Family, visitors and the weather: patterns of flowering in tropical rain forests of northern Australia



    1. Cooperative Research Centre for Tropical Rainforest Ecology and Management, Australian School of Environmental Studies, Griffith University, Nathan, Queensland 4111, Australia
    Search for more papers by this author

    1. Cooperative Research Centre for Tropical Rainforest Ecology and Management, Australian School of Environmental Studies, Griffith University, Nathan, Queensland 4111, Australia
    Search for more papers by this author

    Search for more papers by this author
    • *

      Present address: New Zealand Institute for Crop and Food Research Ltd, Private Bag 4074, Christchurch, New Zealand.

S. L. Boulter (fax + 61 73875 55014; e-mail


  • 1A data base on the flowering phenology of the Wet Tropics bioregion of far northern Queensland, Australia, has been constructed, based upon over 36 774 records from two Queensland-based herbaria.
  • 2Flowering patterns have been analysed against the predictions of three overlapping hypotheses based on climatic, biotic and phylogenetic explanations. No one hypothesis is supported to the exclusion of the others.
  • 3Patterns of flowering in the Wet Tropics show marked seasonal increases and decreases, except in the northern lowlands. In general this seasonality correlates with rainfall and temperature and is exacerbated by increasing latitude and altitude.
  • 4There is little or no statistical evidence for the over-dispersion of flowering times that would indicate a competition-avoidance mechanism: flowering within taxa or morphological groups tends to be clumped (and if not, is random).
  • 5That clumping of flowering within taxa does not coincide with a single season provides support for a mass action hypothesis based on the minimization of generalist predation and/or the avoidance of flower predation.
  • 6Timing of flowering did show some consistency among species within genera and within families, but there was little consistency at higher taxonomic levels. Clear separation of the biotic and phylogenetic hypotheses requires greater knowledge of pollination ecology and phylogeny of this large and diverse flora.
  • 7Understanding of flowering patterns and their underlying determining mechanisms is a key to assessing the ecosystem health of the forest. Our results highlight the importance of competitive interactions and of physical and evolutionary factors as determinants of flowering time, intensity and co-occurrence in tropical forests.


Understanding flowering phenology in tropical rain forests presents problems not encountered in better-known temperate floras. Three characteristics represent particular challenges to the tropical ecologist: geographical position, biological richness and availability of information. The low latitudes, mesic climates and associated year-round biological activity characteristic of many tropical forests means that the mere proposition of seasonality in any phenological feature, including flowering, needs to be established rather than assumed. Secondly, most tropical rain forests exhibit levels of plant species richness an order of magnitude greater than those encountered in temperate regions. This richness includes substantial co-occurrence of congeners and con-familials. Coupled with a high diversity of potential animal pollinators this raises interesting and challenging hypotheses related to resource-partitioning within and among plant taxa. Lastly, there are relatively few tropical rain forest floras for which we have sufficient information to be able to make statistically valid tests of hypotheses at the level of the entire flora.

Flowering is the precursor to the reproductive events that determine the future sustainability of the species. Timing, duration and frequency of flowering are all known to vary amongst individual species in tropical forests (Bawa et al. 2003; Borchert et al. 2004). Deciphering the proximate cues and ultimate causes of phenological variation has led, on one hand, to an expectation that phenological patterns are adaptive, leading to the synchronization of reproductive activity with the availability of biotic resources (i.e. pollinators, predators) and with the peak availability of abiotic resources (e.g. sunlight, water). Alternate theories are based on evidence that phenological patterns are not adaptive, and are therefore conserved among closely related taxa. Numerous hypotheses have been formulated regarding the influence of various factors on flowering phenology (see reviews in Wright & van Schaik 1994; Bawa et al. 2003; Bolmgren et al. 2003) and fall into three categories, as summarized below.

Climatic hypotheses, of which several have been developed, link seasonal increases of phenological activity to predictable seasonal variation in limiting factors. These include the presence of seasonally variable pollinators (Waser 1979; Rathcke & Lacey 1985) and predators (Wright & van Schaik 1994), conditions favourable to dispersal or germination (Frankie et al. 1974; Heideman 1989; Ramírez 2002) and variation in environmental factors that favour or limit plant production, and therefore reproduction (Borchert 1983; van Schaik et al. 1993; Wright & van Schaik 1994). The relative importance of these influences will vary among sites. For example, where forests are dry, moisture availability is expected to become important (Wright & Cornejo 1990), whereas where water is not limited other aspects of seasonality (e.g. solar irradiance, Wright & van Schaik 1994; Graham et al. 2003) may be of greater importance. The phenology of individual species can also be a factor of a tree's functional type (Borchert et al. 2004).

Biotic hypotheses link the activities of pollinator, predator and dispersers to the synchrony of flowering in individual plants (Wright 1996). One such, the pollinator competition hypothesis, assumes that pollinators are a limiting resource and flowering events should be evenly spread through time, a ‘staggered phenological community’ (Pleasants 1980; Bolmgren et al. 2003). This should be more especially the case for species within genera or genera within families where a common origin and morphology may be expected to demand a common suite of pollinators (Pleasants 1980; Primack 1985). The main alternative hypothesis, mass action, suggests that facilitation will be more important than competition and predicts that temporal clumping of flowering periods will both increase the likelihood of successful pollination and decrease the risk of predation upon individual flowers by spreading the risk across more individuals (Rathcke 1983; Sakai 2002). As the hypothesis is based on pollination interactions being less specialized, it is assumed that the risk of receiving foreign pollen is less than the benefits of increased visitation (failure of this assumption, of course, might lead to selection for increased synchrony within species and decreased synchrony among species).

