Integrating herbicide and mechanical control treatments with fire and biological control to manage an invasive wetland shrub, Mimosa pigra


  • Present address: Kangaroo Island Natural Resources Board, PO Box 520, Kingscote, SA 5223, Australia.

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


  • 1In Australia, biological control is a promising long-term management strategy for the woody weed mimosa Mimosa pigra but does not yet provide adequate control. Other management techniques, including herbicides and fire, can be ineffective and their impact on biological control agents is unknown. We investigated the potential of integrating control techniques, including biological control, to provide cost-effective management.
  • 2A large-scale (128-ha) split-plot experiment was performed to measure the impact of single and repeated applications of herbicide and crushing by bulldozer, either alone or in combination, on both mimosa and five introduced biological control agents that were abundant at the site. Herbicides were applied over three seasons (1997–99) and all plots were burned in 2000.
  • 3The impact of control options on mimosa cover, biomass, number of stems per ha, stand size structure and seedling regeneration was determined by aerial photography and by sampling permanent and random quadrats. Biological control agent abundance was also quantified.
  • 4In isolation, herbicide, bulldozing and fire were not effective, but several combinations of techniques cleared mimosa thickets and promoted establishment of competing vegetation that inhibited mimosa regeneration from seed.
  • 5Depending on the species, biological control agent abundance on surviving mimosa plants was either unchanged or increased following herbicide and/or bulldozing treatments. All agents recolonized regenerating mimosa within 1 year of the fire treatment, and Neurostrota gunniella increased dramatically. Carmenta mimosa abundance, however, was reduced by fire.
  • 6The abundance of N. gunniella increased in response to all treatments, which we attribute to attack by this species being most common along stand edges. Control treatments separated monocultures of mimosa into smaller patches, thereby increasing the ratio of ‘edge’ to ‘thicket’ plants. The proportion of plants susceptible to N. gunniella attack increased as a result.
  • 7Synthesis and applications. We conclude that integrating control techniques can successfully control dense mimosa thickets. Biological control integrates well with other control options and should lead to significant cost reductions for mimosa management. To maximize this benefit, integrated weed management plans should be designed to integrate biological control fully with other methods, rather than separate them spatially or temporarily.


Invasion of natural communities by introduced plants is considered one of the most serious threats to biodiversity (Heywood 1989). Woody legumes comprise some of the world's most important invasive weeds. In Australia, a recent ranking exercise identified 20 weeds of national significance (Thorp & Lynch 2000), of which five, including mimosa Mimosa pigra L. (Mimosaceae), are introduced woody legumes.

Many woody legume weeds infest rangelands of low economic value. This constrains chemical and mechanical control because of the relative costs involved, so that biological control often represents the only economically practical long-term management option (Paynter, Downey & Sheppard 2003).

Classical biological control has had mixed results against woody legumes. Notable successes have been achieved (Dennill 1990; Hoffmann & Moran 1998), while some long-term programmes have not yet delivered adequate control (e.g. Acacia nilotica in Australia; Kriticos et al. 1999), even where biological control agents dramatically reduced plant fecundity (e.g. Ulex europaeus in New Zealand; Rees & Hill 2001). This may be because reduced suppression by natural enemies is not the only cause of invasiveness: even native populations of shrubs can be invasive depending on a range of factors, including fire and grazing (Paynter et al. 1998; Roques, O’Connor & Watkinson 2001).

Modelling is an increasingly important component of invasive weed management programmes (Buckley, Briese & Rees 2003a,b). Rees & Paynter (1997), Rees & Hill (2001) and Buckley et al. (2004) modelled the consequences of integrating biological control with other control methods, such as herbicide application, mechanical control and fire, against three woody legume species. These simulations all predicted that biological control should enhance the impact of other control options. However, they also assumed that additional control methods do not affect biological control agent abundance, which may not be the case. Indeed, scientists originally assumed that classical biological control and techniques such as chemical weed control were incompatible (Harris 1991). More recently it has been recognized that interactions between fire (Briese 1996), herbicides (Harris 1991; Messersmith & Adkins 1995; Lindgren, Gabor & Murkin 1999; Paynter 2003) or mechanical control (Kluth, Kruess & Tscharntke 2003) and biological control are not necessarily unfavourable and their integration may result in improved weed control.


Mimosa is native to tropical America but is now a pantropical weed that poses the most serious of all invasive threats to tropical wetlands (Cronk & Fuller 1995). In Australia, it forms impenetrable nearly monospecific thickets, up to 6 m tall, over more than 800 km2 of the Northern Territory (NT) within the wet–dry tropics, where the climate is typified by well defined wet (December–April) and dry (May–November) seasons (Lonsdale 1992). Mimosa greatly reduces biodiversity (Braithwaite, Lonsdale & Estbergs 1989), competes with pastures, hinders livestock movement and prevents access to water (Lonsdale 1988).

Mimosa control

biological control

In 1979, a biological control programme against mimosa was established in Australia (Forno 1992), where a paucity of specialist insect herbivores (Harley et al. 1995) may explain dramatically superior growth and fecundity compared with native populations (Lonsdale & Segura 1987).

