Integrating multiple disturbance aspects: management of an invasive thistle, Carduus nutans

Abstract

Background and Aims

Disturbances occur in most ecological systems, and play an important role in biological invasions. We delimit five key disturbance aspects: intensity, frequency, timing, duration and extent. Few studies address more than one of these aspects, yet interactions and interdependence between aspects may lead to complex outcomes.

Methods

In a two-cohort experimental study, we examined how multiple aspects (intensity, frequency and timing) of a mowing disturbance regime affect the survival, phenology, growth and reproduction of an invasive thistle Carduus nutans (musk thistle).

Key Results

Our results show that high intensity and late timing strongly delay flowering phenology and reduce plant survival, capitulum production and plant height. A significant interaction between intensity and timing further magnifies the main effects. Unexpectedly, high frequency alone did not effectively reduce reproduction. However, a study examining only frequency and intensity, and not timing, would have erroneously attributed the importance of timing to frequency.

Conclusions

We used management of an invasive species as an example to demonstrate the importance of a multiple-aspect disturbance framework. Failure to consider possible interactions, and the inherent interdependence of certain aspects, could result in misinterpretation and inappropriate management efforts. This framework can be broadly applied to improve our understanding of disturbance effects on individual responses, population dynamics and community composition.

INTRODUCTION

Disturbances are ubiquitous in both natural and managed ecosystems. While there are many related definitions of disturbance (Sousa, 1984; Grime, 2001), most focus on any relatively discrete event that disrupts ecosystem, community or population structure, and changes resources, substrate availability or the physical environment (White and Pickett, 1985). Despite various origins and forms (White, 1979; Sousa, 1984), disturbances can be characterized using the following five aspects: (1) intensity relates to the vigour of the perturbing force; (2) frequency addresses how often a disturbance event happens; (3) timing refers to when a disturbance happens relative to the life cycle of the disturbed organism; (4) duration describes how long a single disturbance event lasts; and (5) extent represents the spatial scale of the effects of a disturbance (Roxburgh et al., 2004; Shea et al., 2004; Miller et al., 2011). These aspects apply to both natural disturbances, such as hurricanes and fires, and to management actions, such as herbicide application and mowing, as all of these can be considered as disturbances (Buckley et al., 2007; Lockwood et al., 2007).

There is a rich literature on disturbance effects, for example effects on plant vegetative regeneration (Iwasa and Kudo, 1997; Simons et al., 1999; Klimesova et al., 2007; Peguero and Espelta, 2011), invasion success (DiTomaso, 1999; Kercher et al., 2007) and community structure (Haddad et al., 2008; Belote et al., 2009). However, many studies are limited to just one aspect, for example only the frequency of the disturbance (Shea et al., 2004). Despite a few recent case studies where multiple aspects are considered (McCabe and Gotelli, 2000; Laterra et al., 2006; Rinella and Hileman, 2009), relatively few studies explicitly address the interactions between aspects, and the implications of their possibly confounding natures. This can have serious repercussions for our understanding of the effect of disturbance on individuals, populations, communities and ecosystems. At the community level, recent theoretical work demonstrates that small changes in disturbance intensity can lead to extremely different disturbance–diversity relationships over the same range of frequencies (Miller et al., 2011). At the level of individuals, plant resprouting ability is less affected by frequency at low than at high intensity levels (Bellingham and Sparrow, 2000). Such interactions between aspects are probably common; aspects may also be inherently interdependent. For example, infrequent floods are often associated with large extent and long duration, while key timing is more likely to be included in disturbances with higher frequency or longer duration. In such cases, failure to consider multiple aspects may result in misinterpretation of results, attributing effects of one aspect to its confounding aspect.

We feel that conclusions from many disturbance studies could be strengthened with the use of a multiple-aspect framework. This framework should illuminate the relative importance of each main aspect and allow us to examine the potentially complex interactions between aspects (Collins et al., 1995; Shea et al., 2004). Furthermore, it should help tease out potential interdependence of aspects. Such a framework is useful both for examining responses to natural or anthropogenic disturbances, and for designing management approaches.

