistent with a smaller durationand less severe drought for this region. The return peri-ods for Labasa and Udu Point are similar, but generallysmaller than Nausori for values less than 35 years. It isalso clear that, for return periods of up to 13 years, Nadiand Ono-I-Lau demonstrated the highest return period forany given rainfall. However, for greater than 13 years,the return period for Nadi was the highest, providing fur-ther support to the suggestion that the western regions ofFiji are significantly prone to extreme and severe droughtevents on a long-term scale.

    Based on a 60 year drought history, it is obvious that

    Fijis climate has shifted towards drier conditions sincethe 1950s, but the extent of the shift is distinctly differentthroughout Fiji. These differences appear to be correlatedwith geographic locations (Derrick, 1951). The domi-nant southeast trade winds which bring moist air fromthe ocean, aiding in cloud formation over mountainousregions (Lal, 2004). The mountains separate the wind-ward and leeward side. Hence, Nadi and Nacocolevu,located on the leeward side, receive less rain, and eas-ily fall into drought episodes, while Nausori, being onwindward side, does not. By contrast, Rotuma whichhas no mountainous barriers enjoys surplus oceanic rain.

    With an average annual rain of about 3550 mm, droughtepisodes last only 3 weeks (Dawe, 2001). This is fur-ther supported by the results of Figure 9, which showed

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    8/10

    R. C. DEO

    -2

    0

    2

    (a)

    SPI

    SPI

    -2

    0

    2

    (b)

    -2

    0

    2

    (c)

    -2

    0

    2

    SPI

    -2

    0

    2

    (d)

    -2

    0

    2

    (e)

    -2

    0

    2

    (f)

    (g)

    2010

    2000

    1990

    1980

    1970

    1960

    1950

    2010

    2000

    1990

    1980

    1970

    1960

    1950

    2010

    2000

    1990

    1980

    1970

    1960

    1950

    Figure 6. The time-series of standardized precipitation index (SPI) of yearly rainfall totals over the period 19482008. (a) Nadi (Sens Slope

    0.02, Z-value 9.7, p-value 0.000); (b) Nausori (Sens Slope 0.01, Z-value 7.6, p-value 0.000); (c) Rotuma (Sens Slope 0.001, Z-value3.4, p-value 0.000; (d) Labasa (Sens Slope 0.01, Z-value 9.4, p-value 0.000); (e) Udu Point (Sens Slope 0.001 mm year1, Z-value6.9, p-value 0.000); (f) Ono-I-Lau (Sens Slope 0.002 mm year1, Z-value 8.4, p-value 0.000); (g) Nacocolevu (Sens Slope 0.002 mmyear1, Z-value 8.5, p-value 0.000). Note: dashed line denotes Sens Slope, negative Z-value shows a downward trend and p-value shows

    statistical significance of trend using the non-parametric MannKendall test.

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    -3 -2 -1 0 1 2 3

    moderate

    severe

    extreme

    SPI

    probability(SPI)

    Figure 7. The probability distribution function of SPI derived from monthly rainfall accumulation. The thresholds for different drought categories

    (extreme, severe and moderate droughts) are shown and PDF is normalized such that area under the curve given by

    probability(SPI)

    d(SPI) = 1. Symbols: Rotuma; Nausori; Nadi; Labasa.

    the smallest return period for Rotuma compared to otherstations for a given rainfall amount.

    4. Conclusions

    Analysis of observed monthly rainfall for Fiji overthe period 19492008 showed downward trends at the99% level of significance and decreases in rainfall of

    1347 mm year1 as determined by the Sens Slope.The strongest drought-impacted regions were western andnorthern Fiji (Nadi, Nacocolevu and Labasa). Compar-isons of trends over the periods 19491968, 19691988and 19892008 showed serious rainfall deficiency during

    the mid period and a marked increase in number ofextreme-severe droughts as determined by the Standard-ized Precipitation Index (SPI). While rainfall increased

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    9/10

    DROUGHTS IN TROPICAL PACIFIC ISLANDS A CASE STUDY OF FIJI

    (a) (c)

    (d)(b)

    Figure 8. The total duration of droughts (shown in number of drought months) of different intensities computed from monthly SPI over

    20-year time-slice periods of 1949 1968, 1969 1988 and 1989 2008. The drought intensities are (a) Extreme droughts (SPI 2.0);(b) Severe droughts (2.0 < SPI 1.5); (c) Moderate droughts (1.5 < SPI 1.0); (d) All types of droughts (SPI 1.0). Symbols:

    19491968; 19461988; 19892008.

