Evaluating the impact of a biological control agent Carmenta mimosa on the woody wetland weed Mimosa pigra in Australia

Authors


Present address and correspondence: Dr Q. Paynter, Landcare Research, Private Bag 92170, Auckland, New Zealand (fax +64 09 5744101; e-mail PaynterQ@LandcareResearch.co.nz).

Summary

  • 1Evaluation of the success of biological control agents is essential to improve the efficiency and safety of future programmes. This study assessed the impact of Carmenta mimosa, a stem-mining moth, introduced into northern Australia as a biological control agent of mimosa Mimosa pigra. Litter fall, seed banks, vegetation cover, density and age structure of mimosa stands were compared using data collected at nine sites where Carmenta mimosa was present and eight sites where it was initially absent.
  • 2Mimosa seed rain was negatively correlated with Carmenta mimosa damage and declined by more than 90% at the highest Carmenta mimosa densities. Seed banks also declined with Carmenta mimosa damage.
  • 3Percentage cover of competing vegetation was significantly higher under stands defoliated by Carmenta mimosa and this inhibited mimosa seedling establishment and apparently increased the susceptibility of mimosa to fire, by increasing fuel loads beneath stands.
  • 4Four of the eight stands where Carmenta mimosa was absent expanded and none contracted. In contrast, none of the nine stands where Carmenta mimosa was present expanded and three contracted. Analysis of the age structure of mimosa stands indicated that contracting stands were typified by an absence of seedling regeneration.
  • 5In contrast to previous studies, no impact could be attributed to the flower feeder Coelocephalapion pigrae or to the stem-mining moth Neurostrota gunniella, whilst the bruchid Acanthoscelides puniceus consumed up to only c. 10% of seed. However, it is argued that, while these agents are unlikely to suppress dense mimosa thickets, they may reduce the rate mimosa can expand.
  • 6Synthesis and applications. By preventing stand regeneration, Carmenta mimosa is predicted to cause widespread reductions in mimosa populations and should therefore be redistributed to areas of mimosa where it is absent. Differences between the age structures of stands with Carmenta mimosa present may highlight how biological control could succeed for a suite of woody legume weeds with long-lived seed banks.

Introduction

Evaluating the impact of biological control agents on their target weeds can help prioritize the redistribution of proven agents (Nordblom et al. 2002) and the selection of additional complementary agents. Furthermore, for woody legumes, successful biological control could take decades (Buckley et al. 2004). Therefore, if evaluation forecasts the ultimate success of a programme, work on additional potential agents could cease. This could significantly reduce costs, given that an average estimated cost for each weed biological control agent released was $460 000 (McFadyen 1998).

Current host-specificity testing protocols are considered to assess reliably the risk of direct non-target impacts (Paynter et al. 2004) and there is as yet no evidence of evolutionary change in the fundamental host range of weed biocontrol agents after release (van Klinken & Edwards 2002). Nevertheless, safety concerns have resulted in tighter controls over the release of biological control agents (Sheppard et al. 2003). Future biocontrol programmes are likely to be more selective so that fewer agents are released, and prioritized according to predicted effectiveness (Louda et al. 2003). Agent evaluation is therefore essential to improve the efficiency and safety of future programmes.

Mimosa pigra L. (Mimosaceae), henceforth mimosa, is a woody neotropical legume that forms impenetrable thickets over more than 800 km2 of the Northern Territory (NT) of Australia, within the wet–dry tropics (Lonsdale 1992). Mimosa invades both open floodplains, which would normally be dominated by grass and sedge, and the understorey of paperbark (Melaleuca L. spp.) woodland, greatly reducing biodiversity (Braithwaite, Lonsdale & Estbergs 1989). It also competes with pastures, hinders livestock movement and prevents access to water (Lonsdale 1988).