When we commenced this study, five insect species had established in Australia. The stem-mining moths Neurostrota gunniella Busck and Carmenta mimosa Eichlin & Passoa, the flower-feeding weevil Coelocephalapion pigrae Kissinger and the seed-feeding bruchid Acanthoscelides puniceus Johnson are widespread through much of the introduced range of mimosa. Chlamisus mimosae Karren, a leaf-feeding chrysomelid, has only established on the Finniss River catchment, where it causes only minor damage to mimosa (Wilson & Forno 1995).

It is too soon to tell if biological control will succeed. Acanthoscelides puniceus destroyed only 0·8% of seed and was considered a failure (Wilson & Flanagan 1991). Neurostrota gunniella, however, reduced fecundity by 58–78% and stunted growth of both mature plants and seedlings (Lonsdale & Farrell 1998; Paynter & Hennecke 2001). Recent work indicates Carmenta mimosa is now very damaging locally but widespread control could still be decades away (Paynter & Flanagan 2002; Buckley et al. 2004).

physical and chemical control

Eradicating small satellite infestations is a priority for managing invasive weeds (Moody & Mack 1988). Several control options were tried for small infestations in Kakadu National Park (NT, Australia), including hand-pulling, cutting, burning and herbicides (Cook, Setterfield & Maddison 1996). Many infestations (c. 30%) were eradicated within 1 year, although 20% required sustained control for 7 years or more to prevent regeneration from the seed bank. For stands that may cover thousands of hectares, less labour-intensive techniques are necessary, including the following.

Aerial herbicide application

This is most effective in the wet season, although spraying may not achieve 100% kill. Follow-up control is required due to regeneration from the seed bank (Miller & Siriworakul 1992).

Mechanical control

Siriworakul & Schultz (1992) outlined ploughing, chaining and bulldozing methods, noting that re-growth and regeneration from the seed bank generally occurs. Contaminated machinery may disperse mimosa seed.


Green mimosa is difficult to burn. Even if a fire carries through a stand, burnt plants often regrow from buds at the base of stems and fire can enhance mimosa germination (Lonsdale & Miller 1993).

The mimosa integrated control experiment

Miller et al. (1992) suggested that an integrated approach should provide the most effective mimosa management strategy.

We define integrated weed management (IWM) as a sustainable approach to managing weeds that combines biological, cultural, physical and chemical methods in a way that maximizes their effectiveness while minimizing economic, health and environmental harm. It is a form of ecosystem management, and requires sufficient knowledge of the ecology of the weed and the invaded system to allow prediction of the outcome of control efforts. Studying weed population ecology (Briese 1993; Paynter et al. 1998; Higgins, Richardson & Cowling 2001; Sheppard et al. 2002) is therefore a vital component of an IWM programme.

We designed a split-plot experiment to measure impacts of herbicide, crushing by bulldozer and burning on both mimosa and its biocontrol agents. We addressed the following questions. (i) Which measures, alone or in combination, most effectively destroy established mimosa plants and limit regeneration from seed? (ii) How do these measures affect introduced biological control agents? (iii) What is the best technique or combination of techniques to manage mimosa?

study area

The experiment was performed at Wagait Aboriginal Reserve, in the Finniss River catchment, NT, Australia (12°56′S, 130°33′E, altitude c. 20 m a.s.l.). The site contained a c. 250-ha mimosa stand, on a finger of ‘black soil’ (black cracking clay) floodplain, bounded to the north by a billabong and c. 8000 ha of Melaleuca forest infested with a dense mimosa understorey, and to the south, east and west by uplands too dry to support mimosa. Mimosa was first recorded at Wagait in 1979. During the mid-1980s, a continuous stand developed in the study site, which was heavily grazed by c. 300–400 feral water buffalo Bubalis bubalis Lydekker, until they were culled between 1986 and 1989 (C. Deveraux, personal communication).

allocation of treatments

The experimental plots covered c. 128 ha of dense mimosa, recreating the scale required for practical mimosa control. The original design consisted of four blocks, each divided by bulldozed access tracks into two c. 400 × 400-m subblocks, one to be burnt, one to remain unburnt (randomly assigned), within which we applied the following treatments.

Aerial herbicide application

In April 1998, two 100 × 400-m strips were sprayed with 300 g L−1 fluroxypyr (Starane 300®; DowElanco Co., Frenchs Forest, Australia) diluted to 0·5% v/v at 1·5–2 L ha−1 (Miller & Siriworakul 1992), leaving two 100 × 400-m unsprayed strips. In January 1999, fluroxypyr was randomly reapplied to one sprayed strip and one unsprayed strip to give a control, two single herbicide applications (April 1998 or January 1999) and a double herbicide application (April 1998 + January 1999).


In October–November 1998, approximately one half of each subblock (randomly assigned) was bulldozed at right angles to the sprayed strips, giving eight 100 × 200-m plots per subblock. Initially two bulldozers were used in tandem, dragging a heavy chain between them to crush the mimosa. However, following mechanical failure of one bulldozer, similar results were achieved by driving the remaining bulldozer over the mimosa.

