Hydrological Sciences-Journal-des Sciences Hydrologiques, 46(3) June 2001 435

Tropical cyclones and floods in Fiji

RAY KOSTASCHUK
Department of Geography, University ofGuelph, Guelph, Ontario NIG 2W1, Canada
e-mail: [email protected]

JAMES TERRY
Department of Geography, University of the South Pacific, Suva, Republic of Fiji
e-mail: terry [email protected]

RISHI RAJ
Division of Hydrology, Fiji Public Works Department, Suva, Republic of Fiji
e-mail: [email protected]

Abstract Daily flow records, rainfall data and tropical cyclone maps during 1970-
1998 are used to document the impact of tropical cyclones (TCs) on floods in the
Rewa River system, Viti Levu, Fiji. Floods are large, brief, isolated events caused by
TCs and non-TC tropical rainstorms. More floods are caused by tropical rainstorms
Downloaded At: 23:51 28 September 2010

than by TCs, but TC floods are larger. The log Pearson Type III distribution
consistently provided the best fit to partial duration flood series and the widely-
recommended generalized Pareto distribution performed very poorly, underscoring the
need to test a variety of distributions for a particular geographic location. Tropical
cyclones occur more often in Fiji during negative values of the Southern Oscillation
Index (SOP) and all TCs that occurred during El Nino conditions caused floods. Peak
flood discharges caused by TCs are inversely correlated with the SOI, reflecting
possible links with tropical cyclone frequency and precipitation intensity.
Key words Fiji; tropical cyclones; floods; partial duration series; ENSO
Cyclones tropicaux et crues aux Fidji
Résumé Les données de débits journaliers et de précipitations, et les cartes des
cyclones tropicaux (CT) des années 1970-1998, ont été utilisées pour étudier
l'influence des cyclones tropicaux sur les crues de la rivière Rewa, Viti Levu, Fidji.
Les crues sont des événements intenses, brefs et isolés, causés soit par des cyclones
tropicaux soit par des orages tropicaux non-cycloniques. La plupart des crues sont
causées par des orages tropicaux mais celles qui sont causées par les cyclones
tropicaux sont plus intenses. La distribution log Pearson Type III a fourni les meilleurs
ajustements, tandis que la distribution de Pareto, largement recommandée, se révèle
peu pertinente, ce qui souligne la nécessité de tester plusieurs distributions lors de
l'étude d'une station particulière. Les cyclones tropicaux interviennent plus souvent
aux Fidji lorsque l'indice de l'oscillation australe prend des valeurs négatives. En
outre, tous les cyclones tropicaux intervenus pendant une phase d'El Nino ont généré
des crues. Les pics de débit générés par les cyclones tropicaux sont inversement
corrélés avec les valeurs de l'indice d'oscillation australe, suggérant ainsi des relations
avec la fréquence d'occurrence des cyclones tropicaux et avec l'intensité des
précipitations.
Mots clefs Fidji; cyclones tropicaux; crues; séries de durées partielles; ENSO

INTRODUCTION
Large flood events in rivers have a major impact on channel stability and on environ-
mental and socio-economic conditions on floodplains. Most studies of river floods
have focused on humid temperate climates (Newson, 1994), even though large floods

Open for discussion until 1 December 2001

 436 Ray Kostaschuk et al.

are probably more frequent and significant in humid tropical regions (Gupta, 1988).
Tropical cyclones (TCs), also known as "hurricanes" and "typhoons", are a major
cause of floods in the humid tropics (Gupta, 1988) and the recent destruction caused by
Hurricane Mitch in Central America and a series of TCs in southern Africa under-
scores the hazard presented by these storms.
There are an average of five to six TCs annually in the South Pacific (Gupta, 1988)
and the Republic of Fiji has experienced 37 TCs during the period 1970-1998, an
average of less than two per year (Fiji Meteorological Service, 1998). The intense
precipitation associated with these storms generated river floods that caused wide-
spread damage to agriculture, homes and businesses. One of the most destructive
historical floods on Viti Levu (Fig. 1), the largest and most densely populated Fijian
island, was associated with TC Kina in 1992-1993. In the Rewa River watershed alone
(Fig. 1), the damage estimate for TC Kina exceeded F$88 million (c. US$50 million)
(JICA, 1997). Despite their importance to Fiji and other South Pacific Islands, TC-
induced floods are poorly understood. This investigation examines the impact of TCs
on the flood hydrology of the Rewa River watershed, the largest drainage basin in Viti
Downloaded At: 23:51 28 September 2010

Fig. 1 Location of the Rewa River watershed and gauging stations examined in this
study.
 Tropical cyclones and floods in Fiji 437

Levu (Fig. 1). Three issues of importance to watershed planning and flood mitigation
in Fiji in particular and the South Pacific in general are studied herein: (a) the role of
TCs and other tropical rainstorms in flood generation, (b) the partial duration series of
flood flows, and (c) temporal trends in TC occurrence and flooding. For the purposes
of this paper, a flood is understood as a flow that exceeds the capacity of the channel
banks.

