Climate Change and Plant Invasions: Restoration Opportunities (2009 PDF)

Global Change Biology (2009) 15, 1511–1521, doi: 10.1111/j.1365-2486.2008.01824.x Climate change and plant invasions: restoration opportunities ahead? B E T H A N Y A. B R A D L E Y *, M I C H A E L O P P E N H E I M E R *w and D AV I D S. W I L C O V E *z *Woodrow Wilson School of Public and Inter...

Global Change Biology (2009) 15, 1511–1521, doi: 10.1111/j.1365-2486.2008.01824.x Climate change and plant invasions: restoration opportunities ahead? B E T H A N Y A. B R A D L E Y *, M I C H A E L O P P E N H E I M E R *w and D AV I D S. W I L C O V E *z *Woodrow Wilson School of Public and International Affairs, Princeton University, Princeton, NJ 08544, USA, wDepartment of Geosciences, Princeton University, Princeton, NJ 08544, USA, zDepartment of Ecology and Evolutionary Biology, Princeton University, Princeton, NJ 08544, USA Abstract Rather than simply enhancing invasion risk, climate change may also reduce invasive plant competitiveness if conditions become climatically unsuitable. Using bioclimatic envelope modeling, we show that climate change could result in both range expansion and contraction for five widespread and dominant invasive plants in the western United States. Yellow starthistle (Centaurea solstitialis) and tamarisk (Tamarix spp.) are likely to expand with climate change. Cheatgrass (Bromus tectorum) and spotted knapweed (Centaurea biebersteinii) are likely to shift in range, leading to both expansion and contraction. Leafy spurge (Euphorbia esula) is likely to contract. The retreat of once- intractable invasive species could create restoration opportunities across millions of hectares. Identifying and establishing native or novel species in places where invasive species contract will pose a considerable challenge for ecologists and land managers. This challenge must be addressed before other undesirable species invade and eliminate restoration opportunities. Keywords: bioclimatic envelope model, climate change, invasive species, Mahalanobis distance, model ensemble, restoration ecology, species distribution Received 13 June 2008; revised version received 22 September 2008 and accepted 3 November 2008 from invasive species due to climate change has re- Introduction ceived scant attention (Bradley, 2008; Mika et al., 2008). Invasive plant species threaten native and managed At global and regional scales, invasive plant distribu- ecosystems worldwide. They are increasingly expensive tions are limited by climate (e.g. Guisan & Zimmer- to control (Pimentel et al., 2000) and have become a mann, 2000; Pearson & Dawson, 2003; Thuiller et al., major component of global change (Vitousek et al., 2005). Bioclimatic envelope modeling is an approach for 1996). Global climate change is expected to further predicting potential species distributions based on the expand the risk of plant invasion through ecosystem geographical relationship between occurrences and cli- disturbance and enhanced competitiveness due to ele- mate conditions. Although land use, soils, and species vated CO2 (Dukes & Mooney, 1999; Weltzin et al., 2003; interactions are important for assessing invasion risk at Thuiller et al., 2007). However, climate change may also local and landscape scales (Davis et al., 1998; Bradley & reduce invasive plant competitiveness if conditions Mustard, 2006), climate change is expected to lead to become climatically unsuitable, creating opportunities large-scale range shifts in species distribution (Hughes, for restoration in areas currently dominated by intract- 2000; Peterson et al., 2002; Pearson & Dawson, 2003; able invasive species. Although expanded risk from Root et al., 2003; Thomas et al., 2004; Hijmans & Graham, invasive plants due to climate change has been identi- 2006). Biological conservation and ecosystem restora- fied for several species (Beerling, 1993; Sutherst, 1995; tion face increasing challenges in light of climate change Zavaleta & Royval, 2002; Kriticos et al., 2003; Thuiller as native species become less viable under future cli- et al., 2007; Bradley, 2008; Mika et al., 2008), reduced risk mate conditions (Harris et al., 2006; Millar et al., 2007). Climatically suitable habitat, in either the present or future, can be defined as all areas with similar climate Correspondence: Bethany A. Bradley, tel. 1 1 609 258 2392, fax 1 1 conditions to lands currently occupied by the target 609 258 0390, e-mail: [email protected] species (Kearney, 2006) (Fig. 1a). For invasive species, r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd 1511 1512 B. A. B R A D L E Y et al. climate change can create both expanded risk, when more land area becomes climatically suitable (Fig. 1b) and/or reduced risk, when land area currently at risk becomes climatically unsuitable for certain plant inva- ders (Fig. 1c). In some areas, currently invaded lands may also become climatically unsuitable, creating po- tential retreat areas which may provide opportunities for ecological restoration (Fig. 1c). We assess the relationship between climate and spe- cies distribution for five prominent invasive plants in the western United States: cheatgrass (Bromus tectorum), spotted knapweed (Centaurea biebersteinii; Syn. Centaur- ea maculosa), yellow starthistle (Centaurea solstitialis), tamarisk (Tamarix spp.), and leafy spurge (Euphorbia esula). All of these species are defined as invasive because they originate outside of North America, they are able to dominate ecosystems, outcompete native species, and alter ecosystem function, and they are currently widespread and expanding in range. We then project range shifts due to climate change for each invasive species based on an ensemble of 10 atmo- sphere–ocean general circulation models (AOGCMs). Both expanded risk and restoration potential are likely for invasive species in the western United States. Background C. solstitialis is an annual forb that dominates California grasslands and has become a serious agricultural pest, particularly due to its use of water resources (DiToma- so, 2000; Pitcairn et al., 2006). C. solstitialis was acciden- tally introduced as a seed contaminant in the mid-1800s. Tamarix is a shrubby tree that occurs primarily in riparian ecosystems in the western United States, where it displaces native plants and threatens scarce water resources (Zavaleta, 2000). Tamarix was intentionally introduced as an ornamental throughout the western United States beginning in the early 1800s (Zavaleta, 2000). B. tectorum is an invasive annual grass that dominates shrublands of the intermountain west, lead- ing to increased fire frequency and topsoil erosion Fig. 1 Schematic representation of climate-change scenarios for (D’Antonio & Vitousek, 1992; Knapp, 1996). B. tectorum invasive species, showing the frequency distributions of invasive was accidentally introduced as a grain contaminant species (black dots) and all land cover (black line) relative to a across western rangelands in the late 1800s (Knapp, hypothetical climate variable. (a) Under current conditions, risk from invasion (gray fill) can be defined as all lands climatically 1996). C. biebersteinii (Syn. C. maculosa) is a perennial similar to lands occupied by the invasive species. (b) The worst- forb that invades grasslands and forests of the western case scenario is one in which conditions shift to increase the land United States and outcompetes native plant species area climatically suitable for invasion. (c) The best-case scenario (DiTomaso, 2000). C. biebersteinii was accidentally intro- is one in which climate conditions shift to decrease land area duced as a seed contaminant in alfalfa in the late 1800s. climatically suitable for invasion, potentially leading to a retreat E. esula is an invasive perennial herb that dominates in some currently invaded areas. northern prairies (DiTomaso, 2000; Leistritz et al., 2004). E. esula was first introduced in the northeast United These species were selected because they represent States in the early 1800s, and was found across the some of the most problematic invaders in the western country by the early 1900s (Dunn, 1979). United States (DiTomaso, 2000). Regional distributions r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521  E F F E C T S O F C L I M A T E C H A N G E O N I N VA S I V E P L A N T S 1513 of invaded range exist for these species, which is rare resolution ( 4.5 km pixel size). This resolution was for invasive plants in the United States. Further, these selected to match the Parameter-elevation Regressions species have been widely introduced throughout North on Independent Slopes Method (PRISM) interpolation America and have been established since the 1800s. One of current climate data (Daly et al., 2002). The PRISM limitation of envelope modeling for invasive species is climate interpolation is derived from US weather sta- that they may not have not fully invaded their potential tions and takes into account topographic influences on range, and hence may not be in equilibrium with the precipitation and temperature. Currently available cli- environment. This is a particular concern for modeling mate data include monthly and annual averages of recently introduced invasive species with small popula- precipitation and temperature for the 1970–2000 time tions. The widespread and longstanding introduction of period. these five species makes it plausible to assume that Before creating an envelope model, it is important to currently invaded ranges approximate equilibrium select climate variables that best predict species pre- conditions with current climate. This increases our sence. The best climate predictors were selected by confidence that bioclimatic envelope modeling is appro- identifying the ones that most constrained species dis- priate for these species. tribution. To determine this, we compared the standard deviation of each climate variable within pixels where the species was present to the standard deviation of all Materials and methods pixels (Hirzel et al., 