Abscission of seeds in some plant species occurs as a result of strong wind gusts that exert sufficient drag forces to liberate the seed from its parent. This may be an adaptive feature, as release into stronger wind gusts has been shown to lead to greater dispersal distances, which is likely to have evolutionary advantages.
We test the hypotheses that (i) seeds released into upward wind gusts will, on average, travel further than those seeds released into wind gusts with horizontal or downward orientations and (ii) that the preferential abscission of seeds into upward wind gusts will result in the dispersal of seeds over greater distances.
As a case study, we studied the abscission dynamics of Conyza bonariensis (L.) Cronquist (fleabane), which is an important weed with global distribution. Using abscission data obtained through a series of seed release experiments, we confirm that abscission of seeds in C. bonariensis is most likely to occur during strong and upward wind gusts.
We demonstrate that, for a given wind speed, seeds released into upward wind gusts will, on average, travel further than those seeds released into wind gusts with horizontal or downward orientations. We also show that preferential release into upward wind gusts has some influence on the distance travelled, but the strength of this influence is dependent on the correlation between wind orientation and wind speed. For this particular study, the sensitivity of release to the horizontal wind speed seems to have a larger effect on the distance travelled than sensitivity to wind orientation.
Abscission of seeds in some trees and herbaceous species occurs in response to specific environmental conditions, resulting in a non-random timing of seed release (e.g. Greene & Johnson 1992; Greene 2005; Skarpaas, Auhl & Shea 2006; Soons & Bullock 2008; Greene & Quesada 2011). This non-random release results from the formation of abscission layers that are more likely to break during strong wind gusts and when the orientation of the wind is such that the drag force exerted by the wind pulls the seed away from the plant, rather than pushing it back against its supporting structure (Greene 2005; Greene & Quesada 2011). As a consequence of this non-random release, seeds are more likely to experience turbulent wind conditions as they undergo wind-assisted dispersal and are therefore likely to experience upward drafts over the course of their journey, leading to dispersal over greater distances than if released at random (Greene & Johnson 1992; Schippers & Jongejans 2005; Bohrer et al. 2008; Soons & Bullock 2008).
Previous research into the dynamics of seed abscission and its effects on wind-assisted dispersal have often focused on horizontal wind speed and the existence of thresholds that horizontal wind speed must exceed before abscission will take place (e.g. Soons et al. 2004; Schippers & Jongejans 2005; Bohrer et al. 2008). However, the orientation of wind gusts has been shown to also influence the release of seeds, with release preferentially occurring into upward wind gusts (Greene & Johnson 1992; Greene & Quesada 2011; Borger et al. 2012). Moreover, Skarpaas, Auhl & Shea (2006) demonstrated that slower turbulent winds are more effective at releasing seeds than faster laminar flow. While this result may be attributed to the occurrence of turbulent wind gusts whose speed exceeded that of the laminar flow (Skarpaas, Auhl & Shea 2006), it may also reflect the fact that wind gusts occurred with differing vertical orientation to the flower head during turbulent, but not laminar flow.
In addition to the strength and the orientation of turbulent wind gusts, abscission dynamics are also affected by the time since flowering. Over time, the abscission layer between a seed and its parent weakens, resulting in a decrease to any wind speed threshold affecting the seed's release. In one simulation of abscission dynamics, Schippers & Jongejans (2005) used a simple linear function to describe the relationship between the time since flowering and the horizontal wind speed required to cause abscission. They demonstrated that the rate at which the abscission layer decayed, and the release threshold decreased, could affect subsequent wind-assisted dispersal, particularly the tail of the resulting dispersal distribution. This suggests that accurate modelling of the effects of abscission dynamics on wind-assisted dispersal will need to take the time since flowering into account.
In this paper, we extend previous work investigating the manner in which abscission dynamics can effect dispersal in our case study, Conyza bonariensis (L.) Cronquist (fleabane), which is an important weed with global distribution (Powles 2008; Wu 2009). The seeds of C. bonariensis are individually contained in small, hard achenes equipped with a pappus, and are predominantly spread through wind-assisted dispersal. The abscission of seeds in C. bonariensis is dependent not only on the speed of wind gusts, but also on their vertical orientation relative to the flower head as well as the time since flowering (Borger et al. 2012).