Overgrazing by feral Asiatic water buffalo Bubalis bubalis Lydekker was considered a major factor promoting a rapid expansion of mimosa populations in the 1970s and 1980s (Lonsdale, Harley & Gillett 1988; Cook, Setterfield & Maddison 1996). The rate that new mimosa infestations were discovered declined when feral buffalo were subsequently culled (Lonsdale 1993), and removing competing grass enhanced mimosa seedling establishment (Lonsdale & Farrell 1998). Buffalo eradication resulted in dense vegetation again covering floodplains. Land managers now commonly burn floodplain vegetation early in the dry season, to prevent more destructive fires later in the year (Braithwaite & Roberts 1995). The impact of these fires on mimosa is unknown, although mature healthy mimosa is resistant to fire (Lonsdale & Miller 1993; Paynter & Flanagan 2004).

Six biological control agents are currently established on mimosa in Australia; the twig and stem-mining moths Neurostrota gunniella Busck and Carmenta mimosa Eichlin & Passoa (both first released in 1989), the flower-weevil Coelocephalapion pigrae Kissinger (released in 1994) and the seed-feeding bruchid Acanthoscelides puniceus Johnson (released in 1983) are relatively widespread. Chlamisus mimosae Karren, a leaf-feeding chrysomelid that was released in 1985, only established on the Finniss River catchment, where it inflicts minor damage (Wilson & Forno 1995). The chrysomelid Malacorhinus irregularis Jacoby, first released in 2000, established at Beatrice Lagoon (Table 1) during the course of this study (Heard et al. 2004).

Table 1.  Summary information for the field sites used
SiteHabitatLocation Carmenta mimosa presence (November 2000)Stand edge June 2003, compared with November 2000Frequency of fire, 2001–03
Black JunglePaperbark12°32′S, 131°14′EPresentRetreated 100 m2002
Beatrice LagoonFloodplain12°36′S, 131°19′EPresentRetreated 10 m0
Djukbinj National ParkFloodplain12°39′S, 131°26′EPresentUnchanged2003
Auld's LagoonFloodplain12°44′S, 131°13′EPresentUnchanged2001, 2002, 2003
Dot and Dash LagoonFloodplain13°03′S, 131°14′EAbsentUnchanged2001, 2003
Tortilla FlatFloodplain13°05′S, 131°13′EAbsentUnchanged0
Haines RoadPaperbark13°12′S, 131°09′EAbsentUnchanged0
East of Finniss RiverFloodplain12°50′S, 130°40′EPresentUnchanged0
Finniss RiverPaperbark12°50′S, 130°36′EAbsentUnchanged2003
Sweet's LagoonPaperbark12°52′S, 130°35′EPresentUnchanged0
Wagait Aboriginal ReserveFloodplain12°52′S, 130°36′E PresentUnchanged2001, 2003
Wadjigan 1Floodplain12°58′S, 130°12′EAbsentAdvanced 25–30 m2002
Werat 1Floodplain13°02′S, 130°24′EPresentUnchanged0
Wadjigan 2Floodplain13°03′S, 130°14′EAbsentAdvanced 10 m2001
Wadjigan 3Floodplain13°03′S, 130°20′EAbsentAdvanced 15–20 m2002
Werat 2Floodplain13°03′S, 130°27′EPresentRetreated 10 m2002
Wadjigan 4Floodplain13°05′S, 130°15′EAbsentAdvanced 10–20 m2002

Neurostrota gunniella herbivory was correlated with a 60% reduction in seed rain and reduced radial canopy growth by 14% over one growing season (Lonsdale & Farrell 1998), and one generation of larvae reduced seedling growth by 30% (Paynter & Hennecke 2001). However, Neurostrota gunniella alone was considered unlikely to control mimosa (Lonsdale & Farrell 1998).

Two species of Acanthoscelides (Acanthoscelides puniceus and Acanthoscelides quadridentatus) were released and initially established, but only Acanthoscelides puniceus was recorded in recent collections. They destroyed only 0·8% of mimosa seed (Wilson & Flanagan 1991). Recent priorities have therefore been to determine the impact of Carmenta mimosa and Coelocephalapion pigrae.

Materials and methods

Use of insecticide exclusion was not appropriate because it reduces seed set, which has been attributed to disruption of pollinating insects (Lonsdale & Farrell 1998), and fails to exclude Carmenta mimosa (Paynter 2004). Alternatively, comparison between sites with and without the biocontrol agent was possible for Carmenta mimosa, which is spreading slowly from its original release sites (Ostermeyer 2000).

