Figures
Abstract
Global climate change and invasions by nonnative species rank among the top concerns for agents of biological loss in coming decades. Although each of these themes has seen considerable attention in the modeling and forecasting communities, their joint effects remain little explored and poorly understood. We developed ecological niche models for 1804 species from the European flora, which we projected globally to identify areas of potential distribution, both at present and across 4 scenarios of future (2055) climates. As expected from previous studies, projections based on the CGCM1 climate model were more extreme than those based on the HadCM3 model, and projections based on the a2 emissions scenario were more extreme than those based on the b2 emissions scenario. However, less expected were the highly nonlinear and contrasting projected changes in distributional areas among continents: increases in distributional potential in Europe often corresponded with decreases on other continents, and species seeing expanding potential on one continent often saw contracting potential on others. In conclusion, global climate change will have complex effects on invasive potential of plant species. The shifts and changes identified in this study suggest strongly that biological communities will see dramatic reorganizations in coming decades owing to shifting invasive potential by nonnative species.
Citation: Peterson AT, Stewart A, Mohamed KI, Araújo MB (2008) Shifting Global Invasive Potential of European Plants with Climate Change. PLoS ONE 3(6): e2441. https://doi.org/10.1371/journal.pone.0002441
Editor: Jerome Chave, Centre National de la Recherche Scientifique, France
Received: February 15, 2008; Accepted: March 10, 2008; Published: June 18, 2008
Copyright: © 2008 Peterson et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The authors have no support or funding to report.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Considerable recent concern has focused on the potential negative effects of global climate change and species invasions on native biodiversity [1], [2]. Although each of these themes has seen extensive monitoring and predictive forecasting [3], [4], [5], [6], [7], [8], [9], their joint effects remain largely unexplored [10]. Although single species’ likely responses have been analyzed [11], we here provide a first exploration of climate change effects on global trends in invasive potential, based on a large sample of European plant species.
Ecological niche modeling (ENM) provides a predictive framework for anticipating spatial consequences of global change phenomena for biodiversity [12], [13]. Regarding climate change, extensive methodological testing has produced not just consistent and robust projections across future climate projections [13], [14], [15], [16], [17], [18], [19], [20], [21], but also a growing understanding of the sensitivity, assumptions, and limitations of the approach [13], [22], [23], [24]. Similarly, application of ENM to forecasting potential geographic distributions of invading species has seen extensive testing [25], [26], [27], [28], [29], [30], [31], [32], [33]. Although each topic has attracted attention individually, their combined effects have seen little or no attention [11], [34].
This study provides a first survey of likely changes in invasive potential under changing climates of a significant sample of biodiversity—in this case, we assess likely changes in the global invasive potential of >1800 species of European plants. This data set, which has been explored regarding biodiversity patterns within Europe [20], [35], here provides the basis for development of ENMs that can be used to learn about trends in global geographic potential of species under present and future climate regimes. This study builds on the foundation of ENM applications for assessment of global invasive potential [28], [33], but extends it significantly in assessing interactions between invasive potential and current dramatic changes in climate.
Results
Validation of Model Results
Of the European plant species included in this study, many are already known to be present as introduced and possibly invasive species on other continents. In spite of difficulties of making taxonomic equivalencies, we identified at least 65 species from the pool under consideration that are considered invasive in the United States (see Text S1); many more are known as alien species in other regions and continents. A first challenge for the ENMs developed in this study is thus that of predicting present-day distributions of European plant species invasive on other continents. That is, if the models hold significant predictive ability on other continents with distinct biotic communities, then independent occurrence information should be more coincident with projections than expected by chance.
Hence, based on European occurrences, can we project a niche model to another continent, and anticipate the species’ distributional potential there? An example analysis, illustrated in Figure 1, is that of projections of the North American potential distribution of Ranunculus ficaria. Of the 3111 counties in the United States considered, 1520 were predicted as within the potential distribution of the species; of the 52 counties where this species is known to occur in the United States, 48 were among those predicted present. This level of coincidence of model projections of the species’ North American distributional potential based on European occurrences is significantly higher than expected by chance (P = 1.8×10−12). Of the 10 species tested in this way, 9 projections were statististically significant (Table 1).
