Kakadu National Park Landscape Symposia Series 2007–2009

səhifə	13/16
tarix	25.06.2016
ölçüsü	5.42 Mb.

1 ... 8 9 10 11 12 13 14 15 16

7 Incorporating dispersal ecology and simulation modelling into the management of plant invasions

HT Murphy, DA Westcott & C Fletcher^¹⁵

7.1 Introduction

Dispersal processes determine the proportion of seeds moved any given distance from the source plant and for invasive species the outcome of these processes may be the most important determinant of invasion success (Lockwood et al 2005). An understanding of dispersal processes is necessary for predicting the potential rate and pattern of invasive spread (Kot et al 1996); this information can be fundamental in the design and implementation of management programs (Trakhtenbrot et al 2005). For fleshy fruited plants, dispersal curves are in large part the product of the movement and foraging behaviour of the animals that feed upon their fruits. These dispersers determine the quantity, distance and direction seeds are dispersed as well as the quality of that dispersal (Wang & Smith 2002, Levin et al 2003).

In many contexts, search and eradication efforts are reliant on on-ground efforts being distributed in the right places and at the right time. Currently these decisions must often be made in the absence of complete information from the field. For example, in the rainforest habitats of eastern Australia search and eradication efforts are severely hampered by dense vegetation, high species diversity, difficult terrain, trying climatic conditions and the remoteness of much of the area from vehicle access. These factors can conspire to make operational decisions about the investment of effort more akin to guesswork than considered and information based. Very similar issues confront the weed managers and control crews in Kakadu National Park. Use of predictive weed spread models incorporating realistic dispersal curves will greatly increase search efficiency through better prediction of the extent and likely location of weed infestations.

We describe here how we have developed dispersal curves for fleshy-fruited weed species occurring in rainforest habitats in North Queensland and how these have been incorporated, along with life-history parameters, into models of weed spread in order to improve management efficiency.

7.2 Generating dispersal curves for weeds

An important descriptor of dispersal processes is the dispersal kernel, which is the frequency distribution of dispersed seeds relative to distance from a parent plant (Levin et al 2003). Relatively little is known about the dispersal of weed species and, in particular, the proportion of seeds being dispersed over long distances is poorly understood and is probably often under-estimated. Seeds dispersed short distances tend to contribute to increased density or abundance in current infestations while those dispersed long distances contribute to spread. Because containment is a primary goal of most weed programs, understanding long-distance dispersal is critical for effective management. For most weed species, the frugivores actually removing and dispersing seeds are unknown. Furthermore, for highly invasive species, when mature individuals are identified they are quickly eradicated, and observing seed removal by fruigivores is not an option. Although new to a particular area, invasive species rarely rely on novel dispersal processes; instead co-opting dispersal processes already existing in the new environment (Westcott & Dennis 2006). Therefore a surrogate or analogue native species with fruit characteristics similar to the weed species can be used to generate the dispersal kernel because it can be assumed that the dispersers will ultimately be very similar.

There are several steps involved in generating dispersal curves for fleshy-fruited species:

Identifying potential dispersers based on fruit characteristics and characteristics of the dispersers themselves (eg gape size).
Determining relative contributions to dispersal by different dispersers (ie per cent of fruit crop removed by different dispersers). This is achieved by a variety of methods including observations of disperser behaviour at fruiting trees, measuring removal rates of fruits and seeds placed on the forest floor, and measuring fruit production in the canopy and fruit fallen to the ground.
Determining seed retention times (‘beak to bum’ times) for dispersers. Seed retention times are measured by recording time from ingestion to defecation during observation of captive animals (Fig 1a).
Measuring disperser movement patterns (Fig 1b & 1c). Disperser movement in space as a function of time is measured through the use of continuous radio-telemetry. A radio-tagged disperser’s location at any given point in time, a ‘fix’, is determined by triangulating bearings from GPS-mapped stations.

7.3 Dispersal ecology and weed management

Dispersal estimates resulting from the methods described above, while not being direct or exact measures of actual dispersal, provide managers with an immediate indication of the nature of dispersal the new invasive is likely to receive, and consequently an indication of the kind of response that is appropriate (Westcott & Dennis 2006). For example, the most immediate and effective use of a well-described dispersal kernel for a weed species is in determining how far on-ground teams should be searching for individuals from any source individual or population. Based on the example dispersal kernel below (Fig 2), searching out to a radius of 250 m from a source will likely account for about 90% of dispersed seed. However, the remaining 10% of seed will be dispersed up to 1400 m from the source. These relatively rare long-distance seed dispersal events which subsequently develop into outlying, or satellite populations, may be critical in contributing to weed range expansion and spread. In addition, an understanding of dispersal agents or vectors that contribute to the long-distance tail of the dispersal kernel may allow targeted management strategies for those vectors – for example focusing searches at known roost sites or frequented habitats.

