DOI: 10.1111/j.1365-3180.2009.00703.x Assessment of management options for Salsola australis in south-west Australia by transition matrix modelling C P D BORGER* , J K SCOTTà, M RENTON , M WALSH  & S B POWLES  *Department of Agriculture and Food, Western Australia, Merredin, WA, Australia,  Western Australian Herbicide Resistance Initiative, Faculty of Natural and Agricultural Sciences, School of Plant Biology, The University of Western Australia, Crawley, WA, Australia, and àCSIRO Entomology, PO Wembley, WA, Australia Received 15 September 2008 Revised version accepted 29 December 2008 effective weed management depended on reducing seed Summary dispersal. The model determined that burning all A matrix model of the life cycle of Salsola australis senesced, mobile plants in late autumn, combined was constructed, based on population ecology data with herbicide control of the largest cohorts of collected from the district of Lake Grace, Western S. australis in summer and autumn, reduced popula- Australia. The model was used to assess potential tion growth rate to 0.79. This control strategy resulted control strategies for this summer annual weed within in a 66.1% chance of the population becoming extinct the Western Australian broad acre grain cropping over 25 years. Management strategies are proposed system. The population growth rate (k) of S. australis based on the results of the models and further in the absence of weed control strategies was 1.49 and research is required to validate their effectiveness and was virtually unaffected by the dormant seedbank. practicality in the field. However, k increased to 8.21 if it was assumed that a Keywords: Russian thistle, tumbleweed, weed control, constant number of seed immigrated into the area in management, model, population growth rate, seedbank, question from neighbouring populations of S. aus- seed dispersal. tralis, through farm-scale seed dispersal. As a result, BORGER CPD, SCOTT JK, RENTON M, WALSH M & POWLES SB (2009). Assessment of management options for Salsola australis in south-west Australia by transition matrix modelling. Weed Research 49, 400–408. chemical and biological control measures used against Introduction morphologically similar species of the Salsola genus in Salsola australis R.B. (Chenopodiaceae; Russian thistle arable cropping systems (Young & Gealy, 1986; Young or tumbleweed) (=in part Salsola tragus as previously & Whitesides, 1987; Blackshaw & Lindwall, 1995; applied in Australia, Borger et al., 2008) is a common Anderson et al., 1998; Hasan et al., 2001; Smith, 2005). weed in the annual grain cropping and ley pasture The population ecology of S. australis within the systems in the broad-scale farming region (wheat-belt) South-west Australian agricultural system, including of south-west Australia (Wilson, 1984; Hussey et al., characteristics of the life cycle, seed biology and seed 1997). There is little research available on the effective- dispersal, has been the subject of recent study. Salsola ness, or long-term impact, of weed control measures australis is an annual plant with an indeterminate applied to S. australis in Australia. However, there is growth habit that produces seeds that shed into the soil much international research on the effect of physical, seedbank or are retained for some weeks or months on Correspondence: Catherine Borger, Department of Agriculture and Food, Western Australia, PO Box 432, Merredin, WA 6415, Australia. Tel: (+61) 08 9081 3105; Fax: (+61) 08 9041 1138; E-mail: cborger@agric.wa.gov.au  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408  Management options for Salsola australis 401 the mature plant. Total viable seed production is include the fate of the seedbank, to determine the impact moderate and long-term dormancy is low (Borger et al., of potential management options on dormant seed. The 2007, 2009). Population ecology data on other weed data used to construct this model were predominantly species has been used to construct models, to determine obtained from trials conducted in the district of Lake how best to manipulate population dynamics and to Grace, Western Australia, from 2004 to 2006 (Borger estimate the long-term effects of weed management et al., 2007, 2009). The S. australis populations in strategies (Caswell, 2001; Holst et al., 2007). Matrix question were growing as weeds in a wheat ⁄ barley population models have previously been used to suc- system. The model assumes an annual cropping event, as cessfully simulate the growth rate of species with that is the predominant land use in the Western multiple cohorts and indeterminate growth habits (Cou- Australian wheat-belt. sens & Mortimer, 1995; Caswell, 2001). These models can also be used to investigate the impact of each specific Within-year life cycle dynamics part of the life cycle, such as the seedbank, on population growth rate (Kalisz & Mcpeek, 1992; Vid- Salsola australis seeds emerge during most months of the otto et al., 2001; Davis, 2006). Davis (2006) found that year. Seedlings progress unevenly through the stages of for annual species with modest fecundity and low long- juvenile vegetative growth, adult vegetative growth, term dormancy, control of freshly produced seed (i.e. the reproduction and senescence, over an average span of short-term seedbank) had a larger impact on