STUDY AREA AND DATA SOURCES

The western portion of the Rewa watershed is located on a plateau where maximum
elevations exceed 900 m, although most of the drainage basin falls within hills of
300-600 m in elevation. River channels are steep in their upper reaches (mean slope
= 0.033) and decrease abruptly downstream (mean slope = 0.002) in the alluvial
channels where the gauging stations are located. The northern part of the Rewa
watershed is composed mainly of relatively young basaltic rocks and the southern part
of older plutonic rocks (JICA, 1997). Most of the basin is covered by dense, tropical
rainforest, except for agricultural areas along river floodplains.
Downloaded At: 23:51 28 September 2010

Annual rainfall on Viti Levu is distinctly seasonal, with a wetter season during
December-May and a drier season from June to November. Mean annual rainfall is
higher on the windward eastern side of the island (Fig. 1) and also increases with
elevation. Overall precipitation on the windward side of the island is dominated by the

Vanua Levu

Extent of winds

> I/UKIII/II

90- 119 km/h

180° - 63-89krrtfi

Fig. 2 Track of Cyclone Gavin in 1997 (data from the Fiji Meteorological Service).
 438 Ray Kostaschuk et al.

constant Southeast Trade winds and orographic effects (JICA, 1997). However, storm
events associated with tropical depressions, including TCs, produce the most intense
rainfalls (JICA, 1997). The TC season occurs between October and May.
Maps of TC paths (Fig. 2) were provided by the Fiji Meteorological Service; Fiji
Public Works Department records of total daily rainfall were obtained for several
locations in the Rewa watershed. This investigation focuses on mean daily discharge
from Fiji Public Works Department gauging stations at Nabukaluka on the Waidina
River, Nairukuruku on the Wainimala River, and Navolau on the Rewa River (Fig. 1).
These gauging stations were selected because they have reasonably long and
continuous records and are well upstream of any tidal influence. The Nabukaluka
gauging station drains the smallest area at 253 km2, the Nairukuruku station drains
790 km 2 and the Navolau station, just downstream of the confluence of the Wainimala
and Rewa rivers, drains 1960 km2 (Fig. 1). Bankfull discharge is estimated at
1200 m3 s"1 at Nairukuruku, 2000 m3 s"1 at Navolau and 450 m3 s"1 at Nabukaluka.

FLOODS, TROPICAL CYCLONES AND TROPICAL RAINSTORMS

Downloaded At: 23:51 28 September 2010

The annual hydrograph for 1997 at Navolau (Fig. 3) clearly shows that flow in the
Rewa system is dominated by a series of large, brief and isolated events which increase
in frequency in the wet season. The "flashy" nature of the flood hydrographs and the
rapid response of flow to precipitation (e.g. Fig. 4) reflect both the rugged topography
and the intensity of the storms. The two largest flow events of the year were caused by
TCs Gavin (Fig. 2) and June. There were five other major flow events during the year
produced by non-TC events. Non-TC events are caused by tropical rainstorms that are
common during the wet season.

5000

J F M A M J J A S O N D
1997
Fig. 3 Mean daily discharge at the Navolau gauging station, Rewa River (Fig. 1).
 Tropical cyclones and floods in Fiji 439

250

200

150 E

«
o
05 ioo 'i

15 20 25 30
Downloaded At: 23:51 28 September 2010

December 1992 January 1993

Fig. 4 Rainfall and discharge associated with Cyclone Kina at the Nabukaluka
gauging station, Waidina River (Fig. 1).

Rainfall and discharge records at Nabukaluka (Fig. 4) illustrate the impact of TC

Kina on the Waidina River. Tropical Cyclone Kina tracked to the southeast, with the
eye of the storm passing along the northeast coast of Viti Levu. Intense precipitation
from the TC first appears at Nabukaluka on 31 December 1992, following sporadic
rainfall earlier in the month, and lasts until 4 January 1993. Minor flow peaks around
16 December and 23 December reflect rainfall events preceding TC Kina. Discharge
begins to rise rapidly on 27 December until 30 December, falls slightly on
31 December, then rises again and peaks on 2 January. Discharge remains high for a
single day and rapidly falls to pre-TC values by 5 January. These data show that
discharge at Nabukaluka associated with TC Kina precedes recorded rainfall at the
same site, indicating unrecorded precipitation in the upper part of the Waidina basin
(Fig. 1), either from Kina or another rainstorm.
Table 1 presents the chronology of TCs that have caused overbank floods in the
Rewa River system during 1970-1997. I^ess than half of the TCs that enter Fijian
waters result in floods in the Rewa watershed. Maps of TC paths (Fig. 2) clearly show
that the TCs which have the greatest impact on river flows are those passing closest to
the watershed. The centres of TCs Wally, Bebe and Kina (Table 1), for example, all
tracked within 50 km of the centre of the Rewa basin.
Floods in the watershed are caused by either TCs or non-TC tropical rainstorms
(Fig. 3), so flood observations have been separated into three groups for analysis: all
data, non-TC and TC. Table 2 summarizes the descriptive statistics for all floods at
each station. All three stations have more floods caused by tropical rainstorms than by
TCs, a result of the lower frequency of TCs in the watershed compared to tropical
rainstorms. However, TC floods are larger than non-TC floods, which reflects both the
high intensity of TCs and the direction of movement of the storms. Tropical rainstorms
 440 Ray Kostaschuk et al.

