Malin J. Hansen, Department of Biology, University of Regina, 3737 Wascana Parkway, Regina, Saskatchewan, Canada S4S 0A2 (fax +306 337 2410; e-mail email@example.com).
1The control of invasive species is a challenge heightened by the dependency of management outcomes on environmental variation. This is especially true for plants invading semi-arid habitats, where growing season precipitation varies greatly among years.
2Agropyron cristatum is an invasive grass widely introduced in the Great Plains of North America. We studied its demographic responses to management using field experiments and matrix population models. Plants were clipped to simulate grazing, treated with herbicide or left unmanaged, at three levels of water availability, for 2 years.
3Growth rates (λ) were high in unmanaged populations. Clipped populations were mainly stable, whereas λ for herbicide-treated populations varied greatly with water availability and between years. Low λ in clipped and herbicide-treated populations was mainly the result of low seed production, and these populations were the most sensitive to water availability.
4Clipped populations produced no seeds in the second year, indicating a cumulative negative effect of defoliation. In general, seed production, germination and juvenile survival all increased with water supply, suggesting that invasion may increase under wet conditions.
5Population projections revealed a steady increase in population size for unmanaged populations, whereas clipped populations were more stable. Herbicide-treated populations mainly decreased. Life-cycle stages associated with recruitment contributed the most to the rapid growth of unmanaged populations, whilst the persistence of managed populations relied on the survival of established tussocks.
6Synthesis and applications. These results demonstrate strong management effects on A. cristatum invasion in spite of significant population responses to water availability. Management can therefore have large effects on the invasion of native grasslands regardless of among-year variability in precipitation.
We used size-structured matrix population models and elasticity analysis to evaluate management effects on demographic patterns and to project growth of Agropyron cristatum (L.) Gaertn. populations invading native North American grassland. This method allows for key stages in the life cycle of invasive species to be identified and targeted, and the results of management to be evaluated. Matrices for experimental A. cristatum populations were produced using three management treatments: clipping to simulate grazing, herbicide application, and no management.
We asked two main questions. (i) Can A. cristatum invasion be controlled by management over the long term? (ii) Does water availability affect the impact of management practices? We conducted the study over 2 years in order to examine how the results depend on the duration of study.
Agropyron cristatum is a C3 perennial tussock grass native to south Siberian semi-desert regions (Hubbard 1949; Looman & Henrichs 1973), an environment much like that of the Great Plains of North America. In its native range, it co-exists with genera that also dominate Great Plains grasslands, i.e. Stipa, Festuca, Koeleria and Artemisia (Looman & Henrichs 1973).
Agropyron cristatum has been planted in abandoned fields across the northern Great Plains for pasture restoration and soil protection purposes since the 1930s (Looman & Henrichs 1973). This species has been used because of its rapid establishment and spread, high cover and ability to improve pastures and decrease erosion (Dillman 1946; Rogler 1960). Today as much as 10 million ha of the northern Great Plains may be covered by A. cristatum (Lesica & DeLuca 1996). Agropyron cristatum spreads solely by seed and seed production is copious (Pyke 1990), resulting in an extensive seed bank (Marlette & Anderson 1986). Seed bank density declines exponentially with age so that few viable seeds remain after 5 years (Moriuchi et al. 2000). In addition, seed predation in the seed bank is high in semi-arid grasslands, as much as 50% over the growing season (Pyke 1990). Agropyron cristatum invades native grassland adjacent to planted fields and roadsides (Hull & Klomp 1967; Heidinga & Wilson 2002), where it establishes persistent monotypic stands with low diversity (Marlette & Anderson 1986; Bakker & Wilson 2004).
We worked in Grasslands National Park (49°22′N, 107°53′W) in south-western Saskatchewan, Canada. The native vegetation is mixed-grass grassland dominated by blue grama grass Bouteloua gracilis, needle-and-thread grass Stipa comata and spikemoss Selaginella densa (Christian & Wilson 1999). The average precipitation during the growing season (April–August) was 195·6 mm during 1937–2001 (data from the nearest meteorological station, about 6 km west of the study site; Environment Canada, unpublished data). However, because of high among-year variation in precipitation in this region (Briggs & Knapp 1995), annual precipitation typically differs significantly from the average. We studied population dynamics in 2002, when the total growing season precipitation was 326·8 mm, and in 2003, when the total growing season precipitation was 116·6 mm, representing 167% and 60%, respectively, of the average growing season precipitation.
