Short-term effects of cyclone impact and long-term recovery of tropical rain forest on Kolombangara, Solomon Islands


D.F.R.P. Burslem (fax 01224 272703; e-mail:
‡Present address: Office for National Statistics, 1 Drummond Gate, London, SW1V 2QQ, UK.


1 We evaluate the effects of large-scale disturbance on tropical tree communities by examining the population dynamics of all individuals > 4.9 cm in diameter at breast height (d.b.h.) of 12 tree species over 30 years (1964–94) in lowland tropical rain forest on Kolombangara, Solomon Islands.

2 During the study period Kolombangara was struck by four cyclones between 1967 and 1970. The last cyclone caused most damage to canopy structure. Mortality in the 6-month interval spanning the first cyclone was 7.0% of all trees, while mean annual mortality for all other intervals (including those spanning other cyclones) was 1.4–2.2% year−1. Mortality varied between species but was independent of topography and geographical location.

3 Recruitment increased from very low rates (median 0.0% year−1) before the first cyclone to median values of 1.6–3.2% year−1 during 1971–79, i.e. following a lag period of 3.5–8 years after the first cyclone. Recruitment rates were higher on plots showing greater mortality rates during this cyclone. Recruitment and mortality rates were still higher in 1994 than they had been before the 1967–70 cyclones.

4 Mean annual mortality rates were positively correlated with mean annual recruitment rates across species. This relationship reflects a continuum of life-history characteristics and contributes to constancy in the relative abundance of the 12 species when the same sets of plots are compared over all measurement intervals up to 30 years.

5 We conclude that cyclone impacts have only short-term effects on the relative abundance of common tree species on Kolombangara, and do not therefore prevent the establishment of an equilibrium rank abundance hierarchy or create spatial variation in tree species composition. Differences in forest composition across Kolombangara are more likely to have been caused by differential anthropogenic disturbance linked to settlement patterns.


Studies of damage and mortality to trees by the catastrophic windstorms in the Pacific and Caribbean variously known as cyclones and hurricanes (Whitmore 1974, 1989; Walker et al. 1991; Everham & Brokaw 1996), as well as studies of the impacts of drought, fire, landslides and earthquakes in a variety of tropical forests (reviewed by Whitmore & Burslem 1998), all suggest that tropical rain forests are non-equilibrium plant communities in which background tree mortality rates are high (1–2% year−1), and that community composition may be influenced strongly by rare, but large-scale, disturbance events.

Cyclones or hurricanes, which occur in two belts 10–15° north and south of the equator, cause massive canopy damage to forests (reviewed by Everham & Brokaw 1996). The heavy rain and high winds associated with these catastrophic windstorms often cause high rates of defoliation, uprooting and snapping of stems and branches to trees in their path (e.g. Wadsworth & Englerth 1959; Unwin et al. 1988; Brokaw & Walker 1991; Bellingham et al. 1994; Zimmerman et al. 1994). However, their impacts on tree population dynamics and community composition are less clear, and therefore more controversial, because the evidence needed to assess these processes can come only from long-term monitoring of tree populations, and long-term data are mostly lacking. In tropical forests, short-term mortality in response to catastrophic windstorms is relatively low (1–25%, Everham & Brokaw 1996), compared with the impacts of other natural landscape-level disturbance factors such as fire or landslides. However, comparisons between studies are complicated by the differing intervals after disturbance during which mortality has been recorded and the (mostly unquantified) importance of delayed mortality of damaged trees (Walker 1995).

Recovery of forest structure after a severe windstorm may occur by one or more mechanisms. These are re-sprouting of damaged stems or crowns, recruitment of new individuals from seed arriving after the disturbance or previously buried in the soil, and release of seedlings and saplings that are present in the forest understorey. At one extreme it is possible that the open, defoliated canopy conditions created by severe windstorms allow mass germination of the seeds of pioneer species (sensuSwaine & Whitmore 1988) which become a recognizable cohort of larger trees within a short period. Then, as in the process of forest regeneration on abandoned agricultural fields, shade-tolerant species might become established and grow up beneath the canopy of the pioneers, and the forest gradually revert to its original composition. This mechanism of forest recovery, however, seems rare, although the early stages have been described in Nicaragua, where a pulse of recruitment of Cecropia spp. followed hurricane Joan in 1988 (Ferguson et al. 1995).

A more general outcome is that short-term recovery of forest structure takes place by re-sprouting of damaged stems and branches (Walker 1991; Yih et al. 1991; Bellingham et al. 1994; Zimmerman et al. 1994). The prevalence of re-sprouts among the stems damaged by hurricane Joan in Nicaragua led Yih et al. (1991) and Boucher et al. (1994) to propose a ‘direct regeneration’ model of forest recovery, in which ‘species dominant in the first years after the disturbance will be the same as the species which were dominant before the disturbance’ (Boucher et al. 1994). If validated, this model would challenge the view that massive disturbances prevent the establishment of an equilibrium species composition.

Observations on two of the species affected by hurricane Joan have supported the ‘direct regeneration’ model (Boucher et al. 1994). However, rigorous validation requires much longer-term data on a wider range of species because of the considerable time-lag between seedling establishment after disturbance and the first record of a stem above the minimum size used in tree population surveys (usually 5 or 10 cm diameter), and because different species in the community may behave differently.