The phylogenetic hypothesis proposes that flowering patterns are in some way influenced or constrained by phylogeny and as a result, taxonomically related species will tend to show similar flowering times (Waser 1979; Kochmer & Handel 1986; Johnson 1992; Bolmgren et al. 2003; but see Ollerton & Lack 1992). Similarities in related groups may include flowering duration and intensity (Johnson 1992). Kochmer & Handel (1986) showed, for example, that certain angiosperm families flowered at particular times of the year between continents. Again, this hypothesis relaxes the supposition of pollinators as a limiting resource and may imply, in consequence, less specialized co-evolutionary interactions between the trees and their associated pollinators.

We make three further points regarding studies of flowering phenology. First, the predicted outcomes of the putatively separate hypotheses, such as temporal clumping of flowering activity, may, in fact, be indistinguishable (Rathcke & Lacey 1985). Secondly, care is needed to distinguish between the ultimate cause of phenological patterns and the proximate cues that individual plants respond to. Theoretical explanations of phenological diversity generally address ultimate causes. For example, in certain tree species, environmental change is known to trigger the phenological events that precede flowering. In this case, climatic cues are ‘proximate mechanisms’ although these may also have been selected for in response to environmental or biotic adaptive pressures (Borchert et al. 2004). Thirdly, the operation of all the various hypotheses may operate, or be detected, at a variety of levels, from microsite to landscape or community level (Borchert et al. 2004). The resulting phenological community pattern is likely, by default, to partially represent variation at smaller scales (see, for example, discussions in Bolmgren et al. 2003; Borchert et al. 2004).

We examine the community-wide patterns of flowering phenology within the tree flora of the wet tropical rain forests of far northern Queensland, Australia (referred to as the Wet Tropics bioregion) using herbarium records. We examine the resultant data base for evidence of the main predictions of the alternate hypotheses concerning the evolution and co-evolution of flowering patterns.

We ask the following:

  • 1Does flowering activity coincide with seasonal variation in climate at the level of the entire community and, in consequence, are flowering patterns accentuated with increased latitude and altitude (Burger 1974; Borchert et al. 2004)?
  • 2Is there evidence of the coincidence (‘clumping’) or divergence (‘staggering’) of flowering among species that would be expected to share pollinators?
  • 3Is there evidence that relatedness affects the timing, concentration and duration of flowering?


the region

The Wet Tropics bioregion (15°39′ to 19°17′ S, 144°58′ to 146°27′ E) lies along the tropical north-east coast of Queensland, from Black Mountain in the north to approximately Townsville in the south. It is recognized for its high floristic diversity (Myers et al. 2000). The 1.8 m ha bioregion is dominated by rain-forested mountains with extensive plateau areas along its western margin and lowland coastal plains along the eastern margin (Sattler & Williams 1999). Approximately 3000 species of plants are recorded in the bioregion, with more than 700 species (23%) endemic to the area (Sattler & Williams 1999). The region has a considerable range of climates associated with gradients of altitude and annual rainfall. Mean annual rainfall ranges from more than 3000 mm to less than 1600 mm (Tracey 1982). The Wet Tropics can be classed as ‘seasonally dry’, with at least 5 months of the year receiving less than 60 mm average rainfall (van Schaik et al. 1993). November is the first month to receive greater than 60 mm on average at most sites, with a summer wet season continuing into March.

data base construction

A data base has been constructed that collates data on the flowering phenology of individual species of trees and shrubs recorded from rain forest of the Wet Tropics bioregion. Records of the month, altitude and latitude of collection of all specimens possessing reproductive structures (flowers or buds) were taken from herbarium specimen sheets at the North Queensland herbarium. Additional records of the same species from the same region have been accessed through HERBRECS, the collection data base of the Queensland Herbarium, Brisbane. A total of over 112 000 records were examined, of which some 36 774 were flowering specimens when collected. All native tree and shrub species from the far north Queensland Wet Tropics bioregion described in Hyland et al. (1999) have been included. Species from the families Asteraceae, Commelinaceae, Ericaceae and Plumbaginaceae that were listed by Hyland et al. (1999) as ‘predominantly herbaceous with occasional shrub forms’ were not included. All species from the families Cyperaceae, Juncaceae, Liliaceae, Orchidaceae, Poaceae and Zingiberaceae were also excluded. The tree flora that we have investigated comprises 1267 species. These include members of 118 families.

Individual flowering records were subdivided into biogeographical units based on latitude and altitude of collection point. Records were deemed to be ‘northern’ if they were recorded at less than 17° south or ‘southern’ if they were collected at latitudes greater than 17° south. This division coincides with the Black Mountain Corridor or Gap, an area that divided the Wet Tropics rain forest into two discrete refugia during the most recent glacial period (Nix & Switzer 1991). This divide is believed to have influenced the distribution of vertebrates in the bioregion (Winter 1997; Schneider & Moritz 1999). Records were also categorized by altitude using a modification of the classification described in Tracey (1982). Records of specimens collected at < 400 m were classified as ‘lowland’, at 400–800 m as ‘upland’ and at > 800 m as ‘highland’.