We studied the responses of A. cristatum tussocks to three management treatments (clipping to simulate grazing, application of herbicide and no management) at each of three water treatments (dry, average and wet), producing a factorial design with two factors, each with three levels. We established 360 study plots (60-cm diameter, separated by > 4 m), each containing a single tussock of A. cristatum in an area of native grassland (2 ha) where A. cristatum had invaded from an adjacent planted field.
Grazing was simulated by clipping all biomass in study plots 6 cm above ground three times early in the growing season (May and June) of 2002 and 2003, allowing regrowth before the autumn. Herbicide (2 : 1 glyphosate) was applied to A. cristatum tussocks once in May of both 2002 and 2003.
Water was applied every 2 weeks from April to August of 2002 and 2003. Plots in the wet water treatment received water equivalent to the precipitation during the wettest growing season (381·5 mm in 1993) for the period for which data were available (1937–2001). The average water treatment received water equivalent to the average growing-season precipitation (195·6 mm) during the same period. Monthly water addition reflected the average seasonal variation in precipitation. Ambient precipitation for each 2-week period was subtracted from the amount applied to the plots. During periods when the ambient precipitation was greater than the treatment amount, no additional water was applied. Each plot was surrounded by a metal circle (60-cm diameter, 20 cm tall, buried 10 cm deep) centred on the tussock, which helped to prevent added water from leaking out in the wet and average water treatments, and discouraged soil water from entering the dry treatments.
In the dry treatment, plots received no additional water and were covered by rain-out shelters, clear acrylic cones (60-cm diameter at the base, 1 m high), raised 10 cm above the ground, with a vent at the top to allow air flow. Soil moisture was measured 2 days after each water application using resistance blocks (Rundel & Jarrell 1991) buried 8 cm below the soil surface in four plots of each water treatment, and was significantly reduced by shelters (dry treatment mean −0·18 MPa, average treatment mean −0·085 MPa, wet treatment mean −0·07 MPa, F2,113 = 8·70, P= 0·0003). Soil moisture did not differ significantly (P < 0·05) between wet and average treatments. Evaporation and thus soil moisture are highly influenced by air and soil temperature (Breshears et al. 1998). We therefore measured both air and soil temperature in sheltered and unsheltered plots. Air temperature (measured 50 cm above the ground three times a day for several days in 2002) was significantly higher in sheltered plots (28·6 °C) than in plots without shelters (26·7 °C; F1,338 = 15·18, P < 0·0001), as was soil temperature (measured 10 cm below ground; sheltered plots, 21·1 °C; no shelter, 20·1 °C; F1,338 = 9·78, P= 0·0019). Tussocks in the dry treatment therefore experienced slightly warmer temperatures, which are typical of dry growing seasons (Breshears et al. 1998) and increase the effect of water stress (Shah & Paulsen 2003). Photosynthetic photon flux density, measured at midday using a Sunfleck Ceptometer (model SF-40; Decagon Devices, 1991, Pullman, WA), was significantly lower in sheltered plots than in plots without shelters (mean in sheltered plots, 1411·2 µmol photons m−2 s−1; unsheltered, 1795·4 µmol photons m−2 s−1; F2,460 = 181·35, P < 0·0001). This reduction of light probably had no or little effect on photosynthesis because the photon flux density remained above light-saturated levels for prairie plants (Foggo & Warrington 1989; Turner & Knapp 1996). In summary, the additional effects of the rain-out shelters on temperature and light were significant but small. While building shelters over all tussocks would allow the strongest interpretation of the water effect, experimental resources were limited and we maximized the number of replicates (n = 10) instead of equipping all plots with shelters.
In summary, there were nine experimental treatments (three management treatments at each of three levels of water availability). Each treatment was applied to each of four size classes of tussocks (see below), for a total of 36 experimental units, each of which was replicated 10 times.
To construct life cycles and matrix population models, tiller dynamics of A. cristatum were measured over 2 years. The initial size of each tussock was determined by counting 2001 tillers that were alive. This was done in April 2002 before new tillers emerged. 2001 tillers were easily distinguished from older tillers by colour and firmness. Tiller number was also counted in August 2002 and 2003.