The mechanism by which tropical forests recover from catastrophic windstorms is important because it determines the species composition of the ensuing forest and hence the long-term response of the community to disturbance. Where forests recover primarily by regrowth of damaged stems or release of shade-tolerant seedlings pre-existing under the canopy, then relative abundance hierarchies and forest floristic composition are less likely to change in response to disturbance than when regrowth occurs by recruitment of pioneer species (Vandermeer et al. 1996). A number of studies in tropical forests have found that windstorms do not have major impacts on tree species composition, but again long-term post-disturbance data are lacking (Dittus 1985; Bellingham et al. 1995). The one study with decades of data on post-disturbance tropical forest recovery from a hurricane (Crow 1980 working in Puerto Rico) is complicated by additional human impacts on the forest. Thus it is usually impossible to determine the relative importance of delayed mortality of damaged stems (including those which have re-sprouted), regrowth by sprouting and recruitment by individuals which established in the disturbed forest shortly after the storm.

Populations of the 12 most common big tree species have been studied on natural forest plots on Kolombangara in the Solomon Islands since 1964 (Whitmore 1974, 1989). There is circumstantial evidence that some of these plots occur in forests that have grown up on land close to human settlements that were abandoned approximately 100 years ago (Burslem & Whitmore 1999). In this paper we present data on disturbance history and tree population dynamics over 30 years, during which all the forests were subjected to four cyclones in the period 1967–70. Our focus here is to answer the following questions:

  • • What are the short-term effects of a cyclone on canopy structure and tree mortality?
  • • How long does it take for stem density and basal area to recover to pre-cyclone levels?
  • • Is the relative abundance of the common big tree species influenced in the long term by the impact of cyclones?
  • • What are the relative contributions of anthropogenic disturbance and cyclone impact to the determination of variation in tree species composition across Kolombangara?


Study site and species

Kolombangara in the Solomon Islands (8°S, 157°E) is an extinct Pleistocene volcano with a roughly circular outline which rises from sea-level to a maximum altitude of 1420–1580 m a.s.l. on the crater rim (Fig. 1). The rocks underlying most of the island and all of the plots described in this paper are olivine basalt breccias and lavas (Anonymous 1984). Topsoil is strongly acidic (pHH2O mostly 4.1–5.3 at 0–31 cm; n = 24). A general description of the soils is given by Hansell & Wall (1975) with some further details in van Baren (1961), Lee (1969) and Burslem & Whitmore (1996a).

Figure 1.

Map of Kolombangara showing the location of the 22 permanent sample plots. Roman numerals represent forest type sensu Greig-Smith et al. (1967) and Whitmore (1974).

Kolombangara has an ever-wet tropical aseasonal climate (Neumann 1986 and unpublished data). Long-term rainfall records from the west and north coasts give mean annual values of 3196 mm (range 2571–4012 mm) during 1965–93 and 3035 mm (range 2046–4345 mm) during 1981–93, respectively. On both coasts mean monthly rainfall exceeded 165 mm. There is no regular dry season, but occasionally rain-free periods occur, lasting about a week, or rarely more. On the nearby island of New Georgia mean daily temperature varied between 23.4°C in August and 26.1°C in December, during the period 1962–85 (Neumann 1986).

The vegetation of Kolombangara below 300 m a.s.l. is lowland evergreen tropical rain forest (sensuWhitmore 1975). The 12 most common big tree species were selected for study. Collectively they make up 42% (range 13–69%) of the basal area of all trees > 9.7 cm diameter (1 ft girth) at 1.3 m height and 72% (range 23–100%) of the basal area of all trees > 70 cm d.b.h. on the 22 permanent sample plots (T.C.W., unpublished data). The species are described briefly in Table 1.

Table 1.   Characteristics of the 12 species studied since 1964 in lowland tropical rain forest on Kolombangara, Solomon Islands. Maximum diameter (cm) is given for all stems observed in the study, with total numbers of each species encountered given in parentheses. Species ordered by seedling shade tolerance class
Species and authority (two-
letter code in parentheses)
FamilySeedling shade-
tolerance class1
Max. stem diam. (cm)
(sample size)
density (kg m3)2
Distribution on
  1. 1 Class 1 (most shade tolerant) to Class 4 (most gap-demanding, Whitmore 1974, 1989); 2data from Anonymous (1976) and Anonymous (1979); 3Greig-Smith et al. (1967); 4Parinari salomonensis of Whitmore (1974, 1989); 5Calophyllum vitiense of Whitmore (1974, 1989); 6Calophyllum kajewskii of Whitmore (1974, 1989); 7Elaeocarpus sphaericus of Whitmore (1974, 1989).

Dillenia salomonensis
(C.T. White) Hoogl. (Ds)
Dilleniaceae1158 (227)550West coast lowland
Maranthes corymbosa Bl. (Mc)Chrysobalanaceae1118 (50)720North coast
Parinari papuana ssp. salomonensis
(C.T.White) Prance4 (Ps)
Chrysobalanaceae1142 (234)660General
Schizomeria serrata
(Hochr.) Hochr. (Ss)
Cunoniaceae1152 (57)490Upland
Calophyllum neo-ebudicum
Guillaumin5 (Cn)
Guttiferae2229 (104)500North coast upland
Calophyllum peekelii
Lauterb.6 (Cp)
Guttiferae2164 (154)480Lowlands, avoiding slopes
Pometia pinnata Forst. (Pp)Sapindaceae2145 (137)590North coast, more abundant
in lowlands
Campnosperma brevipetiolatum
Volkens (Cb)
Anacardiaceae3133 (217)330General; in the north mainly
Elaeocarpus angustifolius Bl7 (Ea)Elaeocarpaceae3122 (94)350Mainly north coast
Endospermum medullosum
L.S. Smith (Em)
Euphorbiaceae4107 (46)370Mostly north coast
Gmelina moluccana
(Bl.) Backer (Gm)
Verbenaceae4134 (38)410General
Terminalia calamansanai
(Blco.) Rolfe (Tc)
Combretaceae4124 (86)460Mainly north coast