Unfortunately, following 2 years of above-average rainfall, safe access on foot was not possible until late in the 1999 dry season (due to the presence of saltwater crocodiles Crocodylus porosus Schneider). By the time insect and plant sampling (see below) was completed, the first storms of the 1999–2000 wet season began flooding the plots, so the fire treatment could not be conducted. Additional bulldozing work was also impossible, so firebreaks between the subblocks became overgrown. The experiment had to be redesigned.

We reapplied fluroxypyr (late December 1999) to the whole of one of the two subblocks (selected at random) in each replicate, resulting in four replicates of the following herbicide treatments (Fig. 1): (i) control (no herbicide); (ii) single herbicide applications (April 1998, January 1999, December 1999); (iii) double herbicide applications (April 1998 + January 1999, April 1998 + December 1999, January 1999 + December 1999); (iv) a triple herbicide application (April 1998 + January 1999 + December 1999). The April 1998, January 1999 and December 1999 herbicide treatments, corresponding to the 1997–98, 1998–99 and 1999–2000 wet seasons, are henceforth referred to as the 1997, 1998 and 1999 wet season treatments, respectively.

Figure 1.

Summary of treatments for one replicate of the final design. Bulldozed = plots crushed by bulldozer. Herbicide applications: no shading = unsprayed; vertical shading = 1997 wet season; diagonal shading = 1998 wet season; stippled = 1997 + 1998 wet seasons. Grey shading = plots also sprayed in 1999 wet season. Both subblocks were burnt. Numbers within plots refer to positions of permanent quadrats (m).

Following construction of a firebreak around the study site perimeter, the burn treatment was conducted on 3 November 2000, when most of the site had dried out and the weather was hot (37–41 °C) and windy. A 25% petrol : 75% diesel-filled ‘firebug’ drip torch (RAPP Australia Pty Ltd, Lara, Australia) was used to ignite vegetation around the site perimeter. Fire passed through all plots, burning for at least 2 weeks, until it was extinguished by heavy thunderstorms.

Thus the experiment tested different herbicide application frequencies, with and without bulldozing, against a background of natural populations of biocontrol agents and fire applied across all treatments in 2000.

The approximate costs of each treatment are given in Table 1.

Table 1.  Approximate control treatment costs ha−1 (Australian $) (M. Ashley & C. Deveraux, personal communication). For the fire treatment, the cost of creating and supervising a firebreak around the perimeter of a 100-ha mimosa stand was estimated
TreatmentNo bulldozingBulldozing
Fire alone30 90
Single herbicide application + fire50110
Two herbicide applications + fire70130
Three herbicide applications + fire90150


In November 1997, before allocating treatments, permanent 1 × 5-m quadrats were located every 25 m on transects running along two edges of each subblock. The first quadrat was located at 10 m, so that each 100 × 200-m treatment plot contained four permanent quadrats (Fig. 1), aligned towards the subplot centre. Within each quadrat, number and diameters (at ground level) of ‘mature’ (> 50 cm tall) mimosa stems were recorded. Plants > 50 cm tall were considered mature because plots were accessed soon after they dried out, so newly germinating seedlings were small (< 10 cm). Plants > 50 cm tall should have been more than 1 year old and mimosa can flower within 6–8 months of germination (Lonsdale 1992). Above-ground biomass was estimated, using 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, T. J. Schatz & W. J. Müller, unpublished data]. Subsequent sampling was performed in November–December 1999, after the first two herbicide treatments, in November–December 2000, after the final herbicide treatment and the fire, and in November 2001, 1 year after the fire. The data from the four quadrats in each plot were averaged, prior to analysis. The stem diameter data were also used to investigate the impact of treatments on the size distribution of the mimosa populations recorded in the November–December 1999 survey. A chi-squared test was performed, with size class as the classifying variable and using the frequencies of stems in each size class for each treatment (using mean stem numbers from the control plot as the ‘expected’ numbers).

Estimates of percentage cover of mimosa, competing plant species, litter and bare ground were taken at each sample date using a random quadrat (1 × 1 m; four replicates per plot). A smaller random quadrat (0·5 × 0·5 m) was used to sample mimosa seedlings (four replicates per plot). Quadrats were rerandomized for each sample date to give statistically independent samples across dates. As above, data were averaged for each plot before subsequent analyses.

The percentage cover of mimosa was also estimated from aerial photographs taken from a helicopter flown approximately 300 m above the plots during April 1998 (pre-treatments), in December 1999, after the bulldozing and first two herbicide treatments, November 2000 (post-herbicide and post-bulldozing treatments) and December 2000 (post-fire treatment).

Biological control agents were sampled in November–December 1999 and November 2001 to indicate presence and relative abundance as follows.

Neurostrota gunniella

Ten plants per plot were sampled as described by Smith & Wilson 1995). A single living branch from each plant was cut 50 cm from the tip, and the number of frass holes in the stem, which is closely correlated to the number of larvae within, was counted.

Carmenta mimosa

Frass extruding from exit holes in the stem indicated presence of larvae (Eichlin & Passoa 1983), which are considerably larger and confined to larger stems than those of N. gunniella. At the start of this study, Carmenta mimosa was uncommon and difficult to sample quantitatively because when plots could be accessed, damage was extremely cryptic and concentrated beneath skirts of adventitious roots formed during wet-season flooding. Therefore, individual branches were not sampled as for N. gunniella, but observers walked slowly within each plot and counted Carmenta mimosa frass holes over a 5-min time period.