    0

    1000

    2000

    3000

    4000

    5000

    0 10 20 30 40 50 60

    return period (years)

    yearlyrainfalltotals(mm)

    Figure 9. The return periods of total yearly rainfall. Symbols: filled triangle () Rotuma; empty circle ( ) Nadi; filled circle ( ) Labasa; filled

    square ( ) Nausori; solid line with no symbol ( ) Udu Point; cross ( ) Ono-I-Lau.

    over the last period, this did not outweigh the amountsduring first 20 years. In addition, monthly rainfallpatterns showed a substantial decline during the dri-

    est month, particularly for Nadi and Labasa. Althoughdroughts ranging from moderate to extreme in inten-sity may become common in the future, the scarcity

    of economic data and modelling expertize is inhibit-ing proper diagnosis and prediction Therefore, empiricalstudies using observed rainfall records can be used to

    derive operational drought indices such as the SPI fordiagnosis of droughts in small tropical Pacific islandssuch as Fiji.

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)

  • 7/31/2019 Deo 2010 - On Meteorological Droughts in Tropical Pacific Islands

    10/10

    R. C. DEO

    Acknowledgements

    The rainfall datasets, provided by the Environment Ver-ification and Analysis Centre (EVAC) and Fiji Meteoro-logical Services, is greatly acknowledged. We thank thetwo anonymous reviewers who have provided valuableinputs to increase the clarity of this article.

    References

    Agrawala S, Ota T, Risbey J, Hagenstad M, Smith J, van Aalst M,Koshy K, Prasad B. 2003. Development and climate change. Fiji:Focus On Coastal Mangroves, Working Party on Global andStructural Policies, Working Party on Development Co-operation andEnvironment: Cedex 16, France; 156.

    Basher RE, Zheng X. 1998. Mapping rainfall fields and their ENSOvariations in data-sparse tropical Southwest Pacific Ocean region.International Journal of Climatology 18: 237251.

    Christensen JH, Hewitson B, Busuioc A, Chen Gao X, Held I,Jones R, Kolli RK, Kwon WT, Laprise R, Magana Rueda V,Mearns L, Menendez CG, Raisanen J, Sarr A, Whetton P. 2007.Regional climate projections. In Climate Change 2007: The PhysicalScience Basis, Contribution of Working Group I to the Fourth

    Assessment Report of the Intergovernmental Panel on ClimateChange, Solomon S, Qin D, Manning M, Chen Z, Marquis M,Averyt KB, Tignor M, Miller HL (eds). Cambridge University Press:Cambridge, United Kingdom and New York, NY.

    Dawe P. 2001. Review of the rotuma water supply and distributionsystem. Preliminary Report 131, SOPAC: Suva, Fiji.

    Derrick RA. 1951. Weather and climate. The Fiji Islands. FijiGovernment Press: Suva, Fiji; 102119.

    Feresi J, Kenny G, de Wet N, Limalevu L, Bhusan J, Ratukalou I.2000. Climate change vulnerability and adaptation assessment forFiji. IGCI Technical Report 1, International Global Change Institute(IGCI), University of Waikato: Hamilton, New Zealand.

    Griffiths GM, Salinger MJ, Leleu I. 2003. Trends in extreme dailyrainfall across the South Pacific and relationship to the SouthPacific Convergence Zone. International Journal of Climatology23(8): 847869.

    Hayashi Y, Golder DG. 1993. Tropical 4050 and 2530 dayoscillations appearing in realistic and idealized GFDL climatemodels and ECMWF dataset. Journal of Atmospheric Science 50:464494.

    Hay JE, Mimura N, Campbell J, Fifita S, Koshy K, McLean RF,Nakalevu T, Nunn P, de Wet N. 2003. Climate Variability andChange and Sea-level Rise in the Pacific Islands Region AResource Book for Policy and Decision Makers, Educators and OtherStakeholders. South Pacific Regional Environment Programme:Apia, Samoa; 108 p.

    He Y, Barnston AG. 1996. Long-lead forecasts of seasonal rainfallin the tropical pacific island using CCA. Journal of Climate 9:20202035.

    Hirsch RM, Slack JR. 1984. Non-parametric trend test for seasonal datawith serial dependence. Water Resource Research 20(6): 727732.

    Intergovernmental Panel on Climate Change. 2007. The physicalscience basis. In Contribution of Working Group I to the Fourth

    Assessment Report of the Intergovernmental Panel on ClimateChange, Solomon S, Qin D, Manning M, Chen Z, Marquis M,Averyt KB, Tignor M, Miller HL (eds). Cambridge University Press:Cambridge, United Kingdom and New York, NY.