We counted and collected all seed heads from each tussock in August 2002 and 2003. The average number of flowers per reproductive tiller for each tussock was determined by counting the spikelets on a maximum of five reproductive tillers on each tussock. Each spikelet had on average 4·7 flowers, which agrees with earlier reports (McGregor & Barkley 1986). As many flowers never develop seeds, we calculated the percentage of fully developed seeds for 10 tussocks for each management and water treatment combination. The average percentage of developed seeds for each combination was then used to calculate the actual seed production for each tussock. As A. cristatum is self-incompatible (Dewey 1963), very few seeds were developed under the rain-out shelters. In order to estimate seed development for tussocks in the dry water treatment, we used the percentage of developed seeds of tussocks in 2001, when precipitation during the growing season was 168·8 mm, which is well below average.
germination, juvenile survival and seed bank
Germination of A. cristatum was studied in 90 plots (three water treatments × 30 replicates) in native grassland about 500 m from the tussock-containing plots. Germination plots were separated by > 1 m. Each germination plot was paired with an adjacent control plot in which no seeds were sown, allowing us to measure potential establishment from the seed bank. Germination plots were enclosed in PVC tubes (10 cm in diameter) to retain added water, inserted into the ground to a depth of 4 cm (Ambrose & Wilson 2003). The tops of tubes were left 4 cm above ground to prevent seeds from blowing away. Fifty seeds were broadcast in each plot in September 2003. The same water treatments applied to plots containing tussocks were applied to seed plots. Each plot of the dry treatment was covered by a sheet of transparent acrylic (20 × 20 × 0·3 cm) fixed 4 cm above the top of the PVC tube (Ambrose & Wilson 2003). Germinated seeds were counted and marked during May–August 2004.
Juvenile survival was examined in 90 additional plots (three management treatments × three water treatments × 10 replicates) interspersed among the tussock-containing plots. Juvenile survival plots were the same size as the germination plots. Seeds were broadcast in May 2002. A third of the plots were clipped on three occasions from May to June in 2003. A third of the plots were treated with herbicide once in May 2003. Plots were watered according to the same schedule as germination and tussock-containing plots. Surviving juveniles were counted in July 2004.
Survival of seeds in the seed bank was measured using non-germinated seeds collected from the juvenile survival plots in August 2003. Non-germinated seeds were germinated on Petri dishes in a greenhouse under suitable light and moisture conditions. We used the survival rate of these seeds to calculate annual survival of seeds in the seed bank for the life-cycle models.
life cycle ofa. cristatum
A life cycle was constructed for A. cristatum using the probabilities of tussocks changing in size, average seed production, germination, juvenile survival (progression to size class 1) and survival of seeds in the seed bank, for each management and water treatment combination over two complete years, 2001–02 and 2002–03 (Fig. 1).
Tussocks were assigned to one of four size classes according to the number of live tillers (class 1, 1–46 tillers; class 2, 47–77; class 3, 78–140; class 4, > 140). Size classes were determined according to the tussock size distribution prior to the experiment (M. Hansen, unpublished data). The distribution error increases as the range of a size class increases because the probability of changing size class for tussocks of different sizes within a size class differs. On the other hand, the sampling error increases as the range of a size class decreases because the size class will contain fewer individuals. Initial size classes were therefore adjusted to minimize sampling and distribution error using Moloney's algorithm (Moloney 1986).
Juveniles were defined as vegetative 1-year-old-tussocks, generally with one tiller. Seed bank turnover was calculated using survival of non-germinated seeds after 1 year on the ground. Seed survival was tested by germinating these seeds in the greenhouse. Annual seed survival of A. cristatum in the seed bank was 28–58%, or 2–12% over 3 years. In order to keep the population matrix models to a reasonable size, we assumed that the survival of seeds after 3 years was negligible and that the life span of seeds entering the seed bank was 3 years (Fig. 1).
matrix population model of a. cristatum
A size-structured matrix population model (Caswell 2001) was constructed from the life-cycle data for each treatment combination (three management treatments × three water treatments) in each year, resulting in 18 matrices.
Dominant eigenvalues (finite growth rates, λ; see Appendix S1 in the Supplementary material) and elasticity for each matrix element (relative contribution to λ; see Appendix S2 in the Supplementary material) were calculated for each of the 18 matrices. A 95% confidence interval was generated for each λ with a bootstrap procedure using 10 000 iterations.