Table 1 Chronology of tropical cyclones that have caused overbank floods in the Rewa River system,
1970-1997.
Tropical Storm duration Nabukaluka: Nairukuruku: Navolau:
cyclone (dd/mm/yy) QP Flood Q„, Flood Flood
(m3 s"1) duration (m s ) duration (m3 s"1) duration
(days) (days) (days)
Priscilla 14/12/70-18/12/70 < < 2003 1
Bebe 19/10/72-06/11/72 < 1 6711 3
Lottie 05/12/73-12/12/73 < < 2304 2
Tina 24/04/74-28/04/74 < < 3100 1
Val 29/01/75-05/02/75 < < 2806 2
Wally 01/04/80-06/04/80 798 2 1853 2 4218 2
Hettie 27/01/82-30/01/82 < < 1609 2 3311 2
Oscar 28/02/83-02/03/83 < < 2759 2 4533 3
Nigel 19/01/85-20/01/85 < < 1368 2 2734 1
Gavin 04/03/85-07/03/85 < < 2140 2 3960 2
Sina 24/11/90-30/11/90 < < < < 2117 1
Joni 06/12/92-13/12/92 < < 2089 1 3581 2
Kina 26/12/92-05/01/93 602 2 7334 4 6923 2
Gavin 04/03/97-11/03/97 < < 6384 3 3927 1
Downloaded At: 23:51 28 September 2010

June 03/05/97-05/05/97 < < 4015 3

Storm duration is the time that the tropical cyclone occupied Fijian waters.
Qp is the peak daily discharge caused by the cyclone; and flood duration is the period that the flow
exceeded bankfull stage.
< indicates no overbank flow, blank cell indicates no data.
Cyclone data from the Fiji Meteorological Service.

usually pass through the watershed from the southeast, whereas TCs frequently
approach from the northwest. This TC path maximizes peak flows, at least in the upper
Rewa and Wainimala sub-watersheds, by causing precipitation in the upper portions of
the watersheds before the lower reaches, thereby concentrating discharge over a
shorter period of time at Navolau and Nairukuruku (Fig. 1). The lower frequency of
TC flooding at Nabukaluka may be due to the westerly orientation of the Waidina sub-
watershed, which would reduce the discharge concentration effect that occurs at
Nairukuruku and Navolau (Fig. 1). Sample standard deviation is higher for TC floods
compared to non-TC floods, a result of the larger mean flows for TC floods and the
variability in TC paths. The coefficient of variability (standard deviation/mean) shows
that flows are most variable at Nairukuruku and least variable at Nabukaluka,
reflecting the weaker influence of TCs at Nabukaluka. Skewness is lower for TC
floods compared to non-TC floods, indicating a more equal distribution of large and
small TC flows compared to non-TC flows.
Table 2 shows that the Navolau station experiences the highest frequency of
flooding for both storm types, followed by Nairukuruku and Nabukaluka. There are no
obvious differences in topography, vegetation, geology or land use (JICA, 1997) that
would account for these variations. The most likely explanation for the high frequency
of flooding at Navolau is that the station is located just downstream of the Wainimala
confluence where flood waves from the upper Rewa and Wainimala would interact
constructively. The low frequency of non-TC flooding at Nabukaluka is more difficult
to explain, but it may reflect a "rain shadow" effect caused by the ridge of hills on the
southern boundary of the Waidina basin (Fig. 1). Tropical rainstorms approaching
 Tropical cyclones and floods in Fiji 441