Previous surveys in Australia (Ostermeyer 2000) identified 10 field sites on the Finniss River catchment (five with Carmenta mimosa and five without), which were set up in early November 2000. In July 2001, eight more sites (four with Carmenta mimosa and four without, one of which was subsequently destroyed by herbicide and fire) were set up on the Adelaide River catchment (Table 1). Sites were selected so that, for Carmenta mimosa comparisons, sites included both open floodplain and paperbark forest habitats, to test whether agent impact varied according to habitat. Access to eight of the most remote sites could only be made by helicopter, so a protocol was designed that allowed all fieldwork at a site to be conducted by three people within c. 40 min. Five litter trays were placed, at least 10 m apart, just inside the edge of the mimosa stand at each site. The trays were circular, made of a fine plastic mesh of c. 60 cm circumference and c. 40 cm deep, and attached to steel fence pickets 1·5 m above the soil surface. The mesh was sufficiently fine to trap mimosa litter but was permeable to water, minimizing decomposition between collections. Large stands with a closed canopy (where the branches of adjacent plants overlapped) were selected, so that variation in stand density would not be correlated with litter fall. Indications of fire (presence of ash and burnt vegetation) were noted each year. In 2001, 25 plants were labelled at each site, to monitor their survival over time. Litter samples were collected annually, during the dry season, when access was easiest and after most of the annual litter fall had occurred (Lonsdale 1988). Litter trays that had been destroyed were replaced. At the laboratory, the contents of each litter tray were dried, separated into fractions (leaves, woody material, seeds, pods and prematurely abscised inflorescences) and weighed. Carmenta mimosa abundance was assessed by randomly selecting 10 of the 25 marked plants at each site and using vernier callipers to measure the trunk (for small plants) or a branch diameter (for large plants) and counting the number of larval frass holes on the 10 sampled plants/stems. The mass of mimosa sampled was estimated according to a correlation between stem diameter (mm) and dry weight [loge weight (g) = −2·92 + 2·783(loge diameter); r= 0·971 n= 358, P < 0·001] so that the number of Carmenta mimosa frass holes per kilogram of mimosa could be estimated (Paynter & Flanagan 2004). If fewer than 10 marked plants survived at a site, additional unmarked plants were sampled. Neurostrota gunniella stem damage was quantified by removing 10 50-cm stem samples (one per plant) and counting the number of frass holes per stem (Smith & Wilson 1995).

Acanthoscelides puniceus seed feeding was quantified by examining subsamples of 100 seeds from each litter tray sample (or all seeds if fewer that 100 seeds were in a particular tray) from each site and counting the proportion that contained holes made by adult beetles emerging from the seeds. Coelocephalapion pigrae abundance was sampled by beating the tips of 10 stems (approximately 20 cm long) from the nearest plants to each litter tray against the edge of a 21-cm diameter funnel attached to a c. 7-cm diameter, 7·5-cm deep plastic jar containing 70% alcohol, to collect and preserve insects dislodged from the vegetation.

In 2002, plant densities and percentage cover at ground level were estimated at each site by setting up transects between two of the litter trays (selected at random) within the mimosa stand. Plants growing within the sample area (initially a 0·5 × 0·5-m quadrat located at the start of the transect) were counted until at least 30 mature (at least 50 cm tall; Paynter & Flanagan 2004) plants were sampled. If fewer than 30 mature plants occurred within a quadrat, the area sampled was increased sequentially along the transect. Numbers of seedlings and mature plants, and the percentage cover of mimosa, litter, bare soil and competing vegetation at ground level within the sample area, were recorded. The area sampled was then measured so the density of plants per m2 could be calculated. In addition, 15 soil cores were taken from beneath the mimosa canopy at each site, using a 7-cm diameter auger to a depth of 5 cm, where the majority of seeds in the seed bank occur (Lonsdale, Harley & Gillett 1988). In the laboratory, the core samples were dried in an oven (48 h at 70 °C). The soil was then pulverized and the seeds extracted using a 1-mm mesh sieve and counted.

