Weeds - Phase 1 - Research Highlights

Predicting Hotspots of Plant Invasion

What the report is about

The invasion of Australian ecosystems by weeds poses a major conservation threat and is occurring at an unprecedented rate. If these invasions are to be minimised and their impact mitigated, a better understanding is needed of the mechanisms that facilitate the invasion.

Invasion of ecosystems by weeds is determined by the degree of presence of weed propagules (parts of a plant that have the ability to reproduce), the non-living chemical and physical characteristics of the ecosystem or site (known as abiotic factors and usually includes light, temperature and atmospheric gases) and the origin, growth, reproduction, structure, and behaviour of the invader plant and the recipient ecosystem (ie, their biological characteristics).

Weed risk assessment, bio-security and quarantine procedures traditionally focus on individual species that pose an invasion threat. This is known as a species-based approach and it has been the key approach in predictions of weed spread and of areas that are vulnerable to future invasion.

Most predictions of the ability of specific weeds to invade ecosystems relate to the characteristics of invading species and of the invaded ecosystems. However, at a broad landscape scale, there may be multiple plant invaders of a specific ecosystem and researchers are looking for predictions about the ability of those weeds in general to invade an ecosystem. This requires an approach to generalise from single-species studies.

This paper presents a simple, yet novel, and non-species specific method for characterising and predicting the ability of weeds to invade ecosystems at the landscape scale. This is known as ecosystem invasibility.

Using spatially referenced historical data on the locations of exotic plant species, modelling was conducted of the expected cover of species as a function of environmental conditions and the geographic location of a site. Models were built as ‘boosted regression' trees. On average, the boosted regression trees explained 38 per cent of the variation in the distribution and abundance of all exotic taxa and exotic forbs (herbaceous flowering plants that are not grasses, sedges or rushes) in the study region.

Click here to view the full report.

Who is the report targeted at?

This is believed to be the first multi-species model based on occupancy and abundance data that can be used to predict habitat invasibility. The approach is flexible and can be applied in different biomes (ecosystems), at multiple scales and to different groups of taxa. Quantifying general processes of plant invasion and predicting invasion risk will increase the understanding of invasion and the efficiency of weed surveillance and control.

Given the threat posed by increasing numbers of invaders, such a general approach to invasion risk might be an extremely valuable tool for ecologists, land managers and regulators.

Background

The primary aim of this study was to develop, at the landscape scale, a non-species specific method for characterising and predicting ecosystem invasibility.

In a case study region surrogates were used of propagule pressure and environmental conditions to predict habitats and locations vulnerable to invasion by exotic species. The approach was tested by modelling relationships between two response variables-all exotic plant taxa and exotic forbs-and 17 predictor variables using boosted regression trees (Elith et al. 2008).

This study provides a quantitative analysis of the environmental predictors that explain historical patterns of exotic plant invasion, offering insights into and evidence about ecological processes that mediate invasion and a template for mapping landscape susceptibility to invasion as a basis for weed risk assessment and surveillance.

Methods used

The study centred on the Corangamite Catchment Management Authority region in Victoria, which covers an area of 13,340 square kilometres and encompasses a wide range of environmental and anthropogenic conditions, among them the Great Otway National Park and the regional cities of Geelong and Ballarat.

Data on plant distribution and abundance were obtained from 3,118 quadrats (sampling squares) surveyed in the Corangamite CMA by the Victorian Department of Sustainability and Environment between 1972 and 2006. Foliar projective cover of plant taxa found in 30-square-metre quadrats was estimated using the Braun-Blanquet scale. Of the 3118 quadrats surveyed, exotic species were present in 1,521 quadrats and exotic forbs in 1,266 quadrats.

Modelling of the relationship between the distribution and abundance of exotic species and surrogates of propagule pressure and habitat suitability was carried out using boosted regression trees (BRTs). Two types of dependent variable were analysed: the occurrence (presence or absence) of exotic taxa and the relative abundance of exotic taxa.

Because many sites that are currently unoccupied by exotic plants might offer suitable habitat, separate models were used to analyse occurrence and abundance data. Instead of integrative mixture models, a two-stage modelling process was chosen for three reasons: the results are easier to interpret; the results from both types of models will be of interest since they might reflect aspects or stages of invasion; and the models are compatible with available BRT software.

Results/key findings

Invasion was highest near the edges of vegetation fragments and areas of human activity. Sites with high vegetation cover had a greater probability of occupancy but a lower proportional abundance of invaders.

Variables indicating propagule pressure and human impacts, abiotic characteristics, and the state of the recipient community (that is, vegetation and forb cover) were rated in the top four most influential variables in each model. Invasion patterns varied, but the overall consistency of invasion trends across response variables illustrates that general patterns of invasion exist. In the study region the abundance of exotic plants was consistently predicted to be highest at the edges of vegetation fragments where native vegetation cover was relatively low.

As well as at the edges of vegetation fragments, areas most vulnerable to general exotic plant invasion (presence and abundance) were near towns and along roads. Habitat fragmentation presumably facilitated invasion because of higher propagule pressure and the availability of light, water and nutrients near the edge of vegetation fragments. If one takes naturally disturbed areas as a quasi-control, the lower level of invasion along watercourses and in recently burnt areas, compared with around towns and roads, suggests that human-initiated spread of exotic plants facilitates rapid invasion more than disturbance itself does. Lower watercourse density and more frequent fire in the forested uplands (which were not accounted for by land use variables included in the analysis) could, however, have contributed to the unusual trends observed.

Implications for relevant stakeholders

The study showed that it is possible to build a reliable model of landscape susceptibility to exotic plant invasion. The general invasion models outperformed single-species models overall in terms of cross-validated predictive accuracy (unpublished data), and the high AUC scores indicated that the models correctly rank locations in terms of their probability of containing exotic taxa and exotic forb distributions (0.87 and 0.91 respectively).

This multi-species model identifies areas susceptible to invasion and points to dominant pathways of plant invasion. Although the approach can be applied to individual species, there are currently 441 exotic species in the study region alone and, despite the fact that many will not pose an ecological or economic threat, predicting the potential distribution of even a quarter of the species would be labour intensive. Further, unless the species have a high prevalence and habitat specificity, model performance will be lower than would be the case with multi-species models, which are built on more data. Interpretation of a collection of (potentially less accurate) single-species models and subsequent allocation of management resources would be difficult. Moving beyond species-based predictions will enable more efficient management and surveillance, will ensure that the majority of species causing an exotic plant incursion are dealt with, and can suggest areas vulnerable to future invasion by functionally similar species.

This work holds out the promise that, by quantifying processes that influence the distribution and abundance of exotic plants, it is possible to predict habitats and regions that are vulnerable to invasion irrespective of biome or geographic region. The capacity to generalise and compare findings gathered from a range of systems will help advance invasion ecology theory. However, further validation and testing is required to ascertain the robustness of the method and its predictive capability. Predictions of ecosystem invasibility can be used to spatially prioritise weed surveillance and control. This development of priorities will increase the efficiency of management by highlighting factors that cause and facilitate invasion.