Beat samples

Tips (approximately 20 cm long) of 20 stems plot−1, selected randomly, were beaten (three sharp taps) over 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 Coelocephalapion pigrae and Chlamisus mimosae (which feed externally). Samples were sorted and counted, using a binocular microscope. Acanthoscelides puniceus was not quantified (Wilson & Flanagan 1991) because mimosa seeds most prolifically shortly after the wet season (Lonsdale 1988), when the field site could not be accessed due to seasonal flooding.

The 2001 survey work was confined to plots, or parts of plots, that had been burnt, to assess the level of regeneration of mimosa and competing vegetation and the abundance of biological control agents following fire (see below).

data analysis

General analyses of variance were performed using genstat® 5.1. (Numerical Algorithms Group Ltd, Oxford, UK). Analyses were structured with replicates (REP) declared with four levels; ‘block’ was declared with two levels, corresponding to the two subblocks per replicate, within which were nested ‘half-blocks’ with two levels. To denote the two halves (left or right), each subblock was divided by the bulldozing treatment and plot, with levels (1–4) corresponding to the position of each plot within each block. Two herbicide treatments (1997 and 1998 wet seasons) and the bulldozer treatment were declared as factors for analyses of 1999 survey data, collected prior to the final herbicide treatment, otherwise three herbicide treatments (corresponding to the 1997, 1998 and 1999 wet-season treatments) were declared. All interactions were included in the analyses. The following analyses were performed.

Percentage cover of mimosa using aerial photographs

Analyses were performed for April 1998 (pre-treatments), December 1999, November 2000 (post-herbicide and post-bulldozing treatments) and December 2000 (post-fire treatment).

Percentage cover of competing vegetation and seedling counts, using random quadrat data

An analysis of variance was performed to determine the impact of herbicide and bulldozing on competing vegetation, using 1999 survey data and, with ‘year’ as an additional factor, using 2000 and 2001 survey data. A similar analysis, with year (1997, 1999, 2000 and 2001) as an additional factor, was performed to determine the impact of the treatments on the number of germinating mimosa seedlings.

Mimosa biomass and survival using permanent quadrat data

General analyses of variance were performed to determine the impact of treatments on the biomass and number of mimosa stems in the plots.

Biological control agent abundance

Analyses of variance were performed to determine the impact of herbicide and bulldozing on biological control agent abundance, using 1999 and 2001 survey data.

Proportion data were arcsine (angular) transformed, prior to analysis. Residuals of other analyses were examined and appropriate transformations performed if necessary.


percentage cover, biomass and number of mimosa stems

There were no significant differences between plots prior to allocation of treatments. Mimosa was present as a virtual monoculture: mean percentage cover was 96·3% and number of mature stems was estimated to be 15137 ha−1. Mimosa biomass (dry weight) was estimated to be 39·3 t ha−1, which agrees with the 35–45 t ha−1 recorded on the Adelaide River floodplain (NT, Australia) (Presnell 2004) and is similar to invasive woody legumes in other systems (Van Wilgen & Richardson 1985; Bossar d & Rejmánek 1994).

Repeat herbicide applications were generally more effective than single applications, as were combined herbicide and bulldozing treatments, compared with either treatment in isolation (Figs 2a and 3a). For example, aerial photographs taken in December 1999 indicated that bulldozing significantly reduced mimosa cover to c. 22% (Fig. 2a). However, by 2000, cover recovered to c. 50% of control levels (Fig. 3a). Efficacy of single herbicide applications was similar, with c. 30–35% cover recorded in both the 1997 and 1998 treatment plots. However, by November 2000 mimosa cover in plots left untreated following single herbicide applications in the 1998 wet season was c. 75% of levels in the control plots (Fig. 3a).

Figure 2.

Effect of herbicide and bulldozing on (a) percentage cover, (b) mimosa biomass (kg ha−1) and (c) number of stems ha−1 (± SE) recorded in 1999. Columns with the same letter are not significantly different (LSD). Significant factors: bulldozer, F1,7 = 111·85, P < 0·001; herbicide 1997, F1,21 = 49·17, P < 0·001; herbicide 1998, F1,21 = 89·21, P < 0·001; herbicide 1997 + 1998, F1,21 = 12·23, P < 0·01; bulldozer + herbicide 1997, F1,21 = 14·36, P < 0·001; bulldozer + herbicide 1998, F1,21 = 21·08, P < 0·001.

Figure 3.

(a) The effect of herbicide and bulldozing treatments on percentage cover of mimosa (± SE) recorded in November 2000. Significant factors: herbicide 1999, F1,3 = 38·80, P < 0·01; bulldozer, F1,6 = 12·93, P < 0·05; herbicide 1999 + bulldozer, F1,6 = 7·07, P < 0·05; herbicide 1997, F1,18 = 25·86, P < 0·001; herbicide 1998, F1,18 = 14·01, P < 0·001; herbicide 1998 + herbicide 1999, F1,18 = 11·38, P < 0·01; herbicide 1997 + herbicide 1999, F1,18 = 51·76, P < 0·001; herbicide 1997 + bulldozer + herbicide 1999, F1,18 = 6·39, P < 0·05. (b) The effect of herbicide, bulldozing and fire treatments on percentage cover of mimosa (± SE) recorded in December 2000. Columns with the same letter are not significantly different (LSD).