In 2003, similar transects were conducted (rerandomized to give statistically independent samples across dates). As this was the last year of observations, mature plants were sampled destructively. Smaller plants were cut with secateurs, larger plants with a chainsaw, so that one stem section from each plant was taken from as close to the soil surface as possible. All stem sections were taken back to the laboratory, rubbed smooth with sand paper and moistened with water, which improved the visibility of the growth rings. The age of each plant was determined according to the number of growth rings.

Finally, because the litter trays were initially located just inside each stand, it was possible to measure the approximate distance the stands had advanced or retreated during the course of the experiment by measuring from the edge of the litter tray to the stand edge.

analysis

For each site, a mean value per m2 was calculated for each litter component using data collected from the five litter trays. Where sites had been affected by fire, an average was calculated for the surviving litter trays. Analyses of covariance were performed using GENSTAT® to investigate variation in mean annual seed rain and total litter fall (leaves, wood and reproductive material) against Carmenta mimosa and Neurostrota gunniella damage and Coelocephalapion pigrae abundance, with habitat (paperbark woodland and open floodplain) as a grouping factor. Year was not used as a grouping factor; the same litter trays were used each year so there was no cross-year independence of data points, and separate models were fitted for each year. Further analyses of variance were performed to investigate the impact of habitat on the proportion of seed consumed by Acanthoscelides puniceus each year and variation in seed banks measured in 2002.

Analyses of covariance were performed to investigate (i) the impact of Carmenta mimosa abundance and habitat on the percentage cover of competing vegetation growing beneath mimosa stands and the survival of tagged plants, and (ii) the impact of Carmenta mimosa abundance, habitat and percentage cover of competing vegetation on the numbers of seedlings growing beneath mimosa stands. Year was used as a grouping factor for these analyses, as the quadrats used to estimate percentage cover and seedling densities were randomized each year. Finally, an analysis of variance was performed to investigate the impact of Carmenta mimosa (treated as a grouping factor, present or absent) on the population density of mimosa recorded in 2002 and 2003 (separate analyses were performed for all plants, seedlings and mature plants (plants > 50 cm tall).

In the above analyses, an angular transformation was used on all proportion (and percentage) data, and litter fall and agent abundance data were log(n + 1) transformed to normalize data, where appropriate.

Results

annual litter fall and pre-dispersal seed predation by acanthoscelides puniceus

Carmenta mimosa feeding damage was conspicuous at several sites, with many plants heavily defoliated and dead and dying branches riddled with frass holes. Seed rain varied between habitats, being significantly lower in paperbark woodland in all years (F1,6 = 27·4, P < 0·01; F1,13 = 27·63, P < 0·001; F1,11 = 43·43, P < 0·001; for 2001, 2002 and 2003, respectively). In 2001 (F1,6 = 17·7, P < 0·01) and 2003 (F1,11 = 20·78, P < 0·001) there was a significant negative correlation between Carmenta mimosa and seed rain. Significant interactions between Carmenta mimosa damage and habitat in 2002 (F1,13 = 9·30, P < 0·01) and 2003 (F1,11 = 6·23, P < 0·05) indicated that the impact of Carmenta mimosa was greater in paperbark woodland compared with open floodplain habitats (Fig. 1a). In both habitats, the highest level of attack by Carmenta mimosa was associated with an order of magnitude reduction in seed rain. In open floodplain sites without Carmenta mimosa, seed rain averaged 9013 seeds m−2, which was almost identical to levels recorded by Lonsdale (1988) prior to the introduction of biological control agents. At the highest level of Carmenta mimosa, recorded seed rain was only 336 seeds m−2.

Figure 1.

Correlations between Carmenta mimosa (mean number of frass holes per kilogram of stem) and (a) seed rain (seeds m−2) and (b) total litter fall (g m−2). (a) Floodplain sites: 2001 (black circles, unbroken line), seeds m−2 = 7430·7e− 0·1498 (frass holes kg−1); 2002 (white circles, NS); 2003 (grey circles; dot and dash line), seeds m−2 = 4253·1e− 0·0762 (frass holes kg−1). Paperbark sites: 2001 (black triangles, NS); 2002 (white triangles, dotted line), seeds m−2 = 1687·2e− 1·8174 (frass holes kg−1); 2003 (grey triangles; dashed line), seeds m−2 = 635·42e− 0·2996 (frass holes/kg). (b) Symbols as above. Regression formulae: 2003 floodplain sites, litter fall = 379·9e− 0·023 (frass holes kg−1); 2003 paperbark sites, litter fall = 277e− 0·3793 (frass holes kg−1).

