sign in sign up
<<
Home , Invasive species, Native species

The of Invasive – Projecting Range Shifts 15 with

Bethany A. Bradley

Environmental Conservation, University of Massachusetts, Amherst, Massachusetts, USA

Abstract invasions (Moody and Mack, 1988; Strayer, 2009). Projections of potential range shifts Correlative models between occur- with climate change could provide a valuable rences and climate (here referred to as tool for informing monitoring ‘ suitability models’) have become and targeting high-risk species (Stohlgren increasingly popular for forecasting risk and Schnase, 2006). In response to this from invasive plants under current and need, spatial models of species habitat under future climate scenarios. Th ese models have current and future climate conditions have the potential to inform management and become increasingly prevalent in recent monitoring eff orts by prioritizing land- years, with many investigators focusing on scapes considered at highest risk under a invasive plants. changing climate. However, a wide range of Conceptually, the approach is relatively choices regarding climatic predictor straightforward. Habitat suitability model- variables, model ling approaches and even ling uses empirical relationships between distributional data sets infl uences the species occurrence and environmental resulting projections. Th e eff ects of these variables (e.g. climate, topography, land use) choices are seldom defi ned explicitly, which to defi ne habitat spatially across landscapes reduces their utility for scientists and or regions (Fig. 15.1). Typically, this process managers alike. Th is chapter reviews involves collecting locations where an common practices of habitat suitability invasive is present along with spatial modelling as they apply to invasive plants. maps or models of the environmental Th e chapter also reviews major fi ndings of conditions (Fig. 15.1a). Next, presence recent projections of range shifts in invasive locations are used to specify environmental plants. In both cases, the aim is to explore conditions in which the invasive plant can how diff erent choices of predictors, models establish successfully, defi ning its potential and input data can infl uence conclusions in habitat in environmental space (Fig. 15.1b). a habitat suitability modelling framework Finally, environmental habitat is projected and develop recommendations for best back on to geographic space to determine practices. the areas where the species could potentially establish (Fig. 15.1c). For , the potential range usually includes Introduction substantial land area that has not yet been invaded. Identifying and eradicating early infestations Th ere are numerous synonymous or near- is the most eff ective strategy for preventing synonymous terms used to describe this

240 © CAB International 2014. Invasive Species and Global Climate Change (eds L.H. Ziska and J.S. Dukes) The Biogeography of Invasive Plants 241

(a) (b) (c)

Suitable for species 1 Precip. variable Low High 2 variable Environmental Suitable habitat • Species 1 present Environmental variable 1

Fig. 15.1. Schematic representation of habitat suitability modelling. (a) Geographic occurrences of an invasive species are compared to climatic variables to calculate (b) climatic suitability for that species. (c) Climatic suitability is mapped back on to geographic space to create a model of suitable habitat, or potential invasive species range. modelling process (Franklin, 2009). Th is Model Limitations chapter refers to it as ‘habitat suitability modelling’. Other common terms include How important is climate for modelling ‘ modelling’, ‘ecological plant invasion risk? At a regional scale, niche modelling’ and ‘bioclimatic envelope climate can limit plant growth and com- modelling’. All of these models aim to predict petitive ability, defi ning broad boundaries the potential distribution of species, on where species can establish, persist and although the latter term typically implies potentially become problematic. However, that the environmental predictor variables at landscape and local scales, which are related to climate only. are much more management relevant, Habitat suitability models (HSMs) do not other environmental conditions such as contain any physiological information about topography, soils and may be far invasive plant species, nor do they include more important (Pearson and Dawson, any mechanistic information about spread 2003). Moreover, the locations of initial or . Th ey are purely empirical. introduction and dispersal ability infl uence As a result, they have been criticized, how quickly non- are able to because correlation between climate and become an important component of regional distribution does not necessarily imply a fl ora (Stohlgren et al., 2011; Stohlgren et al., causal relationship between climate and Part II, Chapter 10, this volume). All of these distribution (Dormann, 2007). Environ- components play important roles in mental variables often have a high degree of understanding invasion risk. Here, the focus correlation – for example, locations that is only on the approaches to modelling have high annual precipitation also tend to regional invasion risk in response to climate be wet in the summer. So, the environmental and climate change. variables that are empirically the best predictors of species distribution may not actually be the ones infl uencing plant Predictor Variables growth. For a hypothetical plant, annual precipitation might provide the best fi t Under current climate conditions, switching empirically, even though summer pre- one climatic predictor for another correlated cipitation actually aff ects how quickly the predictor will have little eff ect on the over- plant can grow and how extensive the all projection. Studies have shown that invasion could ultimately be. empirical models based on current climate 242 B.A. Bradley