Percentage cover was reduced to very low levels in plots treated with herbicide in both 1997 and 1998. Herbicide and bulldozing also significantly reduced mimosa biomass, by c. 99% (herbicide: 1997, F1,20 = 6·51, P < 0·05; 1998, F1,20 = 9·46, P < 0·01; bulldozing: F1,7 = 11·48, P < 0·05; Fig. 2b), and number of live mimosa stems per quadrat recorded in 1999 (herbicide: 1997, F1,20 = 24·72, P < 0·001; 1998, F1,20 = 11·82, P < 0·01; bulldozing: F1,7 = 10·47, P < 0·05; Fig. 2c). Further analysis of the 1999 data revealed the size structures of the mimosa population varied significantly between the treatments (chi-squared = 84·34, d.f. 18, P < 0·001). Single herbicide applications had little impact on size structure, perhaps indicating low herbicide efficacy or patchy application (Fig. 4). In combination with bulldozing, repeat herbicide applications eliminated plants from all but the smallest size class (Fig. 4).

Figure 4.

The size structure of mimosa populations under the following treatments: (a) control; (b) bulldozer; (c) herbicide 1997; (d) herbicide 1997 + bulldozer; (e) herbicide 1998; (f) herbicide 1998 + bulldozer; (g) herbicide 1997 + 1998; (h) herbicide 1997 + 1998 + bulldozer, recorded in the permanent quadrats during the 1999 survey, and (i) all treatments combined, recorded one year following the fire treatment (2001). Size classes: 1, 0–30 mm stem diameter at ground level; 2, 31–60 mm; 3, 61–90 mm; (4) ≥ 91 mm.

Analysis of November 2000 (pre-fire) data (Fig. 3a) indicated efficacy of single herbicide applications varied between years, with almost no living mimosa recorded in the 1999 treatment plots, but percentage covers of 33% and 72% were recorded in plots treated in 1997 and 1998, respectively. Three significant interactions occurred. First, the combined 1998 + 1999 wet season herbicide treatment was more effective than the single 1998, but not the single 1999, application. Secondly, plots sprayed in both 1997 and 1999 wet seasons had more mimosa cover than plots sprayed in the 1999 wet season only, but less than plots sprayed in 1997 only (Fig. 3a). Finally, bulldozing was more effective in plots sprayed in both 1997 and 1999 wet seasons than in plots sprayed in the 1997 wet season, but not compared to plots sprayed in the 1999 wet season only.

Analysis of December 2000 (post-fire) data produced similar results (Fig. 3b). Bulldozing (F1,6 = 9·15, P < 0·05) remained effective: fire reduced the percentage cover from c. 42% to c. 2% in unsprayed bulldozed plots, but percentage cover remained at 70% in unsprayed plots that were not bulldozed. Fire was not particularly effective in non-bulldozed plots treated with a single herbicide application. The three significant interactions encountered in the November 2000 analysis were present again. There was a significant interaction between the 1999 and 1997 herbicide treatments (F1,18 = 10·52, P < 0·01), between the 1999 and 1998 herbicide treatments (F1,18 = 4·78, P < 0·05) and between the 1999 and 1997 herbicide application treatments and the bulldozing treatment (F1,18 = 5·61, P < 0·05; Fig. 3b). There was also a significant interaction between the 1998 herbicide treatment and the bulldozing treatment. Mimosa cover was almost eliminated from bulldozed plots that were sprayed with herbicide in 1998, but it was not eliminated from plots that were sprayed in 1998 and not bulldozed (F1,18 = 5·16, P < 0·05).

In the 2000 and 2001 surveys there were no significant treatment effects on mimosa biomass (mean biomass 7·89 kg ha−1 and 0·42 kg ha−1, respectively) and number of stems measured (mean 1333·5 stems ha−1 and 1261·8 stems ha−1, respectively). Most permanent quadrats were burnt, including those in the control plots. However, because fire tended to burn plot edges, where the permanent quadrats were located, but did not penetrate far into plots containing green mimosa, the permanent quadrats greatly overestimated the efficacy of the fire treatment compared with estimates of percentage cover from aerial photographs. Fire was generally most effective in bulldozed plots. For example, although mimosa biomass was reduced by a similar extent in the single herbicide and bulldozed-only treatment plots (Fig. 2b) (where dead stems provided the same amount of fuel), fire was much more effective in bulldozed-only plots (Fig. 3b).

percentage cover of competing vegetation and seedling regeneration

Plants were identified according to Cowie, Short & Ostercamp Madsen (2000). Grasses and sedges were the most important competing species, in terms of percentage cover. In areas that retained standing water until late in the dry season, Hymenachne acutigluma was abundant. In drier areas the introduced pasture grass Urochloa mutica was common, with Leersia hexandra and Pseudoraphis spinescens occurring in the driest spots. Sedges were mostly Cyperus and Fimbristylis spp. Common broad-leaved species included Ludwigia hyssopifolia, Glinus oppositifolius, Eclipta prostrata, Heliotropium indicum, Oldenlandia galioides, Dentella sp. and the introduced floating weed Salvinia molesta.