population 4.6 months (Table 1, Fig. 1) (Borger et al., 2009). The growth rate than control of seedlings or plants. senesced plants break free of their root systems to A matrix population model of S. australis was become mobile, with some seed shedding prior to and constructed using the available population ecology data, following plant release (hereafter referred to as easily to explore the relative importance of different aspects of shed seed). However, some seed remains attached the life cycle for population growth rate, including the (retained seed), even after several months of movement, short- and long-term seedbank and seed dispersal. The aging and weathering of the senesced plant (Borger model was used to assess the effectiveness of various et al., 2007, 2009). The mobile plants can travel for over physical, chemical and biological control techniques on a kilometre, but over half of the plants move <100 m the population growth rate of this weed. (Borger et al., 2007). In the agricultural system, mobile plants are routinely crushed and incorporated into the soil seedbank before or during crop sowing. So, both Model formation easily shed and retained seeds can germinate in the year following seed production or remain dormant. The Model selection likelihood of seeds growing in the following year (for The population dynamics of S. australis were investi- both seed shed into the soil seedbank and or seed gated using an age-structured population matrix model, dispersing while retained on the mature plants) is based on the model developed by Kalisz and Mcpeek dependant on initial seed viability, seed survival and (1992). The model considers the annual life history of probability of successful establishment, determined by S. australis, but the annual life cycle is expanded to Borger et al. (2007, 2009). Table 1 Mean and standard deviation of Parameter Definition Mean SD each parameter used to describe the life cycle of S. australis defined in Fig. 1 SES Total number of viable, easily shed seeds per plant 72.1 69.09 SR Total number of viable, retained seeds per plant 101.0 96.65 PESG Probability of easily shed seed germinating in the year 0.146 0.0446 following seed production PESD Probability of easily shed seed staying dormant or dying 0.85 0.0446 in the year following seed production PRG Probability of retained seed germinating in the year 0.026 0.0134 following seed production PRD Probability of retained seed staying dormant or dying in 0.974 0.0134 the year following seed production PDG Probability of dormant seeds germinating 0.018 0.0036 PDD Probability of dormant seeds remaining dormant 0.01 0.002 PS Survival probability of seedlings 0.90 0.129 PV Survival probability of vegetative plants 0.28 0.011 PR Survival probability of reproductive plants 0.39 0.054  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408 402 C P D Borger et al. PDD Between-year life cycle dynamics Modelling of within-year population dynamics is difficult to manage accurately because of the overlapping gener- Dormant seeds ations of plants. However, in autumn, when the crop is PDG sown, all plants have finished producing seed or are killed by the cropping enterprise. At this stage of the year, the PESD Seedlings PR life cycle can be collapsed on an annual time span (Fig. 2). During year one, mature plants produced seed rain prior PS to or during autumn. By autumn of year two, some of the PESG PRG seeds from this seed rain event have grown into mature Vegetative plants (first year cohort, Cohort 1), some have remained growth stage dormant (Dormant Seed) or died. By autumn of the third year, the Dormant Seed have germinated to produce a Easily shed Retained PV second cohort (Cohort 2), remained dormant (to con- seeds seeds tribute to Cohort 2 in future) or died. The mature Reproductive Cohort 1 and Cohort 2 plants produce subsequent seed SES growth stage SR rain events to continue the cycle. This model assumes that the proportion of seed remaining dormant for more than PR 2 years (i.e. the persistent seedbank, PDD, Table 1) is 1%, as Borger et al. (2007) found that viability of S. australis Senescence seed was reduced to <2% in the first year after seed production and similar results have been found for Fig. 1 Diagrammatic life table of S. australis. The rectangles morphologically similar species of the Salsola genus represent stages of the life-cycle, including seedlings, plants in the vegetative, reproductive or senesced stage of the life cycle, seeds (Evans & Young, 1972, 1980). However, as long-term retained on the plant, seeds that are easily shed from the plant and dormancy was unknown, the impact of increasing the dormant seeds. The triangles represent the survival probabilities of proportion of seed remaining dormant for over 2 years each life cycle stage, including seedlings to vegetative plants (PS), was investigated, to determine the impact of this param- vegetative to reproductive plants (PV), reproductive to senesced eter on population growth. plants (PR), probability of easily shed or retained seeds germinating in the year following seed production (PESG and PRG), probability It was assumed that Cohort 1 and Cohort 2 plants (in of easily shed or retained seeds becoming dormant or dying (PESD any given year) would produce seed that had an equal and PRD) and the probability of