In contrast, no correlations were found between Coelocephalapion pigrae abundance and seed rain. Similarly, there was no correlation between Neurostrota gunniella abundance and seed rain, even when sites where Carmenta mimosa was present were excluded from the analysis.

The impact of habitat and Carmenta mimosa on total litter fall was less dramatic. Litter fall was significantly lower in paperbark woodland in 2003 only (F1,11 = 25·82, P < 0·001). Furthermore, only in 2003 was there was a significant negative correlation between Carmenta mimosa and total litter fall (F1,11 = 9·18, P < 0·05; Fig. 1b). A significant interaction between Carmenta mimosa damage and habitat in 2003 (F1,11 = 34·46, P < 0·001) again indicated that the impact of Carmenta mimosa was greatest in paperbark woodland (Fig. 1b).

Overall, c. 6% of seeds were consumed by Acanthoscelides puniceus. In 2002, seed consumption was higher in open floodplain habitats compared with paperbark woodland (10·45% vs. 0·95%; F1,14 = 6·91, P < 0·05).

mimosa seed banks

Overall, seed banks averaged 3710 (± 755) seeds m−2, somewhat lower than values recorded by Lonsdale, Harley & Gillett (1988). Carmenta mimosa presence had a significant negative impact on seed banks (F1,13 = 5·00, P < 0·05). However, a significant (F1,13 = 7·25, P < 0·05; Fig. 2) interaction between Carmenta mimosa and habitat indicated impact varied between habitats. No difference was apparent between Carmenta mimosa treatments in paperbark woodland sites, where seed banks were significantly lower than in open floodplain habitats (F1,13 = 48·75, P < 0·001; Fig. 2).

Figure 2.

Soil seed banks (± SE) recorded in open floodplain and paperbark woodland habitats at sites where Carmenta mimosa was present (diagonal bars) or absent (no fill) in 2002.

competing vegetation beneath mimosa stands

Percentage cover of competing vegetation varied significantly between years (F1,28 = 16·23, P < 0·001) and was positively correlated with Carmenta mimosa abundance (F1,28 = 21·11, P < 0·001; Fig. 3a), whilst the number of mimosa seedlings m−2 was negatively correlated with the percentage cover of competing vegetation (F1,27= 5·02, P < 0·05; Fig. 3b).

Figure 3.

Relationship between (a) Carmenta mimosa damage, log(mean number frass holes per kilogram of stem), and percentage cover of competing vegetation, and (b) percentage cover of competing vegetation and mimosa seedling numbers, log(seedlings m−2) recorded at each site in 2002 and 2003.

stand densities and age structures

Densities of all plants (F1,23 = 30·74, P < 0·001), seedlings (F1,23 = 27·26, P < 0·001) and mature plants (F1,23 = 10·72, P < 0·01) were significantly lower at sites where Carmenta mimosa was present. The greatest difference was for seedlings (c. 95% reduction) but even the mean density of mature plants was c. 70% lower at sites with Carmenta mimosa present (Fig. 4). Age structures of expanding and contracting stands are shown in Fig. 5. In all of the expanding sites, large numbers of seedlings were present. Sites where mimosa was contracting were typified by a complete absence of seedlings.

Figure 4.

Mean population densities of all mimosa plants, seedlings and mature broom plants at sites where Carmenta mimosa was absent (no fill) or present (diagonal bars). Columns within the same category with the same letter are not significantly different (LSD).

Figure 5.