The 1999 survey indicated both 1997 (F1,21 = 10·64, P < 0·01) and 1998 (F1,21 = 5·85, P < 0·05) herbicide treatments had a significant positive effect on percentage cover of competing vegetation, which increased from c. 15% to more than 60% in plots treated with herbicide (Figs 5 and 6a,b). Competing vegetation also increased in bulldozed plots but this was not statistically significant (Fig. 5).

Figure 5.

Effect of herbicide and bulldozing treatments on percentage cover of competing vegetation (± SE) recorded in 1999. Columns with the same letter are not significantly different (LSD).

Figure 6.

Effect of herbicide, bulldozing and burn treatments on percentage cover of competing vegetation (± SE) recorded in (a) 2000 and (b) 2001. Columns with the same letter are not significantly different (LSD). Significant factors: herbicide 1997, F1,18 = 5·39, P < 0·05; herbicide 1998, F1,18 = 7·24, P < 0·05; herbicide 1997 + herbicide 1998 + herbicide 1999, F1,18 = 6·63, P < 0·05; year, F1,35 = 18·93, P < 0·001; year + bulldozer, F1,35 = 5·15, P < 0·05.

In the 2000 and 2001 surveys, the only individual treatments to affect percentage cover of competing vegetation significantly were the 1997 and 1998 herbicide applications (Fig. 6a,b). Year was significant, with percentage cover of competing vegetation being significantly higher in 2001, 1 year after the fire, than in 2000, immediately after the fire. The significant interaction between year and bulldozing indicated that percentage cover of competing vegetation was significantly lower in bulldozed plots in 2000, although it recovered within a year (Fig. 6a,b).

Seedling numbers only varied significantly between years, with the highest numbers in 2000, after the fire treatment (F1,118 = 7·55, P < 0·001; Fig. 7). Significant negative correlations between seedling numbers and percentage cover of competing vegetation were detected in 3 of the 4 years in which surveys were conducted (Fig. 8). Furthermore, the 2000 survey revealed that, compared with levels recorded in 1999, mimosa did not increase in plots treated with repeat (1997 + 1998) herbicide applications or combined herbicide and bulldozing treatments (Figs 2a and 3a), even though sufficient mimosa seedlings germinated each year to guarantee a monoculture had they survived (Fig. 7).

Figure 7.

Number of seedlings ha−1 (± SE) recorded each year. Columns with the same letter are not significantly different (LSD).

Figure 8.

Correlations between percentage cover of competing vegetation and the mean number of mimosa seedlings present in the 0·5 × 0·5-m quadrats for each plot. (a) 1997, loge(no seedlings + 1) = −0·003x + 1·04, R= 0·095, NS; (b) 1999, loge (no seedlings + 1) = −0·013x + 1·14, R= 0·40, n= 54, P < 0·01; (c) 2000, loge(no seedlings + 1) = −0·031x + 1·81, R= 0·46, n= 54, P < 0·001; (d) 2001, loge(no seedlings + 1) = −0·01x + 1·03, R= 0·36, n= 55, P < 0·01.

biological control agent abundance

Neurostrota gunniella

In 1999 abundance was significantly higher in the 1997 (F1,20 = 6·28, P < 0·05) and 1998 (F1,20 = 15·22, P < 0·001) herbicide treatment plots (Fig. 9a) and in bulldozed plots (Fig. 9a; F1,7 = 9·23, P < 0·05) and was negatively correlated with the percentage cover of mimosa (Fig. 9b).

Figure 9.

(a) Effect of herbicide and bulldozing treatments on Neurostrota gunniella abundance [mean number frass holes per 50 cm stem (± SE)] in 1999 and overall abundance in 2001 following the fire. Columns with the same letter are not significantly different (LSD). (b) Neurostrota gunniella abundance vs. percentage cover of mimosa recorded in 1999. Loge(no frass holes + 1) = 2·793e−0·0063x, R= 0·91, P < 0·001. (c) Effect of herbicide and bulldozing treatments on Carmenta mimosa abundance [mean number frass holes per 5 min count (± SE)] in 1999 and overall abundance in 2001 following the fire. (d) Carmenta mimosa abundance vs. percentage cover of mimosa recorded in 1999. Loge(no frass holes + 1) = y = 3·2571e−0·0025x, R= 0·82, P < 0·001.

No differences between treatments were detected following the fire treatment. The mean attack rate, at 20·7 frass holes 50 cm stem−1, was significantly higher than the 3·65 holes 50 cm stem−1 recorded in unsprayed, non-bulldozed plots in 1999 (Fig. 9a; t= 6·68, n= 68, P < 0·001).

Carmenta mimosa

Despite a trend for increased abundance of Carmenta mimosa in plots treated with herbicide and bulldozing (Fig. 9c), the abundance in 1999 was not significantly affected by bulldozing (F1,7 = 0·05, P = 0·84) or the 1997 and 1998 herbicide treatments (F1,20 = 1·15, P = 0·29 and F1,20 = 2·25, P = 0·15, respectively). However, like N. gunniella, abundance was negatively correlated with mimosa cover (Fig. 9d).