To determine the relative contribution of stage transitions in the life cycle to λ, elasticity was summarized for regions of the matrix representing recruitment (seed production, and transitions to and from the seed banks and juveniles) and for regions representing changes in tussock size (transitions from each size class). These regions, rather than transitions, will therefore be referred to when discussing elasticity. Similar approaches have been used in previous studies (Silvertown et al. 1993; Parker 2000). Matrices for some treatment combinations, i.e. clipped populations in 2003 and herbicide-treated populations in the average and dry treatments in 2002 and 2003, had element values that were too low for the calculation of elasticity to be valid. Matrices for herbicide-treated populations in the wet treatment in 2002 and 2003 were condensed to contain one size class for tussocks in order to give valid λ and elasticity values.
Analyses of the transition matrices and the calculation of bootstrap 95% confidence intervals were performed using MATLAB (version 6·5; The MathWorks, 2002, Natick, MA). Analyses to compare tiller number and seed production among management and water treatments were performed with anova using JMP (version 3·1·5; SAS Institute, 1994, Cary, NC).
population projections ofa. cristatum
We calculated 5-year projections of A. cristatum populations separately for unmanaged, clipped and herbicide-treated treatments. Projections were done using population matrices from each management, and among-year variation in precipitation was incorporated into projections by using sequentially matrices from populations under different water treatments in the order average, dry, average, wet, average. The decision to make 5-year projections was based on the fact that the study spanned 2 years only and long-term projections often fail to make accurate predictions (Menges 2000; Lindborg & Ehrlén 2002). The results from these 5-year projections agreed, however, with A. cristatum population trends projected over longer terms using real precipitation data (M. Hansen, unpublished data). Projections started with 10 tussocks from each of the four size classes, a seed bank containing 220 seeds and 83 juveniles. These numbers of seeds and juveniles are likely to result from the 40 tussocks after 1 year (M. Hansen, unpublished data).
Projections of matrices were performed in Excel (version mac 2001; Microsoft, 2001, Redmond, WA).
Unmanaged tussocks significantly increased in tiller number in both 2002 (F1,338 = 12·01, P= 0·0006) and 2003 (F1,338 = 22·35, P < 0·0001; Fig. 1a,b). Because of the large variation in seed production for unmanaged populations, λ varied greatly, but was significantly > 1 in all water treatments in 2002 and 2003 (Fig. 2). As a result, unmanaged populations of A. cristatum increased rapidly when projected over 5 years using data from either 2002 or 2003 (Fig. 3). λ for unmanaged populations was significantly higher than λ for clipped populations, except in dry treatments in both 2002 and 2003 (Fig. 2). λ for unmanaged populations was significantly higher than λ for herbicide-treated populations, except in wet treatments (Fig. 2).
Clipping had no significant effect on tiller number in 2002, but significantly decreased tiller number in 2003 (F1,338 = 40·55, P < 0·0001; Fig. 1c,d). Clipping significantly reduced seed production in 2002 (F1,338 = 24·46, P < 0·0001) and resulted in no seed production in 2003. For clipped populations, λ was not significantly different from 1 in 2002 in any of the water treatments, but was significantly < 1 for wet and dry treatments in 2003 (Fig. 2). Consequently, clipped populations slowly increased when projected over 5 years using data from 2002, but decreased when using data from 2003 (Fig. 3). λ for clipped populations was not significantly different from λ for herbicide-treated populations (Fig. 2).
Herbicide significantly reduced tiller number from 2001 to 2002 (F1,338 = 165·08, P < 0·0001) but tiller number stayed the same between 2002 and 2003 (Fig. 1e,f). λ for herbicide-treated populations was significantly < 1 in average and dry treatments in both years, but not in the wet treatment (Fig. 2). Herbicide-treated populations decreased when projected over 5 years using data from 2002, but fluctuated in 2003 (Fig. 3).
Juvenile survival was significantly higher in unmanaged plots than in clipped and herbicide-treated plots (F2,87 = 13·06, P < 0·0001; Fig. 1).