Table 2 Descriptive statistics for overbank flood flows in the Rewa watershed, 1970-1997.
Nabukaluka: Nairukuruku: Navolau:
All data Non-TC TC All data Non-TC TC All data Non-TC TC
(n = 9) (7i = 7) (n = 2) (7i = 22) (n = 15) (n = 7) (n = 57) (n = 42) (n = 15)
p. (m3 sl) 616 592 700 2643 2351 3192 3293 3130 3756
ô (m3 sl) 94.3 73.1 - 1643 1118 2315 1140 1095 1470
ô/p. 0.153 0.124 - 0.622 0.476 0.725 0.346 0.350 0.391
Sk 0.820 0.854 - 1.600 1.164 0.893 1.261 1.257 0.902
X 0.32 0.25 0.07 1.28 0.83 0.44 2.04 1.50 0.54

n : number of floods; p, : sample mean; ô : sample standard deviation; Sk: sample skewness; À : mean
annual frequency of floods.

from the southeast would lose much of their precipitation orographically on the
windward side of the ridge, resulting in less intense rainfall over the leeward Waidina
basin. It is also possible that the Nabukaluka site is undergoing degradation of the bed,
which would result in fewer overbank flows. However, there is no indication of overall
bed degradation at this site, at least in the last 30 years.
Downloaded At: 23:51 28 September 2010

PARTIAL DURATION FLOOD SERIES

Two approaches are commonly used in hydrological frequency modelling—the annual
maximum series (AMS) (e.g. Mkhandi et al., 2000) and the partial duration series
(PDS). Madsen et al. (1997) note that the AMS approach, although widely used, has
serious limitations. First, it uses the largest annual value, and does not consider
secondary events which may exceed the annual maxima of other years. Second, the
AMS may include annual peaks from very dry years, which can bias the analysis of large
values. The PDS considers flows above a certain threshold and avoids these limitations.
It is also more flexible in flood representation and provides a more complete description
of flood-generating processes (Lang et al, 1999; Waylen, 1991). For this analysis of
Fijian flows over the "bankfull" threshold that result from different processes (TCs and
tropical rainstorms), the PDS approach is more appropriate than the AMS.
The choice of PDS values for modelling involves selection of a threshold and
ensuring that selected peaks are independent of each other (Lang et al., 1999; Madsen
et al, 1997). Threshold selection can be based on physical or mathematical criteria
(Lang et al, 1999). This study is concerned with flows that exceed the river bank
threshold, which is a physical criterion. Several procedures have been proposed to
determine the independence of PDS peaks (see Lang et al, 1999, for a detailed
review), although this can be a complex and somewhat subjective procedure.
Independence of values has been maintained by ensuring that all peaks in the data sets
were from separate storm events and that flows between events had dropped below the
critical bankfull discharge. Floods in the Rewa basin are very flashy and driven by
individual storms (e.g. Fig. 3), so basin saturation by previous storms has a limited
impact on subsequent peaks.
A number of probability distributions have been proposed for PDS, although the
generalized Pareto (GP) distribution has recently gained broad acceptance and has
been recommended in several studies (Birikundavyi & Rouselle, 1997; Rosbjerg et al,
 442 Ray Kostaschuk et al.

1992). Waylen (1991) found that an exponential model, which is a special case of the
GP distribution, provided a good fit to TC-induced floods in Florida. The measured
PDS frequencies (Fig. 5) were compared with the GP distribution (see Madsen et al,
1997, for computational procedures), as well as five other extreme value probability
distributions (2-Parameter lognormal, 3-Parameter lognormal, Pearson Type III, log
Pearson Type III, Gumbel Type 1 Extremal) commonly used for flood frequencies.
The Kolmogorov-Smirnov test was used to select the best distribution for each set of
measured flood flows. Figures 6 and 7 show the best distribution in each case, plus the
GP for comparison. The log Pearson Type III distribution, which was recommended by
the Work Group on Flood Flow Frequency (1977), consistently provided the best fit to
the measured data (Figs 6 and 7), although it usually underestimated the largest flows.
The GP distribution performed very poorly, except for non-TC floods at Nairukuruku,
although it was still worse than the log Pearson Type III for probabilities less than

Nabukaluka
Downloaded At: 23:51 28 September 2010

-
a 1 data non-TC

600 720 840 500 587 673 760

Nairukuruku
12.
o -i

9- 5-
O all data non-TC TC
4-
6 4 3-
2-
E 3 2A
1-
n
0 0
500 3500 6500 500 3000 5500 500 3000 5500 8000

Navolau
25- 25- 8-
20 20'
6-
all data non-TC TC
15 15 4-
10
2-
5-
n-
1500 3900 6300 1500 3700 5900 1500 4500 7500

Discharge (m7s)
Fig. 5 Flood frequencies for the partial duration series.
 Tropical cyclones and floods in Fiji 443

Nabukaluka: ail data

0 0,2 0.4 0.6 0.8 1

Weibull Probability

Nairukuruku: ail data

„« 8000 -r
E
Downloaded At: 23:51 28 September 2010

0-j i j i j i | i | i |

0 0,2 0,4 0,6 0.8 1

Weibull Probability

Navolau: ail data

0 0.2 0.4 0.6 0.8 1

Weibull Probability
Fig. 6 Measured and theoretical cumulative frequency distributions for the partial
duration series. GP is the generalized Pareto distribution.

about 0.75. This result underscores the need to test a variety of distributions for PDS
from a particular geographic location, rather than assuming a particular distribution
(such as the GP) for all PDSs.