dormant seeds germinating or remaining dormant (PDG and PDD). The diamonds represent C1-C1 C2-C1 fecundity, i.e. total viable easily shed or retained seed produced per plant (SES and SR). Cohort Cohort As a result of the uneven emergence and growth of 1 2 C2-DS S. australis, plants in all life cycle stages are present throughout the year (Borger et al., 2009). However, within the wheat system, seedlings can establish from C1-DS DS-C2 Dormant August (i.e. in winter, within the crop) to May of the Seed following year (i.e. autumn, prior to sowing the next crop). The majority of plants establish over spring and summer and complete their life cycle by late autumn, DS-DS when the wheat crop is sown. The plants that have not completed the life cycle are killed by cultivation and Fig. 2 Life cycle graph for S. australis. Each arrow represents an autumn to autumn transition. Cohort 1 and 2 indicates the herbicides when the crop is sown in April or May. first and second year cohort of plants, resulting from a seed rain Herbicides applied at the time of sowing, and in-crop event. Cohort 2 plants result from Dormant Seed, i.e. seed from herbicides applied later in the season, prevent further the seed rain event that remained dormant for over 1 year. The cohorts from establishing until July–August. The tran- labels associated with each arrow are abbreviations for the sition probabilities (i.e. the probabilities of surviving Cohort 1 and 2 or Dormant Seed stages at the start and end of each transition. The arrows (and values of the arrow labels) from one stage to the next) shown in Table 1 are indicate the probability of individuals at one stage passing into averages taken from all cohorts growing throughout a the next stage. These values are the products of survival year from July–August to April–May (Borger et al., probabilities or survival probabilities and fecundity, identified in 2009). Fig. 1 and defined in Eqn (1), (2), (3) and (4).  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408  Management options for Salsola australis 403 probability of becoming Cohort 1 plants or Dormant absolute change in k resulting from an additive change Seeds in the subsequent year, (i.e. C1-C1 = C2-C1 and to a parameter (i.e. slope of k as a function of the C1-DS = C2-DS, Fig. 2). Given that these values were parameter). Elasticity indicates the relative change in k assumed to be equal, they could have been combined in resulting from a proportional change to a parameter (i.e. the model. They are kept separate, to distinguish the slope of log k as a function of the log of the parameter). sensitivity and elasticity of the population growth rate to The sensitivity and elasticity of k to both transition each parameter and so determine the impact of the probabilities within the matrix model (A) and the lower dormant seedbank on population growth rate (as level parameters used to calculate the matrix transition Cohort 1 plants result from non-dormant seed and probabilities (in Eqns (1), (2), (3) and (4), Fig. 1) were Cohort 2 plants result from Dormant Seeds). This calculated according to the methods in Caswell (2001). allows later investigation of the impact of management options on the seedbank. The transition probabilities are Environmental and demographic stochasticity calculated from the lower level demographic parameters defined in Fig. 1, as shown in Eqns (1), (2), (3) and (4). To simulate population growth under various manage- ment scenarios, stochasticity was incorporated into the C1-C1 ¼ C2-C1 ¼ ððSES  PESG Þ þ ðSR  PRG ÞÞ model, according to the methods of Akcakaya (1991);  PS  PV  PR ð1Þ Akcakaya et al. (1997), utilised by Hiraldo et al. (1996). To simulate environmental stochasticity, seed produc- tion and probabilities (for survival, seed dormancy and C1-DS ¼ C2-DS ¼ ðSES  PESD Þ þ ðSR  PRD Þ ð2Þ seed germination) were considered as random variables with normal distributions. For each simulation, the DS-C2 ¼ PDG  PS  PV  PR ð3Þ value of the parameters was calculated using Eqn (6), where P is the parameter value for a given year in a given DS-DS ¼ PDD ð4Þ simulation, M and S are the mean and standard deviation of each parameter (Table 1) and D is the The transition parameters for each stage in the life normal standard deviation (which varied randomly for cycle of a S. australis population (defined in Fig. 2 and each simulation). The model assumes a perfect correla- Eqns (1), (2), (3) and (4)) were incorporated into the tion between survival probabilities and seed production, transition matrix A. i.e. if the probability of a plant surviving is higher than average then seed production will also be higher than 0 Dormant Seed Cohort 1 Cohort 2 1 average (to simulate a year with favourable environ- Dormant Seed DS-DS C1-DS C2-DS mental conditions). A¼ Cohort 1 @ 0 C1-C1 C2-C1 A Cohort 2 DS-C2 0 0 P ¼ M þ ðS  DÞ ð6Þ Demographic stochasticity was simulated by ran- The transition matrix was used to construct the domly sampling Poisson distributions to determine the transition matrix model shown in Eqn (5), which can be number of seeds produced per plant and binomial used to project the population size. The column vector, distributions to determine the number of plants that describing the number of Dormant Seed, Cohort 1 and survive and number of seeds that become dormant or Cohort 2 individuals