λ differed significantly among water treatments for managed (clipped and herbicide-treated) populations, but not for unmanaged populations. Because λ was > 1 in all water treatments, unmanaged populations increased rapidly regardless of water supply. Unmanaged populations in the wet water treatment had significantly higher seed production than those in average and dry water treatments in both 2002 (F2,117= 18·27, P < 0·0001) and 2003 (F2,117 = 12·02, P < 0·0001).
Clipped tussocks in the dry treatment had significantly fewer tillers than tussocks in the average and wet treatments in 2003 (F2,117 = 5·71, P < 0·0001) but not in 2002. This suggested that the effect of clipping was more pronounced in dry conditions. Clipped tussocks had significantly higher seed production in the wet treatment than in the average and dry treatments in 2002 (F2,117 = 3·30, P= 0·0386) but produced no seeds in any water treatment in 2003 (Fig. 1).
For herbicide-treated populations, λ was significantly < 1 in average and dry treatments in both years, but not in the wet treatment (Fig. 2). Herbicide-treated populations varied when projected for 5 years using data from 2003 (Fig. 3) but decreased when using data from 2002.
Germination (wet, 0·134; average, 0·112; dry, 0·024; F2,87 = 15·04, P < 0·0001) and juvenile survival (wet, 0·089; average, 0·099; dry, 0·026; F2,87 = 6·26, P= 0·0029) were significantly higher in wet and average plots than in dry plots.
contribution of life-cycle stages to population growth
High elasticity values identify transitions of the life cycle that contribute the most to population growth (Caswell 2001). Recruitment (seed production, transitions to and from seed banks and juveniles) as well as transitions from small tussocks (1–45 tillers) had the highest elasticity in unmanaged wet and average water treatments of A. cristatum in both 2002 and 2003 (Fig. 4; see Appendix S2 in the Supplementary material). Dry unmanaged populations, however, demonstrated less variable elasticities across the life-cycle stages, but the elasticity for recruitment and transitions from larger tussocks (> 76 tillers) was slightly higher (Fig. 4; see Appendix S2 in the Supplementary material).
Clipped populations in the wet and average treatments demonstrated high elasticity for transitions from small tussocks, while the elasticity for recruitment was lower. On the other hand, populations in the clipped dry treatment showed high elasticity for transitions from larger tussocks, while the elasticity for recruitment was very low.
Herbicide-treated populations showed high elasticity for recruitment in 2003 but more similar elasticities across the life-cycle stages in 2002 (Fig. 4; see Appendix S2 in the Supplementary material).
cana. cristatuminvasion be controlled by management over the long term?
The high growth rate of unmanaged populations (Fig. 2) was mainly the result of life-cycle stages associated with recruitment (Fig. 4). For clipped and herbicide-treated populations, however, the survival of larger tussocks was more important for maintaining the population, because of low seed production, low germination and low juvenile survival rates. Similar results were found in populations of invasive scotch broom Cytisus scoparius in Washington, USA: seedling survival and growth rate were high for the invading edge population and recruitment parameters, such as seed production and the survival and growth of seedlings, had higher elasticity values than the survival of large individuals. Seedling survival was low, however, in the middle of the stand because of density-dependence (non-invasive population), and the survival of large plants had high elasticity (Parker 2000). Our results also agree with a study of a non-invasive perennial tussock grass Danthonia sericea in North Carolina, USA, which had low seed production (compared with A. cristatum), low λ and high elasticity for the survival of larger individuals (Moloney 1988). Together these results suggest that invasive populations are successful because of seed production and seedling survival, while the survival of adult individuals is more important in order to maintain stable and non-invasive populations. Thus the management of invasions should focus on limiting recruitment.
Clipping had no effect on A. cristatum tussock size in 2002, but reduced seed production. On the other hand, clipped tussocks decreased in size and set no seed in 2003 (Fig. 1). Furthermore, in contrast to unmanaged populations, in which dynamics varied little between years, clipped populations slowly increased in size when projected using data from 2002, but decreased in size when using data from 2003 (Fig. 3). This suggests that the negative effect of defoliation on tussock size and seed production in 2003 is a result of cumulative defoliation. This has been shown in earlier studies on other grass species (Knapp et al. 1999), but studies on A. cristatum have shown little or inconsistent cumulative effects of defoliation (Sneva 1973). Reduced root growth, as a consequence of repetitive defoliation (Caldwell et al. 1981; Richards 1984; Roundy et al. 1985), and thus an inability to take up sufficient nutrients and water, may explain the cumulative effect of clipping on A. cristatum.