    Jones C, Waliser DE, Gautier C. 1998. The influence of the Maddenand Julian Oscillation on ocean surface heat fluxes and very highsea surface temperature variability in the warm pool region. Journalof Climate 11: 1057 1072.

    Jones C, Weare BC. 1996. The role of low-level moisture convergenceand ocean latent heat fluxes in the Madden and Julian Oscillation:

    an observational analysis using ISCCP data and ECMWF analyses.Journal of Climate 9: 3086 3104.

    Kendall MG. 1975. Rank Correlation Methods. Griffin: London, UnitedKingdom.

    Kumar V, Deo RC, Ramachandran V. 2006. Total rain accumulationand rain-rate analysis for small tropical Pacific islands: a case studyof Suva, Fiji. Atmospheric Science Letters 7(3): 5358.

    Lal M. 2004. Climate change in small island developing countriesof the South Pacific. Fijian Studies Special Issue on SustainableDevelopment 2(1): 1531.

    Lal M, Harasawa H, Takahashi K. 2002. Future climate change and itsimpacts over small island states. Climate Research 79: 179192.

    McKee TB, Doesken NJ, Kleist J. 1993. The relationship of droughtfrequency and duration to time scales. Eighth Conference onApplied Climatology. American Meteorological Society: Anaheim,CA; 179184.

    Mann HB. 1945. Nonparametric tests against trend. Econometrica 13:245261.

    Manton MJ, Della-Marta PM, Haylock MR, Hennessy KJ, Nicholls N,Chambers LE, Collins DA, Daw G, Finet A, Gunawan D, Inape K,Isobe H, Kestin TS, Lefale P, Leyu CH, Lwin T, Maitrepierre L,Ouprasitwong N, Page CM, Prahalad J, Plummer N, Salinger MJ,Suppiah R, Tran VL, Trewin B, Tibig I, Yee D. 2001. Trend inextreme daily rainfall and temperature in southeast Asia and theSouth Pacific: 19611998. International Journal of Climatology 21:

    269284.Mataki M, Koshy KC, Lal M. 2006. Baseline climatology of Viti Levu

    (Fiji) and current climatic trends. Pacific Science 60(1): 4968.Mimura NL, Nurse RF, McLean J, Agard J, Briguglio L, Lefale P,

    Payet R, Sem G. 2007. Small islands. In Climate Change 2007:Impacts, Adaptation and Vulnerability, Contribution of WorkingGroup II to the Fourth Assessment Report of the IntergovernmentalPanel on Climate Change, Parry ML, Canziani OF, Palutikof JP,van der Linden PJ, Hanson CE (eds). Cambridge University Press:Cambridge, United Kingdom and New York, NY.

    Nicholls N, Wong KK. 1990. Dependence of rainfall variability onmean rainfall, latitude, and the Southern Oscillation. Journal ofClimate 3: 163170.

    Onoz B, Bayazit MC. 2003. The power of statistical tests for trenddetection. Turkish Journal of Engineering Environment Science 27:247251.

    Peterson T, Daan H, Jones P. 1997. Initial selection of a GCOS

    surface network. Bulletin of American Meteorological Society 78:21452152.

    Risbey JS, Lamb PJ, Miller RL, Morgan MC, Roe GH. 2002.Exploring the structure of region climate scenarios by combiningsynoptic and dynamic guidance and GCM output. Journal of Climate15: 1036 1050.

    Salinger MJ, Renwick JA, Mullan AB. 2001. Inter decadal PacificOscillation and South Pacific climate. International Journal ofClimatology 21: 1705 1722.

    Sen PK. 1968. Estimates of the regression coefficient based onKendalls tau. Journal of American Statistics Association 63:13791389.

    Theil H. 1950. A rank-invariant method of Linear and polynomialregression analysis, I, II, III. Nederlandse Akademie WetenschappenProceedings 53: 386392, 512525, 13971412.

    Wigley TML. 1994. MAGICC (Model for the Assessment of

    Greenhouse-gas Induced Climate Change). Users Guide andScientific Reference Manual. National Centre for AtmosphericResearch: Boulder, CO.

    World Bank. 2000. Cities, Seas and Storms: Managing Change inPacific Island Economies, Adapting to Climate Change, Vol. 4 . WorldBank: Washington, DC; 135.

    Yue S, Hashino M. 2003. Temperature trends in Japan: 1900 1996.Theoretical and Applied Climatology 75: 15 27.

    Copyright 2010 Royal Meteorological Society Meteorol. Appl. (2010)