To check for any bias in collection times that would determine the patterns of flowering seen in the records, we collated the collection month of all specimens (both sterile and reproductive) for all species selected, from the two herbarium collections. The number of species collected by month was markedly different from the number of flowering specimens collected (Fig. 1). The total number of specimens (both flowering and sterile) collected in a month was randomly distributed across the course of the year (Runs test P = 0.113).

Figure 1.

(a) The total number of species collected and (b) the number of those species recorded flowering by month for the Wet Tropics rain forest flora, North Queensland.


The analysis uses statistical techniques widely used in other phenological studies (e.g. Wright & Calderon 1995; Davies & Ashton 1999; Hamann 2004). To estimate the mean flowering times or flowering midpoint, circular vector statistics were employed as flowering events for individual species frequently span the break in years, making it inappropriate to use linear models based on a simple numbering of months.

For all species with 10 or more records the flowering midpoint was calculated as the angle of the mean vector, φ:

φ = arctan(y/x) if x > 0


φ = 180° + arctan(y/x) if x < 0( eqn 1 )



ni is the number of flower records in month i and φi is the midpoint of month i expressed as an angle. The first of January was chosen as 0°, with the midpoint of January expressed as 15° and the midpoint of each succeeding month arbitrarily assigned at 30° increments from that point.

We also calculated the length of the mean vector length, r, as a measure of the concentration of flowering times for all species for which flowering midpoint was calculated (Batschelet 1981):

image( eqn 2 )

We tested the relationship between mean vector length and sample size and found no correlation.

Distributions of flowering times, peak flowering and number of species collected over time, were tested for randomness using Runs tests (Sokal & Rohlf 1995).

climate hypotheses

Patterns of peak flowering were examined collectively across the flora using all species for which the flowering midpoint was determined, and these were compared with climatic patterns for the region to determine if flowering peaks coincide with particular climatic events. Records were then further divided into northern and southern and the flowering peak re-examined. Finally, this was repeated after further dividing the flowering records among the three altitudinal categories. Runs tests were used to determine if seasonal trends were demonstrated (Sokal & Rohlf 1995).

biotic hypotheses

To determine if there is evidence that flowering of Wet Tropics species was selected to avoid pollinator competition or enhance pollinator visitation, the overlap of species that probably share pollinators was quantified and tested to determine if this differed from random expectation. Pollination data were available for no more than a dozen or so of the Wet Tropics flowering plants, and we therefore used information on closely related species groups to assign species to groups that are likely to share pollinators. The overlap of flowering times for selected species was calculated using a pairwise overlap model (Pleasants 1980, 1990; Wright & Calderon 1995). For each species, the proportion of flowering recorded in each month was calculated. The overlap between all pairs of species was calculated as:

image( eqn 3 )

where pik and pjk are the proportion of flowering records for species i and j recorded in the kth month. The mean temporal overlap value was calculated for all pairwise combinations. To generate the null model, the month of the start of flowering was selected randomly for each species in a taxonomic group (family or genus). The shape of the flowering curve for each species was retained in these simulations. In the case of those groups of species where flowering did not occur year-round, the start date was generated to ensure flowering would not extend beyond the flowering season. The mean of all pairwise combinations was then calculated. This procedure was repeated 1000 times. The actual mean overlap index was then compared with the mean of the simulated values. If the actual mean was greater than the median 95 percentile (i.e. greater than 97.5% of the simulated values, equivalent to a P-value of 0.05 in a two-tailed test) of the simulated values, flowering was seen as consistent with an aggregated distribution. If the actual mean fell below this 95 percentile (i.e. less than 97.5% of simulated values), then flowering was categorized as staggered. This test was performed for the 20 families that had at least eight species with eight or more flowering records (Table 1) and for all genera with at least five species with eight or more flowering records (Table 2). To determine if a sample-size effect would bias the outcome of this test, the data set for all families was stratified into those species with greater than eight records, those with greater than 14 records and those with greater than 20 records and the test re-run to see if a greater sample size improved the chances of a significant result.

Table 1.  Index of flowering overlap calculated for the 20 largest families that had at least eight species with a minimum of eight records. The upper and lower limits of the 95% median of 1000 random simulations of overlap are shown. The distribution of flowering is classed as ‘clumped’ for an overlap index above the 95% maximum and ‘staggered’ if less than the 95% minimum
FamilyNo. speciesActual overlap index95% minimum95% maximumDistribution
Annonaceae 90.42770.22680.3444Clumped
Malvaceae 90.42080.37580.4580Random
Monimiaceae 80.25180.23230.3611Random
Sapotaceae 90.38870.35700.4191Random
Sterculiaceae 90.34950.29090.3580Random
Table 2.  Index of flowering overlap calculated for those genera with at least five species for which a minimum of eight records existed. The upper and lower limits of the 95% median of 1000 random simulations of overlap are shown. The distribution of flowering is classed as ‘clumped’ for an overlap index above the 95% maximum and ‘staggered’ if less than the 95% minimum
FamilyGenusNo. speciesActual mean overlap95% minimum95% maximumDistribution
ClerodendrumChlerodendrum 50.46820.28370.4831Random
CombretaceaeTerminalia 60.44950.25270.4405Clumped
EuphorbiaceaeCroton 50.44830.35240.4881Random
Mallotus 70.44340.32010.4373Clumped
Phyllanthus 50.46950.33510.4688Clumped
LamiaceaePlectranthus 70.48150.35090.4553Clumped
LauraceaeBeilschmiedia 70.23530.12250.3047Random
Litsea 60.32790.16950.4322Random
MeliaceaeDysoxylum 60.24760.14370.3606Random
MyrtaceaeMelaleuca 50.43890.26860.4690Random
Rhodomyrtus 70.48150.35090.4572Clumped
PittosporaceaePittosporum 50.45110.11870.3680Clumped
ProteaceaeHelicia 60.32970.20910.3258Clumped
RutaceaeAcronychia 80.29670.23950.3611Random
Flindersia 70.29940.21950.3824Random
SolanaceaeSolanum 80.35420.29760.4091Random
SymplocaceaeSymplocos 90.30870.17620.3146Random