TEMPORAL TRENDS IN TROPICAL CYCLONES AND FLOODING

It has been shown that TCs have a major impact on floods in the Rewa watershed, so it
is useful to consider trends in TC occurrence, especially in the context of global
 444 Ray Kostasch.uk et al.

Nairukuruku: nonTC Nairukuruku: TC

„ 50001
log Pearson f
Type r

4000

2000 -

0.2 0,4 0,6 0.E 0.2 0,4 0.6 0,8

Weibull Probability Weibull Probability

Navolau: nonTC Navolau: TC

6000- measured F,
f 6000 log Pearson
es
Type III
4000- | 4000 -
h
| 2000 -
2000- C
Downloaded At: 23:51 28 September 2010

- i — i — i — | — i — r I 0
0,2 0.4 0.6 0.! 0.2 0,4 0.6
Weibuli Probability Weibull Probability

Fig. 7 Measured and theoretical cumulative frequency distributions for the partial
duration series. GP is the generalized Pareto distribution.

warming and the possible increase in TC-induced floods (e.g. Nunn, 1994). The annual
frequency of TCs has decreased in recent decades in the Atlantic (Landsea et al, 1996)
and Australian regions (Nicholls et al, 1998), but increased in the North Pacific (Chan
& Shi, 1996). Figure 8 shows the time between TCs, or inter-event period, of all TCs
in Fijian waters (N = 37: Fig. 8(a)) and TCs that have caused floods at Navolau on the
Rewa River (N = 15: Fig. 8(b)) (Table 1). Linear regression analysis shows that both
TC frequency in Fiji (Fig. 8(a)) and the frequency of flood-generating TCs on Viti
Levu (Fig. 8(b)) do not change significantly over time (R2 « 0.1). This result must be
treated with caution, however, because of the limited geographical extent of the
analysis (i.e. the boundaries of Fiji) and the relatively small number of TCs considered.
In addition, TCs are not segregated into intensity classes (e.g. Nicholls et al, 1998),
which could alter the direction and significance of the trends.
The formation of TCs, and their frequency, depends on several factors including
sea surface temperature (SST), the vertical lapse rate of the atmosphere, vertical wind
shear, mid-tropospheric relative humidity, and the existence of a centre of low-level
cyclonic vorticity (Walsh & Pittock, 1998). Standing oscillations in atmospheric
surface pressure in the South Pacific, termed the Southern Oscillation (SO), have been
shown to affect TC frequency in the South Pacific (e.g. Basher & Zheng, 1995;
Nicholls et al, 1998). The Southern Oscillation Index (SOI), which measures the
surface pressure gradient between Tahiti and Darwin, Australia, is often used as an
indicator of SO behaviour. Generally, SOI is inversely correlated with SST (Neelin &
Latif, 1998). Figure 9(a) shows variations in monthly SOI (based on a three month
running mean) during 1970-1998. Twenty six of the 37 TCs in Fijian waters during
1970-1998 occurred during months with negative SOI, as did nine of 15 flood-
 Tropical cyclones and floods in Fiji 445

- 2.5

a>
•u
o
O 1.5
0)
>

O
0.5
-3
Downloaded At: 23:51 28 September 2010

1970 1975 1980 1985 1990 1995 2000

(b)
7000

- 6000

- 5000
>
2
S O
t| 4000
-o
o
o

3000
O

2000

Fig. 8 (a) Inter-event periods of tropical cyclones (TC) in Fijian waters and mean
monthly Southern Oscillation Index (SOt) for 1970-1998 (data from the Fiji
Meteorological Service), (b) Inter-event periods of tropical cyclones (TC) that have
caused floods in the Rewa River at Navolau and peak discharge (Qp) of those floods
for 1970-1998.

generating TCs (Table 1). Nicholls et al. (1998) found that SOI during August
(SO/AUS), which represents the spring prior to the start of the cyclone season (October-
May) in the Southern Hemisphere, was a useful indicator of TC activity. In Fiji, 22 of
the total of 37 TCs, but only six of 15 flood-generating TCs occurred after negative
 446 Ray Kostaschuk et al.

(a)

cjr

-3 -2
(b) sa
Downloaded At: 23:51 28 September 2010

Fig. 9 Relationships between the number of tropical cyclones during the cyclone
season (TCJ) during 1970-1998 and: (a) mean Southern Oscillation Index of months
when tropical cyclones occurred during the cyclone season ( SOI ), (b) Southern
Oscillation Index during August prior to the start of the cyclone season (SO/Aug).