While both a plant’s demography and dispersal play important roles in determining the rate of invasion, modellers have shown that the speed and pattern of spread of invasive species is extremely sensitive to the shape of the dispersal kernel (Kot et al 1996, Buckley et al 2006). The use of well-parameterised dispersal kernels in models of weed spread is in its very early stages. In the following section we describe a model of spread for a fleshy-fruited woody weed, Miconia calvescens (Melastomataceae) which incorporates both a realistic dispersal kernel and estimates of reproduction and mortality (see also Murphy et al 2008).

(a)

(b)

(c)

Figure 1 Example of data used in generating weed dispersal kernels. (a) Percentage of ingested seeds versus time in gut (b) using radio telemetry to record disperser movement and (c) disperser movement distances (or displacement) over time.

90% of seeds within 250 m

Figure 2 A dispersal curve for a fleshy-fruited rainforest species

7.3.1 Example of a model of spread for Miconia calvescens

Miconia is a declared Class 1 weed (the highest priority category) in Queensland (Land Protection [Pest and Stock Route Management] Act 2002) and is currently the target of a national eradication program. Miconia embodies many of the traits which make for a successful invader; the species can persist in both high and low light environments (Csurches 1998). The small (ca. 6–7 mm), spherical fleshy fruits contain large numbers of tiny seeds (ca. 0.65 mm) (Meyer 1998a) and the small fruit size means they can be consumed by nearly the entire range of frugivores present in tropical forests.

For Miconia in Australia, the eradication program aims to control all individuals in known infestations before they reach maturity. Therefore we used dispersal curves assembled for Miconia based upon two integral components: plant species with similar fruit characteristics for which we have already developed dispersal curves and frugivore movement patterns adapted from functional groups of those animals likely occurring within the range of present Miconia infestation areas (Westcott & Dennis 2006, Metcalfe et al 2006). The model we employed used a dispersal curve based on parameters derived from foraging patterns, seed retention times in the gut, and displacement distances of each species of seed disperser (Westcott et al 2005, Westcott & Dennis 2006, 2007) as described above. These values were generated from hundreds of hours of telemetry data from radio-tracking of avian frugivores in the Wet Tropics region of Australia. The dispersal kernel also included the proportion of fruit estimated to fall directly below the fruiting individual.

We developed a single-species, individual-based model of weed spread in a homogenous landscape, that included the dispersal kernel as well as life-history parameters related to reproduction, mortality and seedling establishment. All the life-history parameters were estimated from a combination of field experience and experimental data. Within the model, up to one million seeds are produced by a mature adult plant per reproductive season. Several types of mortality occur once a seed has been dispersed, including mortality associated with limited seed viability, density dependence, and generally high seedling mortality in the first year. Once a seedling has become established it may die in any year depending on an age-dependant mortality curve, where the probability of dying decreases with age. When an individual in the model reaches maturity it begins to produce seeds, which are themselves dispersed across the landscape. For this particular case study, we began the model with 4 reproductive individuals and let the model run for 30 years; we then compared the number and spatial extent of individuals predicted by the model with an actual infestation occurring in North Queensland that was approximately 30 years old (Fig 3).

Figure 3 Example of the dispersal model. The figure at top right indicates the number of individuals in the infestation. Large dots indicate mature individuals.

We found that the model generally over-predicted the spatial extent and size of the real infestation after 30 years of simulations. There are several possible reasons for this, most notably that the effect of management effort is not included in the model. For example, whereas in the model individuals were allowed to grow and reproduce yearly until they suffered natural mortality, in reality, recent control activities mean that all individuals in the infestation are removed as soon as they are encountered and it is very rare that an individual remains undetected long enough to reproduce for many years. Therefore, in real infestations, mature trees do not provide a regular source of seeds into the population. During the last 20 years local landholders may also have suppressed the population growth by occasionally controlling established individuals. The model also does not include a stored seed bank, however, we know that Miconia seeds may be viable for up to 14 years in the seed bank.

Another factor having a significant impact on the accuracy of the model is that landscape features are not accounted for. The model currently assumes that seeds are dispersed 360° around the source and have an approximately equal probability of survival in all directions (an isotropic distribution) (Fig 4). In reality this is clearly not the case; far from being a homogenous landscape, the area where the infestation occurs is topographically complex and diverse, including various native and human-modified habitats. Landscape features influence both the movement pathways of dispersers as well as the probability of establishment of a dispersed seed. Future work on the model will incorporate data on frugivore habitat use and movement patterns resulting in an anisotropic model (Fig 4). The effect of different spatial and temporal patterns of management investment on population spatial structure and spread will also be incorporated to determine whether it is possible to identify more effective strategies for distributing management effort whilst ensuring a high probability of detecting stray individuals.

Figure 4 Example of an isotropic model (left) and anisotropic model (right) which incorporates the effect of landscape structure on disperser movements and establishment probabilities