Analysis of data subdivided by altitude and latitude tested whether the over-dispersion hypothesis operates locally rather than regionally, but concerns about statistical power meant that this was only possible for the Myrtaceae, Lauraceae, Sapindaceae and Euphorbiaceae.

phylogenetic hypothesis

Evidence in support of phylogenetic constraints on the timing of flowering was examined by looking at the grouping of flowering time and its temporal concentration across families. Species within all families for which the midpoint was calculated were split into ‘wet season’ (flowering mid-point in December, January, February or March) and ‘dry season’ flowerers and a contingency analysis performed. The contribution of higher taxonomic division of plants, according to the phylogeny reported in Bremer et al. (2000), to the timing of flowering was tested using a heterogeneity G-test (Sokal & Rohlf 1995). This phylogeny is based on recent cladistic methods and the introduction of molecular sequence data. Data were partitioned initially into Eudicots and Laurales. The Eudicots were then further divided into Asterids, Rosids and Monocots. Finally, heterogeneity was evaluated among families within the Rosids and Asterids.

In order to determine if the timing of flowering differed among species within genera and within families, repeated measures analyses of variance were used. Months were treated as repeated measures, species as subjects and the chosen taxonomic unit (family or genus) as the grouping variables. A significant interaction between month and taxa is taken to indicate that the timing of flowering was different among the chosen taxa. Those families with three or more species with a minimum of 10 flowering records, and those genera with two or more species with a minimum of 10 flowering records were used for the analysis. Due to the size of the data sets the family and genera data were analysed separately. The Huynh-Feldt Epsilon correction was used to adjust the degrees of freedom.

Using a Kruskal–Wallis test, mean vector length or flowering concentration (equation 2) was compared for species among families to test if flowering intensity was influenced by membership.


The results of this study have generated data at the species level, represented by the species mean vectors, and two types of community-level data, the number of species flowering each month and the number of species with mean vectors falling in each month, which are summarized for the entire flora.

climate hypotheses

Records for 1371 species from 118 families and 533 genera were examined. Thirty-three of the families were represented by a single species. The family Myrtaceae, in contrast, was represented by 129 species. The number of species with a peak of flowering within a month and the total number of species flowering in a month were non-randomly distributed across the course of the year (Runs test, P = 0.011). November had the highest number of species flowering in any month with over 600 species flowering (Fig. 1b). October, December and January were also months of high flowering whilst the lowest number was recorded for April, and July and August are also months of low activity.

Using the calculated flowering midpoint for all species for which we had 10 or more flowering records, 75 of 511 species had peak flowering in November (Fig. 2). Both flowering activity and calculated ‘peak’ flowering therefore showed a general trend of increasing in the lead up to the wet season, peaking in November and continuing for many species into the first half of the wet season. The lowest number of species displaying any flowering activity occurred at the end of the wet season (March), although this was followed by a small peak in flowering during April and May.

Figure 2.

The distribution of mean flowering times (‘peak’ flowering) calculated as an algebraic vector.

The general pattern differs between the two geographical regions (Fig. 3). Flowering in the more tropical north reflects a similar trend to that seen for all species across the entire region, with a distinct peak in October through to December, and the remainder of the year fluctuating around the same, lower level (Fig. 3a). The south, however, increases from a low period through June to August to a distinct peak at the start of the wet season and then a gradual decrease (Fig. 3b). Runs tests show that both the north and south patterns represent very strong evidence against randomness and thus indicate seasonal increasing and decreasing trends (P < 0.01).

Figure 3.

Flowering midpoint for species recorded (a) north of 17° and (b) south of 17° in the Wet Tropics of northern Australia.

Flowering in the northern zone demonstrated greater variation with altitude. Both the upland and highland zones showed a peak in flowering at the start of the wet season, with the proportion of species flowering in highland areas showing a dramatic increase in flowering activity in October/November. In contrast, in the lowland areas, activity increased only slightly across June and July (the dry season) and again in October (just before the start of the wet season) (Fig. 4a). Runs tests were still significantly different from random (P < 0.05) when tested across the altitudinal groups. Flowering in the south shows a similar pattern at all altitudes (Fig. 4b) although only the Uplands records show a significant seasonal trend when tested with the Runs test (P < 0.01). The greater number of species found at lowland sites may be the result of collecting bias, as these rain forests are more accessible and of greater current extent than their higher counterparts.

Figure 4.