2002; Bradley, 2008). Climate vari- We created bioclimatic envelope models based on the ables with the smallest standard deviations across pix- relationship between five invasive plant distributions els with species present relative to all pixels were and current climate conditions. Current plant distribu- considered the most constrained. In cases where the tions were based on regional maps of invaded range top predictor variables were adjacent months (e.g. June derived from remote sensing, or state and regional and July precipitation), the mean of the variables was surveys of county weed coordinators. Using the non- used. Climate variables tested were monthly and an- native range to develop climatic envelopes is appro- nual average precipitation, minimum temperature, and priate for invasive species because native and non- maximum temperature based on the PRISM dataset native distributions often encompass separate and dis- (Daly et al., 2002). tinct climatic envelopes (Broennimann et al., 2007). Bioclimatic envelopes for each invasive species were Using only the native range, or a combination of native created based on the Mahalanobis distance (Farber & and invaded ranges, to predict invasion could be mis- Kadmon, 2003; Tsoar et al., 2007), which is a presence- leading because climatic and competitive conditions in only multivariate technique that defines perpendicular the native range are not the same as the ones that led to major and minor axes and calculates distance from a large-scale invasion in the non-native range. centroid relative to covariance of axes lengths. The Regional presence of B. tectorum relied on a 1 km resulting envelope is ellipsoidal in n-dimensional spatial resolution map produced for the Great Basin shape. A presence only model is most appropriate for using remote sensing, which identified B. tectorum invasive species modeling because absences are of un- based on its unique interannual response to El Niño certain accuracy, potentially indicating either unsuitable (Bradley & Mustard, 2005). Regional presence of C. climate conditions or suitable climate conditions that biebersteinii was based on a survey of species presence have not yet been invaded. Each model was based on within the state of Montana conducted by the Montana four climatic variables. In all cases, four variables were Department of Agriculture (NRIS, Accessed 2007). found to model species distribution equally well as all Regional presence of Tamarix, E. esula, and C. solsti- of the available climate variables (Bradley, 2008). Suita- tialis was based on surveys commissioned by the Wes- ble climatic conditions (land at risk of invasion) were tern Weed Coordinating Committee (Thoene, 2002; calculated based on the Mahalanobis distance that en- WWCC, 2002). Surveys of county level weed coordina- compassed 95% of the current species distribution. tors with the US Department of Agriculture were con- In order to evaluate the relative fits of the models for ducted to gather expert opinion of the acreage of each each species, we compared predicted frequency to ex- invasive species within 1/4 USGS Quadrangles pected frequency across a range of Mahalanobis dis- ( 6 km pixel size) in each county. We transformed these tances (Hirzel et al., 2006). Predicted frequency is the surveys into presence maps for each species using a number of presence pixels within a given Mahalanobis threshold of 4 ha of species presence (40.1% cover) distance threshold divided by the total number of within each pixel. presence pixels in the study area. Expected frequency Presence data for the five invasive plant species were is the number of pixels (both presence and unknown) resampled based on nearest neighbor to a 0.04166 DD within a given Mahalanobis distance threshold divided r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521 1514 B. A. B R A D L E Y et al. by the total number of pixels in the study area. Under a approach because regional climate models and down- random model, predicted frequency would equal ex- scaled results are available for only a small subset of pected frequency, and, hence the ratio of predicted to AOGCMs. expected frequency at a given Mahalanobis distance The bioclimatic envelopes derived from the species would equal 1. Predicted to expected ratios higher than distribution and current climate variables were applied 1 indicate increasingly good model fits. to future climate projections. The Mahalanobis distance Future climate conditions were derived from an en- that captured 95% of the current distribution was used to semble of 10 AOGCM projections of precipitation, mini- project potential future distribution. The projected inva- mum temperature, and maximum temperature change sive species distributions for each of the 10 AOGCMs by 2100 using the SRESa1b scenario (Nakicenovic & were summed to create an ensemble map of invasion Swart, 2000; PCMDI, 2007). The following 10 AOGCMs risk under future climate conditions. We assumed that were used in this study: CCCMA-CGCM3.1, CNRM- future