The lowland forest on Kolombangara shows variation in species composition in relation to geographical location, altitude and topography, decreasing in importance in that sequence (Greig-Smith et al. 1967). Association analysis of the plots considered in this paper identified six floristically distinct ‘forest types’ divided between those on the west coast (forest types I, II and III on Fig. 1) containing Teysmanniodendron ahernianum and Dillenia salomonensis, and those on the north coast (forest types IV, V and VI, Fig. 1) lacking these two species. In addition, plots of Forest type VI, on the north coast, possessed a greater abundance of species with seedlings requiring large canopy gaps for establishment and onward growth, and with size class distributions suggesting that they were not regenerating in situ (Whitmore 1974). There are also ruins of human settlements close to the plots of this forest type and a local oral tradition confirming the existence of villages in this area until the late 19th century, when a civil war and then increased missionary activity led to an emigration of inland human populations to the coast. Therefore, it is possible that forest type VI is a secondary forest that has grown up on abandoned swiddens linked to these settlements.

Plot establishment

Twenty-two plots of 1.5 acres (0.6 ha) each were established in mid-1964 along access lines striking in from the north coast at Shoulder Hill, Lodomae and Rei Cove and from the west coast at Sandfly Harbour and Merusu Cove (Fig. 1). Plots were sited on flat land on ridges or plateaux (nine plots), slopes (nine plots), valleys (three plots) or mixed topography (one plot, excluded from all analyses relating to topographic variation).

Tree and plot assessments

In August 1964, all trees of the 12 tree species recorded in Table 1 that were > 6 inches in girth (4.9 cm d.b.h.) were measured and permanently tagged on all plots. Censuses were repeated in October 1965, March and August 1966, February and August 1967, February and August 1968, February 1971, September/October 1975, February/March 1979, November 1985, April/May 1989, June/July 1991 and February 1994. At each census surviving trees were relocated and new recruits (> 4.9 cm diameter) added to the data-set. On 10 of these occasions all trees were assessed for crown exposure and crown form according to the five-point scales devised by Dawkins (1958). Crown exposure was scored from 1 (stems ‘entirely shaded vertically and laterally by other crowns’) to 5 (‘entirely exposed’). Dawkins' index was used by Davies et al. (1998; their Table 1) working in Malaysia and a slightly modified index was used by Clark & Clark (1992) where further details are available.

The number of plots censused declined from the 22 established in 1964 to nine in 1994: one plot was never re-located after a cyclone in 1967, nine were logged between 1975 and 1985, and three inaccessible plots were deliberately abandoned after 1985 (Whitmore 1974, 1989; Whitmore & Chaplin 1987). Therefore only nine plots, totalling 5.4 ha in area, have records extending over 30 years; these are located along the access lines starting from Shoulder Hill on the north coast and Merusu Cove on the west coast (Fig. 1).

On five occasions (August 1966, April 1970, February 1971, November 1985 and February 1994), all surviving plots were mapped for the stage of the forest growth cycle, on a three point scale, gap, building or mature phase (Whitmore 1974, 1975, 1989). Gap phase forest is defined as possessing an open canopy and potentially containing tree seedlings and saplings up to 0.3 m girth; building phase is a forest of pole size trees (stems 0.3–0.9 m girth); and mature phase is high forest containing trees in all size classes. On the first three sampling occasions, one assessment was made per 20 × 20 m subplot (15 per plot), but these were increased to five and nine assessments per subplot in 1985 and 1994, respectively. Stage of the forest growth cycle was also scored in either August 1968 (13 plots) or March 1969 (five plots not scored in August 1968), but three plots were not recorded at all between August 1966 and April 1970.

Data manipulation and analysis

The data set was scanned for anomalies (such as changes of species identity or gross changes in an individual's size), and trees for which these could not be clarified in 1994 were excluded from the analysis. To control for the possibility that new recruits might have been missed at earlier censuses we estimated what the diameter of each newly recorded stem would have been at the previous census assuming it had grown at the maximum rate recorded for its species and size class in the intervening period (for rates see Burslem & Whitmore 1996a). If this value was above the minimum value for inclusion in the data set its recruitment was re-assigned to the earlier census date. Using this procedure 43 stems (12.3% of all new recruits) were moved to an earlier census date, and an additional 27 stems (7.7% of all new recruits) were thought to have recruited before 1964 and were therefore eliminated from the data-set.

Mean annual mortality rates, m, were calculated according to the following equation (Alder 1995; Sheil et al. 1995):

inline image

where N0 and N1 are number of stems at the beginning and end of the interval t (years). This measure of mortality is derived from an exponential model of population decline and assumes a constant probability of mortality during interval t. An analogous formulation was used for estimating rates of canopy disturbance, d, between intervals, i.e.

inline image

where N0 is the number of mapped points per plot and N1 the number of mapped points that were not disturbed during the interval t′ (years). The same assumptions apply as for the mortality rate calculation discussed above.

Mean annual recruitment rates, r, were estimated as follows (Sheil 1996, 1998):

inline image

where nr is the number of recruits and Nt is the number of stems present at the end of the measurement interval t. This formulation provides recruitment rate estimates that are equivalent in form to the mortality rate estimates (Sheil 1998). In order to minimize the analytical problems associated with comparing mortality and recruitment estimates over intervals of unequal length (Sheil 1995; Sheil & May 1996), we collapsed the five intervals prior to the first cyclone (i.e. August 1964–August 1967) into a single 3-year period. The remaining intervals are all in the range 2.2–4.6 years except the 6-month intervals August 1967–February 1968 and February–August 1968, and the 6.7 years between 1979 and 1985.