Succeeding panels show geographic trends at a global scale that are expected with future climate change under 2 scenarios each from 2 general circulation models (GCMs): gray = current distributional area projected to be lost with global climate change, green = current distributional area projected to be retained with global climate change, blue = areas projected to become suitable for the species with global climate change.
Global Forecasts of Changing Invasive Potential
Predicted change in potential distributional areas for the entire set of species within Europe was variable. Across the 4 scenarios of climate change analyzed, species averaged 2.4–3.1% decline in potential distributional area (Table 2). The Hadley and Canadian scenarios yielded projections that were closely similar, with the Canadian datasets slightly less drastic in projections than the Hadley datasets (Figure 2). The “A” scenarios, however, were considerably more drastic in their projections than the “B” scenarios. Interestingly, the former scenarios predicted both more drastic expansions and more drastic contractions, as is noticeable in the broader variation in predicted area losses and gains (Figure 2).
Looking at projections of change in potential distributional areas for each species as an invasive on continents beside Europe, Southern Hemisphere continents tended to have most species increasing in invasive potential, whereas Northern Hemisphere continents generally saw species decreasing somewhat in invasive potential (Table 2). Africa showed greatest losses in potential area (Figure 2; Table 2), but South America shows a more mixed response—increasing in potential distributional area under 3 scenarios and decreasing under one (Canadian B scenario). Curiously, all three Southern Hemisphere continents have average declines in potential area, yet around two-thirds of species increased in habitable area, suggesting that the declining third declined dramatically (Table 2).
Changes in potential distributional area in Europe (native range) relate to potential distributional area on other continents in odd ways (Figure 3). In general, species increasing in potential distributional area in Europe also increased in distributional potential on Northern Hemisphere continents, but relationships were more complex in Southern Hemisphere continents—species declining in potential distributional area in Europe might increase or decrease in potential distributional area in the Southern Hemisphere, but species increasing in European potential distributional area generally showed no change in potential distributional area in the Southern Hemisphere.
Discussion
In general, most European plant species are anticipated to decline somewhat in their invasive potential on other continents. Because many European plant species reach their tolerance limits for warm climates in southern Europe, further warming of climates is unlikely to allow these species to invade broadly into the Subtropics and Tropics, and rather are pushed farther north. With northward shifts, most species see shrinking distributional areas. As a consequence, species with shrinking distributional potential in Europe should also see reductions in potential distribution in North America and Asia as well, whereas translation of these effects into the Southern Hemisphere may prove more complex. Similarly, a recent study of Argentine ant global invasive potential [11] found global invasive potential declining somewhat, but with potential for some regional expansions. Nonetheless, more European species were predicted to see expanded possibilities in the Southern Hemisphere, but those declining declined dramatically, thus creating opposing signals.
If nothing else, this study serves to illustrate the complexities of the likely effects of climate change on biodiversity. Even the relatively simple native-range projections that have been made about climate change effects on biodiversity have been complex—landscapes, taxa, and particular climate change scenarios all affect the predicted implications of climate change for biodiversity [13], [20], [36], [37], [38], [39]. The early, crude generalities that were made about how climate change will affect species’ invasions [10] clearly underestimated this complexity.
This study does—of course—have limitations. Most prominently, the 50×50 km resolution of the European occurrence data on which the models were based limits results to a fairly generalized characterization of niches of species. Second, Europe being a fairly small region and one without great environmental diversity, these analyses may not illustrate the full diversity of likely responses by species to climate change processes. Finally, focusing only on plants may also limit the diversity of phenomena that can be appreciated in a survey such as this one.
Still, the general picture painted herein is probably robust. That is, the invasive potential of species in the face of changing climates is very complex, and is not immediately predictable based on simple generalizations. Individual species each have their own particular ecological needs, and those needs make for distinct patterns of invasive potential. This individuality parallels that observed in studies of climate change effects on native species [37], and is echoed even more clearly in our explorations of interactions between climate change effects and invasive potential of species. In this broad survey, invasive potential on average declined with warming climates, yet particular species in certain nonnative regions actually increased in potential distributional area.