climatic suitability in 50% of the AOGCMs tested CM3, GFDL-CM2.1, GISS-AOM, INM-CM3, IPSL-CM4, indicates continued high risk of invasion. MIROC3.2(hi-res), MPI-ECHAM5, NCAR-CCSM3, UKMO-HadCM3. Average monthly and annual climate Results change for each of the 10 AOGCMs was calculated by subtracting the average conditions from 1970 to 2000 Our analysis indicates the distribution of C. solstitialis in from the average conditions from 2090 to 2100. Modeled the western United States is most constrained by sum- climate change was added to the PRISM interpolation of mer precipitation, spring precipitation, winter mini- 1970–2000 average monthly and annual climate condi- mum temperature, and spring minimum temperature. tions. Although the spatial resolution of the AOGCMs Climatically suitable habitat currently includes much of was much coarser, the finer resolution of the PRISM California, eastern Oregon, and parts of eastern Wa- dataset creates a higher-resolution estimate of local shington (Fig. 2a). Climate change is likely to expand variation due to latitudinal and topographic effects. invasion risk from C. solstitialis to include more This method was most practical for an ensemble of California and Nevada (Fig. 2b). Lands currently Fig. 2 Climate change is likely to expand invasion risk of Centaurea solstitialis, creating minimal retreat potential by 2100. (a) C. solstitialis dominated lands in the western United States and climatically suitable habitat based on Mahalanobis distance. (b) Change in future invasion risk based on the number of atmosphere–ocean general circulation models (AOGCMs) that project maintained climatic suitability. Colors represent risk of invasion based on the number of AOGCMs that project climatic suitability; black lines denote regions of expanded risk. (c) Retreat potential of currently invaded lands. Note that most areas currently suitable for C. solstitialis maintain their climatic suitability in five or more of the 10 AOGCMs tested. r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521  E F F E C T S O F C L I M A T E C H A N G E O N I N VA S I V E P L A N T S 1515 Table 1 Currently invaded land area with restoration potential under 2100 climate scenarios Percent of currently invaded area Number of AOGCMs Restoration that project future Tamarix Centaurea Bromus Centaurea Euphorbia potential climatic suitability spp. solstitialis tectorum biebersteinii esula High 0 2 1 13 17 18 1 1 1 21 22 13 2 2 2 30 35 16 3 1 3 18 19 16 4 3 5 10 7 18 Low 5 91 88 8 0 19 AOGCMs, atmosphere–ocean general circulation models. occupied by invasive populations of C. solstitialis in AOGCMs tested, showing potential for retreat (Fig. California, Oregon, and Washington have low potential 4c). Of the currently invaded lands in the Great Basin, for restoration (Fig. 2c). Of the currently invaded lands, 13% are no longer climatically suitable by 2100 in any of only 1% are no longer climatically suitable by 2100 in the 10 AOGCMs tested, and 21% are only climatically any of the 10 AOGCMs tested. Eighty eight percent of suitable in one of the 10 AOGCMs. These areas encom- currently invaded lands maintained climatic suitability pass 40 000 km2 and have the greatest potential for in five or more of the 10 AOGCMs (Table 1). restoration. Only 8% of invaded lands are highly likely Tamarix distribution is poorly constrained by climatic to remain at risk, maintaining climatic suitability in five conditions; however, the best predictors are fall preci- or more of the 10 AOGCMs tested (Table 1). pitation, summer precipitation, spring precipitation, C. biebersteinii distributions in Montana are most and winter precipitation. Temperature changes are constrained by summer precipitation, fall minimum unlikely to affect Tamarix distribution. Climatically sui- temperature, winter maximum temperature, and sum- table habitat currently at risk is widespread, encom- mer minimum temperature. Climatically suitable habi- passing most of the land area of the western United tat currently at risk includes the foothills of the Rocky States (Fig. 3a). Climate change has little effect on risk Mountains and the Colorado Plateau (Fig. 5a). Climate of Tamarix invasion, with the majority of land areas change is likely to shift suitable C. biebersteinii habitat to remaining climatically suitable (Fig. 3b). However, the higher elevations, leading to both expanded and con- current distribution is concentrated in riparian corri- tracted risk in parts of Montana, Wyoming, Utah, and dors (Thoene, 2002), suggesting that actual invasion risk Colorado (Fig. 5b). C. biebersteinii populations in eastern is likely limited. Similar to C. solstitialis, there is little Montana and lower elevations in western Montana do potential for restoration of invaded areas (Fig. 3c). Of not remain climatically suitable in the majority of the the currently invaded lands, only 2% are no longer AOGCMs tested, showing potential for retreat (Fig. 5c). climatically suitable by 2100 in any of the 10 AOGCMs Of the currently invaded