Proportion of species recorded flowering by month at the altitudinal categories lowland (0–400 m), upland (400–800 m) and highland (> 800 m) for rain forest (a) north of 17° and (b) south of 17° in the Wet Tropics of northern Australia.

biotic hypotheses

Analyses of flowering overlap between species within 19 of the 20 largest families were conducted. Moraceae was excluded as only one of its 21 species had the minimum number of records required. Of the remaining 19 families, Lauraceae, Euphorbiaceae, Myrtaceae, Mimosaceae, Rubiaceae and Annonaceae demonstrated an observed mean overlap greater than 97.5% of the simulated values (Table 1). Flowering for species within these families across the entire Wet Tropics appears to be aggregated. In contrast, no family for which the observed mean overlap was calculated, demonstrated a value less than 97.5% of the simulated values and therefore no families demonstrated the staggered flowering distribution predicted by the pollinator competition hypothesis. Our results show a trend of clumped flowering patterns in the more speciose families but this relationship is not significant (logistic regression, P > 0.05). Groups of species with more records were not more likely to be significantly non-random than those with fewer records. Eighteen of the 19 families tested were stratified according to species number and retested (Moniaceae did not have sufficient species with greater than 14). The results remained the same, except for four families when species with greater than 20 records were used, with two showing significant clumping, and two no longer showing a significant result. We conclude that increasing the minimum number of records used does not increase the chances of a significant result.

At the generic level, 25 genera had at least five species with sufficient flowering records and Acacia, Cryptocarya, Endiandra, Mallotus, Phyllanthus, Pittosporum, Plectranthus, Syzygium, Terminalia, Helicia and Rhodomyrtus showed that flowering was aggregated for species (Table 2). Again, there was no evidence of staggered distributions of flowering times at the generic level.

Using the null model technique described for the pollinator competition model, species within the families Myrtaceae, Lauraceae, Sapindaceae and Euphorbiaceae were analysed according to the biogeographical unit (combined latitude and altitude of collection) in which the records were made. Some variation was evident among units (Table 3). Species recorded for Myrtaceae in the south showed a clumped distribution, whereas in the north their flowering overlap could not be distinguished from random. When further subdivided into altitudinal categories, only the highland species in the south showed any deviation from random, having an aggregated distribution. In contrast, species in the family Sapindaceae showed a staggered distribution in the south, although this trend was not reflected in data when further divided into altitudinal categories.

Table 3.  Index of flowering overlap calculated for the four largest families divided into individuals recorded north of 17° and south of 17° and at altitudes < 400 m (lowlands), 400–800 m (uplands) and > 800 m (highlands)
FamilyLatitudeAltitudeNo. speciesActual overlap95% minimum95% maximumDistribution
Uplands 40.19520.07140.4000Random
Uplands 50.27010.11260.3645Random
Highlands 70.25780.12860.3147Random
Uplands + Highlands 50.27760.16890.3439Random
Highlands 60.24590.12890.2889Random
Lowlands 50.19330.08830.3200Random
Uplands + Highlands 50.21220.16220.4689Random
Lowlands 90.23330.14860.3015Random
Uplands + Highlands220.16100.14710.1925Random

the phylogenetic hypothesis

The Annonaceae were excluded from the heterogeneity analysis as this was the only family representing the higher clade, Magnoliales. In this instance, it would be impossible to determine if heterogeneity occurs within or among the higher clade. When flowering midpoints were divided into wet and dry season flowering, a number of families flowered exclusively in one season (Oleaceae, Combretaceae and Annonaceae in the wet and Pittosporaceae, Caesalpiniaceae and Capparaceae in the dry season), while other families appeared to be dominated by either wet or dry season flowering (Table 4). Not surprisingly, the distribution of flowering across seasons varied significantly among families (GHf = 122.01, d.f. = 38, P < 0.001). There was significant heterogeneity among families within the Asterids and Rosids (Table 5). The majority of families within each of these higher clades showed dominance of one or other season (Table 4), but there was no consistent pattern of flowering peaks within each clade. Further, when Asterids and Rosids were compared within the yet higher clade, the Eudicotyledones, no significant difference was indicated.

Table 4.  Number of species whose calculated flowering peak falls in the ‘wet’ or ‘dry’ season in each family
  OrderFamilyNumber of species
 EumagnoliidsMagnolialesAnnonaceae 0 6
Eudicots ProtealesProteaceae1511
EudicotsAsteridsApiaceaeAraliaceae 1 4
EudicotsAsteridsApialesPittosporaceae 6 0
EudicotsAsteridsEricalesEpacridaceae 2 2
EudicotsAsteridsEricalesMyrsinaceae 4 1
EudicotsAsteridsEricalesSapotaceae 2 6
EudicotsAsteridsGentianalesApocynaceae 3 6
EudicotsAsteridsGentianalesRubiaceae 711
EudicotsAsteridsLamialesAcanthaceae 4 1
EudicotsAsteridsLamialesLamiaceae 6 1
EudicotsAsteridsLamialesOleaceae 0 5
EudicotsAsteridsLamialesVerbenaceae 2 2
EudicotsAsteridsSolanalesBoraginaceae 2 2
EudicotsAsteridsSolanalesSolanaceae 5 1
EudicotsRosids Elaeocarpaceae 6 5
EudicotsRosidsBrassicalesCapparaceae 4 0
EudicotsRosidsCelastralesCelastraceae 3 5
EudicotsRosidsFabalesCaesalpiniaceae 5 0
EudicotsRosidsFabalesFabaceae 9 4
EudicotsRosidsFabalesMimosaceae12 2
EudicotsRosidsMalpghialesFlacourtiaceae 1 4
EudicotsRosidsMalvalesMalvaceae 5 1
EudicotsRosidsMalvalesSterculiaceae 7 2
EudicotsRosidsMalvalesThymelaeaceae 3 1
EudicotsRosidsMyrtalesCombretaceae 0 6
EudicotsRosidsOxalidalesCunoniaceae 4 1
EudicotsRosidsRosalesRhamnaceae 6 2
EudicotsRosidsRosalesRosaceae 3 1
EudicotsRosidsSapindalesAnacardiaceae 3 2
EudicotsRosidsSapindalesMeliaceae 6 2
EudicotsRosidsSaxifragalesGrossulariaceae 4 1
Laurales LauralesMonimiaceae 2 2
Laurales LauralesLauraceae 836
Table 5.  The contribution of higher taxa to the heterogeneity of flowering season (dry or wet)
Eudicots vs. Laurales32.52213293 1< 0.0001
Asterids vs. Rosids 2.558949257 1NS
Families (Asterids)31.867205413< 0.0001
Families (Rosids)49.5399893219< 0.0001