Age structures of mimosa populations. (a–c) Retreating stands with Carmenta mimosa present: Beatrice Lagoon, Black Jungle and Werat 2. (d–g) Advancing stands with Carmenta mimosa absent: Wadjigan 1–4.

survival of tagged plants

Mimosa survival significantly declined over time (F1,28 = 48·65, P < 0·001). Only 45% of the originally tagged plants were still alive after 2 years, implying a half-life of c. 24 months. This was similar to Lonsdale (1992) but may be a slight underestimate because c. 10% of tags were lost and assumed destroyed in bush fires that killed the plants they were attached to. Eleven of the 17 sites were affected by fire (five with Carmenta mimosa present, six without). Eight sites burnt once, two sites burnt twice and one site burnt every year (Table 1). There was no significant effect of Carmenta mimosa on survival (F1,28 = 2·44, NS).

mimosa and agent expansion

The frequency of sites where mimosa stands expanded, contracted or remained stable appeared to vary according to the presence of Carmenta mimosa (Table 1). A Fisher exact test examined the difference between the categories Carmenta mimosa present and Carmenta mimosa absent, and stand expanded and stand unchanged/contracted, and indicated stand expansion was not independent of the presence of Carmenta mimosa (P = 0·029).

There was little opportunity to study plant performance before and after colonization by Carmenta mimosa, although small numbers of Carmenta mimosa were recorded at two sites where it was originally absent (Wadjigan 1 and Wadjigan 2; Table 1) in 2003.

Discussion

This study indicated that seed rain was negatively correlated with Carmenta mimosa damage and that this relationship was most marked in paperbark woodland. This concurs with Steinbauer (1998) who, in a nursery-based study, found that Carmenta mimosa impact was greatest in shaded treatments. However, although Carmenta mimosa was associated with smaller seed banks, at least in open floodplain habitats, there should still have been sufficient seed to ensure recruitment maintained mimosa populations. The key to Carmenta mimosa's success appears to be the positive correlation between Carmenta mimosa and percentage cover of competing vegetation. Competing vegetation was largely absent from sites where Carmenta mimosa was absent and healthy plants cast dense shade. Damage from Carmenta mimosa kills mimosa stems (Steinbauer 1998), which should have increased light levels at the soil surface so that competing vegetation could colonize beneath stands (Fig. 3a). Competing vegetation was associated with reduced seedling numbers (Fig. 3b) and has already been demonstrated to suppress mimosa germination (Lonsdale & Farrell 1998). The combined impact of reduced seed rain and reduced seedling emergence and survival, as a result of the presence of competing vegetation, has dramatically altered the density and age-structure of stands where Carmenta mimosa is present. This is illustrated by Fig. 5, which compares sites where Carmenta mimosa was absent and the mimosa stands expanded, with sites where Carmenta mimosa was present and the stands contracted. Seedling numbers were very high at expanding sites. In contrast, contracting stands were typified by an absence of seedlings. This indicates that, as mature plants senesce, an absence of recruitment should result in stand decline, following a lag-period of c. 10–12 years, corresponding to the maximum age recorded (Fig. 5).

importance of fire and grazing

The only stands to expand were those where Carmenta mimosa was absent and a fire had occurred. Fire burnt the surrounding floodplain vegetation but did not penetrate far into these stands because healthy mimosa is resistant to fire (Lonsdale & Miller 1993; Paynter & Flanagan 2004). Stand expansion was presumably the result of fire eliminating competing vegetation at the stand edge, enhancing mimosa germination and seedling survival (Lonsdale & Miller 1993). The rate of expansion (Table 1) was considerably lower than during the 1970s and 1980s, when populations advanced at a rate of 76 m year−1 (Lonsdale 1993), indicating that eradication of buffalo and/or biological control agents other than Carmenta mimosa have slowed, but not halted, the mimosa invasion. In contrast, where Carmenta mimosa was present, no stands expanded following fire and two retreated. Competing vegetation, which increased beneath stands colonized by Carmenta mimosa, should have increased fuel loads, leading to more intense fires (Rossiter et al. 2003) that could carry through the stand. This happened most dramatically at Black Jungle (Table 1), where a fire late in the 2002 dry season drove the edge of the stand back by c. 100 m. By 2003, native grasses and sedges recovered dramatically, stifling mimosa regeneration so that only a few surviving mature plants were resprouting.