We have conducted four types of statistical analysis. First, we compared mortality and recruitment rates before the impact of cyclones with their respective values for intervals during and after the cyclones using Wilcoxon matched-pairs signed-rank tests. For comparisons involving a subset of the original plots (because of loss of plots through logging, etc.), values for the same plots were paired in these tests. Second, for the interval spanning the first cyclone species-specific mortality rates were compared by G-tests employing the Williams correction, and confidence intervals for species-specific rates were derived from the binomial distribution (Sokal & Rohlf 1995).

Third, in order to investigate the temporal pattern of recruitment after the cyclones, we compared the observed frequency of recruits within each interval on the surviving nine plots with a hypothetical frequency estimated from long-term post-cyclone (1971–94) recruitment rates (estimated as above). For those intervals showing a significant difference from expectation, recruitment of each species was compared against the species-specific long-term rate using a similar procedure. In both cases, comparisons of observed and hypothetical frequencies were conducted using G-tests employing the Williams correction (Sokal & Rohlf 1995). It is likely that type I error rates are elevated by this procedure, because recruitment rates calculated on the basis of long-term observations underestimate expected numbers of recruits over short intervals (Sheil & May 1996), hence we do not emphasize marginally significant results in our interpretation of these data.

Finally, product-moment correlation coefficients were determined for the relationships between (a) plot-level mortality rates during the first cyclone and recruitment rates during each of the six subsequent inter-census intervals (six Bonferroni-corrected significance tests, Sokal & Rohlf 1995), (b) species-specific mortality and recruitment rates averaged over the entire 30-year period, and (c) mean stem densities and basal area densities of the 12 tree species at the 1964 census and at each of the subsequent 14 censuses (14 Bonferroni-corrected significance tests for each abundance measure).


Disturbance history

Four cyclones passed close to or across Kolombangara during the 30 years of this study (unpublished data, Meteorological Bureau, Brisbane; Burslem & Whitmore 1996a). These were cyclones Annie (11–12 November 1967), Gisela (3 April 1968), Colleen (28 January 1969) and Isa (17 April 1970). Relative to our mapping of forest growth cycle on all plots, the first three cyclones struck in the first interval and Isa in the second.

The cyclones caused elevated rates of disturbance to some plots (as seen in the annualized percentage reversion to an earlier phase in the forest growth cycle, Fig. 2), although the sampling protocol does not allow us to separate the effects of the first two cyclones for 13 plots (mean 13.3% plot area year−1) or the effects of the first three cyclones for five of the remaining plots (mean 24.5% plot area year−1). Mean disturbance rate was 4.9% plot area year−1 (n = 13 plots) during the interval spanning the impact of the third cyclone in January 1969. Mean disturbance rate was 66% plot area year−1 (n = 21 plots) during the interval spanning the impact of the fourth cyclone in late April 1970, but only 0.8% plot area year−1 (n = 12 plots) between 1971 and 1985 and 1.8% plot area year−1 (n = 9 plots) between 1985 and 1994.

Figure 2.

Mean annual disturbance rate (% plot area year−1) in plots on Kolombangara between all assessments of canopy structure from 1966 to 1994, measured as percentage of plot area which reverted to earlier phases of the forest growth cycle between successive assessments. The sampling strategy and the number of plots assessed changed over time as discussed in the text. The three lines for the intervals between August 1966 and April 1970 record the different pattern of assessments made for groups of plots over this interval; three plots (dashed line) were only visited in 1966 and 1970, five plots (dotted line) were also visited in March 1969 and 13 plots (solid line) were also visited in August 1968 but not March 1969. Arrows and letters indicate the approximate timing of cyclones Annie (A), Gisela (G), Colleen (C) and Isa (I).

Mortality and recruitment rates

Of the 1021 trees present in August 1967, 70 (6.9%) were not recorded in February 1968 and are assumed to have died in response to Cyclone Annie of November 1967. Some trees are likely to have died of other causes during this 6-month interval, but mortality rates before the cyclone were extremely low (0.00–2.61% year−1, see below), so the magnitude of error introduced by this assumption is not likely to be great.

Mortality in response to the November 1967 cyclone did not differ significantly between north (7.25%) and west (6.8%) coast forests (Gadj= 1.42, P > 0.05) or between plots located in valleys (6.2%) or on flat (7.5%) or sloping (6.9%) topography (Gadj= 0.412, P > 0.05), but mortality did differ significantly between species (Gadj= 24.8, P < 0.01; Table 2). The plots most affected by the cyclone were those rich in the most susceptible species rather than those located in a particular geographical or topographic position. Across all species, trees that died did not differ significantly from those that survived in terms of median diameter (23.0 vs. 19.7 cm, respectively) or median crown exposure index (3.0 vs. 3.0) at the August 1967 measurement.