have similar levels of accuracy to mechanistic introduced) and spatial models will under- models (Hijmans and Graham, 2006; Estes predict potential establishment. Invasive et al., 2013). However, there is no way to species, by defi nition, are not at equilibrium measure accuracy under future climate (Václavík and Meentemeyer, 2012). Th eir conditions. And, if physiologically important continued spread through landscapes and variables are neglected in HSMs in favour of regions fails the equilibrium assumption. correlated but physiologically unimportant However, for the purposes of modelling, variables, this choice could have dramatic equilibrium does not require that the eff ects on projections of future range under species occur everywhere that it could climate change. To follow our hypothetical potentially establish geographically. Rather, example above, if our plant’s growth is equilibrium requires that the species infl uenced primarily by summer pre- encompass the environmental space where it cipitation, then a model of future habitat could potentially establish. Environmental based on annual average precipitation could equilibrium is achieved much earlier than skew projections of range shifts with climate geographic equilibrium. For example, Welk change. Skewed projections are particularly (2004) showed that predictions of HSMs for likely if seasonal climate predictors respond purple loosestrife (Lythrum salicaria) in- diff erently; for example, if summer pre- vasion in North America stopped changing cipitation decreases but annual precipitation after distribution points from the fi rst 120– remains the same. 150 years of invasion were added. Occurrence Th e best way to work around this problem points from the past 50 years of invasion did is to select climatic predictor variables that not change the overall prediction, even are known to be physiologically important though they doubled the available data. to the target species. Th is information can Th us, it is a reasonable assumption that sometimes be found from experimental and long-established and/or widely introduced observational data. In the absence of invasive species approximate climatic physiological information, climatic variables equilibrium and are appropriate for use in that provide the best empirical fi t can be habitat suitability modelling. If a species is compared to expert knowledge of the species in the early stages of invasion and or ecosystem in order to select appropriate distribution points are only available in a predictors (Bradley et al., 2010a). In all cases, localized area, then it is safe to assume that models of current and future distribution habitat models will underestimate its should be treated as hypotheses (due not potential range vastly. Th e amount of time only to climatic predictor uncertainty but that it takes a species to approximate also to uncertainty in climate projections climatic equilibrium likely varies with the and changes to species–environment inter- introduction and dispersal mechanism. For actions). If the climatic predictors are later example, plants introduced as ornamentals found to represent poorly, then may be widely distributed early on and any projections of future habitat should be approximate equilibrium sooner. treated with caution.

Modelling Considerations Invasive Species Equilibrium Choice of model One underlying assumption of HSMs is that the target species is at equilibrium (or near A variety of diff erent methodological equilibrium) with current environmental approaches have been applied to habitat conditions (Guisan and Zimmermann, suitability modelling. Th ese methods can be 2000). If a species is not in equilibrium, then broken down roughly into presence/absence, distribution points will be missing from presence/pseudo-absence and presence-only locations where climate is actually suitable models. Presence/absence models require (where the species could establish if it were distribution not only for the location of The Biogeography of Invasive Plants 243

species presence but also for locations of background will underpredict future range species absence. Absence data for invasive with climate change and may bias the plants can be problematic, because of the climatic predictor variables if they are not previously mentioned concerns with equi- defi ned physiologically (Acevedo et al., librium. An absence might mean that the 2012). If defi ned too narrowly, pseudo- species cannot establish at that location. absence points are more likely to be placed But, it could also mean that the species can in locations that are actually climatically establish, but has not yet been introduced. suitable, thereby biasing the projection. To As a result, presence/absence models are date, there is no standard approach for typically used only at landscape or local defi ning background in order to minimize scales to ask questions such as how local bias in presence/pseudo-absence models. disturbance infl uences invasion (e.g. Bradley For invasive plants, which are not in and Mustard, 2006). At regional scales, the equilibrium geographically, the likelihood of likelihood of a ‘false’ absence skewing model creating biased models when selecting results is much greater and there are few pseudo-absence points is high (Lobo et al., data sets that include both presence and 2010). absence. Hence, presence/absence models Presence-only models do not assume any are rarely used to predict climatic suitability information about absence locations or or to forecast species range shifts in response background environmental space (see Tsoar to climate change. et al., 2007, for examples). One example of a Presence/pseudo-absence models use presence-only model is Mahalanobis species occurrences and a set of pseudo- Distance, which is a multivariate technique absence points chosen at random from the that defi nes perpendicular major and minor available environmental space, or ‘back- axes within the data and calculates distance ground’. Suitable climate conditions are from the data centroid relative to the identifi ed by comparing the occurrences to covariance of axes lengths (Farber and the available climate conditions. If more Kadmon, 2003). Th is model measures only occurrence points are found than expected distance from the presence data; absences relative to the available climate space, then are unnecessary. Presence-only models will those climate conditions are considered almost always predict a larger area of more suitable for invasive plant establish- suitable habitat than presence/pseudo- ment. Many examples of presence/pseudo- absence models, because they have no direct absence models have been developed (Elith mechanism for excluding some subsets of et al., 2006), but the most widely used environmental space. Jiménez-Valverde et approach currently is based on a maximum al. (2008) note that this larger predicted entropy (or MaxEnt) approach developed by area has often been interpreted as an Phillips et al. (2006). indication of overprediction, and therefore Th ere has been considerable debate poor model performance. However, accuracy within the modelling on how to assessments are all performed relative to defi ne ‘background’ from which to sample current species distribution, and may be pseudo-absence points (VanDerWal et al., unreliable measures of the accuracy of 2009a; Lobo et al., 2010; Acevedo et al., potential future range. It is equally probable 2012). If defi ned too broadly, the model for that presence/pseudo-absence models under- current climate conditions might be very predict future range due to biased defi nition precise, but the predictor variables and of the background extents (Jiménez- probability relationships will be minimally Valverde et al., 2008). A presence-only informative. For example, a broad back- approach to modelling future habitat ground of North America relative to an suitability under climate change is the least invasion in the Sonoran Desert is more likely likely to underpredict potential range or be to describe the unique climatic conditions of biased in unknown ways. that desert (relative to North America) Although presence-only models are less rather than the species itself. Too broad a prone to underprediction (false negatives), 244 B.A. Bradley