Here we apply a multiple-aspect disturbance framework to a two-cohort mowing study. Specifically, we examine how intensity, frequency and timing of mowing interact to bring about differences in growth, survivorship, reproduction and phenology of an invasive plant species. Thus we use a case study of individual plants' responses to disturbance to demonstrate the strengths of our multiple-aspect disturbance framework in interpreting results and guiding management practice.

METHODS

Study species

Carduus nutans L. (musk thistle: Asteraceae) is a common weed in disturbed habitats in Europe, western Siberia, Asia Minor and North Africa (Moore and Frankton, 1974). It has invaded pastures, rangelands, roadsides and disturbed areas in North America, South America, Southern Africa, New Zealand and Australia (Allen and Shea, 2006). Previous work suggests that a broad suite of insect herbivores reduce individual fitness and population growth in the native range (Jongejans et al., 2006). Infestations of C. nutans cause serious ecological and economic problems in pastures, roadsides and natural areas in the invaded range (Desrochers et al., 1988). Carduus nutans is a short-lived monocarpic herb, and can complete its life cycle as a summer annual, winter annual, biennial or short-lived perennial (Desrochers et al., 1988). Seeds germinate in favourable environments to form flat rosettes, which bolt and flower one or more years after germination, depending on vernalization and size (Shea et al., 2006). Plants flower from May to August and die after producing wind-dispersed seeds (Desrochers et al., 1988).

Study site

The experiment was repeated for two independent cohorts (2006–2007 and 2007–2008) on two adjacent areas in an old pasture at The Russell E. Larson Agricultural Research Farm at Rock Springs, Pennsylvania (latitude 40·711°, longitude –77·942°). The pasture was dominated by Arrhenatherum elatius, Dactylis glomerata, Elytrigia repens and Phleum pratense. Other common plant species included Taraxacum officinale, Linaria vulgaris, Plantago spp., Trifolium spp. and Galium spp.

The experimental site was prepared prior to planting each cohort, to mimic the disturbed areas in which the highest population densities are seen in both the native and invaded ranges. A Miller Offset Disk was applied twice to the area to kill above-ground vegetation. Then a Roller Harrow was applied to level the surface when the soil had dried.

Experimental design

Three-week-old seedlings were transplanted to the field in 12 rows at a spacing of 1·5 m in October 2006 and October 2007. To ensure that most of the thistles would bolt in the summer (and therefore would behave as winter annuals), we applied approx. 13 g of slow-release fertilizer (Osmocote Flower/Vegetable Food, N-P-K: 14-14-14) to each individual in the following April.

In the May following transplanting, rosettes were measured, blocked by size, and randomly assigned to one of 15 treatments (Table 1). Mowing is a commonly applied tool to control thistle invasion (DiTomaso et al., 2000). In our study, we manipulated the intensity, frequency and timing of the mowing treatments, because these are the most meaningful aspects for mowing disturbance events. Extent and duration were excluded as the former does not apply to a single plant, and a mowing event is an instantaneous or pulse disturbance (Bender et al., 1984) so manipulation of duration is not possible. Each thistle was hand clipped at different intensities (cut to 5 or 20 cm above ground), frequencies (never cut, cut once, twice or three times), and timings relative to when plants reached a height threshold of 40 cm (0, 2 or 4 weeks after reaching the threshold). Each treatment had 12 replicates, except for the control (never cut) which had 24 replicates to ensure the accuracy of the baseline for comparisons.

Table 1.

Treatment combinations of intensity, frequency and timing used in this study

Treatment	Intensity*	Frequency	Timing†	Replicates
Control	0	0	0	24
He	High	1	e	12
Le	Low	1	e	12
Hm	High	1	m	12
Lm	Low	1	m	12
Hl	High	1	l	12
Ll	Low	1	l	12
Hem	High	2	em	12
Lem	Low	2	em	12
Hel	High	2	el	12
Lel	Low	2	el	12
Hml	High	2	ml	12
Lml	Low	2	ml	12
Heml	High	3	eml	12
Leml	Low	3	eml	12

* For intensity, ‘High’ represents cutting at 5 cm above-ground and ‘Low’ represents cutting at 20 cm above ground.

^† For timing, ‘e’ stands for an early cut (immediately after plants reached 40 cm); ‘m’ stands for a middle cut (2 weeks after plants reached 40 cm); and ‘l’ stands for a late cut (4 weeks after plants reached 40 cm).