lands in Montana, 17% are no tested. The vast majority of currently invaded lands longer climatically suitable by 2100 in any of the 10 (91%) are projected to remain suitable in five or more of AOGCMs tested, and 22% are only climatically suitable the 10 AOGCMs tested (Table 1). in one of the 10 AOGCMs. These areas, and low eleva- B. tectorum distribution in the Great Basin is most tion invasions in other states, have the greatest potential constrained by summer precipitation, annual precipita- for restoration. None (0%) of invaded lands are highly tion, spring precipitation, and winter maximum tem- likely to remain at risk; no invaded lands maintain perature (Bradley, 2008). Climatically suitable habitat climatic suitability in five or more of the 10 AOGCMs currently at risk of invasion includes the majority of tested (Table 1). shrub and grasslands in the intermountain west (Fig. E. esula distribution in the western United States is 4a). Climate change is likely to shift climatically suitable most constrained by winter precipitation, fall minimum B. tectorum habitat northwards, leading to expanded temperature, spring maximum temperature, and an- risk in Idaho, Montana, and Wyoming, but reduced risk nual precipitation. Climatically suitable habitat cur- in southern Nevada and Utah (Fig. 4b). Central Utah, rently includes the majority of northern states west of southern and central Nevada, which currently harbor the Mississippi River and some rangeland west of the extensive land area dominated by B. tectorum, do not Rocky Mountains (Fig. 6a). Climate change is likely to remain climatically suitable in the majority of the reduce risk from E. esula in states such as Colorado, r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521 1516 B. A. B R A D L E Y et al. Fig. 3 Climate change is unlikely to affect the potential distribution of Tamarix spp. by 2100. (a) Tamarix dominated lands in the western United States and climatically suitable habitat based on Mahalanobis distance. (b) Change in future invasion risk based on the number of atmosphere–ocean general circulation models (AOGCMs) that project maintained climatic viability. Colors represent risk of invasion based on the number of AOGCMs the project climatic viability, hashed areas are expanded risk. (c) Retreat potential of currently invaded lands. The majority of areas maintain climatic viability in five or more of the 10 AOGCMs tested. Nebraska, Iowa, and Minnesota (Fig. 6b). However, it in any of the 10 AOGCMs tested, and 13% are only may expand risk into parts of Canada not included in climatically suitable in one of the 10 AOGCMs. Land this study. E. esula is likely to retreat from Nebraska and area with restoration potential encompasses 67 000 km2. parts of Oregon and Idaho, creating strong potential for Only 19% of invaded lands are highly likely to remain restoration (Fig. 6c). Of the currently invaded lands in at risk, maintaining climatic suitability in five or more the west, 18% are no longer climatically suitable by 2100 of the 10 AOGCMs tested (Table 1). r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521  E F F E C T S O F C L I M A T E C H A N G E O N I N VA S I V E P L A N T S 1517 A comparison of model evaluations for the five are, in order of increasing model fit, 2.48 (Tamarix), 3.53 species is shown in Fig. 7. In all cases, model fits are (E. esula), 8.11 (C. solstitialis), 10.32 (B. tectorum), and better than a random model (indicated by the 1 : 1 line). 10.71 (C. biebersteinii). Of the five species, Tamarix has the poorest model fit, while C. biebersteinii has the best. At the Mahalanobis distance threshold that encompasses 95% of occur- Discussion rences, the ratios of predicted to expected frequency In every bioclimatic envelope model, a precipitation variable was the best predictor of invasive plant dis- tributions in the western United States. This point is important for two reasons. First, projecting plant dis- tribution change based on rising temperature alone may produce misleading results, particularly in water-lim- ited ecoregions such as those found in the western United States. Second, AOGCM projections of precipi- tation change are highly inconsistent between models (e.g. Milly et al., 2005); hence, an ensemble approach, such as the one used here, may lead to more robust species distribution forecasts than any one AOGCM alone (Araujo & New, 2007). Projections of species distribution change based only on temperature, or using a single AOGCM projection may be of limited value. Our results suggest that considerable changes in invasive species distribution may result from climate change. We have identified regions of the country that may become prone to invasion by one or more of these plants in the next century, as well as invaded lands which may no longer be climatically suitable for these invasive species. Just as native species are expected to shift in range and relative competitiveness with climate change (Hughes, 2000; Peterson et al., 2002; Pearson & Dawson, 2003; Root