In this study, we test the hypothesis that (i) seeds released into upward wind gusts will, on average, travel further than those seeds released into wind gusts with horizontal or downward orientations and (ii) the preferential release of seeds into upward wind gusts results in dispersal occurring over greater distances than if seed release depended only on horizontal wind speed. To test these hypotheses, we fitted a statistical model to C. bonariensis release data obtained through seed release experiments, describing the manner in which wind speed, wind orientation and seed age influence the proportion of seeds released. These statistical models of seed release were then coupled with a previously validated model of wind-assisted dispersal, allowing us to determine the effects of C. bonariensis abscission dynamics on the distance travelled by wind-dispersed seeds.
Materials and methods
Seed release experiments
Seed release experiments are fully discussed in (Borger et al. 2012). Conyza bonariensis plants were maintained in a glasshouse, hand watered and fertilized as necessary, and protected from wind. Plants were monitored daily and individual seed heads were labelled with the date on which the head opened. Directly following the removal from the plant, seed heads were exposed to air speeds of 0 to 70 m s−1 (increased in increments of 5 m s−1 every ten seconds) until reaching the maximum speed or until all seeds were removed from the head. The per cent of seeds lost from the seed head at each increment in air pressure was visually estimated. Age and orientation of the seed heads were varied. Seed heads of all ages from zero to ten days old were tested, where day zero was the day in which the head opened (11 different ages in total). Orientation of the seed head was varied in relation to the wind direction. Seed heads are approximately hemispherical, and heads were clamped in a position to allow the air to hit the head at one of five angles: −90°, −45°, 0°, 45° and 90° degrees. At −90°, air was directed upwards towards the base (stem) of the seed head, and at 90°, air was directed downwards towards the top of the seed head. An orientation of 0° corresponds to a perfectly horizontal wind. Each combination of seed head age and orientation was replicated three times (total of 11 × 5 × 3 = 165 seed heads). Replicates were taken from different plants.
We fitted three generalized additive models (GAMs) and three equivalent generalized linear models (GLMs) to the release data obtained from the seed release experiments. For each of these models, we assumed a binomial distribution of errors and GAMs were fitted using spline smoothers. For each model, we specified a model structure, whereby the cumulative proportion of seeds released was given as a function of the number of days since the flower opened and (i) both wind orientation and wind speed, (ii) wind orientation but not wind speed and (iii) wind speed but not wind orientation. In fitting the GAMs, we set the basis dimension for the spline smoothers to k = 4. If k was allowed to vary, the models became over-fitted and predictions of the cumulative proportion of seeds released were not monotonic. As shown in Table 1, the fitted GAMs have a lower AIC than the fitted GLMs and can therefore be considered to provide a better description of the data. The fitted GAMs also have a lower AIC than the GLM used for statistical testing in Borger et al. (2012), even though that model contained additional terms besides the wind speed, wind orientation and time since flowering that we were interested in for this study. Therefore, the three GAMs were used for the remainder of this study to simulate the abscission of seeds in response to wind gusts as described in below.
Table 1. Comparison of the generalized additive models (GAMs) used in this paper with equivalent generalized linear models (GLMs) and the model used in Borger et al. (2012) fitted to data collected from seed release experiments. The GAMs and GLMs have equivalent model structures whereby the cumulative proportion of seeds released were given as a function of the number of days since the flower opened and (i) both wind orientation and wind speed, (ii) wind orientation but not wind speed, and (iii) wind speed but not wind orientation
Deviance explained (%)
Borger et al.
We chose to fit three models with differing structure to test how the dependence of abscission on wind speed and wind orientation influenced subsequent dispersal. Both wind speed and wind orientation have previously been shown to influence abscission, but we wanted to address two new questions: (1) Does release into more upward wind gusts lead to greater dispersal, and (2) does the fact that the probability of a seed release increases with wind speed and a more upward wind orientation (Borger et al. 2012) result in greater dispersal than if the probability of seed release did not depend on wind speed or orientation. Conducting simulations based on fitted model (i) enabled us to test how wind speed and orientation influence dispersal given their observed influence on seed release, thus enabling us to address our first question. However, by fitting the three separate models and conducting dispersal simulations with each of these models, we were also able to determine how including or not including a dependency on wind speed or wind orientation (as a proxy for evolution or no evolution of dependence on wind speed and/or orientation) impacted wind-assisted dispersal, thus addressing our second question.