Treatments were initiated as soon as individual plants reached the height of 40 cm (when they had definitively commenced bolting). Surrounding vegetation in a 50 × 50 cm area was clipped to the same height as the thistles. Capitula were bagged using pollen bags before seed dispersal to prevent new infestations.

Stem number, plant height and key developmental dates, such as flowering date, were recorded in weekly field censuses from May until the end of the experiment. The experiments were terminated by removing all the above-ground biomass in early November, after the first frost in the second year of each cohort, when all plants had finished their life cycles. Capitula removed at each mowing event, and capitula from the final destructive census, were separated based on their developmental stages (especially whether they had set seed), and were then counted in the laboratory.

Data analysis

We conducted our data analyses in R (R Development Core Team, 2008). Because C. nutans is monocarpic (plants died after reproducing), we define survival as the probability of surviving until reproduction. Plant reproduction, measured by capitulum production, is a direct measurement of plant fitness in this study. The capitulum production considered in this study included reproduction both before and after the treatments, as plants receiving late treatments produced mature capitula before the initiation of the treatments. We also analysed flowering date, number of flowering stems and plant height at flowering, as they either are indirectly related to fitness (e.g. through interactions with pollinators or seed dispersal) or can help to explain different fitness results. Flowering date was analysed using a non-parametric survival model (coxph function in the survival package) with cohort as a stratification variable (strata function). Plant survival, reproduction, height at flowering and number of flowering stems were analysed using Generalized Linear Mixed Models (GLMMs; Crawley, 2007) in R (lmer function in the lme4 package). Log linear transformation was used for stem number and reproduction prior to fitting the models. Binomial error distribution was specified for survival. Initial rosette size was used as a covariate in all models because phenology and reproduction are strongly size dependent (Shea et al., 2006). We also considered between-year variation and spatial location as random effects in the models.

In our study, frequency and timing are inherently dependent on each other (i.e. high frequency treatments are more likely to include a late cut). To address this, we analysed the effects of intensity and timing, and the effects of intensity and frequency, separately. We then compared Akaike information criteria (AIC) for the minimum adequate models for intensity and timing (specifically, with or without a late cut) with the comparable models for intensity and frequency, when we analysed survival, plant height and reproduction. We also compared the R² values in the models for flowering phenology. Based on the results from the AIC and R² comparisons (specifically, that late timing was more important than frequency, see the Results below), we examined the effects of intensity and frequency only for the treatments that included a late cut, in order to control for the strong effect of late timing. We then compared these results with the results when late timing is ignored.

RESULTS

Intensity and timing

High intensity and late timing together significantly delayed flowering (P < 0·001), leading to delays of up to 7 weeks. Intensity and timing both had significant effects on survival, plant height and reproduction – treatments with high intensity or late timing caused larger reductions in the responses (Fig. 1). The interaction between intensity and timing had significant effects on reproduction, indicating that the effects of late timing were further magnified by high intensity.

Fig. 1.

Effects of intensity and timing on the survival (A), plant height at flowering (B) and reproduction (C) of C. nutans. Treatments on the x-axes include control, treatment with high intensity but without late timing (HI), treatment with high intensity and late timing (HI + Late), treatment with low intensity but without late timing (LI) and treatment with low intensity and late timing (LI + late). The letters denote the statistical difference between treatments at α = 0·05 (z-test based on lmer).

Intensity and frequency

Treatments with high intensity caused significantly longer delays in flowering, and larger reductions in survival, plant height and reproduction than treatments with low intensity (P < 0·001, Fig. 2). However, the effects of frequency depended on whether late timing was considered, especially for treatments with high intensity (Fig. 2). For example, increased frequency did not further delay flowering when not controlling for late timing, whereas it significantly delayed flowering when late timing was considered. Similarly, frequency appeared to have no significant effect on survival at a high intensity level when we did not control for timing (Fig. 2D). However, when timing was considered, high frequency treatments actually led to higher survival at a high intensity level (Fig. 2A, B). Furthermore, although frequency appeared to have a significant negative effect on reproduction in treatments with high intensity (Fig. 2E), this effect faded away when we accounted for the interdependent timing effect (Fig. 2C).