We also showed (Fig. 5) that there was no significant relationship between similarity in flowering patterns (as indicated by correlation coefficients based on the number of flowering records each month) and the phylogenetic distances among those families showing clumped distributions of flowering (as determined in our tests of the biotic hypothesis).

Figure 5.

Phylogenetic distance vs. the correlation coefficient of flowering per month for each pair of the five families that demonstrated a significantly clumped distribution (Table 1).

The number of flowering records differed significantly among months and among taxa, for both genera and families (repeated measures anova, P < 0.0001, Table 6). This result is reflected in the observed patterns of increased flowering at the end of the dry/beginning of the wet season.

Table 6.  Results of a repeated measures anova of monthly flowering within (a) genera and (b) families
SourceSum-of-squaresd.f.Mean squareFP-value
(a) Monthly flowering within genera
Between species
 Genus 1227.46 11510.671.57< 0.01
 Error 1977.67 290 6.82  
Within species
 Month  367.02  1133.377.33< 0.0001
 Month × genus 8679.311265 6.861.51< 0.0001
 Error14517.073190 4.55  
(b) Monthly flowering within families
Between species
 Family 1289.34  4628.032.89< 0.0001
 Error 5106.08 526 9.71  
Within species
 Month  313.04  1128.455.69< 0.0001
 Month × family 4805.43 506 9.51.9< 0.0001
 Error28924.185786 5  

At both the generic and family level, there is a significant difference in flowering rates for species within these taxonomic groupings by month (within- and between-species interaction, Table 6), which accords with the predictions of the phylogenetic hypothesis.

Duration and intensity of flowering, measured as flowering concentration, and indicated by the mean lengths of the flowering vectors, also differed significantly among families (Kruskal–Wallis test statistic = 143.67, d.f. = 41, P < 0.0001). The families, Lauraceae, Annonaceae and Anacardiaceae had the highest mean vector lengths (Table 7), with species flowering intensely over short time periods. Species within the families Annonaceae and Anacardiaceae all flowered in 7 or less months of the year. Of the 51 species from the family Lauraceae used in the analysis, 48 flowered in ≤ 7 months and 28 (more than half) flowered in ≤ 4 months of the year. Families with low mean r-values included the Dilleniacae, Thymelaeaceae and Epacridaceae (Table 7). Flowering across species in these families was of similar intensity across at least 8 months.

Table 7.  The mean concentration of flowering (r) for all species by family
FamilyMean rSE

We conclude that there may well be a phylogenetic dimension to coincidence of flowering of species within genera and families. There is, however, no convincing evidence of this at higher levels.


community-level, seasonal patterns

Community-level flowering in the Wet Tropics of Australia shows a distinct annual rhythm, with many species at peak flowering near the beginning of the wet season. Synchrony of flowering associated with climatic variation is a widespread phenomenon (Frankie et al. 1974; Morellato et al. 2000). While there is a large diversity of phenological patterns amongst trees of seasonally dry tropical forests, late dry-season flowering is common (Frankie et al. 1974; Heideman 1989; Borchert 1994; Ramírez 2002; Borchert et al. 2004). Climactic changes are known proximate cues of vegetative phenology, and this, taken together with the fact that vegetative phenology is a strong determinant of flowering time in seasonally dry forests (Borchert et al. 2004), suggests seasonal climate variation should be an important determinant of community level flowering patterns.

At the community level, peak flowering coincides with the passage of the sun directly over the Wet Tropics. This meets the predictions of van Schaik et al. (1993) and Wright & van Schaik (1994) that peak flowering along a latitudinal gradient closely tracks the position of the sun. Maximal irradiance coincides with the first month for which average rainfall exceeds 60 mm and the strong selective pressure on phenology that this combination is expected to exert appears to be reflected in community-wide flowering patterns.

When the flowering records were partitioned into latitudinal categories, the south, despite a similar seasonal trend to the whole community, had a longer peak flowering season, extending throughout the entire wet season. When the data were further partitioned into altitudinal categories, the trend for increased flowering activity at the end of the dry season accentuated with increased elevation. The flowering patterns of the northern lowlands differed considerably from that of overall wet-season domination as seen across the entire flora. Variation in flowering patterns with changes in latitude and altitude provides secondary evidence for the influence of variations in environmental factors.