et al., 2003; Thomas et al., 2004; Hijmans & Graham, 2006), the same should be expected of invasive species. Depending on the species, this will create both expanded invasion risk and substantial restoration opportunities. For two of the five species, C. solstitialis and Tamarix, our models predict primarily expanded invasion risk with climate change (Figs 2 and 3). Many areas at risk already contain small but not yet dominant populations of these invaders, creating the potential for rapid ex- Fig. 4 Climate change is likely to cause a shift in the range of Bromus tectorum, leading to both expanded and contracted risk as well as substantial retreat potential in southern Nevada and Utah by 2100. (a) B. tectorum dominated lands in the Great Basin and climatically suitable habitat based on Mahalanobis distance. (b) Change in future invasion risk based on the number of atmosphere–ocean general circulation models (AOGCMs) that project maintained climatic suitability. Colors represent risk of invasion based on the number of AOGCMs the project climatic suitability; lines indicate regions of expanded risk. (c) Retreat potential of currently invaded lands. Note that dark blue areas maintain climatic suitability in zero of the 10 AOGCMs tested. r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521 1518 B. A. B R A D L E Y et al. Fig. 6 Climate change is likely to reduce invasion risk of Euphorbia esula, creating substantial retreat potential in several western states by 2100. (a) E. esula dominated lands in the Fig. 5 Climate change is likely to cause a shift in the range of western United States and climatically suitable habitat based Centaurea biebersteinii, leading to both expanded and contracted on Mahalanobis distance. (b) Change in future invasion risk risk as well as substantial retreat potential in eastern Montana by based on the number of atmosphere–ocean general circulation 2100. (a) C. biebersteinii dominated lands in Montana and clima- models (AOGCMs) that project maintained climatic suitability. tically suitable habitat based on Mahalanobis distance. (b) Colors represent risk of invasion based on the number of Change in future invasion risk based on the number of atmo- AOGCMs that project continued climatic suitability; lines (very sphere–ocean general circulation models (AOGCMs) that project little area) denote regions of expanded risk. (c) Retreat potential maintained climatic viability. Colors represent risk of invasion of currently invaded lands. Note that dark blue areas maintain based on the number of AOGCMs that project climatic viability, climatic suitability in zero of the 10 AOGCMs tested. hashed areas are expanded risk. (c) Retreat potential of currently invaded lands. Note that dark blue areas maintain climatic viability in zero of the 10 AOGCMs tested. r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521  E F F E C T S O F C L I M A T E C H A N G E O N I N VA S I V E P L A N T S 1519 could occupy these sites if the invasive species are reduced or eliminated by climate change. (Native plants present before the arrival of the invasive plants may be unable to reoccupy these sites as a result of climate change). What may be required in these areas is ‘trans- formative’ restoration (Bradley & Wilcove, in press), involving the introduction of species native to the larger ecoregion that may not have been present originally but which can maintain ecosystem function (Harris et al., 2006). Integrated modeling and experimental work is needed to develop and test viable species assemblages and approaches for transformative restoration. In the absence of active management, new invasive species may quickly become established in areas where the old invasive species are less competitive. We second recent calls for interdisciplinary thinking in the fields of conservation and restoration ecology to Fig. 7 Evaluation of envelope model fitness relative to a ran- address challenges and opportunities resulting from dom model. Curves for each species show the fraction of climate change (Harris et al., 2006; Millar et al., 2007). occurrences vs. the fraction of total pixels for incremental in- The restoration opportunities associated with the retreat creases in Mahalanobis distance (not pictured). All models have higher predictive accuracy than would be expected from a of currently intractable invasive species are vast in the random selection of pixels. A Mahalanobis distance threshold western United States. The uncertainties associated capturing 95% of presence pixels was used to construct current with these changes, as well as the unknown make-up and future range projections. of viable future vegetation communities, highlight a pressing need for integrated modeling, monitoring, and experimental work to better address the ecological pansion in the face of climate change. Continued inva- consequences of climate change. Without timely human sion of C. solstitialis and Tamarix is likely in climatically intervention, the window of restoration opportunity suitable