Modelling wind-assisted dispersal
To model the wind-assisted dispersal of C. bonariensis seeds, we employed the atmospheric stability correction (ASC) version of the Lagrangian stochastic model presented in Soons et al. (2004) with the modifications described in Savage et al. (2011). This model takes as input a number of variables describing the properties of the seed being dispersed, the vegetation over which dispersal occurs and the meteorological conditions that drive wind-assisted dispersal. The model simulates the dispersal of individual seeds by considering a time-averaged horizontal wind speed and generating a series of autocorrelated random fluctuations about this average, representing the turbulent nature of atmospheric airflow. These fluctuations, together with the height of release and the terminal velocity of the seed being modelled, are used to simulate the trajectory of an airborne seed up until the point at which it is deposited on the ground. By simulating large numbers of individual seeds, the model can be used to build up a realistic representation of wind-assisted dispersal (Soons et al. 2004; Soons & Bullock 2008).
For the purposes of this paper, we parameterized the ASC Lagrangian stochastic model to simulate the dispersal of C. bonariensis over agricultural land in the grain belt of Western Australia, where it is an emerging agricultural pest. Typically, vegetation within fields is scant during the summer/autumn fallow period, when the seeds of C. bonariensis are dispersed and consist mainly of scattered individual weeds such as C. bonariensis, with the only other ground cover consisting of sparse stubble from the previous seasons' crop. Therefore, we parameterized the model using a vegetation height of 0·4 m, but a leaf area index of 1·0 × 10−5 and a deposition height of 0·01 m. Seeds of C. bonariensis were assumed to have a terminal velocity of 0·3 m s−1 (Andersen 1992) and were released from a height of 1·0 m. The height of individual C. bonariensis plants can vary significantly depending on the availability of water; however, a height of ~1·0 m is not uncommon and therefore represents a reasonable value for this particular study.
Meteorological inputs into the model of wind-assisted dispersal used in this study include friction velocity, temperature, sensible heat flux and a time-averaged horizontal wind speed. As we wished to include the effect of the time since flowering on abscission dynamics, time series data for each of these variables were required. These data were obtained for Merredin, Western Australia (a typical agricultural region in the Western Australian grain belt) through simulation using a well-established, validated meteorological model (Hurley, Hill & Blockley 2005). This model takes large-scale synoptic data as input (provided with the model, http://www.cmar.csiro.au/research/tapm) and simulates meteorological conditions at each point on a grid covering a specified area of the landscape. For the purposes of this paper, we simulated meteorological conditions over March–April 2007 using a grid centred on Merredin (−31·3° S, 118·2° E). We considered the hourly averaged values for friction velocity, temperature, sensible heat flux and horizontal wind speed at the central point in the grid for each hour over the period 20 March–10 April 2007, as a large proportion of C. bonariensis seeds are released during autumn in the Merredin district.
Modelling of wind-assisted dispersal for this investigation consists of three steps: simulation of wind gusts acting on the flower head, determining the number of seeds released into each gust and simulating the wind-assisted dispersal of each seed.
Using the Lagrangian stochastic model, we simulated wind gusts acting on an individual flower head that was assumed to have opened at 12:00 am on the 20 March 2007. This was achieved by taking the hourly estimates of meteorological inputs and simulating wind gusts using the Lagrangian stochastic model assuming a stationary seed positioned at a height of 1·0 m above the ground. We generated wind gusts occurring over each hour for a twenty-day period beginning at 12:00 am on the 20 March 2007. As the Lagrangian model involves a dynamic time step (dependent on the timescale of turbulent gusts, see Soons et al. 2004), the number of gusts simulated may differ from hour to hour. However, the time at which each wind gust occurred can be easily calculated, and this is used as an input into the statistical models. For each gust, the orientation, speed and the time since flowering were used as input into the three fitted statistical models, giving the cumulative proportion of seeds that would be expected to be released into that gust given the particular structure of each model. The actual number of seeds released at a given time was calculated as
where Nr is the number of seeds released, Nt is the total number of seeds on the flower head, Nc is the number of seeds that have already been released, and p is the cumulative proportion of seeds predicted to be released by the fitted models at a given time. The calculated number of seeds released was only used if Nr was positive; at many time points, wind conditions were such that pNt was less than Nc, and thus, Nr was negative, in which case, no seeds were released. Note that although this model allows many seeds to be released during each time step, this does not imply that C. bonariensis undergoes clumped dispersal, where multiple seeds stick together and are dispersed as a single unit. Each seed released is assumed to disperse independently of any other seeds released into the same gust of wind.