The coincidence of flowering with seasonal environmental variation also coincides with increased pollinator activity, especially of insects, and it might be argued that climate is both a proximate cue and an ultimate cause of flowering patterns. Certainly insect abundance and biomass peaks during the wet season at an upland site within the Wet Tropics have been correlated with seasonal increases in resources such as flowers, new leaves and fruit (Frith & Frith 1985). Whether the coincidence of greatest insect activity with flowering activity is a cause or an effect, however, remains a moot point (Rathcke & Lacey 1985).

species level flowering: biotic or phenological causes

Asynchronous flowering

Temporally segregated flowering, the key prediction of the pollinator competition hypothesis, was not demonstrated for any group of congeners or confamilials across the entire Wet Tropics of Australia (with one dubious exception in the southern Sapindaceae). Strictly the ‘staggered’ result for individuals of the Sapindaceae found in the south could be taken as an indication that they had evolved flowering patterns in an environment of pollinator scarcity. The isolated nature of this result, both taxonomically and regionally, leads us to treat this outcome with great caution. There can be little doubt that, in general, our analyses give little if any support to the hypothesis that implies competition for pollination resources. This result is not unexpected. Few studies have demonstrated such divergence of flowering times (Stiles 1975; Ashton et al. 1988; Wright & Calderon 1995; Borchsenius 2002). For those that have, there is stronger evidence that staggered flowering may result from avoidance of interspecific pollen transfer (Waser 1983) rather than pollinator competition. Evidence that pollinators are a limiting resource is seldom found (Rathcke & Lacey 1985). While flowering time is ‘somewhat’ variable, if it is accepted that flowering phenology is under strong phylogenetic constraints (Ollerton & Lack 1992; cf. Rathcke & Lacey 1985), then other premating isolating mechanisms may be more effective in reducing interspecific competition and gene flow. Wright & Calderon (1995) warned against interpreting low evidence of staggering amongst species in a community-wide study of flowering on Barro Colorado Island, Panama, as evidence against the shared pollinator hypothesis. They suggested, as an alternative, that selection to avoid pollinator competition simply does not obscure phylogenetic patterns of flowering. It could be argued that ‘random’ flowering times may also lessen competition for pollinators. If this is accepted, then the evolution of asynchronous flowering may not be necessary for successful reproduction. In the present study, failure to demonstrate staggered flowering times may also reflect the temporal and geographical scale at which data have been collected and analysed. While staggered flowering of species has been shown across several months (Rabinowitz et al. 1981), other authors have found that, for many species, anthesis occurs over a matter of days or weeks, rather than months (e.g. Syzygium tierneyanum, Hopper 1980; Shorea spp., Yap & Chan 1990; Shorea parvifolia, Sakai et al. 1999; Uvaria elmeri, Nagamitsu & Inoue 1997). Asynchronous flowering may be more difficult to detect in comparisons of flowering based on monthly analyses.

Synchronous flowering

Evidence of synchronous or clumped flowering among particular taxa satisfies, at least in part, the predictions of the climatic hypothesis, one version of the biotic hypothesis, i.e. ‘mass action’ and the phylogenetic hypothesis. Separating the evidence in support of each of these competing drivers is accordingly difficult. The coincidence of a large number of species with seasonal variation in climate suggests that at the least, environmental factors cue phenological activity in a large number of species. Separating the influence of biotic and phylogenetic factors, however, requires more data than are currently available. This analysis could be greatly enriched by knowledge of the pollinators of a large number of species (currently for the Wet Tropics we are aware of no more than 20 published studies of pollination) and resolution of phylogeny to the level of species. With this information, flowering data could be divided into actual pollinator groups and techniques such as phylogenetically independent contrasts (Felsenstein 1985; Harvey & Pagel 1991) could be used to partition out the effect of phylogeny.

Our analysis thus far, has shown that the coincidence of flowering among the species of each family is limited, with only five of the 19 families examined being significantly ‘clumped’. Further, evidence of clumping was more likely to be found among congeneric species and often, but not always, in genera belonging to ‘clumped’ families. In addition, we demonstrated that species within many families flower preferentially in the wet or dry season. For example, peak flowering for species of the Lauraceae family was restricted to the period between October and June, with over 80% of species at their flowering peak in the wet season. In contrast, 12 of the 14 species tested for Mimosaceae flowered in the dry season. In addition, several smaller families for which the null hypothesis was not tested also showed a strong wet or dry season dominance in flowering (Tables 1 and 4). We note that while community-level data show a seasonal peak in flowering (suggesting climatically driven seasonality), when species are considered community-wide, closely related species (i.e. confamilials) exhibit clumped flowering at times of the year other than at the start of the wet season. This highlights the variation in strategies adopted by plants. Again, we emphasize that the physiology of individual plants, as outlined by Borchert et al. (2004), will be an important determinant in their response to environmental changes.

If the timing of flowering (i.e. wet or dry season) or coincidence of flowering is a factor of phylogeny, then we would expect that phylogenetic distance between families would relate to the timing of flowering. This was not the case. For example, those species that showed clumping often did so at different times of the year and seasonally flowering families showed little recent common ancestry. We did show, however, that when grouped by genera or family, the timing, duration and concentration of flowering, was consistent among members of particular taxonomic groups (i.e. family or genera). Similar results have been found from other floras (Wright & Calderon 1995; Bawa et al. 2003), although Bawa et al. (2003), analysing flowering among confamilial species, found that while phylogeny appeared to constrain the frequency of flowering (i.e. annual, supra annual), timing and duration of flowering were not so constrained. We conclude that any possible ‘phylogenetic’ element of flowering patterns, in the case of the Wet Tropics of Australia, is restricted to the level of family or genus.