areas. Heightened monitoring and treatment of presented by climate change may quickly close. nascent populations (Moody & Mack, 1988) increas- ingly will be necessary in areas where invasion risk Acknowledgements expands with climate change. For three of the five species, B. tectorum, C. bieberstei- This work was supported by the High Meadows Foundation. We thank two anonymous reviewers for their thoughtful sugges- nii, and E. esula, our models predict both reduced tions, which strengthened the manuscript. We gratefully ac- invasion risk and significant range contractions knowledge the modeling groups for providing their data for (Figs 4–6). Lands with reduced invasion risk are less analysis, the Program for Climate Model Diagnosis and Inter- likely to be invaded by the modeled species with comparison (PCMDI) for collecting and archiving the model climate change. However, they may remain at risk from output, and the JSC/CLIVAR Working Group on Coupled Mod- eling (WGCM) for organizing the model data analysis activity. other invasive species or become at risk from invasive The IPCC Data Archive is supported by the Office of Science, US species not included in this study. For example, red Department of Energy. brome (Bromus rubens), a relative of B. tectorum, is more tolerant of high temperatures (Salo, 2005) and may References replace B. tectorum in parts of the southern Great Basin where climate conditions become unsuitable for B. Araujo MB, New M (2007) Ensemble forecasting of species tectorum. Climate change poses a substantial challenge distributions. Trends in Ecology and Evolution, 22, 42–47. to invasive species monitoring and management strate- Beerling DJ (1993) The impact of temperature on the northern gies because of the likely geographical shifts of invasion distribution-limits of the introduced species fallopia–japonica and impatiens–glandulifera in North-West Europe. Journal of risk. Long-term management planning could benefit Biogeography, 20, 45–53. from more spatially explicit projections of invasion risk Bradley BA (2008) Regional analysis of impacts of climate change under current and future climate conditions. on cheatgrass invasion shows potential risk and opportunity. Reduced climatic suitability on currently invaded Global Change Biology, doi: 10.1111/j.1365-2486.2008.01709.x lands may make invasive species less competitive, Bradley BA, Mustard JF (2005) Identifying land cover variability potentially leading to retreat. Modeling and experimen- distinct from land cover change: cheatgrass in the Great Basin. tal work is needed to assess whether native species Remote Sensing of Environment, 94, 204–213. r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521 1520 B. A. B R A D L E Y et al. Bradley BA, Mustard JF (2006) Characterizing the landscape Mika AM, Weiss RM, Olfert O, Hallett RH, Newman JA (2008) dynamics of an invasive plant and risk of invasion using Will climate change be beneficial or detrimental to the invasive remote sensing. Ecological Applications, 16, 1132–1147. swede midge in North America? Contrasting predictions Bradley BA, Wilcove DS (in press) When invasive plants disappear: using climate projections from different general circulation transformative restoration possibilities in the western United models. Global Change Biology, 14, 1721–1733. States resulting from climate change. Restoration Ecology. Millar CI, Stephenson NL, Stephens SL (2007) Climate change Broennimann O, Treier UA, Muller-Scharer H, Thuiller W, Pe- and forests of the future: managing in the face of uncertainty. terson AT, Guisan A (2007) Evidence of climatic niche shift Ecological Applications, 17, 2145–2151. during biological invasion. Ecology Letters, 10, 701–709. Milly PCD, Dunne KA, Vecchia AV (2005) Global pattern of Daly C, Gibson WP, Taylor GH, Johnson GL, Pasteris P (2002) A trends in streamflow and water availability in a changing knowledge-based approach to the statistical mapping of cli- climate. Nature, 438, 347–350. mate. Climate Research, 22, 99–113. Moody ME, Mack RN (1988) Controlling the spread of plant D’Antonio CM, Vitousek PM (1992) Biological invasions by invasions – the importance of Nascent Foci. Journal of Applied exotic grasses, the grass fire cycle, and global change. Annual Ecology, 25, 1009–1021. Review of Ecology and Systematics, 23, 63–87. Nakicenovic N, Swart R (2000) Special Report on Emissions Scenar- Davis AJ, Jenkinson LS, Lawton JH, Shorrocks B, Wood S (1998) ios. Cambridge University Press, Cambridge, UK. Making mistakes when predicting shifts in species range in NRIS http://nris.mt.gov (accessed 2007) response to global warming. Nature, 391, 783–786. PCMDI (2007) http://www-pcmdi.llnl.gov/, Vol. 2007. DiTomaso JM (2000) Invasive weeds in rangelands: species, Pearson RG, Dawson TP (2003) Predicting the impacts of climate impacts, and management. Weed Science, 48, 255–265. change on the distribution of species: are bioclimate en- Dukes JS, Mooney HA (1999) Does global change increase the velope models useful? Global Ecology and Biogeography, 12, success of biological invaders? Trends in Ecology and Evolution, 361–371. 