presence/pseudo-absence models may be as a general rule, there is much higher less prone to overprediction (false positives). uncertainty in projecting range shifts into A balance between minimizing both of these climate conditions with no current analogue errors may ultimately be most useful for (Elith et al., 2010). Identifying and high- guiding regional monitoring and manage- lighting no-analogue climate conditions ment. Hence, an ensemble approach (Araujo (Williams and Jackson, 2007) within habitat and New, 2007) that combines the projections would help to convey uncertainty projections of multiple models and climate in risk assessments. projections could identify priority manage- ment areas. Range shifts with climate change

No-analogue climate Although rising levels of CO2 increase the growth of invasive plants relative to native Th e Earth’s surface is not currently plants (Blumenthal and Kray, Part I, Chapter experiencing all the possible combinations 5, this volume), higher temperatures and of climate conditions that could exist. In altered precipitation do not necessarily some locations, rising temperatures and provide a similar advantage. At the cold edge altered precipitation will change local of invasive plant ranges, warming tem- climate to conditions that historically have peratures might enable plants to expand. not been experienced. Th ese ‘no-analogue’ For example, ’s (Pueraria lobata) range climate conditions (Williams and Jackson, is likely limited at its northern edge in the 2007) may pose a problem for habitat USA by its intolerance of frost. With higher modelling. It is possible to imagine climate temperatures, this range margin is likely to conditions that do not currently exist on move northwards (Bradley et al., 2010b). Earth, but in which a plant could still However, at the warm edge, higher tem- establish. Maybe the conditions are slightly peratures could increase evapotranspiration warmer than current temperatures, or and reduce invasive plant competitiveness. maybe they refl ect an annual monsoon Observations of native species have shown a arriving a month later. In either case, our consistent shift in species distribution lack of invasive plant establishment data towards the colder edge and away from the does not signify that the species cannot warmer edge with warming temperatures tolerate the environmental conditions, just (Parmesan and Yohe, 2003; et al., that the location does not exist. 2003). Similar range shifts are likely for When projecting into these no-analogue invasive plants (Bradley et al., 2009). climate conditions, models must extrapolate A number of projections of invasive plant beyond the available data. Th e more complex range shifts with climate change have shown the model, the greater the chances it will that both expanded and contracted range is produce a problematic relationship beyond likely (Fig. 15.2). In North America, of fi ve the current environmental conditions on problematic invasive plants, two showed which the model is trained. Th is problem is primarily range expansion, two showed more likely in presence/absence or presence/ primarily range contraction and one showed pseudo-absence models, and provides roughly equal expansion and contraction further rationale for including presence- (Bradley et al., 2009). In South Africa, 30 only models when modelling range shifts grass species showed primarily contraction with climate change. Some presence/ of potential range due to climate warming, pseudo-absence models have developed with up to 50% loss of potential range by methods for dealing with no-analogue con- mid-century (Parker-Allie et al., 2009). In ditions. For example, ‘clamping’ in MaxEnt Australia, three hawkweed species showed a (Phillips et al., 2006) sets the value of any loss of potential range of 20% by 2030 location outside the training range as equal (Beaumont et al., 2009). In North America, to the edge of the training range. However, projections of range shifts by 2035 for 12 The Biogeography of Invasive Plants 245