50/Aug- These simple TC counts suggest a weak correspondence between TC

occurrence and negative SOI and SOIp,np but at least part of this result is due to the
general decrease in SOI during the study period (Fig. 8(a)).
Spearman's rank correlation coefficients (r) were used to examine the functional
link between SOI and TC frequency in Fiji in more detail. TC frequency (TC/) in this
analysis is the number of TCs per cyclone season. Two measurements of SOI are used,
SOlAag and SOI, where SOI is the mean SOI of months that TCs occurred in Fiji. The
correlations between TC/ and both SOI and SOIAng are not statistically significant
(r = -0.05 and -0.11 respectively), but as with the earlier analysis of TC trends, these
correlations must also be treated with caution because the analysis only considers TCs
within Fijian waters, which is a small portion of the Southwest Pacific.
 Tropical cyclones andfloodsin Fiji 447

Many investigations have shown that relationships between SOI and TC frequency
in the South Pacific exhibit strong geographical variability. Lander & Guard (1998)
examined the Southern Hemisphere as a whole and did not find a significant
correlation between SOI and TC frequency, nor did Hastings (1990) who focused on
the entire Southwest Pacific. In contrast, Nicholls et al. (1998) found that, in the
Australian region, TC frequency was directly correlated with SOI measured prior to the
start of the TC season. More detailed regional analyses in the Southwest Pacific,
however, provide some insights into these results. Basher & Zheng (1995) determined
correlations between TC frequency and SOI for four sub-regions of the Southwest
Pacific. They found weak positive correlations between SOI and TC frequency in the
western portion of the region, which includes northeastern Australia, and stronger
negative correlations in the eastern sub-region surrounding Tahiti. The weakest
correlations occurred in the central sub-regions that include Fiji, which is consistent
with the results of this study. Basher & Zheng (1995) suggest that in the western sub-
region, which is roughly midway between the SO poles, seasonal atmospheric
conditions are generally suitable for TC formation so there is a weak correspondence
with SOI and a stronger control by pre-season SST. In the sub-regions further east, TC
frequency is related to the extent of the South Pacific Convergence Zone (SPCZ),
Downloaded At: 23:51 28 September 2010

which is dependent on the SO. The SPCZ shifts towards the east during negative SOI
and expands the area subject to TCs. Fiji's "central" location is such that it could be
affected by both western pre-season SST and eastern SPCZ influences, resulting in
poor overall correlations between SOI and TC frequency.
The SO is linked to the El Nino phenomenon (ENSO) (Neelin & Latif, 1998),
which has a broad impact on precipitation patterns and streamflow characteristics
throughout the world. El Nino refers to the large-scale warming of the whole tropical
Pacific that occurs on average every four years. The El Nino phase (strongly negative
SOIs) typically alternates with a cold phase (strongly positive SOIs) called La Nina.
Hastings (1990) compared TCs during El Nino and La Nina phases in the Southwest
Pacific and found that TC origin points were distributed throughout the region during
El Nino phases but were concentrated in the western portion of the region during La
Nina conditions. Hastings' maps of TC tracks (Hastings, 1990, Fig. 4) clearly show
that Fiji was affected by more TCs during El Nino phases compared to La Nina. This
study found that six of the total of 37 TCs in Fiji waters occurred during El Nino
conditions and two during La Nina, which is in agreement with Hastings' findings.
More importantly, all six of the El Nino TCs were responsible for TC-induced floods
in the Rewa River at Navolau (Table 1).
The SO and El Nino influences on TCs in Fiji suggest possible relationships
between SOI and river flood parameters (e.g. Mosley, 2000). This possibility was
examined by regressing peak discharges of TC-induced floods in the Rewa River (Qp:
Table 1) with SOI measured during the month the TC occurred, and SOIAUO prior to the
season that the TC occurred (Fig. 10). A negative exponential model provided the best
fit for Qp vs SOI and a negative linear model for Qp vs SOIAug. The Southern
Oscillation Index, SOI, explains more than 40% of the variance in Qp and is significant
at the 95% level, while SOIAug explains about 23% of the variance in Qp and is
significant at the 90% level. Although these relationships are rather weak and the data
are not normally distributed, they do show that TC-induced flooding in the Rewa River
is inversely related to SOI\ag and SOI. It was also shown that the largest floods in the
 448 Ray Kostasch.uk et al.

(a)
8,000 - "•s.

7,000 - • •
6,000- ^ ^ ^ ^ v
v

-sa.
5,000 -
%
4,000-
Gf
""""---. • ^"""-v^ ~""~--..