Table 2.   Stem numbers and mortality (%) of 12 big tree species and all stems of those species (bold) on Kolombangara during the interval spanning the impact of cyclone Annie (August 1967–February 1968), with 95% confidence intervals in parentheses; mean annual mortality rates (%) averaged over all intervals not spanning a cyclone (i.e. August 1964–August 1967 and all intervals since February 1971), and over all intervals (i.e. 1964–94), with number of deaths recorded in parentheses
SpeciesCyclone AnnieAnnual mortality rate
 Stem number before cycloneStems dying during periodMortality (%)Without cycleTotal period
Endospermum2300.00 (0.00–14.51)4.27 (11)3.24 (13)
C. neo-ebudicum7822.56 (0.71–9.43)3.20 (31)3.06 (43)
Dillenia17852.81 (1.26–6.65)0.84 (32)1.36 (46)
Terminalia7222.78 (0.67–9.84)1.31 (14)1.32 (19)
Gmelina2813.57 (0.19–17.48)1.51 (6)1.19 (7)
Parinari15385.23 (2.47–9.67)1.72 (47)1.83 (60)
Schizomeria4636.52 (2.29–18.86)1.04 (9)2.66 (14)
All stems1021706.86 (5.54–8.78)1.95 (301)4.10 (432)
C. peekelii137118.03 (4.40–13.90)1.21 (18)2.63 (34)
Pometia931010.75 (5.79–19.17)1.93 (24)3.35 (39)
Campnosperma1361511.03 (6.56–17.45)3.91 (71)4.73 (99)
Maranthes34514.71 (6.90–32.71)1.30 (6)3.35 (13)
Elaeocarpus43818.60 (10.80–35.43)7.01 (32)7.45 (45)

Median annual mortality rate across plots was significantly lower in the interval before the impact of the first cyclone than in any interval subsequently (Table 3). The maximum value (10.9% year−1) was obtained during the 6-month interval spanning Cyclone Annie (August 1967–February 1968), and mortality remained high (3.6% year−1) during the subsequent interval (February–August 1968) which spanned the impact of Cyclone Gisela on 3 April 1968. Median mortality rates peaked again in the interval 1975–79 (3.8% year−1). Median values have been on a downward trend since 1979, but the rate during 1991–94 was still significantly greater than that during the pre-cyclone interval (Table 3). The peak of mortality which occurred on plots during 1975–1979 (Table 3) can be explained by the extremely high mortality rates of individuals recruited during 1971–75 (total across all plots, 37.15% year−1) compared with trees that were not new recruits in 1975 (total across all plots 5.30% year−1).

Table 3.  Comparison of median and range of annual mortality rate (% year−1) across plots for the interval before any cyclones (August 1964–August 1967) and the same subset of plots after each of eight census intervals since August 1967. Samples included all trees > 4.9 cm d.b.h. of 12 big tree species on Kolombangara. Significance of the difference between pre-cyclone interval and subsequent intervals (Wilcoxon matched-pairs signed-ranks tests) indicated as follows: *  P < 0.05; **  P < 0.01
IntervalValues during intervalValues for same plots
during precyclone interval
  • 1

    Interval spanning Cyclone Annie;

  • 2

    2 interval spanning Cyclone Gisela;

  • 3

    3 interval spanning Cyclone Colleen and Cyclone Isa.

August 1964– August 19670.000.00–2.61    
August 1967–February 1968110.910.00–41.330.000.00–2.6121**
February 1968–August 196823.610.00–19.870.000.00–2.6121**
August 1968–February 197131.150.00–4.580.000.00–2.6121**
February 1971–September/October 19751.200.00–3.430.000.00–2.6121**
September/October 1975–February/March 19793.750.66––2.6115**
February/March 1979–November 19852.040.35–5.460.000.00–1.4012**
November 1985–April/May 19892.170.00–5.910.000.00–0.769**
April/May 1989–June/July 19911.320.00–7.820.000.00–0.769*
June/July 1991–February 19941.020.00–4.310.000.00–0.769*

There was a high degree of both spatial and temporal variance in mortality among plots (Table 3). The influence of the cyclones was the major factor determining temporal variance in mortality, even though some plots showed zero mortality during the interval spanning the first cyclone (August 1967–February 1968), and intervals not containing any cyclones had median mortality values of up to 3.8% year−1.

Over the 30-year study period, recruitment of 323 new individuals > 4.9 cm d.b.h. was recorded across the 12 study species; median annual recruitment rates across plots varied in the range 0–4.6% year−1 (Table 4). Recruitment has been episodic rather than continuous, with the peaks occurring during 1975–79 (median 2.8% year−1, n = 21) and 1991–94 (median 4.6% year−1, n = 9). Recruitment rates were low both before and during the period of intense cyclone impact (all medians 0.0% year−1) but at all subsequent intervals (i.e. since August 1968) recruitment was significantly higher than before the first cyclone (Table 4).

Table 4.  Comparison of median and range of annual recruitment rate (% year−1) across plots for the interval before any cyclones (August 1964–August 1967) and the same subset of plots after each of eight census intervals since August 1967. Samples included all trees > 4.9 cm d.b.h. of 12 big tree species on Kolombangara. Significance of the difference between pre-cyclone interval and subsequent intervals (Wilcoxon matched-pairs signed-ranks tests) indicated as follows: *  P < 0.05, **  P < 0.01; –, insufficient data available to make comparison
IntervalValues during intervalValues for same plots
during pre-cyclone interval
  • 1

    Interval spanning Cyclone Annie;

  • 2

    2 interval spanning Cyclone Gisela;

  • 3

    3 interval spanning Cyclone Colleen and Cyclone Isa.