Future suitability High Medium Low Unsuitable

Currently unsuitable

Fig. 15.2. Future climate conditions are likely to lead to substantial habitat loss for leafy spurge (Euphorbia esula) in the USA by the end of the century (adapted from Bradley et al., 2009). Black areas on the map are currently climatically suitable, but lose suitability under climate change. grasses and forbs showed more contraction restoration opportunities due to climate of potential land area than expansion change (Bradley et al., 2009). However, the (Holcombe et al., 2010). In Australia, models term ‘restoration’ is poor for this type of for 72 of national signifi cance opportunity, because altered climate con- projected that range will contract for the ditions make it counterproductive to restore vast majority of species, with losses the native that existed prior to averaging 40% of potential range by mid- invasion. Rather, a type of transformative century (O’Donnell et al., 2012). Losses of restoration may be needed to revegetate potential range suggest that establishment landscapes with non-invasive, regionally of these species might not be quite so native species that can survive the novel widespread in the future. climate conditions and will serve the desired Th ese studies have focused primarily on function of the ecosystem (e.g. soil stability, shifts in potential range. However, loss of carbon storage, animal habitat) (Bradley habitat within the currently invaded range is and Wilcove, 2009). Th e development of also likely. For example, of the three species ecological goals and acceptable risks for with reduced potential range in North assisted migration are needed if these America, 30% of currently invaded areas opportunities are to be realized (Harris et were likely to become climatically unsuitable al., 2006; McLachlan et al., 2007; Richardson by the end of the century (Bradley et al., et al., 2009). 2009). Th is loss of habitat suggests that Although a number of studies now climate change may cause some invasive suggest that range contractions will be just plants to become less competitive in, or even as important, if not more so, than range retreat from, areas that they have already expansion for invasive plants, the total invaded. Invasive species retreat highlights sample size still remains small. Th ere are the intriguing possibility of widespread thousands of non-native species that have 246 B.A. Bradley

become invasive, but the potential ranges of industry is the primary source of non-native only a handful have been modelled. For the plant introductions (Reichard and White, majority of invasive plants that have not 2001; Mack and Erneberg, 2002), and trade been modelled, potential range shifts are is on the rise (Hulme, 2009; Bradley et al., likely to mimic those observed in native 2012). Ecoregions with higher rates of trade species. Warmer temperatures are causing (Vila and Pujadas, 2001) and/or wider an upward elevational and a poleward distribution of aliens versus natives directional shift in distributions, and a (Stohlgren et al., 2011) ultimately may see similar response is likely for invasive plants. the most dramatic shifts in invasive species For regional managers attempting to in response to climate change. catalogue future risk, species that have invaded neighbouring regions with slightly warmer climates likely pose the greatest risk Invasion Risk: Establishment Versus under climate change. Impact

Invasive plants are unique in that they have Vulnerable multiple diff erent ranges: the range in which they are native and all of the locations where Th e pool of invasive species with climate they have been introduced and successfully change projections is small relative to all established. Often, the native range and the global invaders and tends to be focused on invaded ranges encompass diff erent climatic the most problematic examples. Th is small space (Broennimann et al., 2007), which sample size limits the potential for greatly expands the potential geographic identifying regions or biomes that might be locations of invasive species establishment. more susceptible under future climate Models based on the invaded range alone are conditions. To date, only one study in likely to underestimate the potential for Australia has attempted to assess whether establishment, and therefore underestimate some biomes have higher risk than others. the potential range under climate change. O’Donnell et al. (2012) modelled 72 weeds However, using all possible distribution of national signifi cance in Australia under data in a suitability model assumes that current and future climate conditions. Th ey establishment is the most important stage projected two main invasion ‘hotspots’, one of invasion risk to predict. For native species, in the south-east and one in the south-west trying to fi gure out where a plant could part of the country. Both hotspots were establish and persist under climate change is primarily temperate ecosystems. However, important for informing conservation this fi nding might refl ect the unique (Th omas et al., 2004; Kremen et al., 2008), of Australia rather than tem- restoration and even assisted migration perate ecosystems as higher risk. Temperate (Richardson et al., 2009). But, for invasive ecosystems are the poleward destination of species, the management applications of all tropical invasive plants in Australia, and understanding potential establishment are the country’s central desert reduces the much less direct. Many invasive plants available range. Interestingly, the authors have wide climatic tolerances (Pysek and note that the invasion hotspots also occur in Richardson, 2007), and could establish areas with intensive land use, so disturbance under a variety of conditions. For example, is likely to create more opportunities for the cheatgrass ( tectorum) is established establishment of novel species in these in all 50 states and most Canadian provinces particular areas. (Fig. 15.3) (USDA-NRCS, 2012). But, it is Identifying the most vulnerable eco- abundant and has a high impact only in regions in the near term may be linked more western states, where its early season closely to trade, particularly imports and growth helps it outcompete native species the availability of ornamental plants, than for limited water resources (Mack, 1981; to climatic suitability. Th e Knapp, 1996) and where its fi ne fuel The Biogeography of Invasive Plants 247