Fig. 2.

Effects of intensity and frequency on the survival (A, D), flowering plant height (B, E) and reproduction (C, F) of C. nutans depend on whether the statistical model accounts for timing (A, B, C) or not (D, E, F). (A–C) present results based on only treatments that included a late cut; (D–F) present results of all treatments in the study (the full data set). The effect of frequency at each intensity level is indicated to the right of the curves. ‘ns’ denotes no significant effect. ‘ + ’ denotes a significant positive effect. ‘–’ denotes a significant negative effect. α = 0·05, z-test based on lmer.

Flowering stem number was positively correlated with reproduction (Fig. 3A). While high intensity reduced the number of induced flowering stems, high frequency significantly increased the number of flowering stems (Fig. 3B).

Fig. 3.

The relationships between number of flowering stems and capitulum production (A), and between number of flowering stem and frequency (B). The number of flowering stems is a significant predictor of capitulum production (P < 0·001). The number of flowering stems significantly increases with frequency at each intensity level (P < 0·001), suggesting that high frequency treatment plants compensate for lost reproduction by increasing stem numbers.

Timing or frequency

In all cases, AIC values were lower (lower AICs indicate better fits) for models that included timing instead of frequency (Table 2). The R² values for non-parametric models for flowering date were an order of magnitude higher (higher R² values indicate better fit) in models with intensity and timing instead of intensity and frequency (0·438 vs. 0·047). This suggested that models which included timing instead of frequency fit better for all these responses, and the frequency results in the full data set were mostly driven by the effect of late timing.

Table 2.

Akaike information criteria (AIC) values for the minimum adequate linear mixed models fitted with intensity and timing (i.e. with or without a late cut) and comparable models fitted with intensity and frequency

Responses*	Factors
	Timing and intensity	Frequency and intensity
Survival†	267	282
Plant height	2120	2191
Capitulum production‡	762	777

* Initial rosette size was used as a covariate in all models. Cohort and spatial location of individual plants were considered as random effects.

^† Binomial error distribution was specified for the survival data.

^‡ Results were based on log-transformed data. The minimum adequate models include the interaction between timing and intensity or between frequency and intensity.

DISCUSSION

Using a multiple aspect framework for disturbance, we manipulated all three relevant aspects in our system – intensity, frequency and timing of management – and examined their effects on the fitness of the invasive plant, C. nutans. Our results show that mowing treatments with higher intensity or late timing caused larger reductions in plant fitness, which broadly agrees with previous studies which addressed the importance of high intensity and relatively late timing in reducing the regrowth of monocarpic species (Lennartsson et al., 1998; Martinkova et al., 2004; Pyšek et al., 2007; Peguero and Espelta, 2011). This is probably because these treatments removed more biomass and imposed harsher constraints (e.g. lack of inactive meristems; Huhta et al., 2000) on the post-injury plant recovery.

Compared with previous investigations that only manipulated a single aspect of disturbance, our experimental design allowed us to examine the system more systematically. We started with all five disturbance aspects and deliberately eliminated two aspects (duration and extent) that were not applicable to our system. Therefore, all three possible disturbance aspects are covered in our study. Our results demonstrate two distinctive advantages of the multiple-aspect framework. First, it enables us to examine the interactions between different aspects of a disturbance. Our results show that multiple aspects of a disturbance can interact with each other to produce complex outcomes. For example, higher intensity and late timing significantly interacted with each other to magnify decreases in capitulum production. On the other hand, the positive effects of high frequency on survival and plant height were only significant at the low intensity level. This indicates that the negative effect of high intensity on plant height was so strong that it masked the potential positive effect of high frequency (Fig. 2B). Thus, considering these interactions provides a more thorough understanding of the disturbances.