August 1964–August 19670.000.00–0.93    
August 1967–February 196810.000.00–7.900.000.00–0.9321
February 1968–August 196820.000.00––0.9321
August 1968– February 197130.000.00––0.9321*
February 1971–September/October 19751.580.00–10.640.000.00–0.9321**
September/October 1975–February/March 19792.830.00–7.450.000.00–0.9315**
February/March 1979–November 19850.940.00––0.9312**
November 1985–April/May 19892.070.00––0.939**
April/May 1989–June/July 19911.070.00–3.700.000.00–0.939*
June/July 1991–February 19944.641.81–6.600.000.00–0.939**

Mortality and recruitment were not independent at the plot scale. Mortality rates during August 1967–November 1968, the interval spanning Cyclone Annie (Fig. 3), were positively correlated with recruitment rates in 1971–75 (r = 0.674, Bonferroni-corrected P < 0.01), but immediate cyclone-induced mortality did not correlate with recruitment during any other post-disturbance period (Bonferroni-corrected P > 0.05). The temporal pattern of post-disturbance recruitment was examined in more detail considering only the nine plots surviving until the end of the study period (Tables 5 and 6). On these plots mean annual recruitment peaked during 1975–79 (3.1% year−1) and 1991–94 (4.3% year−1), when rates were significantly greater than for the whole 23-year post-disturbance interval (1.5% year−1, Table 5). Sample sizes of individual species are mostly too low to detect differential patterns over time. However, it is clear that an increase in the numbers of recruits of the abundant species Dillenia and Parinari accounts for most of the second peak (1991–94) in community-level recruitment rates (Table 6).

Figure 3.

Relationships between mean mortality rate (% year−1) on 22 plots during August 1967–November 1968 (spanning cyclone Annie) and mean recruitment rate (% year−1) on the same plots during 1971–1975. r = 0.674; P < 0.01.

Table 5.  Final stem number (Nt), observed and expected number of recruits > 4.9 cm d.b.h. (nr), inter-census time interval (t, years) and annual recruitment rates (r, % year−1) on the nine plots surviving during all intervals between February 1971 and February 1994. Significance of the difference between observed and expected frequencies of recruits (G-tests) indicated as follows: ***  P < 0.001
t (years)23.014.613.436.733.442.182.61
nr (observed)11733403423844
r (% year−1)1.461.803.061.431.871.054.28
nr (expected) 26.8519.4834.7418.0211.1115.35
Gadj 1.417.6 ***< ***
Table 6.  Mean annual recruitment rates (r, % year−1) and observed and expected number of recruits > 4.9 cm d.b.h. (nr) during 1975–79, 1991–94 and 1971–94, for 10 tree species growing on the nine plots surviving from 1971 to 1994. Expected number of recruits was determined from post-cyclone recruitment rates (1971–94) on the same plots. Significance of the difference between observed and expected frequencies of recruits (G-tests) indicated as follows: ***  P < 0.001. Two species (Calophyllum peekeli and Endospermum) were excluded from this analysis because they showed no recruitment during the interval on the censused plots
r % year−1nr exp.Gadjr % year−1nr exp.Gadjr % year−1
C. neo-ebudicum7.9572.785.1*10.1951.555.2*2.97

Long-term changes in relative abundance

Mean stem density and basal area of the 12 species were both significantly auto-correlated over all intervals up to 30 years (Fig. 4). The correlation coefficients comparing mean stem density on 22 plots over 1.1 years at the start of the study and mean stem density on nine plots over 30 years were 1.00 and 0.98, respectively, and both were highly significant (Bonferroni-corrected P < 0.001). Autocorrelation of mean basal area declined faster over time, but remained significant (P < 0.01) even comparing the 12 species on nine plots over the full 30 years (Fig. 4).

Figure 4.

Correlation coefficients comparing the mean stem densities (closed symbols) and basal areas (open symbols) of 12 tree species on the same plots between the first census in 1964 and the 14 subsequent censuses. Bonferroni-corrected significance levels indicated as horizontal lines on the figure.

There was no substantial change in relative abundance of the more common species over time despite the high rates of mortality and recruitment generated by the intense period of cyclone activity during 1967–70 (Fig. 5). Some species showed a decline in mean stem density and basal area during the period 1967–68 and a partial recovery of both abundance measures by 1971 or 1975. Figure 5 shows clearly that the relative abundance of the 12 species was maintained over time, despite substantial inter-specific differences in cyclone-induced mortality (Table 2). This lack of change follows from the positive relationship between mortality and recruitment rates across the 12 species (Fig. 6).

Figure 5.

Log mean stem density and log mean basal area of 12 tree species in lowland tropical rain forest on Kolombangara, Solomon Islands, comparing all nine plots over 1964–1994. Two-letter codes correspond to the species listed in Table 1.

Figure 6.

Mean annual mortality and recruitment rates over 30 years (% year−1) for 12 tree species of lowland tropical rain forest on Kolombangara, Solomon Islands. Two-letter codes correspond to the species listed in Table 1. The 1 : 1 mortality = recruitment line is also shown.


Short-term effects of cyclones on forest structure and tree mortality

Cyclone Annie induced immediate mortality on approximately 7% of all trees > 4.9 cm d.b.h. of the 12 common big tree species on Kolombangara (Table 2). This value is similar to overall mortality on plots of subtropical wet forest in Puerto Rico following hurricane Hugo (Walker 1991) and in lower montane rain forest in Jamaica in response to hurricane Gilbert (Bellingham 1991; Bellingham et al. 1992), although it is near the bottom of the range for catastrophic windstorms (reviewed in Everham & Brokaw 1996). The high winds and rainfall associated with such storms cause increased rates of stem and branch breakage, uprooting and defoliation (Wadsworth & Englerth 1959; Unwin et al. 1988; Brokaw & Walker 1991; Zimmerman et al. 1994), either by direct wind damage (wind-throw or defoliation), or indirect effects of wind (large trees and branches damaging small trees, Frangi & Lugo 1991) or rainfall (landslides). Forests in Texas and Puerto Rico suffer higher mortality rates among larger diameter trees of some species (Glitzenstein & Harcombe 1988; Zimmerman et al. 1994), but this was not the case on Kolombangara (Whitmore 1974; Fig. 7.4). Indirect causes of mortality may therefore have been important, although significant trends occurring within some species might have been obscured by pooling the sample.