Absent Present Abundant

Fig. 15.3. Comparison of establishment range to /impact range for cheatgrass () in North America. State-/province-level establishment information is available on the USDA PLANTS website (USDA-NRCS, 2012). Point locations of high abundance come from USGS GAP analysis data for the north-west and south-west and from . increases fi re frequency (D’Antonio and Not using all available establishment data Vitousek, 1992; Brooks et al., 2004; Balch et will result in an underestimate of potential al., 2013) (Fig. 15.3). Th ese types of impacts establishment. But, what if the goal is a are not a concern in eastern states, even more management-relevant assessment of though the species is well established. the current and future geographical extents Th e climatic and geographical space where of impact? Th e choice of distribution data in an invasive species has an impact is a smaller this case is not straightforward. Very few subset of its total establishment. Th is space distribution data sets contain any sort of has been termed ‘damage niche’ (McDonald measure of impact, particularly because et al., 2009; DiTommaso et al., Part III, impact itself is notoriously diffi cult to Chapter 16, this volume) or ‘impact niche’ quantify, changes through time as species (Bradley, 2013). Modelling the current and interact and could be measured for a range future geographical extents of the impact of diff erent native competitors and/or niche is much more management relevant ecosystem processes. An easier-to-measure than establishment or range alone (Hulme, proxy for impact in the case of invasive 2006; McDonald et al., 2009). Yet, the two plants is abundance. Some plant species invasion stages of establishment and impact have meaningful impacts at low abundance, (Lockwood et al., 2007) are often confusingly but high abundance always leads to greater lumped together in model projections under impacts (Parker et al., 1999). Unfortunately, the single term ‘invasion risk’. even abundance or cover is rarely available If the goal of a model is to identify the in distribution data sets, particularly at the current and potential range of species coarse (>1 km) spatial resolutions typically establishment, then all available distribution used to model climatic habitat. data should be used (including data from the One thing is clear, distribution data, even native range and any other invaded ranges). if limited to the invaded range, are a poor 248 B.A. Bradley

proxy for impact. As a result, models of early detection and rapid response. Because suitability based on distribution data will control is more eff ective for early infestations overestimate impact (Bradley, 2013). Th is is (Moody and Mack, 1988), treatment tends partially due to the establishment niche to be focused on areas with low abundance. being much larger than the impact niche. Hence, distribution data compiled regionally But, this phenomenon is also a function of also tend to be skewed towards low invasive plant distribution data sets abundance (Marvin et al., 2009; Bradley, themselves. Herbarium collections, often 2013). As a result, distribution data available available online from sources such as the in the invaded range are likely to be a poor Global Information Facility proxy for the geographic and climatic (http://www.gbif.org/), tend to be focused conditions that defi ne the impact or damage on collecting individuals where the species is niche. rare. Hence, invasive plant herbarium records are more likely to be located in sites with low abundance, and therefore low Establishment Models Do Not Predict impact. Indeed, herbarium records for Impact problematic invasive plants in the western USA were located in regions where managers HSMs create a surface of suitability values considered the species to be absent nearly that are related to the probability that a 60% of the time (Bradley, 2013). species could establish at a given location Invasive plant distribution data sets (Fig. 15.4a). MaxEnt, for example, produces collected by managers and compiled a ‘gain’ value ranging from zero to one, regionally, such as the Early Detection and where high numbers indicate higher Distribution Mapping System (http:// probability of establishment. It seems EDDMaps.org/; Wallace and Bargeron, Part intuitive that higher probability of establish- III, Chapter 13, this volume) or the invasive ment should also be related to higher plant atlas of New England (http://www. abundance, and therefore impact. If true, eddmaps.org/ipane/), tend to be focused on distribution data alone could be used to

(a) (b) Species abundance

Suitability score Suitability score Low High

Fig. 15.4. Habitat suitability scores based on distribution data are linked poorly to abundance. (a) Probability of establishment for an invasive plant in the western USA based on MaxEnt gain values (Phillips et al., 2006). White is unsuitable, lighter shades have higher suitability. (b) Typical relationship between suitability scores and species abundance. The Biogeography of Invasive Plants 249