Methods
Input Occurrence Data
Distributional data were available for 2362 plant species [40], [41], comprising ∼20% of the total European flora, sampled between 1972 and 1996. Although not all of the species in the European flora compilation are actually native to Europe, the great majority was, and as such we treat the European continent as a source area for invasive plant populations. The data have a bias towards well-represented groups in western and central Europe whereas taxa in the Mediterranean region are relatively underrepresented [42]. These data document species distributions on a UTM (Universal Transverse Mercator) 50×50 km grid covering most of Europe, but excluding Russia and the Caucasus, where survey effort was less intensive. This data set has seen numerous detailed analyses, including several based on ENM [20], [35]. Of the 2362 plant species in the overall data set, we developed predictive models for all 1804 species for which ≥20 occurrences (i.e., grid squares) were available.
Environmental Data Sets
To characterize present-day climates, we used coarse-scale climate summaries [43], including annual mean precipitation, annual mean temperature, annual mean maximum monthly temperature, and annual mean minimum monthly temperature (more dimensions are available, but this reduced set was necessary owing to the more limited suite of dimensions available for future climates). Future climates were summarized via parallel data sets summarizing general circulation model results (projection to 2055) drawn from the Hadley (HadCM3) and Canadian (CGCM1) climate modeling centers [44], [45], in each case for the A2 and B2 emissions scenarios—these two scenarios bracket the range of most likely future climate conditions (i.e., being relatively liberal and relatively conservative, respectively). Future climate data sets were obtained from the Data Distribution Centre of the Intergovernmental Panel on Climate Change [46]. As such, we analyzed species’ global potential distributions under present-day conditions and 4 scenarios of future conditions (HadCM3 A2 and B2, CGCM1 A2 and B2).
Ecological Niche Modeling
Several studies have compared projections by different models, concluding that complex-fitting algorithms provide generally better projections than simpler analogues [47], [48]. In these studies model performance was assessed using measures of model fit, i.e., measuring how well models fitted the training data or test data on the same landscape as the training data. The problem addressed in this contribution is that of transferability – making projections into situations that are statistically independent from the training data, and such measures of model performance have been shown to be overly optimistic [38]. In practice and despite all model comparisons performed in recent years, little guidance can be provided regarding the selection of ‘best’ ENM algorithms for transferability [49]. Here, we used the Genetic Algorithm for Rule-Set Projection (GARP), a method that has been extensively used in studies involving transferability [26], [49]. The OpenModeller version of the GARP [50], [51] was utilized. GARP is an evolutionary-computing method that builds ENMs based on nonrandom associations between known occurrence points for species and sets of GIS coverages describing variation in several ecological parameters of environments. Occurrence data are used by GARP as follows: 50% of occurrence data points are set aside for an independent filtering to assure predictive ability of models (extrinsic testing data), 25% are used for developing models (training data), and 25% are used for tests of model quality internal to GARP (intrinsic testing data). Distributional data are converted to binary raster layers, and by random resampling from training and intrinsic test data and areas of ‘pseudoabsence’ (areas lacking known presences), two data sets are created, each of 1250 points; these data sets are used for rule generation and model testing, respectively [50], [51].
Within GARP’s processing, the first rule is created by applying a method chosen randomly from a set of inferential tools (e.g., logistic regression, bioclimatic rules). The genetic algorithm consists of specially defined operators (e.g. crossover, mutation) that modify the initial rules, and thus the result are models that have “evolved”—after each modification, the quality of the rule is tested (to maximize both significance and predictive accuracy) and a size-limited set of best rules is retained. Because rules are tested based on independent data (the intrinsic test data), performance values reflect expected performance of rules, an independent verification that gives a more reliable estimate of true rule performance.
The result is a set of rules that can be projected onto a map to produce a potential geographic distribution for the species under investigation—in this particular study, we projected models trained within Europe based on European occurrences and European environmental data to worldwide coverages for present-day and 4 future-climate coverage sets. For all analyses, we used the OpenModeller interface to the GARP algorithm that has been optimized for use in parallel- and grid-computing environments [52], and carried out the modeling on a 128-processor high-throughput parallel computing cluster at the University of Kansas. All model output was in the form of geo-tif image files.