Cyclone-induced mortality was similar on north and west coasts of Kolombangara and did not differ significantly in relation to topography. Bellingham (1991) also showed no effect of topography or aspect on tree mortality in response to the passage of hurricane Gilbert across plots in lower montane rain forest in Jamaica. In contrast, topography or aspect had significant effects on the amounts of damage sustained by trees in response to tropical cyclones and hurricanes in both of these studies and in numerous others (Webb 1958; Wadsworth & Englerth 1959; Gane 1970; Unwin et al. 1988; Reilly 1991; Basnet et al. 1992). A possible explanation is that the high prevalence of re-sprouting among tropical forest trees uncouples canopy damage and tree death (Bellingham et al. 1995; Paciorek et al. 2000).

The factor most strongly influencing the probability of cyclone-induced mortality on Kolombangara was tree species, with individuals of Elaeocarpus being approximately seven times more likely to die in response to Cyclone Annie than those of Dillenia (Table 2). Other studies have shown differences between tree species in susceptibility to death or damage during severe windstorms (Wadsworth & Englerth 1959; Lugo et al. 1983; Foster 1988; Basnet et al. 1992; Zimmerman et al. 1994). These differences may be linked to wood properties (Putz et al. 1983; Zimmerman et al. 1994) or tree architecture (Foster 1988). Wood density correlates closely with wood strength (Anonymous 1974) and has been shown to correlate negatively with percentage mortality due to wind among tree species of lowland tropical rain forest in Panama (Putz et al. 1983). On Kolombangara a similar explanation may account for the difference in mortality between Elaeocarpus and the much denser wooded Dillenia (see Table 1), but the species with the highest density wood (Maranthes) had the second highest cyclone-induced mortality (Table 2).

On Kolombangara, the peaks of maximum damage to the forest canopy did not coincide with maximum mortality rates (Fig. 2 cf. Table 3). The first cyclone, in November 1967, resulted in the highest rates of mortality, while the fourth cyclone, in April 1970, was responsible for causing the greatest canopy damage. It is possible that the low mortality rates during the last cyclone reflect the fact that most susceptible trees had already died as a result of the earlier storms (Burslem & Whitmore 1996b). However, when two cyclones, separated by 22 months, struck lowland forests of Western Samoa, the opposite trend was seen, with the second cyclone causing higher mortality and lower amounts of uprooting than the first (Elmqvist et al. 1994). Further research is required to explore the structural and demographic significance of multiple disturbance events (Whitmore & Burslem 1998).

Recruitment and ecosystem resilience

Recruitment was low during all intervals up to 1971 before rising significantly in the mid-1970s, i.e. 3.5–8 years after the first cyclone (Table 4). Recruitment rates have remained significantly greater than pre-cyclone values during all intervals between 1971 and 1994. The conclusion that recruitment in the mid-1970s was stimulated by cyclone-induced mortality in 1967 is strengthened by the observation that plots showing higher rates of mortality also showed higher rates of recruitment on these occasions (Fig. 3). This phase of recruitment led to the re-establishment of pre-cyclone stem density and basal area by 1971 or 1975 in most species, i.e. after a lag period of 3.5–8 years (Fig. 5).

The speed with which the forest returned to its pre-cyclone structure and species composition defines one component of its ‘resilience’, or ability to respond, to disturbance (Connell & Sousa 1983). Rapid recruitment rates also contribute to the high resilience of forests in New Hampshire to hurricanes (Foster 1988). Few demographic studies of tropical forests have a long enough post-disturbance monitoring period to capture the phase of recovery reported here, and comparisons between studies are complicated by differences in site histories. Changes in the structure and floristics of a 0.72-ha plot in a Dacryodes-Sloanea forest in Puerto Rico between 1943 and 1976 were reported by Crow (1980). At the start, the forest was recovering from earlier cyclones in 1928, 1931 and 1932 and a selective harvest in 1937. Recovery of stem density and basal area density was complete by about 14 years after disturbance, paralleling the rapid recovery of structural characteristics observed on Kolombangara. However, changes in the density, and relative density, of individual species continued throughout the period of study, but comparisons with Kolombangara are complicated by additional natural and anthropogenic disturbances (Crow 1980).

Long-term dynamics of tree populations

Mortality and recruitment rates were both low before the first cyclone (median values across plots 0.0% year−1 in both cases) and have remained significantly higher than these initial values during all subsequent intervals (Tables 3 and 4). As discussed above, the cause of the increases since 1967 is clearly the impact of up to four cyclones, but the extremely low pre-cyclone values warrant examination. Mortality rates of most other lowland tropical forest tree species and all communities lie in the range 1–2% year−1 (Swaine et al. 1987; Phillips et al. 1994; Lugo & Scatena 1996).

On Kolombangara mortality rates have been declining since the mid-1970s, suggesting that rates of tree death slow down as the period of intense disturbance recedes into the past (Table 3). Recruitment rates have varied more erratically since the mid-1970s increase and showed their 30-year maximum during 1991–94 (Table 4). The western Solomon Islands are at the northern edge of the southern tropical cyclone belt and experience cyclones only rarely (Whitmore 1974). It is possible that the low turnover prior to the impact of the first cyclone in our study reflects a long time since the last previous cyclone had struck, although there are no records to check this. Ongoing monitoring of the remaining Kolombangara plots will determine whether mortality and recruitment fall back to pre-cyclone rates before the intervention of another cyclone.