predict impact by identifying locations with locations known to have a high impact. the highest suitability for establishment. McDonald et al. (2009) modelled current Relationships between modelled suit- and future potential damage to crop species ability based on distribution data and by identifying the states with high economic abundance have been tested for a number of losses to two invasive species (Bridges, species, including (VanDerWal et al., 1992), velvetleaf (Abutilon theophrasti) and 2009b), mammals (Tôrres et al., 2012), Johnsongrass (Sorghum halepense), and arthropods (Jiménez-Valverde et al., 2009) using those state locations as distribution and plants (Pearce and Ferrier, 2001; Nielsen data for suitability models. Th is approach et al., 2005; Estes et al., 2013). Rather than could be applied more widely to invasive fi nding a strong relationship between plants based on expert knowledge of where suitability for establishment and measured the species has the highest impact, with the abundance, these studies have found low or results less likely to overestimate impact insignifi cant correlations. In several cases, than distribution data alone. the relationship is represented best as a step Finally, although suitability models based function, where low suitability corresponds on distribution data are correlated poorly to to absence but high suitability corresponds abundance (Fig. 15.3), a recent study to any value of abundance from low to high suggests that suitability models based on (Fig. 15.4b). high-abundance locations are well correlated with abundance. Estes et al. (2013) showed that suitability values derived from a How Can We Predict Impact? MaxEnt (Phillips et al., 2006) model based on locations of high-yield maize in South If distribution data are a poor proxy for Africa were correlated linearly (R2 = 0.6) to impact niche, and suitability for establish- continuous measurements of maize yield. ment is a poor predictor of impact, is there Th is fi nding suggests that models based on any way to predict impact niche under point locations of known high abundance, current and future climate conditions? One such as the ones created by McDonald et al. positive fi nding is that models appear to be (2009), could do better than just modelling able to predict abundance (a proxy for impact/non-impact by predicting levels of impact) using continuous abundance data. potential abundance. Th e study is based on a Kulhanek et al. (2011) showed that observed single plant species and needs wider testing, abundance of invasive in Minnesota but the approach of modelling damage or lakes (USA) could be used to predict the impact niche is very promising for invasive abundance of carp in South Dakota lakes. plant risk assessments. Unfortunately, continuous abundance data are extremely rare, particularly at the regional scales needed to link abundance to Conclusions climate. A few regional abundance data sets have been collected in the USA, including Although habitat suitability modelling is the inventory and analysis (FIA) data conceptually straightforward, the case of set (http://www.fi a.fs.fed.us/) used by the forecasting plant invasions due to climate US Forest Service to inventory forest change presents a unique set of challenges. resources and the national gap analysis While models of establishment risk under program (GAP) data sets used to validate current climate are likely robust, choices of state and regional land cover classifi cations. modelling approach, distribution data and However, these data are focused on native climatic predictors can have substantial species and are likely to overlook all but the eff ects on projections of range shifts with most problematic invasive plants. climate change. Th e limitations and likely One innovative approach to modelling biases of model and data choices are com- the impact or damage niche is to create a municated poorly to managers, and better subset of distribution data based on eff orts towards transparency are needed. 250 B.A. Bradley

Th e sample size of predictive models for Bradley, B.A., Blumenthal, D.M., Wilcove, D.S. and invasive plants remains small relative to the Ziska, L.H. (2010a) Predicting plant invasions in total pool of global invaders. However, an era of global change. Trends in and models have consistently shown both Evolution 25, 310–318. Bradley, B.A., Wilcove, D.S. and Oppenheimer, M. potential for expansion and likely con- (2010b) Climate change increases risk of plant traction of invasive plant range. It is invasion in the Eastern United States. Biological currently unknown whether the impact Invasions 12, 1855–1872. niche is projected to follow similar expansion Bradley, B.A., Blumenthal, D.M., Early, R., Grosholz, and contraction patterns as the total range. E.D., Lawler, J.J., Miller, L.P., Sorte, C.J.B., Likely invasive plant expansion suggests D’Antonio, C.M., Diez, J.M., Dukes, J.S., Ibanez, that managers should be including new I. and Olden, J.D. (2012) Global change, global species in monitoring eff orts, particularly trade, and the next wave of plant invasions. the problematic species from warmer- Frontiers in Ecology and the Environment 10, climate neighbours. Invasive plant con- 20–28. traction due to climate change may open a Bridges, D.C. (1992) Crop Losses Due to Weeds in the United States, 1992. Science Society window of unique opportunity for the of America, Champaign, Illinois. targeted restoration of high-priority invaded Broennimann, O., Treier, U.A., Muller-Scharer, H., ecosystems. Thuiller, W., Peterson, A.T. and Guisan, A. (2007) Evidence of climatic niche shift during biological invasion. Ecology Letters 10, 701–709. References Brooks, M.L., D’Antonio, C.M., Richardson, D.M., Grace, J.B., Keeley, J.E., Ditomaso, J.M., Hobbs, Acevedo, P., Jiménez-Valverde, A., Lobo, J.M. and R.J., Pellant, M. and Pyke, D. (2004) Effects of Real, R. (2012) Delimiting the geographical invasive alien plants on fi re regimes. Bioscience background in species distribution modelling. 54, 677–688. Journal of Biogeography 39, 1383–1390. D’Antonio, C.M. and Vitousek, P.M. (1992) Biological Araujo, M.B. and New, M. (2007) Ensemble invasions by exotic grasses, the grass fi re cycle, forecasting of species distributions. Trends in and global change. Annual Review of Ecology Ecology and Evolution 22, 42–47. and Systematics 23, 63–87. Balch, J.K., Bradley, B.A., D’Antonio, C.M. and Dormann, C.F. (2007) Promising the future? Global Gomez-Dans, J. (2013) Introduced annual grass change projections of species distributions. Basic increases regional fi re activity across the arid and 8, 387–397. western USA (1980–2009). Global Change Elith, J., et al. (2006) Novel methods improve Biology 19, 173–183. prediction of species’ distributions from Beaumont, L.J., Gallagher, R.V., Downey, P.O., occurrence data. Ecography 29, 129–151. Thuiller, W., Leishman, M.R. and Hughes, L. Elith, J., Kearney, M. and Phillips, S. (2010) The art (2009) Modelling the impact of Hieracium spp. of modelling range-shifting species. Methods in on protected areas in Australia under future Ecology and Evolution 1, 330–342. climates. Ecography 32, 757–764. Estes, L.D., Bradley, B.A., Beukes, H., Hole, D.G., Bradley, B.A. (2013) Distribution models of invasive Lau, M., Oppenheimer, M., Schulze, R., Tadross, plants over-estimate potential impact. Biological M.A. and Turner, W.R. (2013) Comparing Invasions 15, 1417–1429. mechanistic and empirical model projections of Bradley, B.A. and Mustard, J.F. (2006) Characterizing crop suitability and : implications for the landscape dynamics of an invasive plant and . Global Ecology and risk of invasion using remote sensing. Ecological Biogeography 22, 1007–1018. Applications 16, 1132–1147. Farber, O. and Kadmon, R. (2003) Assessment of Bradley, B.A. and Wilcove, D.S. (2009) When alternative approaches for bioclimatic modeling invasive plants disappear: transformative with special emphasis on the Mahalanobis restoration possibilities in the western United distance. Ecological Modelling 160, 115–130. States resulting from climate change. Franklin, J. (2009) Mapping Species Distributions: 17, 715–721. Spatial Inference and Prediction. Cambridge Bradley, B.A., Oppenheimer, M. and Wilcove, D.S. University Press, Cambridge, UK. (2009) Climate change and plant invasion: Guisan, A. and Zimmermann, N.E. (2000) Predictive restoration opportunities ahead? Global Change habitat distribution models in ecology. Ecological Biology 15, 1511–1521. Modelling 135, 147–186. The Biogeography of Invasive Plants 251