14, 135–139. Peterson AT, Ortega-Huerta MA, Bartley J, Sanchez-Cordero V, Dunn PH (1979) Distribution of leafy spurge (Euphorbia–Esula) Soberon J, Buddemeier RH, Stockwell DRB (2002) Future and other weedy euphorbia spp. in the United-States. Weed projections for Mexican faunas under global climate change Science, 27, 509–516. scenarios. Nature, 416, 626–629. Farber O, Kadmon R (2003) Assessment of alternative ap- Pimentel D, Lach L, Zuniga R, Morrison D (2000) Environmental proaches for bioclimatic modeling with special emphasis on and economic costs of nonindigenous species in the United the Mahalanobis distance. Ecological Modelling, 160, 115–130. States. Bioscience, 50, 53–65. Guisan A, Zimmermann NE (2000) Predictive habitat distribu- Pitcairn MJ, Schoenig S, Yacoub R, Gendron J (2006) Yellow tion models in ecology. Ecological Modelling, 135, 147–186. starthistle continues its spread in California. California Agri- Harris JA, Hobbs RJ, Higgs E, Aronson J (2006) Ecological restora- culture, 60, 83–90. tion and global climate change. Restoration Ecology, 14, 170–176. Root TL, Price JT, Hall KR, Schneider SH, Rosenzweig C, Pounds Hijmans RJ, Graham CH (2006) The ability of climate envelope JA (2003) Fingerprints of global warming on wild animals and models to predict the effect of climate change on species plants. Nature, 421, 57–60. distributions. Global Change Biology, 12, 2272–2281. Salo LF (2005) Red brome (Bromus rubens subsp. madritensis) in Hirzel AH, Hausser J, Chessel D, Perrin N (2002) Ecological- North America: possible modes for early introductions, sub- niche factor analysis: how to compute habitat-suitability maps sequent spread. Biological Invasions, 7, 165–180. without absence data? Ecology, 83, 2027–2036. Sutherst RW (1995) The potential advance of pests in natural Hirzel AH, Le Lay G, Helfer V, Randin C, Guisan A (2006) ecosystems under climate change: implications for planning Evaluating the ability of habitat suitability models to predict and management. In: The Impact of Climate Change on Eco- species presences. Ecological Modelling, 199, 142–152. systems and Species: Terrestrial Ecosystems (eds Pernetta J, Lee- Hughes L (2000) Biological consequences of global warming: is mans R, Elder D, Humphrey S), pp. 88–98. IUCN, Gland, the signal already apparent? Trends in Ecology and Evolution, 15, Switzerland. 56–61. Thoene JW (2002) Implementation of a GIS for Regional Management Kearney M (2006) Habitat, environment and niche: what are we of Leafy Spurge (Euphorbia esula) and Yellow Starthistle (Centaurea modelling? Oikos, 115, 186–191. solstitialis) in the Western United States. University of Denver, Knapp PA (1996) Cheatgrass (Bromus tectorum L) dominance in Denver. the Great Basin Desert – history, persistence, and influences to Thomas CD, Cameron A, Green RE et al. (2004) Extinction risk human activities. Global Environmental Change-Human and Pol- from climate change. Nature, 427, 145–148. icy Dimensions, 6, 37–52. Thuiller W, Richardson DM, Midgley GF (2007) Will climate Kriticos DJ, Sutherst RW, Brown JR, Adkins SW, Maywald GF change promote alien plant invasions? In: Ecological Studies, (2003) Climate change and the potential distribution of an Vol. 193 (ed. Nentwig W), pp. 197–211. Springer-Verlag, invasive alien plant: Acacia nilotica ssp. indica in Australia. Berlin. Journal of Applied Ecology, 40, 111–124. Thuiller W, Richardson DM, Pysek P, Midgley GF, Hughes GO, Leistritz FL, Bangsund DA, Hodur NM (2004) Assessing the Rouget M (2005) Niche-based modelling as a tool for predict- economic impact of invasive weeds: the case of leafy Spurge ing the risk of alien plant invasions at a global scale. Global (Euphorbia esula). Weed Technology, 18, 1392–1395. Change Biology, 11, 2234–2250. r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521  E F F E C T S O F C L I M A T E C H A N G E O N I N VA S I V E P L A N T S 1521 Tsoar A, Allouche O, Steinitz O, Rotem D, Kadmon R (2007) A WWCC (2002) http://www.weedcenter.org/wwcc. comparative evaluation of presence-only methods for modelling Zavaleta E (2000) The economic value of controlling an invasive species distribution. Diversity and Distributions, 13, 397–405. shrub. Ambio, 29, 462–467. Vitousek PM, D’Antonio CM, Loope LL, Westbrooks R (1996) Zavaleta ES, Royval JL (2002) Climate change and the sus- Biological invasions as global environmental change. American ceptibility of US ecosystems to biological invasions: two cases Scientist, 84, 468–478. of expected range expansion. In: Wildlife Responses to Climate Weltzin JF, Belote RT, Sanders NJ (2003) Biological invaders in a Change: North American Case Studies (eds Schneider SH, Root greenhouse world: will elevated CO2 fuel plant invasions? TL), pp. 277–341. Island Press, Washington, DC. Frontiers in Ecology and the Environment, 1, 146–153. r 2009 The Authors Journal compilation r 2009 Blackwell Publishing Ltd, Global Change Biology, 15, 1511–1521