Following recent best-practices recommendations [53], for each species, we developed 100 replicate random-walk GARP models, and filtered out 90% based on consideration of error statistics, as follows. The ‘best subsets’ methodology consists of an initial filter removing models that omit (omission error = predicting absence at points of known presence) heavily based on the extrinsic testing data, and a second filter based on an index of commission error ( = predicting presence in areas of known absence), in which models predicting very large and very small areas are removed from consideration. Specifically, in GARP, we retained only the 20% of models that showed lowest omission errors, and then retained only the central 50% of the frequency distribution of proportional area predicted present (an index of commission error); the result was 10 ‘best subsets’ models (binary raster data layers) that were summed to produce a best ensemble estimate of geographic projection.
Evaluating Model Projections
Model projections were tested for 10 European plant species already invasive in North America (see Text S1) to assess the predictive ability of these models, based on the example of a previous publication [29], as follows. In these tests, the data used to validate models are completely independent of the data used to train them, as they are exclusively on another continent from where the model was trained [38]. We drew North American occurrence data from the U.S. Department of Agriculture’s Plants Database [54], summarizing known ranges of each species at the level of counties in which they are known to occur by means of adding fields for each species to the attributes table of a vector dataset summarizing boundaries of 3111 counties across the United States.
To summarize projections from the ecological niche model for each species in each county, we plotted 5 random points within each county, and intersected those points with the raster grid from the ENM; we tallied a county as ‘predicted present’ if ≥1 of its random points intersected areas of predicted presence. We calculated (1) the number of counties where the species is known to occur (x), (2) the number of counties in which the species was predicted to occur (y), and the number of counties in which the known occurrence coincides with a county of predicted occurrence (z). From these 3 quantities, we calculated the cumulative binomial probability of obtaining z successes, given x trials, and with a y/3111 probability of success. This calculation, in effect, provides a one-tailed probability of achieving the observed level of coincidence between known and predicted invasive distributional areas by chance alone.
Analysis and Interpretation
Worldwide projections for present and 4 future climate scenarios were summarized as follows. To summarize patterns on more specific spatial scales, we subset the global output grids to focus on each continent except Antarctica (Europe, Africa, Asia, Australia, North America, South America), and calculated areas predicted present in each by reprojecting GARP results from geographic coordinates to a Lambert Equal Area projection, and extracting areas predicted present. Final area estimates were in terms of m2 predicted present on each continent under each climate scenario, which were translated into calculations of percent change between future and present-day climates.
Supporting Information
Text S1.
European Plant Species in Occurrence Data Set and Invasive in North America
https://doi.org/10.1371/journal.pone.0002441.s001
(0.02 MB DOC)
Acknowledgments
We thank the OpenModeller development team at Centro de Referência em Informação Ambiental, Campinas, Brazil, for their innovative programming and development work; thanks also to Monica Papeş for her always-excellent assistance with GIS challenges.
Author Contributions
Conceived and designed the experiments: AP. Performed the experiments: AS. Analyzed the data: AP AS KM. Contributed reagents/materials/analysis tools: MA KM. Wrote the paper: AP MA.
References
- 1. Chapin FSI, Zavaleta ES, Eviner VT, Naylor RL, Vitousek PM, et al. (2000) Consequences of changing biodiversity. Nature 405: 234–242.
- 2. Novacek MJ, Cleland EE (2001) The current biodiversity extinction event: Scenarios for mitigation and recovery. Proceedings of the National Academy of Sciences USA 98: 5466–5470.
- 3. Walther G-R, Post E, Convey P, Menzel A, Parmesan C, et al. (2002) Ecological responses to recent climate change. Nature 416: 389–395.
- 4. Parmesan C, Yohe G (2003) A globally coherent fingerprint of climate change impacts across natural systems. Nature 421: 37–42.
- 5. Carlton JT (1996) Pattern, process, and prediction in marine invasion ecology. Biological Conservation 78: 97–106.
- 6. Lawton JH, Brown KC (1986) The population and community ecology of invading insects. Philosophical Transactions of the Royal Society of London B 314: 607–617.
- 7.
NAS (2002) Predicting Invasions of Nonindigenous Plants and Plant Pests. Washington, D.C.: National Academy of Sciences.
- 8.
Williamson M (1996) Biological Invasions. London: Chapman & Hall.
- 9. Soberón J, Peterson AT (2004) Biodiversity informatics: Managing and applying primary biodiversity data. Philosophical Transactions of the Royal Society of London B 359: 689–698.
- 10. Dukes JS, Mooney HA (1999) Does global change increase the success of biological invaders. Trends in Ecology & Evolution 14: 135–139.