Each seed released was assumed to have an initial velocity given by the wind gust that resulted in its abscission from the plant. This initial velocity was then input into the Lagrangian model, and the subsequent wind-assisted dispersal was simulated. For each of the three statistical models, we simulated release and subsequent dispersal over a period of twenty days. We assumed that the total number of seeds to be released was 10 000, and each simulation was repeated 300 times (because the underlying model is stochastic), resulting in a total of 3 000 000 seeds being simulated for each model. This number of seeds would be representative of the seeds produced by 5–10 plants over the course of a season (Kempen & Graf 1981; Wu et al. 2007). Note that for each of the three release models, the vast majority of seeds had been released by the end of the simulation (more than 99·86% in each case). For subsequent analyses, we did all testing based on proportions of all seeds released, but also checked against proportions of total seeds (released and unreleased); as the numbers of unreleased seeds were so small, results were the same in both cases.
To test the influence of the initial wind speed and orientation on the distance travelled by dispersing seeds, we fitted a linear model to the simulated dispersal of seeds released under model (i). This linear model described the distance travelled as a function of the wind speed, and wind orientation at the time of release, and the interaction between these two variables. To test for differences between dispersal resulting from release using each of the three statistical models, we compared the resulting distributions of distances travelled using a Kolmogorov–Smirnov test and a Mann–Whitney–Wilcoxon test. We also used a series of proportion tests to compare the proportion of seeds dispersing over a given distance x, where x ∈ [0, 250].
In fitting our statistical model (i) to release data obtained from seed release experiments, the time since flowering, wind speed and wind orientation were all significant (P < 0·001 for each). For model (ii), the time since flowering and wind orientation were significant, and for model (iii), the time since flowering and wind speed were significant. Figure 1 provides an example of the outputs from these models, giving the cumulative proportion of seeds predicted by model (i) to be released as the time since flowering increases, using different fixed values for wind speed and orientation. Seeds can be seen to be more readily released during faster winds, or when the orientation is negative, which corresponds to an upward wind gust. From this figure, it appears that wind speed has a greater influence on abscission dynamics than wind orientation.
Effects of abscission dynamics on wind-assisted dispersal
Both wind orientation and wind speed were significant in determining the distance travelled, as was the interaction between these two terms (Table 2). The values of the coefficients in this model show that upward wind gusts (negative orientation) increase the distance travelled, as do faster wind gusts. However, the interaction between wind orientation and wind speed is such that as wind speed increases, the relative influence of wind orientation becomes weaker. From these values, we see that if the wind gust that results in abscission is exactly horizontal, each extra m/s of wind speed would, on average, lead to an extra 2·07 m of distance being covered in the subsequent dispersal. If the wind speed is 5 m s−1, then a wind orientation of −45° (upward) would, on average, lead to an increase in dispersal distance of ~15 m compared with a wind orientation of 45° (downward). However, if the wind speed is increased to 8 m s−1, the relative influence of wind orientation is reduced, and changing the orientation from 45° to −45° would only result in an increase of ~10 m.
Table 2. A linear model fitted to the distance travelled by dispersing seeds as a function of wind speed, wind orientation and wind speed × wind orientation
Speed x orientation
The proportions of seeds dispersing to various distances declined quickly as distance increased and varied depending on which of the three statistical models were used to simulate release (Fig. 2). Note that for each of the three release models, a small proportion of seeds still remained on the plant after the twenty days of simulation; 0·000097, 0·0014 and 0·00015 for model (i), model (ii) and model (iii), respectively. As the seed head ages, and sensitivity to speed and orientation declines, we may expect that these seeds would eventually be released into more unfavourable winds and thus disperse less far than previous seeds; this would only add to the observed differences between the models. However, as noted in our methods section, the fact that these proportions were so small meant that including or excluding them from further analysis made negligible differences to our results.