at time t + 1, is a function of the germinate each year (Akcakaya, 1991; Akcakaya et al., transition matrix A and the column vector describing 1997). the number of individuals at time t. 0 1 0 1 Dormant Seeds DS-DS C1-DS C2-DS Results @ Cohort 1 A ¼@ 0 C1-C1 C2-C1 A B C B C Cohort 2 DS-C2 0 0 Population projections tþ1 0 1 Dormant Seeds This model indicates k (population growth rate) was  @ Cohort 1 A ð5Þ 1.49 (i.e. the population was increasing by approxi- B C Cohort 2 t mately half every year). The stable age distribution was dominated by dormant seeds (i.e. 0.99 Dormant Seed Sensitivity analysis individuals vs. 0.008 Cohort 1 and 0.0012 Cohort 2 The sensitivity and elasticity of k with respect to changes individuals). However, the reproductive values (i.e. in the demographic parameters indicate the impact of contribution of Dormant Seed, Cohort 1 and Cohort 2 the parameters on k. Sensitivity indicates the additive or individuals to future generations relative to that of  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408 404 C P D Borger et al. Plants (from cohort 1or 2) 60 7000 Table 2 Sensitivity and elasticity of k to individual lower level Cohort 1 parameters (Fig. 1) used to calculate the matrix transition 50 6000 Cohort 2 probabilities Dormant seeds 5000 40 Dormant seeds 4000 Parameter Sensitivity Elasticity 30 3000 20 SES 0.014 0.661 2000 SR 0.003 0.224 10 1000 PESG 6.273 0.617 0 0 PRG 8.787 0.154 0 2 4 6 8 10 PESD 0.076 0.044 Time (years) PRD 0.107 0.070 PS 1.464 0.885 Fig. 3 Population growth of S. australis over 10 years, using Eqn PV 4.705 0.885 (5). Population growth is simulated by assuming a single Cohort 1 PR 3.378 0.885 plant is present at time zero. The total number of Cohort 1 and PDG 9.422 0.114 Cohort 2 plants in the population is plotted on the left axis and PDD 0.115 0.001 the total number of Dormant Seeds is plotted on right axis. Dormant Seed individuals) were much higher for the probability of the retained seeds germinating was very Cohorts (95.9 for both Cohort 1 and Cohort 2 individ- low compared with that of the easily shed seeds. uals compared with 0.11 for Dormant Seed individuals). Population growth, assuming that a single Cohort 1 Persistent seedbank plant was found in the region at the beginning of the period, was simulated over 10 years (Fig. 3). The The persistent seedbank (i.e. PDD) at 1% had no impact population expanded exponentially as there are no on population growth rate (compared with PDD set to density dependent regulators in the model. See Caswell 0%) and sensitivity and elasticity were very low (2001) for a full explanation of the theory and calcula- (Table 2, Fig. 4). In fact, this parameter could be tion of k, stable age distribution and reproductive increased to 20% before the population growth rate values. was affected (i.e. k increased from 1.49 to 1.5). At this level, sensitivity and elasticity increased to 0.126 and 0.008, which was still very low compared with the other Sensitivity and elasticity parameters. If PDD is set to 0.98 (1-PDG, i.e. assumes all The sensitivity of k to the matrix transition probabilities dormant seed survives to remain dormant if it does not was greatest in relation to the fate of Dormant Seeds germinate), then k increased to 1.69. (Fig. 4). However, the elasticity of k was greatest in relation to the transition of Cohort 1 plants to Cohort 1 The agricultural system plants. This indicates that all control measures should focus on disruption of the within-year population Seed dispersal in the agricultural system is limited. dynamics that affect the transition C1-C1 (i.e. seed Cropping fields in the Western Australian wheat-belt are production, survival and germination of seed and usually fenced, with a wire mesh fence around 1 m high, transition probabilities between life cycle stages). This to contain livestock (as annual cropping is rotated with is confirmed by the elasticity of k to lower level pasture systems). Most mobile S. australis plants are parameters (Table 2). However, the elasticity of k contained in the field in which they originate, but c. 25% to the number of retained seeds produced or the of the plants are blown over the fence to migrate into the 100 0.7 0.6 80 0.5 Sensitivity Elasticity 60 0.4 40 0.3 Fig. 4 Sensitivity matrix (left) and 0.2 20 elasticity matrix (right) of k to individual 0.1 matrix transition probabilities. The 0 0 C2 columns in the graphs correspond to the DS C1 C2 C1 DS C1 C1 transition probabilities shown in transition C2 DS C2 DS matrix A.  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408  Management options for Salsola australis 405 neighbouring fields downwind of the population (Borger released before senescence and retained seeds, except for et al., 2007). In the population projection above, the the 2525 retained seeds from the neighbouring popula- number of retained seeds is not varied. It is assumed that tion of 100 plants. Under this management scenario, the 25% of plants (with attached seeds) that leave the k was 7.53. If the neighbouring population contained field by moving downwind of the population in question more than four plants (i.e. 101 retained seeds entering is compensated for by the 25% of plants entering from the field), then k was greater than one under this weed the field upwind of the population. control method. However, if