Harris, J.A., Hobbs, R.J., Higgs, E. and Aronson, J. Mack, R.N. and Erneberg, M. (2002) The United (2006) Ecological restoration and global climate States naturalized fl ora: largely the product of change. Restoration Ecology 14, 170–176. deliberate introductions. Annals of the Missouri Hijmans, R.J. and Graham, C.H. (2006) The ability of 89, 176–189. climate envelope models to predict the effect of McLachlan, J.S., Hellmann, J.J. and Schwartz, M.W. climate change on species distributions. Global (2007) A framework for debate of assisted Change Biology 12, 2272–2281. migration in an era of climate change. Holcombe, T.R., Stohlgren, T.J. and Jarnevich, C.S. 21, 297–302. (2010) From points to forecasts: predicting Marvin, D.C., Bradley, B.A. and Wilcove, D.S. (2009) invasive species habitat suitability in the near A novel, web-based ecosystem mapping tool term. Diversity 2, 738–767. using expert opinion. Natural Areas Journal 29, Hulme, P.E. (2006) Beyond control: wider 281–292. implications for the management of biological Moody, M.E. and Mack, R.N. (1988) Controlling the invasions. Journal of Applied Ecology 43, 835– spread of plant invasions – the importance of 847. nascent foci. Journal of Applied Ecology 25, Hulme, P.E. (2009) Trade, transport and trouble: 1009–1021. managing invasive species pathways in an era of Nielsen, S.E., Johnson, C.J., Heard, D.C. and . Journal of Applied Ecology 46, Boyce, M.S. (2005) Can models of presence– 10–18. absence be used to scale abundance? – Two Jiménez-Valverde, A., Lobo, J.M. and Hortal, J. case studies considering extremes in life history. (2008) Not as good as they seem: the importance Ecography 28, 197–208. of concepts in species distribution modelling. O’Donnell, J., Gallagher, R.V., Wilson, P.D., Downey, Diversity and Distributions 14, 885–890. P.O., Hughes, L. and Leishman, M.R. (2012) Jiménez-Valverde, A., Diniz, F., De Azevedo, E.B. Invasion hotspots for non-native plants in and Borges, P.A.V. (2009) Species distribution Australia under current and future climates. models do not account for abundance: the case 18, 617–629. of arthropods on Terceira Island. Annales Parker, I.M., Simberloff, D., Lonsdale, W.M., Zoologici Fennici 46, 451–464. Goodell, K., Wonham, M., Kareiva, P.M., Knapp, P.A. (1996) Cheatgrass (Bromus tectorum Williamson, M.H., Von Holle, B., Moyle, P.B., L.) in the Desert – Byers, J.E. and Goldwasser, L. (1999) Impact: history, persistence, and infl uences to toward a framework for understanding the activities. Global Environmental Change: Human ecological effects of invaders. Biological and Policy Dimensions 6, 37–52. Invasions 1, 3–19. Kremen, C., et al. (2008) Aligning conservation Parker-Allie, F., Musil, C.F. and Thuiller, W. (2009) priorities across taxa in with high- Effects of climate warming on the distributions of resolution planning tools. Science 320, 222–226. invasive Eurasian annual grasses: a South Kulhanek, S.A., Leung, B. and Ricciardi, A. (2011) African perspective. Climatic Change 94, Using models to predict the 87–103. abundance and impact of invasive species: Parmesan, C. and Yohe, G. (2003) A globally application to the . Ecological coherent fi ngerprint of climate change impacts Applications 21, 203–213. across natural systems. 421, 37–42. Lobo, J.M., Jiménez-Valverde, A. and Hortal, J. Pearce, J. and Ferrier, S. (2001) The practical value (2010) The uncertain nature of absences and of modelling relative abundance of species for their importance in species distribution regional conservation planning: a case study. modelling. Ecography 33, 103–114. Biological Conservation 98, 33–43. Lockwood, J., Hoopes, M. and Marchetti, M. (2007) Pearson, R.G. and Dawson, T.P. (2003) Predicting Invasion Ecology. Blackwell Publishing, Malden, the impacts of climate change on the distribution Massachusetts, USA. of species: are bioclimate envelope models McDonald, A., Riha, S., Ditommaso, A. and useful? Global Ecology and Biogeography 12, Degaetano, A. (2009) Climate change and the 361–371. geography of weed damage: analysis of US Phillips, S.J., Anderson, R.P. and Schapire, R.E. maize systems suggests the potential for (2006) Maximum entropy modeling of species signifi cant range transformations. geographic distributions. Ecological Modelling Ecosystems and Environment 130, 131–140. 190, 231–259. Mack, R.N. (1981) Invasions of Bromus tectorum L. Pysek, P. and Richardson, D.M. (2007) Traits into Western North America: an ecological associated with invasiveness in alien plants: chronicle. Agro-Ecosystems 7, 145–165. Where do we stand? In: Nentwig, W. (ed.) 252 B.A. Bradley