Results of the Kolmogorov–Smirnov test indicate that there is significant difference between the overall distribution of distances travelled resulting from release according to each of the three statistical models, and results from the Mann–Whitney–Wilcoxon test indicate that there is a significant difference in the median distance travelled (Tables 3, 4). Figure 3 shows that using different statistical release models results in significant differences in the proportion of seeds travelling farther than a given distance. Ignoring sensitivity to wind speed results in a dramatic decrease in the proportion of seeds travelling farther than all distances shown, while ignoring sensitivity to wind orientation results in a significant decrease in the number of seeds travelling farther than ~120 m or less (this difference appears to continue for greater distances, but the differences in proportions are too small to be significant for the number of seeds simulated).
Table 3. Number of seeds and 50th, 90th, 95th and 99·9th percentiles of distances travelled (metres) by dispersing seeds under the three statistical release models
3 000 000
3 000 000
3 000 000
Table 4. Results of statistical comparisons between the distances travelled by dispersing seeds under the three statistical release models. Note that n = 3 000 000 seeds in all cases
Model (i) vs. model (ii)
Model (i) vs. model (iii)
Model (ii) vs. model (iii)
Model (i) vs. model (ii)
Model (i) vs. model (iii)
Model (ii) vs. model (iii)
When the speed and orientation of the actual wind gusts that resulted in the release of seeds for each of the three release models are considered (Fig. 4), it is evident that faster wind speeds were generally of a more downward orientation, and more upward gusts are generally slower. If wind orientation is ignored (model (iii)), less seeds are released into upward gusts. In contrast, if wind speed is ignored (model (ii)), a far greater numbers of seeds are released into upward wind gusts. Because these upward wind gusts tend to be relatively slow-moving, however, preferential release into these gusts appears to result in higher numbers of seeds being deposited at shorter distances than preferential release into faster moving but more horizontal wind gusts (Fig. 2).
The results of this study demonstrate that abscission dynamics in C. bonariensis are affected by both wind orientation and wind speed as well as the time since flowering and that the orientation and speed of the initial wind gust resulting in abscission have a significant influence on the distance travelled. For a given wind speed, seeds released into upward wind gusts will, on average, travel farther than those seeds released into wind gusts with horizontal or downward orientations. Similarly, for a given wind orientation, seeds released into faster gusts will travel farther than those released into slower gusts. Both abscission and subsequent dispersal appear to be more strongly influenced by the initial wind speed than the initial wind orientation. However, ignoring the influence of wind orientation on abscission dynamics does influence the distance travelled by seeds during subsequent wind-assisted dispersal (Tables 3 and 4, Fig. 3). For our particular case study, the dependence of abscission on wind orientation and the dependence of abscission on wind speed both confer an advantage in terms of distance travelled, as dispersing seeds would be expected to travel shorter distance if either of these dependencies did not exist (Fig. 3).
Previous studies have shown that the preferential abscission of seeds only into wind gusts with speeds above a particular threshold can significantly influence the distance travelled by these seeds as they disperse (Greene 2005; Schippers & Jongejans 2005; Bohrer et al. 2008; Soons & Bullock 2008). Our results agree with these findings, but in addition, demonstrate that wind orientation also plays a significant role in abscission dynamics and that the preferential release of seeds based on wind orientation also increases the distance travelled. If abscission was to occur with no preferential release into fast moving and upward wind gusts, the proportion of seeds travelling father than a given distance would decrease (Fig. 3). Over a large number of dispersal events, this would result in a significantly lower rate of spread and would therefore be particularly important when considering the movement of species such as C. bonariensis into previously uninhabited areas.