- 11. Roura-Pascual N, Suarez A, Gómez C, Pons P, Touyama Y, et al. (2005) Geographic potential of Argentine ants (Linepithema humile Mayr) in the face of global climate change. Proceedings of the Royal Society of London B 271: 2527–2535.
- 12. Soberón J, Peterson AT (2005) Interpretation of models of fundamental ecological niches and species' distributional areas. Biodiversity Informatics 2: 1–10.
- 13. Pearson RG, Dawson TP (2003) Predicting the impacts of climate change on the distribution of species: Are bioclimate envelope models useful? Global Ecology and Biogeography 12: 361–371.
- 14. Bakkenes M, Alkemade JRM, Ihle F, Leemansand R, Latour JB (2002) Assessing effects of forecasted climate change on the diversity and distribution of European higher plants for 2050. Glob Change Biol 8: 390–407.
- 15. Berry PM, Dawson TP, Harrison PA, Pearson RG (2002) Modelling potential impacts of climate change on the bioclimatic envelope of species in Britain and Ireland. Global Ecology and Biogeography 11: 453–462.
- 16. Erasmus BFN, Van Jaarsveld AS, Chown SL, Kshatriya M, Wessels KJ (2002) Vulnerability of South African animal taxa to climate change. Glob Change Biol 8: 679–693.
- 17. Peterson AT, Martínez-Meyer E, González-Salazar C, Hall P (2004) Modeled climate change effects on distributions of Canadian butterfly species. Canadian Journal of Zoology 82: 851–858.
- 18. Peterson AT, Ortega-Huerta MA, Bartley J, Sánchez-Cordero V, Soberón J, et al. (2002) Future projections for Mexican faunas under global climate change scenarios. Nature 416: 626–629.
- 19. Huntley B, Green RE, Collingham YC, Hill JK, Willis SG, et al. (2004) The performance of models relating species geographical distributions to climate is independent of trophic level. Ecology Letters 7: 417–426.
- 20. Thuiller W, Lavorel S, Araújo MB, Sykes MT, Prentice IC (2005) Climate change threats to plant diversity in Europe. Proceedings of the National Academy of Sciences USA 102: 8245–8250.
- 21. Araújo MB, Thuiller W, Pearson RG (2006) Climate warming and the decline of amphibians and reptiles in Europe. Journal of Biogeography 33: 1712–1728.
- 22. Hampe A (2004) Bioclimate envelope models: What they detect and what they hide. Global Ecology and Biogeography 13: 469–471.
- 23.
Peterson AT, Tian H, Martínez-Meyer E, Soberón J, Sánchez-Cordero V, et al. (2005) Modeling distributional shifts of individual species and biomes. In: Lovejoy TE, Hannah L, editors. Climate Change and Biodiversity. New Haven, Conn.: Yale University Press. pp. 211–228.
- 24. Araújo MB, Whittaker RJ, Ladle RJ, Erhard M (2005) Reducing uncertainty in projections of extinction risk from climate change. Global Ecology and Biogeography 14: 529–538.
- 25. Hoffmann MH (2001) The distribution of Senecio vulgaris: Capacity of climatic range models for predicting adventitious ranges. Flora 196/5: 395–403.
- 26. Iguchi K, Matsuura K, McNyset K, Peterson AT, Scachetti-Pereira R, et al. (2004) Predicting invasions of North American basses in Japan using native range data and a genetic algorithm. Transactions of the American Fisheries Society 133: 845–854.
- 27. Martin WK (1996) The current and potential distribution of the common myna (Acridotheres tristis) in Australia. Emu 96: 166–173.
- 28. Peterson AT (2003) Predicting the geography of species' invasions via ecological niche modeling. Quarterly Review of Biology 78: 419–433.
- 29. Peterson AT, Papeş M, Kluza DA (2003) Predicting the potential invasive distributions of four alien plant species in North America. Weed Science 51: 863–868.
- 30. Peterson AT, Vieglais DA (2001) Predicting species invasions using ecological niche modeling. BioScience 51: 363–371.
- 31. Richardson DM, McMahon JP (1992) A bioclimatic analysis of Eucalyptus nintens to identify potential planting regions in Southern Africa. South African Journal of Science 88: 380–387.