Juvenile A. cristatum has been reported to withstand defoliation (McGinnies 1973), but our study and others (Salihi & Norton 1987) suggest that defoliation decreases juvenile survival, especially under dry conditions (Fig. 1).
Herbicide-treated tussocks produced few seeds and decreased in size in 2002. Several herbicide-treated tussocks still remained alive after 2 years; however, and some even produced seeds. This agrees with previous studies in which herbicide application decreased the cover of A. cristatum in the first year (Bakker et al. 1997), but that further applications were unsuccessful in eliminating the species (Romo, Fritz & Delanoy 1994; Wilson & Pärtel 2003). Therefore, if the treatment is not repeated, tussocks may recover the year following herbicide treatment.
Tussocks that survive herbicide treatment produce surprisingly large numbers of seeds (Bakker et al. 2003; Wilson & Pärtel 2003), as occurred in this study in both years (see Appendix S1 in the Supplementary material). Meristems within tussocks that survived herbicide application produced seed, suggesting that the effect of herbicide treatment is not cumulative as in the case of defoliation.
Our results also suggest that herbicide application decreases juvenile survival, especially under dry conditions but, as previous studies have shown that germination is unaffected by herbicide application (Ambrose & Wilson 2003), A. cristatum populations need to be treated for several years because of their persistent seed bank (Marlette & Anderson 1986).
does water availability affect the impact of management practices?
Unmanaged populations of A. cristatum had very high growth rates under wet and average water treatments because of high germination, juvenile survival and seed production (Fig. 4). Life-cycle stages associated with recruitment were, however, low in the dry treatment and the survival of larger tussocks was more important in determining growth rate. This agrees with previous studies, which found increased cover (Currie 1970; Bakker et al. 2003), seed production (Miller, Haferkamp & Angell 1990) and germination (Ambrose & Wilson 2003) of A. cristatum with increased water availability. This suggests that invasion rates may increase during wet years and that control efforts in order to prevent dispersal of large numbers of seeds will be especially important under those conditions.
λ for clipped A. cristatum populations was not significantly < 1 and did not vary significantly with water treatment in 2002 (Fig. 2), but λ was significantly < 1 in both wet and dry treatments in 2003. Several small tussocks died and others decreased in size in the dry treatment in 2003 (Fig. 1d). A decrease in tussock size has previously been observed after defoliation in combination with dry conditions (Roundy et al. 1985; Busso, Mueller & Richards 1989; Pyke 1990). Defoliated tussocks also recover more quickly under wet conditions than under dry (Cook, Stoddart & Kinsinger 1958). The fact that growing season precipitation was above average in 2002 but below average in 2003 may partly explain the difference in performance of clipped tussocks in 2002 and 2003. Our results suggest, however, that reduction in growth by defoliation can be achieved successfully under both wet and dry conditions.
Seed production was generally low for herbicide-treated tussocks. Seed production was, however, larger in the wet treatment in 2003, suggesting that control by herbicide application is less effective in wet years. This has not been observed previously, but is of great importance in management decisions. Wet conditions may require more frequent herbicide applications or a combination of several control methods.
Management can strongly affect the invasion of native grassland by A. cristatum. Management efficacy, however, is contingent on environment conditions.
Clipping and herbicide treatments are both effective in reducing growth and seed production, and thus invasion by A. cristatum, and are suitable for short-term control. Invasive species management is, however, a long-term project, regardless of the control method. Herbicide application is ineffective for long-term control because there is little cumulative effect and because population response varies with precipitation. Herbicides may also harm native species.
Clipping, on the other hand, more consistently reduces cover, seed production and thus invasion rates. As A. cristatum has an earlier growing season than most native species on the northern Great Plains, moderate repetitive defoliation by grazing during the spring before seed development is therefore probably the most effective long-term method for controlling invasion of A. cristatum.
We thank Sara Moar and Amanda Karst for field help, Pat Fargey and Adrian Sturch of Grasslands National Park for logistical support and assistance, Brett Sandercock at Kansas State University for help with MATLAB programming, Phil Hulme, Martin Köchy, Tommy Lennartsson and anonymous referees for comments on earlier versions of the paper, and the Natural Sciences and Engineering Research Council for support.