it was assumed that this Salsola australis in Australia may be controlled by control technique was applied to the whole region (i.e. physical, herbicide or biological management options. no plants can migrate from the neighbouring popula- Weed control was simulated by altering the lower level tion) or if it was assumed that there were no neighbour- parameters in Eqns (1), (2) and (3) as specified below in ing plants, k was reduced to 0.55 and so the population the descriptions of each control technique, which altered was declining towards extinction rather than expanding the parameters in the transition matrix model [Eqn (5)]. (as k was <1). The population trajectory summary over However, for simulations of control techniques, the 25 years under this management regime is shown in number of retained seeds was varied. It was still assumed Fig. 5 and the probability of the population becoming that 25% of the plants migrate out of the field each year. extinct within this period was 100%. However, the field upwind of the population is assumed to contain from 0–100 plants. From Table 1, each plant Herbicide control carries 101 viable, retained seeds. Assuming that 25% of a population of 100 plants migrate into the field, then If it was assumed that there were no neighbouring 2525 retained seeds are added to the field in question. A plants, herbicides reducing seed production by 40% or neighbouring population of 100 plants would be a very more resulted in k of <1 (Fig. 5) and a 22.8% chance of small, sparse population of S. australis in an agricultural the population becoming extinct over 25 years. How- field or a normal population observed on the roadside ever, reducing seed production of the population in next to the field. In the absence of control measures, question by 100% resulted in a k of 7.10, if seed from the assuming that 25% of retained seeds leave the field and 100 neighbouring plants could migrate into the area. 2525 seeds from a neighbouring population of 100 plants The neighbouring population needed to consist of less enter the field, k is increased to 8.21. The increase in than nine plants for k to be <1 (i.e. population k (compared with the previous simulation) results from declining) with 100% seed set control. However, herbi- the fixed number of seed migrating into the field. In the cides applied at several times of the year, or residual prior simulation, seed entering and leaving the field were herbicides applied after crop harvest in November, equivalent (i.e. if the population was small and few seeds could reduce the probability of plant survival. When the leave the area, then few seeds enter the area). In an survival probabilities of plants in the seedling, vegetative agricultural situation, seed production in one location and reproductive stages were reduced by 57% or more, may be low because of control measures, but (even in k was less than 1, in spite of seed migration from the poor seasonal conditions) a neighbouring population of neighbouring population. at least 100 plants is likely to be available to rejuvenate the seedbank. Each model was run for 1000 replications Biological control to determine average population growth under various management scenarios, over a 25 year period. The The rust fungus, Uromyces salsolae Reich., is a potential proportion of the 1000 replications in which the popu- biological control agent of S. australis that is already lation reached zero individuals was the extinction found in Australia, although not observed at Lake probability for a given management option (Akcakaya Grace. Preliminary investigations into the effectiveness et al., 1997). of this biological control agent in controlled conditions in France indicated that for S. kali (possibly S. australis) plants, the rust fungus can increase plant mortality by Physical control 54.5% and prevent seed production in 100% of infected In an annual cropping system, physical control could be plants (Hasan et al., 2001). Biological control was achieved through burning senesced plants in autumn simulated by assuming that half of all S. australis plants before the plants become mobile, instead of using in the field became infected with the fungus. So, survival cultivation (prior to sowing) to crush all senesced or to reproductive maturity was reduced by 27.25% and mobile plants and incorporate them into the soil. It was seed production from the remaining population was assumed that burning (before plants become mobile) reduced by 22.75%. This resulted in a k of 0.94 and a controlled easily shed seeds, except for the 24% that are 35.5% probability that the population would become  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408 406 C P D Borger et al. 140 000 are a common feature of agricultural weed species of the 120 000 Salsola genus (Evans & Young, 1972, 1980). Instead of Total number of individuals relying on a dormant seedbank, these species engage in 100 000 broad scale seed dispersal (Borger et al., 2007), a trait 80 000 has previously been found to develop at the expense of seed dormancy (reviewed by Rees, 1993). The model 60 000 Physical confirmed that seed dispersal had a very large impact on 40 000 population growth rate (i.e. increasing k from 1.49 to Chemical 20 000 Biological 8.21). It indicated that control measures applied to a Physical + chemical single field had to reduce seed dispersal in order to be 0 0 5 10 15 20 25 effective. Alternatively, control measures could be Time (years) applied to an entire farm (including roadsides) or to the entire district, to prevent seed dispersal between Fig. 5 Total number of combined Cohort 1, Cohort 2 and Dormant Seed individuals in a population of S. australis with an populations. This confirms the results of prior studies, initial abundance of 1000 Cohort 1 plants over 25 years, exposed to which indicate that dispersal within and between fields, physical, chemical or biological control measures. Physical control can have a greater impact on growth of weed popula- assumes that all seed is burnt, except for the 24% of seed that tions than seed production (Ghersa & Roush, 1993; comes loose before plants are fully senesced. Chemical control Gonzalez-Andujar et al., 2001). Other weed species have assumes that herbicides reduce seed production by 40%. Biological been shown to have a positive growth rate in spite of low control assumes that the rust fungus reduces plant survival by 27.25% and remaining seed production by 22.75%. Combined seed production (through weed control techniques) physical and chemical control assumes that the 46% of retained where dispersal rates remain high (Ballare et al., 1987; seeds are destroyed and seedling survival is reduced to 60%. All Norris, 1992). simulations (except combined physical and chemical control) Physical weed control through burning crop residue, assume that control measures were applied to the population in in an attempt to burn weed seeds, is a control measure question and to neighbouring populations. currently practiced in the Western Australian wheat-belt (Walsh & Newman, 2007). No data exist on the extinct over 25 years. It was assumed that this control feasibility of burning senesced S. australis plants. How- method was applied to all S. australis populations in the ever, it is likely to be effective, given that the seeds are region (i.e. the population in question and any neigh- only protected by a thin, dry fruiting perianth (CPD bouring plants), since a rust fungus would not be limited Borger personal observation; Wilson, 1984). Some to a single field or population (Fig. 5). farmers currently rake the mobile plants into compacted piles to burn (G Mussel, Department of Agricultural and Food Western Australia, pers. comm.). However, Discussion this may not be effective, as the agitation of the raking The population growth rate indicated that the process would cause seeds to shed from the mobile S. australis population at Lake Grace was expanding plants and thereby escape the burning process. Cultiva- exponentially (because of the lack of density dependant tion is highly effective in controlling species of the regulators in the model). Population growth rate was Salsola genus (Anderson et al., 1998), but was not predominantly influenced by the production and germi- investigated because weed control through cultivation is nation of easily shed seed and plant survival. The not commonly practiced in south-west Australia (Turner dormant seedbank had virtually no effect on population & Asseng, 2005). growth rate, even when set at a much higher level (i.e. Herbicides are not very effective when applied to 20%) than is likely to occur within agricultural systems S. australis plants beyond the seedling stage, with (Borger et al., 2007). Seed viability of S. australis in the commonly recommended herbicides controlling region of Lake Grace was below that of other morpho- 50–90% of mature plants (Young & Whitesides, 1987; logically similar species of the Salsola genus (Young & Mussell & Stewart, 2004). However, if herbicides are Evans, 1979; Young, 1991). However, S. australis applied to all populations in the region, then it was only populations with equally low viability have been necessary to reduce seed production by 40% to cause the observed elsewhere in the wheat-belt (Borger et al., population to approach extinction. As for physical 2007, 2009) and the low seed viability was not causing control, reduction in seed set had virtually no effect if the population to approach extinction. additional seed migrated into the area. Reducing Given the very small impact of the dormant seedbank survival probability of plants by 57% or more resulted on population growth rate, seedbank management is not in the population approaching extinction in spite of important for S. australis control. Short-lived seedbanks immigrating seeds. Given that cohorts of S. australis  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408  Management options for Salsola australis 407 seedlings establish in almost every month of the year has previously been justified by arguing that a declining (Borger et al., 2009), reducing survival of all plants species will be below carrying capacity (i.e. not subject to could most easily be achieved through use of residual density dependence), but Grant and Benton (2000) point herbicides (from the sulfonylurea chemical group). out that even an endangered species may be locally These herbicides are highly effective against species of abundant or show density dependence in part of the life the Salsola genus, including S. australis (Young & cycle. However, for a weed population in an agricultural Gealy, 1986; Mussell & Stewart, 2004). However, field, this effect is minimised, as control techniques are residual herbicides cannot be used in all systems, applied to the entire population, to reduce density in a because of crop compatibility issues. Unfortunately, on uniform manner. Further, part of the density-dependent alkaline soils where S. australis is abundant, residues of competition is accounted for by considering demo- residual herbicides are of particular concern (Mitchell & graphic stochasticity. Wilcox, 1988; Hollaway et al., 2006). This research concludes that a combination of The biological control agent Uromyces salsolae has physical control, through burning mobile plants at already been found on Salsola populations in Australia the end of autumn, and chemical control removing the as well as other continents (Hasan et al., 2001). How- largest seedling cohorts over summer, is likely to prove ever, this fungus has only been evaluated in controlled the most effective control. Burning mobile plants at the glasshouse conditions. To reduce S. australis popula- end of autumn (rather than senesced, non-mobile tions to extinction, it would need to act as effectively in plants earlier in autumn), would only destroy 46% of field conditions as it does in controlled conditions. Other seeds, since these plants have spent the autumn possible biological control agents are in use or under shedding seed through movement, aging and weath- evaluation for use against various species of the genus ering (Borger et al., 2007). However, burning at this Salsola (reviewed by Smith, 2005). However, it is point would also remove any mobile plants that have unlikely that exotic biological control agents would be immigrated into the field from neighbouring popula- released against S. australis, as it is likely this species is tions, reducing seed dispersal. Salsola australis seed- native to Australia (Borger et al., 2008). lings establish throughout the year, but in a crop, most The most successful control measures identified here seed production results from cohorts that establish in rely on killing plants on a regional scale, but this is not summer. Borger et al. (2009) found that one large (21 always practical. It requires all farmers and land seedlings m)2) and one moderate (6 seedlings m)2) managers in the district to be equally concerned about germination event occurred over the summer fallow controlling S. australis. While some districts of Western period of 2004 ⁄ 2005. Chemical control, through appli- Australia have co-ordinated ÔFarmer GroupsÕ that act cation of herbicides to these cohorts, would reduce together to control weeds in their region, it is not seedling survival from 90% to 60%. Under this common. Therefore, the combination of physical con- combination of control techniques, k was 0.79 and trol (through burning mobile plants) and chemical the model population had a 66.1% probability of control (through killing the largest cohorts), is the most becoming extinct over 25 years (Fig. 5). practical recommendation for control of S. australis This research is based on trials that were conducted populations. Burning plants in late autumn destroyed in a single year. Clearly, seasonal variation could have a plants (seeds) that dispersed into the region, and so significant impact on population dynamics of S. australis allowed the population to approach extinction in spite and further research is required to validate the assump- of neighbouring populations. Using herbicides to kill the tions of the model. However, the purpose of the model largest summer cohorts of S. australis is a simple control was to identify profitable areas of future research into technique to apply. However, the occurrence of summer the control of S. australis, rather than conclusively cohorts is directly dependent on rainfall. The cohorts identify successful control techniques. Further research noted by Borger et al. (2009) in 2004 ⁄ 2005 established is required to determine the effectiveness and practicality after rainfall greater than 15 mm. As summer rainfall is of these weed management techniques in field condi- erratic, in some years, rainfall >15 mm may occur in tions, against populations with varying growth rates. every month and in some years, there may be no summer rainfall. Therefore, the cost of this control technique will Acknowledgements be highly variable. This model does not include density-dependent reg- The authors thank the Grains Research and Develop- ulation. Grant and Benton (2000) concluded that not ment Corporation and the Department of Agriculture including density dependence in population analyses and Food, Western Australia, for the funding of a PhD may lead to misleading conclusions for population scholarship to C. Borger. Thanks to Dr Kate Stokes management. Lack of density dependence in a model (CSIRO, Australia) for reviewing this manuscript.  