Roques, O’Connor & Watkinson (2001) demonstrated how fire, drought, browsing, grazing and density-dependence drive encroachment of a legume shrub in north-eastern Swaziland. They found that high grazing pressure had a negative effect on fire frequency, facilitating scrub encroachment. It would appear that by altering the susceptibility of mimosa to fire, Carmenta mimosa has the potential to reduce dramatically the abundance of mimosa on NT floodplains, provided overgrazing does not reduce the level of fuel for fires.

impact of acanthoscelides puniceus, coelocephalapion pigrae and neurostrota gunniella

In most successful weed biological control programmes, several agents were introduced against the target weed and it has been debated whether successful control was the result of their ‘accumulative stress’ or to one agent alone (Denoth, Frid & Myers 2002). Hoffmann & Moran (1998) concluded that a combined attack from a stem-mining weevil, a flower-feeder and a seed-feeder resulted in the effective control of Sesbania punicea (Cav.) Benth, a woody legume that was invasive in South African wetlands, whereas any one agent alone would have probably been ineffective. In contrast, the present study indicates that Carmenta mimosa alone suppresses mimosa populations.

The lack of impact attributable to Neurostrota gunniella was unexpected because Lonsdale & Farrell (1998) found that insecticide treatment increased radial canopy growth of mature plants by 14%, and at low Neurostrota gunniella densities seed rain was 50% of the level recorded by Lonsdale (1988), declining by a further 30% at the highest Neurostrota gunniella densities. Perhaps increased vegetative growth was the result of reallocation of resources (insecticide treatment reduced seed set, presumably by disrupting pollinators). Competition with Carmenta mimosa does not explain the lack of impact recorded in the present study, as there was still no correlation between Neurostrota gunniella and seed production when sites with Carmenta mimosa present were excluded from the analysis. Unlike the present study, which exclusively sampled large mimosa stands, Lonsdale & Farrell (1998) sampled a variety of patch sizes (some patches were just 10 m across; W. M. Lonsdale, personal communication). Smith & Wilson (1995) and Paynter & Flanagan (2004) demonstrated that Neurostrota gunniella attack was much higher on edge plants, isolated plants and small patches compared with plants within dense thickets. Perhaps compensation (i.e. a density-dependent increase in growth and flowering of stems that escape attack) is greater in dense mimosa thickets, where Neurostrota gunniella attack is relatively low, compared with on isolated plants where every stem may be infested. Therefore, Neurostrota gunniella has no impact on dense, established mimosa thickets but may have a significant impact on the rate mimosa invades. Indeed, Paynter & Flanagan (2004) noted that mimosa did not rapidly re-invade plots treated with herbicides, bulldozing and fire, which largely eliminated Carmenta mimosa but resulted in Neurostrota gunniella damage to regenerating mimosa plants that was an order of magnitude higher than on untreated mimosa thickets. As mimosa occupies only a small fraction of its potential range in Australia (Lonsdale 1992), one should not therefore underestimate the possible benefits arising from Neurostrota gunniella herbivory reducing mimosa spread into new areas and increasing the efficacy of other control options in areas where mimosa is already abundant. Furthermore, while very high levels of seed predation are often assumed to be necessary for a seed feeder to control established thickets, if establishment at the stand edge is seed limited then even the relatively low proportion of seeds consumed by Acanthoscelides puniceus should have an impact on the rate of mimosa invasion (Paynter et al. 1996). This could be tested by a simple seed addition experiment. However, despite the potential benefits of Neurostrota gunniella and Acanthoscelides puniceus, even in the absence of biological control 96·7% of inflorescences are often prematurely abscised (Lonsdale 1988), and thus mimosa is likely to be able to compensate for attack from a flower feeder such as Coelocephalapion pigrae.

Carmenta mimosa has a major impact on mimosa. To maximize benefits, this slowly dispersing agent should be redistributed throughout mimosa infestations that it has not yet colonized. This study validates both demographic (Parker 2000) and spatial models (Rees & Paynter 1997; Buckley et al. 2004) that indicated that seedling establishment and the probability of self-replacement is a major factor influencing the potential dominance of exotic woody legume populations.

Acknowledgements

I thank Merrilyn Paskins, Grant Flanagan, Bruce Hitchins (NT DIPE), Magen Geyer, Mathew Hoschke and Robert Eager (CSIRO) for logistical support and/or technical assistance. I thank Mark Lonsdale and Yvonne Buckley for comments on an earlier draft of the manuscript. The Natural Heritage Trust supported this work.

Ancillary