3,000 -
Q0=3388ëec*isa ' ^ ^ ^ \
2
R =0.409 ""-^ ^ ^ « L
2,000 - P=0.01 • ,^Xs ^ > > .
1 1 1 1 1 1
-3.5 -2,5 -1,5 -0.5 0.5 2,5 3.5
sa
(b)

B.UUU
Downloaded At: 23:51 28 September 2010

7,000- •

6,000 ~

5,000-
4,000-
~~~~~~»_
3,000-
~"~~~*-——_ ^~*-~-~_^
2,000- q , =3798-801 SQ^
~ ^ ' ^
R2 =0.233
1,000-
p=0.07
o i i i i i i i i

-2,5 -2.0 -1.5 -1.0 -0.5 0,0 0,5 1.0

sa,
Fig. 10 Regression analyses for the peak discharge of tropical cyclone-induced floods
in the Rewa River at Navolau during 1970-1998 (Q„) vs: (a) Southern Oscillation
Index during the month when the cyclone occurred (SOI), (b) Southern Oscillation
Index during August prior to the start of the cyclone season (S0/Aug). (R2 is the
coefficient of determination and p is the probability that the result occurred by chance;
the dashed lines are 95% confidence limits on the regression.)

Rewa system are caused by TCs that pass close to the watershed. This suggests that the
most likely explanation for the inverse relationships between Qp and SOIAU$ and SOI is
simply that there are more TCs in the Fiji region during negative SOIs (Hastings,
1990), increasing the probability of a TC passing close to Viti Levu and generating a
flood. A related explanation is that TC precipitation intensity in Fiji (e.g. Rodgers et
al, 1994) may also relate to SOIAu„ and SOI, although no data are available to test this
hypothesis. Regardless of the cause, these regressions clearly show that the TC-
induced flood hazard in the Rewa River is influenced by the SO.
Several studies on the Indian peninsula (e.g. Ely et al., 1996) suggest that the
frequency of major floods has increased significantly in recent decades, probably
 Tropical cyclones and floods in Fiji 449

because of an increase in the intensity, rather than the frequency, of TCs. The analyses
described herein indicate that neither TC frequency (Fig. 6(a)) nor the frequency of
flood-generating TCs in Fiji have increased during 1970-1998, although this may be
biased by the small geographical area sampled. Linear regression analysis also shows
that the magnitude and frequency of TC-induced floods in the Rewa River have not
changed significantly over time. The patterns in TC frequency in Fiji thus support the
interpretations from the Indian peninsula, but the Rewa flood data do not demonstrate
an increase in flood frequency over time. The present investigation is over a much
shorter time period compared to those on the Indian peninsula and it is possible that
these flood observations simply do not capture the longer term trends.
This study has shown that TCs have a major impact on flood flows in Fiji.
Concerns have been raised that flooding could increase in the future in response to the
effect of global warming on TC frequency (Nunn, 1994). There is limited empirical
evidence that increases in SST above the 26°C threshold will result in more frequent
TCs and it is likely that changes in TC frequency will be related to changes in
regional-scale circulation patterns, not simply warmer oceans (Hulme & Viner, 1998;
Royer et al, 1998). Some Global Climate Model experiments, however, suggest that
TC intensity could increase and TC frequency could decrease slightly in the Southern
Downloaded At: 23:51 28 September 2010

Hemisphere (Hulme & Viner, 1998; Walsh & Pittock, 1998). If these predictions are
correct, then future floods in Fiji could be less frequent but more severe.

SUMMARY

Floods in the Rewa River system, Fiji, are large, brief, isolated events caused by
tropical cyclones (TCs) and non-TC tropical rainstorms. During 1970-1998, fifteen
TCs caused floods. More floods are caused by tropical rainstorms than by TCs, but TC
floods are larger because of the high intensity of TC precipitation and the direction of
movement of the storms. Tropical cyclones usually travel along the length of the
watershed, which concentrates discharge over a short period of time.
Several probability distributions were tested for partial duration series of TC and
non-TC floods, including the generalized Pareto (GP) distribution which has recently
gained broad acceptance. The log Pearson Type III distribution consistently provided
the best fit to the measured data, underscoring the need to test a variety of distributions
for PDS from a particular geographic location, rather than assuming a specific
distribution (such as the GP).
Tropical cyclones occur more often in Fiji waters during negative values of the
Southern Oscillation Index (SOI) and all TCs during El Nino conditions caused floods.