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408 408 C P D Borger et al. References Salsola kali in the USA. Biocontrol Science and Technology 11, 677–689. AKCAKAYA HR (1991) A method for simulating demographic HIRALDO F, NEGRO JJ, DONAZAR JA & GAONA P (1996) A stochasticity. Ecological Modelling 54, 133–136. demographic model for a population of the endangered AKCAKAYA HR, BURGMAN MA & GINZBURG LR (1997) lesser kestrel in southern Spain. Journal of Applied Ecology Applied Population Ecology: Principles and Computer Exer- 33, 1085–1093. cises using RAMAS(r) EcoLab 2.0. Applied Biomathemat- HOLLAWAY KL, KOOKANA RS, NOY DM, SMITH JG & ics, Setauket, NY, USA. WILHELM N (2006) Crop damage caused by residual aceto- ANDERSON RL, TANAKA DL, BLACK AL & SCHWEIZER EE lactate synthase herbicides in the soils of south-eastern (1998) Weed community and species response to crop Australia. Australian Journal of Experimental Agriculture 46, rotation, tillage and nitrogen fertility. Weed Technology 12, 1323–1331. 531–536. HOLST N, RASMUSSEN IA & BASTIAANS L (2007) Field weed BALLARE CL, SCOPEL AL, GHERSA CM & SANCHEZ RA (1987) population dynamics: a review of model approaches and The population ecology of Datura ferox in soybean crops: a applications. Weed Research 47, 1–14. simulation approach incorporating seed dispersal. Agricul- HUSSEY B, KEIGHERY G, COUSENS R, DODD J & LLOYD S (1997) tural Ecosystems & Environment 19, 177–188. Western Weeds: A Guide to the Weeds of Western Australia. BLACKSHAW RE & LINDWALL CW (1995) Management systems The Plant Protection Society of Western Australia, Perth, for conservation fallow on the southern Canadian prairies. WA, Australia. Canadian Journal of Soil Science 75, 93–99. KALISZ S & MCPEEK MA (1992) Demography of an age- BORGER CPD, WALSH M, SCOTT JK & POWLES SB (2007) structured annual: resampled projection matrices, elasticity Tumbleweeds in the Western Australian cropping system: analyses and seed bank effects. Ecology 73, 1082–1093. seed dispersal characteristics of Salsola australis. Weed MITCHELL AA & WILCOX DG (1988) Plants of the Arid Research 47, 406–414. Shrublands of Western Australia. University of Western BORGER C, YAN G, SCOTT JK, WALSH M & POWLES S (2008) Australian Press and Western Australian Department of Salsola tragus or S. australis (Chenopodiaceae) in Australia – Agriculture, Perth, WA, Australia. untangling the taxonomic confusion through random MUSSELL G & STEWART V (2004) Prickly Saltwort Control. [on- amplified microsatellite polymorphism (RAMP) and cyto- line]. Available at: http://www.agric.wa.gov.au/ (accessed 13 logical analysis. Australian Journal of Botany 56, 600–608. August 2008). BORGER CPD, SCOTT JK, WALSH M & POWLES SB (2009) NORRIS R (1992) Case history for weed competition ⁄ popula- Demography the weed Salsola australis in the agricultural tion ecology: barnyardgrass (Echinochloa cruss-galli) in region of south-west Australia. Weed Research 49 (in press). sugarbeets (Beta vulgaris). Weed Technology 6, 220–227. CASWELL H (2001) Matrix Population Models: Construction, REES M (1993) Trade-offs among dispersal strategies in the Analysis and Interpretation. Sinauer Associates Publishers, British flora. Nature 366, 150–152. Sunderland, MA, USA. SMITH L (2005) Host plant specificity and potential impact COUSENS R & MORTIMER M (1995) Dynamics of Weed of Aceria salsolae (Acari: Eriophyidae), an agent proposed Populations. University Press, Cambridge, UK. for biological control of Russian thistle (Salsola tragus). DAVIS AS (2006) When does it make sense to target the weed Biological Control 34, 83–92. seed bank? Weed Science 54, 558–565. TURNER NC & ASSENG S (2005) Productivity, sustainability, EVANS RA & YOUNG JA (1972) Germination and establishment and rainfall-use efficiency in Australian rainfed Mediterra- of Salsola in relation to seedbed environment. II. Seed nean agricultural systems. Australian Journal of Agricultural distribution, germination and seedling growth of Salsola and Research 56, 1123–1136. microenvironment monitoring of the seedbed. Agronomy VIDOTTO F, FERRERO A & DUCCO G (2001) A mathematical Journal 64, 219–224. model to predict the population dynamics of Oryza sativa EVANS RA & YOUNG JA (1980) Establishment of barbwire var. sylvatica. Weed Research 41, 407–420. Russian thistle in desert environments. Journal of Range WALSH M & NEWMAN P (2007) Burning narrow windrows for Management 33, 169–173. weed seed destruction. Field Crops Research 104, 24–30. GHERSA CM & ROUSH ML (1993) Searching for solutions to WILSON PG (1984) Chenopodiaceae. Flora of Australia 4, weed problems: do we study competition or dispersion? 313–317. BioScience 42, 104–109. YOUNG JA (1991) Tumbleweed. Scientific American 264, 58–63. GONZALEZ-ANDUJAR JL, PLANT RE & FERNANDEZ-QUINTA- YOUNG JA & EVANS RA (1979) Barbwire Russian thistle NILLA C (2001) Modeling the effect of farmersÕ decisions on seed germination. Journal of Range Management 32, the population dynamics of winter wild oat in an agricultural 390–394. landscape. Weed Science 49, 414–422. YOUNG FL & GEALY DR (1986) Control of Russian thistle GRANT A & BENTON TG (2000) Elasticity analysis for density- (Salsola iberica) with chlorsulfuron in a wheat (Triticum dependent populations in stochastic environments. Ecology aestivum) summer-fallow rotation. Weed Science 34, 318– 81, 680–693. 324. HASAN S, SOBHIAN R & HERARD F (2001) Biology, impact and YOUNG FL & WHITESIDES RE (1987) Efficacy of postharvest preliminary host-specificity testing of the rust fungus, herbicides on Russian thistle (Salsola iberica) control and Uromyces salsolae, a potential biological control agent for seed germination. Weed Science 35, 554–559.  2009 The Authors Journal Compilation  2009 European Weed Research Society Weed Research 49, 400–408