As noted by Pazos et al. (2013), simulations of seed release and subsequent dispersal performed in a number of previous studies have tended to assume discrete, binary behaviour for the release of seeds into particular wind gusts. That is, wind gusts above a particular wind speed are assumed to result in release, while wind gusts below this threshold do not. Moreover, previous studies incorporating the ripening of seeds over time and the subsequent weakening of the abscission layer have assumed a linear relationship between the time since flowering and the threshold value (e.g. Schippers & Jongejans (2005)) and do not take into account the cumulative effects of wind gusts exerting stress on the vascular bundle (see Pazos et al. 2013). In this study, we have improved on this situation by using a statistical model fitted to release data, with no prior assumptions as to the relationship between wind speed, wind orientation, time since flowering and the proportion of seeds released other than a binomial distribution of errors. As shown in Fig. 1, the resulting statistical model describes a nonlinear relationship between the time since flowering, wind speed, wind orientation and the proportion of seeds released and indicates that the use of simple release thresholds does not capture the underlying relationship between wind speed and release. Therefore, we suggest that our model provides a more realistic representation of abscission that has been previously used to study the influence of abscission dynamics on wind-assisted dispersal.
In addition to the factors affecting release considered in this study, other factors may also play an important role in determining abscission dynamics. As noted by Greene & Quesada (2011), a unidirectional air flow is relatively inefficient at removing seeds, and by rotating the stem in release experiments, so that the airflow hit all sides of the seed head, we would expect that more seeds would be released. Clearly, this would have some influence on our fitted models. These experiments could also be improved by applying bursts of air, rather than a continuous stream. Abscission of seeds appears to occur during the start of a wind gust (Greene 2005), and so using short bursts of air may also result in larger numbers of seeds being removed. Moreover, short bursts of air may result in greater movement of the achene within its sheath (as force is applied and then relaxed), which would also influence the rate at which seeds are removed (Pazos et al. 2013). It would be interesting to consider these factors in further studies on the relationships between abscission dynamics and subsequent dispersal.
The distances travelled by C. bonariensis seeds in the presented simulations roughly agree with observed seed dispersal in the related C. canadensis (release height 1·12 m, terminal velocity 0·323 ms−1, see Dauer, Mortensen & Humston 2006; Dauer, Mortensen & Vangessel 2007), where seeds were regularly found to disperse over 500 m (Dauer, Mortensen & Humston 2006). Previous attempts to model the dispersal of C. canadensis have tended to under-predict deposition over the tail of the resulting dispersal kernel (Dauer, Mortensen & Humston 2006; Dauer, Mortensen & Vangessel 2007), and this has been largely attributed to the inability of the models used to adequately represent dynamic, turbulent airflow (Dauer, Mortensen & Humston 2006; Dauer, Mortensen & Vangessel 2007). However, results presented in this paper clearly suggest that abscission dynamics in C. bonariensis affect the distance travelled by dispersing seeds, and a similar result is likely to hold for C. canadensis. Therefore, the under-prediction of dispersal distances in previous models of C. canadensis may have also resulted from the omission of abscission dynamics, as well as the particular representations of turbulent air flow used for modelling.
In many parts of the world, C. bonariensis is an important weed both in urban and in agricultural environments. As illustrated in this research, the plant has evolved a mechanism of seed release that acts to maximize the distance travelled by dispersing seeds, and wind-assisted dispersal can be expected to regularly occur over distances >500 m. C. bonariensis is therefore capable of rapid spread into previously uninhabited areas, and given that each plant is capable of producing over 300 000 seeds (Kempen & Graf 1981; Wu et al. 2007), even small numbers of individuals introduced into an area are likely to result in wide spread infestation. This research helps explain the ease with which C. canadensis spread over hundreds of kilometres of agricultural land in California in just a few years (Powles 2008). Further, the genetically diverse Conyza genus is highly prone to develop resistance to herbicides (Powles 2008), and the genetic elements underlying this resistance are likely to be distributed faster, and over far wider areas than in other important resistance-prone weeds with relatively low seed dispersal, such as grasses. Developing effective resistance management strategies in wind-dispersed weeds such as Conyza will therefore require understanding and prediction of seed dispersal, and this will in turn depend on abscission dynamics, as this study shows.