There are no temporal trends in TC frequency or in TC-induced flood frequency and
magnitude. Peak flood discharges caused by TCs are inversely correlated with the SOI,
however, reflecting possible links with tropical cyclone frequency and precipitation
intensity.

Acknowledgements Funding for this project was provided by grants from the
University of the South Pacific to James Terry and from the Natural Sciences and
Engineering Research Council of Canada to Ray Kostaschuk.
 450 Ray Kostaschuk et al.

REFERENCES
Basher, R. E. & Zheng, X. (1995) Tropical cyclones in the Southwest Pacific: spatial patterns and relationships to the
Southern Oscillation and Sea Surface Temperature. /. Climate 8, 1249-1260.
Birikundavyi, S. & Rouselle, J. (1997) Use of partial duration series for single-station and regional analysis of floods.
J. Hydrol. Engng 2, 68-75.
Chan, J. C. L. & Shi, J. (1996) Long term trends in interannual variability in tropical cyclone activity over the western
North Pacific. Geophys. Res. Lett. 23, 2765-2767.
Ely, L. I., Enzel, Y., Baker, V. R., Kale, V. S. & Mishra, S. (1996) Changes in the magnitude and frequency of late
Holocene monsoon floods on the Narmada River, central India. Geol. Soc. Am. Bull. 108, 1134-1148.
Fiji Meteorological Service (1998) Tropical Cyclone Paths: 1970-1997. Government of The Republic of Fiji.
Gupta, A. (1988) Large floods as geomorphic events in the humid tropics. In: Flood Geomorphology (ed. by V. R. Baker,
R. C. Kochel & P. C. Patton), 301-315. John Wiley & Sons, New York, USA.
Hastings, P. A. (1990) Southern Oscillation influences on tropical cyclone activity in the Australian/Southwest Pacific
region. Int. J. Climatol. 10, 291-298.
Hulme, M. & Viner, D. (1998) A climate change scenario for the tropics. Climatic Change 39, 145-176.
JICA (1997) The Study on Watershed Management and Flood control for the Four Major Viti Levu Rivers in the Republic
of Fiji: Interim Report. Japan International Cooperation Agency, Ministry of Agriculture, Fisheries, Forests and
ALTA, the Republic of Fiji.
Lander, M. A. & Guard, C. P. (1998) A look at global tropical cyclone activity in 1995: contrasting high Atlantic activity
with low activity in other basins. Mon. Weath. Rev. 126, 1163—1173.
Landsea, C. W., Nicholls, N., Gray, W. M. & Avila, L. A. (1996) Downward trend in the frequency of intense Atlantic
hurricanes during the last five decades. Geophys. Res. Lett. 23, 1697-1700.
Downloaded At: 23:51 28 September 2010

Lang, M., Ouarda, T. B. M. J. & Bobée, B. (1999) Towards operational guidelines for over-threshold modeling. J. Hydrol.
225, 103-117.
Madsen, FL, Rasmussen, P. F. & Rosbjerg, D. (1997) Comparison of annual maximum series and partial duration series
methods for modeling extreme hydrologie events. Wat. Resour. Res. 33, 747-757.
Mkhandi, S. H., Kachroo, R. K. & Gunasekara, T. A. G. Flood frequency analysis of southern Africa. II. Identification of
regional distributions. Hydrol. Sci. J. 45(3), 449^164.
Mosley, M. P. (2000) Regional differences in the effects of El Nino and La Nina on low flows and floods. Hydrol. Sci. J.
45(2), 249-267.
Neelin, J. D. & Latif, M. (1998) El Nino dynamics. Physics Today, December 1998, 32-36.
Newson, M. D. (1994) Hydrology and the River Environment. Oxford University Press, Oxford, UK.
Nicholls, N., Landsea, C. & Gill, J. (1998) Recent trends in Australian region tropical cyclone activity. Met. Atmos. Phys.
65, 197-205.
Nunn, P. D. (1994) Oceanic Islands. Blackwell, Oxford, UK.
Rodgers, E. B., Baik, J. & Pierce, H. F. (1994) The environmental influence on tropical cyclone precipitation. J. Appl. Met.
33, 573-593.
Royer, J-F., Chauvin, F., Timbal, B., Araspin, P. & Grimai, D. (1998) A GCM study of the impact of greenhouse gas
increase on the frequency of occurrence of tropical cyclones. Climatic Change 38, 307-343.
Rosbjerg, D., Madsen, H. & Rasmussen, P. F. (1992) Prediction in partial duration series with generalized Pareto-
distributed exceedences. Wat. Resour. Res. 28, 3001-3010.
Walsh, K. & Pittock, A. B. (1998) Potential changes in tropical storms, hurricanes, and extreme rainfall events as a result
of climate change. Climatic Change 39, 199-213.
Waylen, P. R. (1991) Modeling the effects of tropical cyclones on flooding in the Santa Fe River basin, Florida.
GeoJournal 23(4), 361-373.
Work Group on Flood Flow Frequency (1977) (Chairman: Thomas S. Kleppe) A Uniform Technique for Determining
Received
Flood10 Flow
May 2000;
Frequency.
accepted
Water
21 December
Resources 2000
Council. Bull. no. 15, Water Resources Council, USA.