Biological Invasion, Ecological Studies. presence-only methods for modelling species Springer-Verlag, Berlin and Heidelberg, pp. distribution. Diversity and Distributions 13, 397– 97–126. 405. Reichard, S.H. and White, P. (2001) Horticulture as USDA-NRCS (2012) The PLANTS Database (http:// a pathway of invasive plant introductions in the plants.usda.gov). National Plant Data Center, United States. Bioscience 51, 103–113. Baton Rouge, Lousiana. Richardson, D.M., et al. (2009) Multidimensional Václavík, T. and Meentemeyer, R.K. (2012) evaluation of managed relocation. Proceedings Equilibrium or not? Modelling potential of the National Academy of Sciences 106, distribution of invasive species in different stages 9721–9724. of invasion. Diversity and Distributions 18, Root, T.L., Price, J.T., Hall, K.R., Schneider, S.H., 73–83. Rosenzweig, C. and Pounds, J.A. (2003) VanDerWal, J., Shoo, L.P., Graham, C. and Williams, Fingerprints of global warming on wild animals S.E. (2009a) Selecting pseudo-absence data for and plants. Nature 421, 57–60. presence-only distribution modeling: How far Stohlgren, T.J. and Schnase, J.L. (2006) Risk should you stray from what you know? Ecological analysis for biological hazards: What we need to Modelling 220, 589–594. know about invasive species. Risk Analysis 26, VanDerWal, J., Shoo, L.P., Johnson, C.N. and 163–173. Williams, S.E. (2009b) Abundance and the Stohlgren, T., et al. (2011) Widespread plant environmental niche: environmental suitability species: natives versus aliens in our changing estimated from niche models predicts the upper world. Biological Invasions 13, 1931–1944. limit of local abundance. American Naturalist Strayer, D.L. (2009) Twenty years of zebra mussels: 174, 282–291. lessons from the mollusk that made headlines. Vila, M. and Pujadas, J. (2001) Land-use and socio- Frontiers in Ecology and the Environment 7, economic correlates of plant invasions in 135–141. European and North African countries. Biological Thomas, C.D., et al. (2004) risk from Conservation 100, 397–401. climate change. Nature 427, 145–148. Welk, E. (2004) Constraints in range predictions of Tôrres, N.M., De Marco, P., Santos, T., Silveira, L., invasive plant species due to non-equilibrium De Almeida Jácomo, A.T. and Diniz-Filho, J.A.F. distribution patterns: purple loosestrife (Lythrum (2012) Can species distribution modelling salicaria) in North America. Ecological Modelling provide estimates of densities? A 179, 551–567. case study with jaguars in the Neotropics. Williams, J.W. and Jackson, S.T. (2007) Novel Diversity and Distributions 18, 615–627. climates, no-analog communities, and ecological Tsoar, A., Allouche, O., Steinitz, O., Rotem, D. and surprises. Frontiers in Ecology and the Kadmon, R. (2007) A comparative evaluation of Environment 5, 475–482.