- 32. Skov F (2000) Potential plant distribution mapping based on climatic similarity. Taxon 49: 503–515.
- 33. Thuiller W, Richardson DM, Pysek P, Midgley GF, Hughes GO, et al. (2005) Niche-based modelling as a tool for predicting the risk of alien plant invasions at a global scale. Glob Change Biol 11: 2234–2250.
- 34. Benning TL, LaPointe D, Atkinson CT, Vitousek PM (2002) Interactions of climate change with biological invasions and land use in the Hawaiian Islands: Modeling the fate of endemic birds using a geographic information system. Proceedings of the National Academy of Sciences USA 99: 14246–14249.
- 35. Araújo MB, Thuiller W, Williams PH, Reginster I (2005) Downscaling European species atlas distributions to a finer resolution: Implications for conservation planning. Global Ecology and Biogeography 14: 17–30.
- 36. Thomas CD, Cameron A, Green RE, Bakkenes M, Beaumont LJ, et al. (2004) Extinction risk from climate change. Nature 427: 145–148.
- 37. Peterson AT (2003) Projected climate change effects on Rocky Mountain and Great Plains birds: Generalities of biodiversity consequences. Glob Change Biol 9: 647–655.
- 38. Araújo MB, Pearson RG, Thuiller W, Erhard M (2005) Validation of species-climate impact models under climate change. Glob Change Biol 11: 1504–1513.
- 39. Midgley GF, Hannah L, Millar D, Thuiller W, Booth A (2003) Developing regional and species-level assessments of climate change impacts on biodiversity in the Cape Floristic Region. Biological Conservation 112: 87–97.
- 40.
Jalas J, Suominen J (1972–1996) Atlas Florae Europaea. Helsinki: Committee for Mapping the Flora of Europe and Societas Biologica Fennica Vanamo.
- 41. Lahti T, Lampinen R (1999) From dot maps to bitmaps–Atlas Florae Europaeae goes digital. Acta Botanica Fennica 162: 5–9.
- 42. Araújo MB (2003) The coincidence of people and biodiversity in Europe. Global Ecology and Biogeography 12: 5–12.
- 43.
New M, Hulme M, Jones P (1997) A 1961–1990 mean monthly climatology of global land areas. Norwich, U.K.: Climatic Research Unit, University of East Anglia.
- 44.
Pope VD, Gallani ML, Rowntree VJ, Stratton RA (2002) The impact of new physical parametrizations in the Hadley Centre climate model - HadAM3. Bracknell, Berks, UK: Hadley Centre for Climate Prediction and Research.
- 45. Flato GM, Boer GJ, Lee WG, McFarlane NA, Ramsden D, et al. (1999) The Canadian Center for Climate Modelling and Analysis Global Coupled Model and its climate. Climate Dynamics 16: 451–467.
- 46. IPCC (2001) Climate data archive. Intergovernmental Panel on Climate Change, http://ipcc-ddc.cru.uea.ac.uk.
- 47. Segurado P, Araújo MB (2004) An evaluation of methods for modelling species distributions. Journal of Biogeography 31: 1555–1568.
- 48. Elith J, Graham CH, Anderson RP, Dudik M, Ferrier S, et al. (2006) Novel methods improve prediction of species' distributions from occurrence data. Ecography 29: 129–151.
- 49. Peterson AT, Papeş M, Eaton M (2007) Transferability and model evaluation in ecological niche modeling: A comparison of GARP and Maxent. Ecography 30: 550–560.
- 50. Stockwell DRB, Noble IR (1992) Induction of sets of rules from animal distribution data: A robust and informative method of analysis. Mathematics and Computers in Simulation 33: 385–390.
- 51. Stockwell DRB, Peters DP (1999) The GARP modelling system: Problems and solutions to automated spatial prediction. International Journal of Geographical Information Science 13: 143–158.
- 52. CRIA (2007) OpenModeller, version ****.Campinas, Brazil: Centro de Referência em Informação Ambiental.
- 53. Anderson RP, Lew D, Peterson AT (2003) Evaluating predictive models of species' distributions: Criteria for selecting optimal models. Ecological Modelling 162: 211–232.
- 54.
USDA (2007) Washington, D.C.: U.S. Department of Agriculture. Plants Database, http://plants.usda.gov.