Carmenta mimosa was the only biological control agent that declined following the fire treatment (t = 6·78, n= 128, P < 0·001; Fig. 9c).

Coelocephalapion pigrae

In 1999, numbers of Coelocephalapion pigrae were not significantly affected by herbicide treatment but were significantly higher in bulldozed plots (Fig. 10a; F1,7 = 9·25, P < 0·05).

Figure 10.

Effect of herbicide and bulldozing treatments on (a) Coelocephalapion pigrae abundance in 1999 and (b) Chlamisus mimosae abundance in 1999 and overall abundance in 2001 following the fire [mean number of beetles per plot (± SE)]. Columns with the same letter are not significantly different (LSD).

In 2001, overall abundance was similar to levels recorded in control plots, prior to the fire treatment (t = 0·98, n= 68, NS; Fig. 10a).

Chlamisus mimosae

In 1999, both 1997 and 1998 herbicide treatments (Fig. 10b; F1,20 = 5·33, P < 0·05 and F1,20 = 11·87, P < 0·05, respectively) significantly affected abundance of Chlamisus mimosae. There was a significant interaction between bulldozing and the 1998 herbicide treatment (F1,9 = 6·08, P < 0·05), indicating abundance was increased by bulldozing unsprayed plots but reduced in bulldozed plots that were treated with herbicide in 1998 (Fig. 10b).

In 2001, abundance was similar to levels recorded in control plots prior to the fire (t = 0·88, n= 68, NS; Fig. 10b).

Acanthoscelides puniceus

Too few A. puniceus were collected for statistical analysis. However, the presence of this species in the 2001 survey indicated it had recolonized plots following the fire.


Impenetrable mimosa thickets were, in some cases, turned into productive, biologically diverse, grassland within just a few years, although not all treatments were equally effective. Treatments were least effective when used in isolation, including bulldozing, which may have been exceptionally destructive to mimosa in this study. Siriworakul & Schultz (1992) noted that regrowth generally occurs following bulldozing. Unusually rapid flooding in early December 1998, when cyclone Thelma produced more than 500 mm of rain in less than 48 h, may have drowned mimosa that was regenerating following the bulldozing treatment. Mimosa is tolerant of flooding if some leaves remain above the water surface, but total immersion for 3 months will drown mature plants (Shibayama et al. 1983).

This study supports previous evidence (Lonsdale & Farrell 1998) that interspecific competition inhibits seedling regeneration. Fluroxypyr probably reduced regeneration directly, by killing plants, and indirectly because it is selective for dicotyledons, allowing monocotyledons to compete with mimosa seedlings.

Lonsdale & Miller (1993) noted follow-up control of seedlings might be necessary following fire, which can eliminate competing vegetation. However, the flush of mimosa germination that we recorded after the November 2000 fire treatment did not result in extensive regeneration because most seedlings apparently drowned during the subsequent wet season. In Australia, mimosa occurs in habitats that flood seasonally every year. Seedling survival should therefore vary according to the extent and duration of flooding. During the course of this study, the annual wet season rainfall (recorded at Darwin Airport; 12°42′S, 130°89′E) varied between a maximum of 2499·4 mm (in 1997–98) and a minimum of 1384·6 mm (in 2000–01). Therefore recruitment failed following the 2000 fire treatment even though the 2000–01 rainfall was the lowest recorded and much lower than the average 1702 mm annual rainfall recorded at Darwin. This indicates that the timing of the fire treatment ensured there was insufficient time for seedlings germinating after the fire to grow tall enough to prevent immersion during subsequent wet season floods. If mimosa stands are burnt earlier in the season, so that seedlings have a longer growing season prior to flooding, then follow-up control treatments are more likely to be needed. For example, seedling growth rates of 1·33 cm day−1 were measured in the field in NT (Lonsdale 1992). Lonsdale & Miller (1993) conducted their burn treatment on 27 September 1988, 38 days earlier in the season than burning was conducted in this study, so that seedlings germinating after their burn treatment would have potentially been c. 50 cm taller at the onset of wet season flooding.

Biological control agent populations were remarkably resilient to the various treatments. Herbicide should not directly affect Coelocephalapion pigrae abundance because treated plants typically take longer to die (Paynter 2003) than the c. 7 days required for larval development (Wilson et al. 1992), but it is not clear why both Coelocephalapion pigrae and Chlamisus mimosae increased in bulldozed plots. Bulldozing may have stressed surviving plants or resulted in enhanced nutrient levels in plants regenerating following control, both of which can favour biological control agents (Briese 1996; Willis, Berentson & Ash 2003).

Neurostrota gunniella abundance, relative to live mimosa biomass, should increase following herbicide application because N. gunniella larvae complete development before the treated plants die (Paynter 2003). However, this could not explain why N. gunniella increased following fire (Fig. 9a). Smith & Wilson (1995) found N. gunniella attack was greatest at stand edges. It seems that, by reducing mimosa populations from monocultures to smaller patches or individual plants, control treatments increased the ratio of ‘edge’ plants to ‘thicket’ plants and therefore the proportion of plants susceptible to N. gunniella attack. This would explain why Carmenta mimosa (which is also most abundant at stand edges; Q. Paynter, unpublished data) also declined in abundance exponentially in relation to mimosa percentage cover.