Secondly, our multiple-aspect experimental design allows us to clarify the interdependence between disturbance aspects. In particular, our finding that timing has more explanatory power than frequency was only possible due to our factorial, multiple-aspect design. Our results thus imply that high frequency mowing may in fact take effect through late timing, because high frequency treatments are more likely than low frequency treatments to include key timings relative to the life history of the plants. For C. nutans, a late timing is crucial because late mowing probably depletes meristem and resource pools that are both critical for plant recovery. In our study, we controlled for the strong effect of timing when analysing the effect of frequency, and thereby addressed the distinction between the two (Fig. 2A–C). However, if we had only considered frequency without addressing timing (Fig. 2D–F), we would have attributed the impact of late timing to high frequency mowing. Such limitations, as seen in many single-aspect studies, may not only conceal significantly positive effects (e.g. the positive effect of high frequency on survival at higher intensity and on plant height at lower intensity; Fig. 2D, E), but may also lead to false-negative results (e.g. the false-negative effect of high frequency on plant height and reproduction at higher intensity; Fig. 2E, F).

The confounding of disturbance aspects is likely to lead to potentially incomplete conclusions or wasteful management decisions unless both interdependent aspects are evaluated. For example, our work suggests that high frequency mowing treatments (requiring several individual mowing events throughout the growing season) can be replaced by fewer, well-timed mowing events. Only after carefully controlling for the overwhelming timing effect were we able to notice the ineffectiveness of high frequency in reducing plant fitness (Fig. 2A–C) and further explore the underlying mechanism. This result is further verified in a study where this species managed to survive ten consecutive weekly cuts at 5 cm (Zhang et al., 2011). The ineffectiveness of high frequency treatments is counterintuitive, and in contrast to previous studies which emphasized high frequency in weed management (Peters and Lowance, 1978; Sullivan, 2004; Nielsen et al., 2006). This is likely to be due to higher numbers of flower stems (Fig. 3A) induced by frequent removal of apical meristems in these treatments, which was highly correlated with capitulum production (Fig. 3B). In these high frequency treatments, the first cut happened before plants invested substantially in growth and reproduction, and therefore failed to remove much plant biomass or deplete meristem pools. Furthermore, the time intervals between two sequential cuts in the high frequency treatments were too short for plants to regrow significantly; thus subsequent cuts also failed to cause major damage to the plants. In our case, failure to address timing when examining frequency may lead to the wrong conclusion that high mowing frequency is beneficial to control of musk thistle. Consequently, management decisions simply to increase frequency will be inefficient and wasteful in terms of both time and labour.

Although aspects of a disturbance may be independent of each other (e.g. spring grazing every 2 years vs. summer grazing every 3 years), often this is not the case. Potential interdependence in disturbance aspects exists in that three (frequency, timing and duration) out of the five aspects are inherently intertwined because all three are temporal in nature. For example, disturbances of high frequency or long duration are prone to include key timings as well. This is especially true when disturbance return intervals are shorter than plant life cycles or seasonal cycles (e.g. multiple disturbance events happen at different times within a single growing season; Hogg et al., 1995; Nielsen et al., 2006; Schooler et al., 2008; Gao et al., 2009). In such cases, it is impossible to increase frequency without introducing effects on timing or duration. Moreover, disturbance aspects such as intensity and frequency may be biologically interdependent. Considering these interdependent aspects will help us to understand natural disturbances where correlations are difficult to avoid. For example, infrequent wild fires are often more intense because of fuel load accumulation (Keith et al., 2010). In this case, if we want to evaluate community responses to fire regimes with different fire return intervals (frequency), we need to account for fire temperature (intensity) carefully. Fire events with similar temperatures can be sub-sampled from the whole data set, and analyses on frequency based on these data can therefore provide unbiased conclusions.

Here we use mowing management of an invasive species as a demonstration of the utility of our theoretical framework for multiple aspects of a disturbance. Although our study was conducted in a relatively simple system, our results have broad implications for examining disturbance in more complex systems and on various levels (e.g. population dynamics, community composition and ecosystem processes): interactions between aspects can produce synergistic outcomes, some aspects are interdependent, and these facts may lead to misinterpretation if they are not distinguished. We suggest that researchers initially consider all five possible aspects (frequency, intensity, duration, extent and timing) in their systems. In cases where some disturbance aspects are impossible or difficult to manipulate (e.g. hurricanes, flooding or fire), recording such information for use as covariates in data analysis is highly desirable. In conclusion, our results show that considering multiple aspects can provide a more thorough understanding of disturbance. This framework can be used to improve the design of manipulative experiments for basic research and for management purposes, and to provide a sound basis for meaningful interpretation of disturbance outcomes.