This study has considered the abscission dynamics and subsequent dispersal of C. bonariensis seeds in the agricultural region of Merredin, Western Australia. As shown in Fig. 4 (column 3), wind conditions at Merredin over the selected simulation period exhibit a tendency for upward wind gusts to occur most frequently with low wind speeds. Under these conditions, sensitivity to wind orientation has a significant effect on abscission dynamics and significantly increases wind-assisted dispersal. However, given the fact that our fitted linear model showed quite a strong influence of orientation on dispersal distance, when considered independently of speed, the effect of sensitivity to orientation appeared quite small. The reason for this is evident in Fig. 4; for our case study at least, more upward gusts tended to be slower. Thus, a tendency to release into upward gusts tended to be advantageous in terms of wind orientation at release but tended to be disadvantageous in terms of wind speed. The small difference in dispersal distances predicted by models (i) and (iii) indicates that these positive and negative effects almost cancel out in this case study. More generally, the importance of sensitivity to wind orientation can be expected to vary depending on the correlated frequency of high speed wind gusts and upward wind gusts. For example, if upward wind gusts occurred more frequently with higher wind speeds, then sensitivity to wind orientation would be expected to have a stronger beneficial effect on long-distance dispersal than shown in this study. This is likely to be the case during summer storms, which occur regularly at our study location and which are expected to increase in frequency due to climate change. At a more local scale, the position of individual plants relative to surrounding landscape features may also result in particular plants experiencing differing frequencies of wind gusts with high speeds or upward orientations.
Previous studies of wind-assisted dispersal have shown that turbulent updrafts are a major driver of dispersal distance. Indeed, in one study, the authors suggested that turbulent updrafts are the driver of dispersal distance and that wind speed is irrelevant (Tackenberg, Poschlod & Kahmen 2003). So long as the seed is held aloft by updrafts, it will eventually disperse over a relatively large distance. If this was indeed the case, then there should be no selective advantage for individual plants that preferentially release their seeds into strong winds. However, the results from this study clearly show that preferential release of seeds into stronger wind gusts does significantly influence the distance travelled. Therefore, we speculate that plants such as C. bonariensis have adapted in two separate but related ways to increase the distance travelled by dispersing seeds.
Because abscission layers form more easily during periods of low humidity, and humidity varies at both seasonal and daily timescales, patterns in the abscission of seeds also occur at both seasonal (Nathan et al. 1999) or daily (Greene & Johnson 1992; Greene, Quesada & Calogeropoulos 2008) timescales. Abscission layers form during particular seasonal or daily weather events, and as periods of low humidity are often associated with warmer temperatures, there is a higher likelihood that the formation of abscission layers will be correlated with turbulent wind conditions. These conditions are characterized by significant periods of sustained, upward wind gusts (Soons & Ozinga 2005), meaning that even if seeds were then released at random within these low humidity periods, the higher probability of upward wind gusts would result in seeds being dispersed over greater distances than if released at other times of the day (e.g Kuparinen et al. 2009; Savage et al. 2010, 2012). This preferential release of seeds during periods of low humidity represents one adaptation by the plant that results in dispersal over larger distances. However, in addition to this adaptation, as shown in this and previous studies, seeds are in fact not released at random during low humidity periods, but are preferentially abscised during both strong and upward drafts. Consequently, seeds can be expected to travel even farther again, and non-random abscission occurring at this finer temporal scale represents a second adaptation by the plant that increases dispersal distance.
Abscission dynamics in wind-dispersed plants such as C. bonariensis depend on wind orientation relative to the flower head, as well as wind speed and the time since flowering. The seeds are preferentially released into fast moving or upwards wind gusts, although this preference is relaxed as the seed ages ensuring that the majority of seeds are eventually released from each flower head. As shown in this study, abscission dynamics can influence subsequent wind-assisted dispersal, increasing the distance over which dispersing seeds are likely to travel. These results suggest that plants such as C. bonariensis may have evolved preferential release strategies to maximize the distance travelled by individual seeds and to rapidly disseminate over wide areas. This has important implications for understanding, predicting and managing the spread of weeds in natural, urban and agricultural landscapes and the spread of genes for herbicide resistance or other unwanted characteristics into new populations.
The authors acknowledge the support of the Cooperative Research Centre for National Plant Biosecurity, established and supported under the Australian Government's Cooperative Research Centres Programme.