In contrast to N. gunniella, Carmenta mimosa declined following the fire (Fig. 9c), which reduced the proportion of plants large enough to support larvae (Fig. 4). Carmenta mimosa frass holes are most prevalent in larger woody stems and were never found in stems < 8 mm diameter (Q. Paynter, unpublished data). While N. gunniella can disperse rapidly over many kilometres (Wilson & Forno 1995), Carmenta mimosa is less mobile, spreading at a rate of c. 2 km year−1 (Ostermeyer 2000), so redistribution of this agent may be necessary when very large or isolated mimosa stands are treated.

optimal treatment combinations

On cattle-grazing properties, fire is required to clear mimosa thickets (including deadwood) and enable vehicular access so that cattle herding, property maintenance and ground control of regenerating seedlings can be conducted. As noted by Lonsdale & Miller (1993), burning was an unreliable method for clearing untreated mimosa. Similar results with Acacia saligna were attributed to the high moisture content of the foliage (van Wilgen & Richardson 1985). Although the 1999 herbicide treatment, followed by fire during the subsequent dry season, cleared plots of living mimosa (Fig. 3b), it left sharp stumps that could puncture car tyres. Herbicide application, followed by bulldozing and then fire, cleared stands more effectively. Bulldozing compacts dead mimosa branches, so hotter, more destructive fires can occur and plants in smaller size classes, which were more prevalent in bulldozed plots (Fig. 4), are more susceptible to fire (Lonsdale & Miller 1993). Prior crushing similarly enhanced the use of fire to control gorse (Rees & Hill 2001).

If vehicular access is unnecessary and fire is not required to clear deadwood, control costs are considerably reduced and the potentially adverse effect fire has on competing vegetation can be avoided. This study indicates two aerial herbicide applications in consecutive years (costing c. Australian $40 ha−1; Table 1) would greatly reduce the mimosa infestation (Fig. 3a), allow competing vegetation to rapidly regenerate (Fig. 5) and promote N. gunniella and Carmenta mimosa attack on surviving plants whilst having no detrimental impacts on other biological control agents.

importance of biological control

An unexpected feature of this study was the low degree of mimosa reinfestation following control. For example, in the November 2000 survey mimosa remained at very low levels in bulldozed plots treated with repeat herbicide applications in 1997 and 1998, almost 2 years after any control measures had been applied to those plots (Fig. 3a). This is remarkable considering the highly invasive nature of mimosa during the 1970s and 1980s, when populations doubled in size every 1·2 years and advanced at a rate of 76 m year−1 (Lonsdale 1993).

Buckley et al. (2004) demonstrated that disturbance is an important factor regulating mimosa populations. Feral water buffalo eradication should have reduced the ability of mimosa to invade (Lonsdale 1993). Nevertheless, untreated mimosa thickets did not decline in the decade following feral buffalo eradication (C. Deveraux, personal communication), perhaps because, like broom Cytisus scoparius (Paynter, Downey & Sheppard 2003), untreated thickets cast such dense shade that competing vegetation is often excluded. Following senescence, the probability of regeneration from the seed bank is high, even in the absence of additional disturbance. Biological control alone might therefore succeed by defoliating plants and allowing competing vegetation to persist and inhibit mimosa regeneration. However, this will only occur if disturbance is maintained at low levels, and widespread control could still take decades (Buckley et al. 2004).

Buckley et al. (2004) demonstrated that biological control enhances other control options and increases the range of options that can succeed. By greatly increasing the proportion of plants susceptible to N. gunniella herbivory, herbicide application, bulldozing and fire will have enhanced the impact of biological control. Indeed, several 2-year treatments (e.g. 1999 herbicide application followed by fire in the subsequent dry season) were successful. Buckley et al. (2004) concluded that these would only succeed at low disturbance levels and with biological control present. These studies therefore demonstrate the value of integrating biological control with other options to reduce significantly the cost of managing weeds.

In the past, as for many weed control programmes (Cullen 1996), integration of techniques was limited to using herbicides in areas where complete control was required, and biological control in areas where eradication was no longer an option. This study demonstrates that much greater integration of techniques is possible. To maximize the benefits of biological control, IWM plans should be designed to integrate biological control fully with other methods, rather than to separate them spatially or temporarily.


We thank White Eagle Aboriginal Corporation, especially Margy Daiyi, Linda Ford and Colin and Wayne Deveraux. Mark Lonsdale helped develop the initial experimental design. Warren Müller helped analyse the increasingly complicated experimental design. We especially thank CSIRO and NT DBIRD technical staff, in particular Magen Geyer, Bruce Hitchins, Mathew Hoschke, Bert Luckitsch, Nicole Ostermeyer Merrilyn Paskins, John Ross and Tim Schatz, who helped conduct sampling in often very trying conditions. We thank David Briese and Andy Sheppard for reviewing an earlier version of this manuscript. This work was supported by the Natural Heritage Trust.