Differential responses between species

The 12 species are arrayed along a continuum linking mean mortality and mean recruitment rates (Fig. 6), from Elaeocarpus with the highest mortality and recruitment rates to a group of three species (Dillenia, Terminalia and Gmelina) with particularly low rates. A positive relationship between mortality and recruitment rates implies that ‘resistance’ (the ability to withstand disturbance) and ‘resilience’ (the ability to respond to disturbance) were inversely correlated across our sample of 12 species, as predicted by theoretical work (Harrison 1979). Similar differences between species have been noted in other tropical forests recovering from cyclones or hurricanes (Boucher et al. 1994; Bellingham et al. 1995) and in other ecosystems (e.g. Halpern 1988). The strength of this relationship determines the ‘stability’ of the system to disturbance, although Connell & Sousa (1983) caution against inferences of ecosystem stability before the replacement of all individuals in the community has been recorded. For tropical forest communities this would require data collection over several centuries rather than decades (Condit et al. 1995).

The mechanism by which species with high mortality rates achieve high recruitment rates were not studied, but must be linked to greater fecundity per adult plant, increased survival rates of seeds or seedlings or more rapid diameter growth rates. Comparison of demographic data for adults with an earlier study on seedling ecology (Whitmore 1974) shows that there is no simple relationship between seedling shade tolerance and turnover rates of adult plants (Whitmore 1998).

Long-term changes in species composition

Catastrophic windstorms, where they occur, are one of the factors preventing the establishment of equilibrium tropical forest communities, i.e. communities in which a consistent hierarchy of relative abundance is maintained over time (Vandermeer et al. 1996; Whitmore & Burslem 1998). However, as a result of the positive relationships between mortality and recruitment among the 12 common species observed in this study (Fig. 6) their rank hierarchy of relative abundance did not change significantly between 1964 and 1994 (Fig. 5), despite the impact of four cyclones which caused massive amounts of damage to the forest canopy between November 1967 and April 1970 (Fig. 2).

We have no data on changes in relative abundance of any other species in these Kolombangara forests, and it is certainly likely that the rank abundance hierarchies of the rarer species have been less stable. For example, it is probable that a pulse of recruitment of pioneers such as Macaranga spp. would have occurred after the cyclones on Kolombangara, much as Cecropia spp. recruit heavily following the impact of some hurricanes in the Caribbean (Crow 1980; Guzmán-Grajales & Walker 1991; Ferguson et al. 1995). However, of the three species with the most strongly light-demanding seedlings included in the Kolombangara survey (Endospermum, Gmelina and Terminalia, Whitmore 1974) only Terminalia showed a significant increase in recruitment soon after these cyclones (Table 6).

Other studies in Sri Lanka, Puerto Rico, Nicaragua and Jamaica have suggested that tropical windstorms have few long-term effects on forests (Dittus 1985; Walker 1991; Yih et al. 1991; Bellingham et al. 1995). The longer-term record for forests on Kolombangara supports these findings. The Caribbean research has shown that recovery after hurricanes occurs by a combination of vegetative regrowth (‘direct regeneration’) and seedling recruitment from advance regeneration, which together account for the stability in species composition (Walker 1991; Bellingham et al. 1994; Boucher et al. 1994; Zimmerman et al. 1994).

It is now possible to evaluate the potential causes of variation in tree species composition across Kolombangara identified by Greig-Smith et al. (1967) following their census of all stems > 9.7 cm d.b.h. on the 22 plots in 1964. These authors suggested that the floristic differences between north and west coast forests could have arisen because the two coasts had been impacted differentially either by cyclones, or by past human activity. The analyses presented here, and other evidence, lead us to conclude that the latter explanation is more likely (Burslem et al. 1998; Burslem & Whitmore 1999).

Firstly, the consistency in species rank abundance hierarchies over time (Fig. 5), despite the intervening phase of cyclone-induced disturbance, does not support the suggestion that cyclones are responsible for generating the spatial variation in species composition. Secondly, cyclone tracks reconstructed from satellite images made since the late 1960s provide no evidence that one coast is more susceptible to cyclones than the other (see Fig. 2 of Burslem & Whitmore 1996a). Of the four cyclones that have passed close to Kolombangara in the last 30 years, two came from the north-east, one from east-south-east and one from the south (unpublished data, Meteorological Bureau, Brisbane).

By contrast, the physical evidence of past human activity in the inland north coast forests provides strong support for the alternative hypothesis that the variation in species composition derives from anthropogenic disturbance. The species composition of forests in New England, southern Georgia (USA) and Puerto Rico varied more in response to differential land use histories than to more recent hurricances (Foster 1992; Foster et al. 1992; Bratton & Miller 1994; Zimmerman et al. 1995). We suggest that the same may be true for Kolombangara.


The Kolombangara Ecological Survey is a research project of the Solomon Islands Forest Division. We thank the Commissioner for Forests, the Meteorological Bureau, Brisbane, and the Solomon Islands Meteorological Service for access to unpublished data; the Research Division of the Solomon Islands Forest Department for field assistance in 1994; and numerous individuals for contributions to previous phases of the project. The censuses have been funded by the Department for International Development, the National Geographic Society and the Solomon Islands Government. We are particularly grateful to Dr Peter Bellingham, Dr Joe Wright, Dr Nick Brokaw and anonymous referees for comments on the manuscripts and to Dr Douglas Sheil for advice on statistics.

Received 29 September 1999
revision accepted 6 July 2000