For plants that rely on the transfer of pollen between individuals, the influence of flowering time on success has led to a common expectation that this trait should be subject to selection. The evidence that plant species in the Wet Tropics of Australia, at least at the level of genus and family, show some relationship between taxonomy and flowering phenology may be evidence of this selection. Closely related species demonstrate considerable morphological similarity, however, and morphological characters of flowers, such as size, shape, scent and nectar production will determine the attraction and success of pollinators. While the operation of biotic selection and phylogenetic conservancy is impossible to clearly distinguish, it may not make sense to try. It is not unreasonable to expect that evolution generates patterns of trait variation that are both correlated with phylogeny and maintained by selective forces, in this case biotic pollinators (Westoby et al. 1995). Bolmgren et al. (2003) argue that a lack of clear evidence of biotic influences cannot simply be assumed to demonstrate phylogenetic conservancy (cf. Ollerton & Lack 1992), and we would agree with this position and the corollary.

Distinction between the relative influence of phylogeny and proximate factors may rely on secondary evidence. For example, if phylogenetic constraints are stronger than local pressures, then species of a family should flower at similar times regardless of their geographical location. In addition, flowering times across families should more or less coincide depending upon their taxonomic affinities, particularly within orders (Kochmer & Handel 1986). In contrast, if causal relationships between abiotic factors and phenology exist (i.e. seasonality is simply climate driven) then timing and duration of flowering should vary among individuals of a species collected at different geographical locations (Borchert 1996), and increased latitudes (van Schaik et al. 1993) and altitude (Burger 1974). As discussed above, community-level flowering patterns were certainly influenced by increases in latitude, although increased altitude simply accentuated the peaks and troughs. Whatever the alternate explanation, the variation of flowering patterns with increased latitude and altitude supports the notion that, where abiotic processes favour phenological convergence (in this case different rainfall and temperature gradients), it is the strength of these processes that determine the ‘shape’ or pattern of flowering trends.

data constraints

This analysis is based on extensive herbarium records. Accordingly the data may reflect collecting biases. Few studies have utilized herbarium records or other mass collections to supplement phenological information derived from direct observation in the field (e.g. Burger 1974; Croat 1975; Wright & Calderon 1995; Borchert 1996). Flowering periods derived from herbarium records, although in some cases longer, have been demonstrated to reflect those from field surveys (Borchert 1996). The Atherton herbarium from which these data were principally compiled has samples from over 100 years of collecting and contains substantial collections made within ecologically or biogeographically motivated surveys. In this case, the collection provides an extensive knowledge of flowering for a large proportion of species found across the Wet Tropics, for which there are few published field studies. Data from two separate studies, from one lowland and one upland site, recorded peak flowering in September–October from 4 years of data (Hopkins & Graham 1989), and October and January in two consecutive years (Frith & Frith 1985). These studies offer results not dissimilar to the conclusion here that peak flowering occurs on average in the period of October–November. We show that there is little similarity between collecting patterns and flowering pattern, with the flowering data showing a marked increase in activity at the end of the dry season, not reflected in collecting data (Fig. 1), and conclude that the herbarium data are a fair reflection of flowering patterns.

Studies conducted at the species and population levels can show wide variation in flowering patterns (Sakai 2002), with annual rhythms observed at the community level not reflected by all individual species (Newstrom et al. 1994). We suggest that the results of the current study encompass the variation between years detected in field studies and represent general regional rather than strictly local patterns. The clear disadvantage of herbaria in the current study is that supra-annual flowering will not be detected. This may be especially important when considering asynchronous flowering, as different mast flowering species may simply not flower in the same year, even if they do so at the same time of year.

delineating the evidence: some conclusions

The precise timing of flowering by individual plants and species is likely to be the result of a combination of abiotic, biotic and evolutionary factors. Distinguishing the ultimate and proximate causes of phenological events is difficult and interpretation of community-level patterns in particular must be made with caution. We have shown here that the flora of the Queensland Wet Tropics shows a diversity of flowering patterns, although a general peak in activity coincides with the end of the dry season/start of the wet season. Whether thought to be the result of phylogenetic constraints or pollinator association, coincidence of flowering patterns is particular only to each individual family at best. Like a number of proceeding community studies (e.g. Wright & Calderon 1995; Bawa et al. 2003), the structure of community flowering patterns of the Australian Wet Tropics flora is unresolved. Precise partitioning between phylogenetic, biotic and abiotic factors is not possible. Different species respond differently to a variety of factors, and the influence of factors on the flowering behaviour of individual species is likely to represent a continuum across the flora.


We are grateful to the Rainforest CRC for financial support. We would like to thank Dr Bernie Hyland, Mr Bruce Gray and Ms Rebel Elick for allowing us access to the Atherton Herbarium specimens and assistance at the Herbarium. We also thank Ms Nat Simpson for assistance with data collection. Dr James McBroom and Mr Alex Creagh provided advice and assistance with the data analysis. Dr John Kanowski and Mr Rod Eastwood are thanked for their patient and helpful discussions of the data analysis. We offer special thanks to Dr Joe Wright for helpful comments on the first draft, and to Dr Kamal Bawa and an anonymous referee whose thoughtful comments on a later draft greatly improved this manuscript.