Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Present and potential distribution of invasive garlic mustard ( Alliaria petiolata ) in North America

Present and potential distribution of invasive garlic mustard ( Alliaria petiolata ) in North... Introduction Occurrences of non‐indigenous plant species have become components of the vegetation of most regions of the world as a result of increasing species exchange between continents and vegetation transformation by man. At a global scale, the problem of invasive alien plants has increased greatly over the past decades ( Perrings ., 2000 ). Procedures for assessing the potential invasiveness of non‐indigenous plants are urgently needed for objective and scientifically based quarantine regulations for preventing introduction or further spread, or for making informed decisions about deliberate introductions. Recent systems for predicting invasiveness of plants have been developed for parts of the non‐indigenous floras of North America ( Reichard & Hamilton, 1997 ), Australia ( Pheloung ., 1999 ), and South Africa ( Tucker & Richardson, 1995 ). They are based on the analysis of a number of biological and ecological characters of the species in question (life history, habitat characteristics, invasiveness elsewhere and biogeography). However, according to ( Cronk & Fuller, 1995 ) environmental factors, such as climate, and particularly seasonality of climate, are of special importance because the relationship between climate and the distribution of plants is well documented ( Grace, 1987 ; Woodward, 1987 ). Climate may be considered as setting the broad limits for plant distribution, while other factors such as geology, soils and competition will determine the presence or absence of a species in a particular area and on a finer regional or local scale. Accepting that climate usually limits the range of a species, analysis of the climatic preferences of a species can be used to predict areas where a species might be expected to occur. Previously, some climatic models have been developed to define the potential distribution of different plant forms ( Box, 1981 ), of forest trees for optimized cultivation ( Booth, 1991 ; Booth & Jones, 1998 ), of possible locations of species ( Box ., 1993 ; Skov & Borchsenius, 1997 ; Skov, 2000 ) and of weeds in newly colonized areas ( Howden, 1985 ; Panetta & Dodd, 1987 ; Panetta & Mitchell, 1991 ). The cited prediction systems use comparisons of weather station data with simple distribution data of where the species is recorded to construct climate profiles for the species. Inadequate sampling, especially where climatic gradients are steep or complex, is a possible problem inherent to these correlative models. A set of localities, or even considerable parts of the known range, often do not contain data representative of the full climatic tolerance of the species. Recently, some of these model systems became available as user‐friendly computer programs (Climate™: developed from concepts contained in the Bioclim™ Prediction System, Climex™, or floramap ™© 1999 International Centre for Tropical Agriculture [CIAT]). For Bioclim™ see Busby (1991 ), for Climex™ see Sutherst & Maywald (1985 ). The main advantage of the bioclimatic modelling systems mentioned above is that they are easy to handle and thus they allow decision makers or land managers to obtain acceptable models for considerable numbers of plant species in a reasonable time frame. However, once a plant is recognized as behaving aggressively or invasively, decision makers, conservationists or land managers may be interested in assessments which are geographically more accurate for both their selected species and the region for which they are responsible, even if the method is more labour consuming. For this special purpose, we would like to introduce a method for determining the spatially different likelihood of long‐term establishment of introduced species. In order to achieve more detailed results, more time and data are necessarily required. A case study for Alliaria petiolata (M. Bieb.) Cavara & Grande in North America is used to present this bioclimatic modelling method, which is based not only on climatic tolerances at a set of localities but also on frequency distributions along climatic gradients in the general native distribution area. Species and methods Long‐term probability zones of potential distribution in North America were assessed for Alliaria petiolata (M. Bieb.) Cavara & Grande (garlic mustard). Alliaria petiolata is a tall, short‐lived herb in the Brassicaceae, native to western Eurasia. Used as a culinary herb and because of its perceived medicinal value, garlic mustard was probably introduced to North America by early European settlers and has become one of the most rapidly expanding invasive plants of woodland habitats in eastern North America. It invaded and now dominates the forest ground layer in many regions from New England through the Midwest, and from southern Ontario to Tennessee. Its spread through forests of the eastern and Midwestern United States and Canada has caused great concern ( Blossey ., 2001 ), and thus garlic mustard is an example of a species that may justify the use of greater time and data intensive methods to determine its potential distribution, like the approach we present here. In Europe, A. petiolata is most common in habitats of relatively high air humidity. Slightly shaded places beside rivers and at roadsides and tracks are reported as optimal habitats from regions with a humid maritime climate (British Isles, Grime ., 1988 ). In more arid regions (Sicily and Greece) the species occurs mainly in shaded habitats and mountainous areas where it exhibits a marked bias towards north facing slopes ( de Halácsy, 1901 ). In North America, A. petiolata most frequently occurs in moist, shaded soils of river floodplains, forests, roadsides, edges of woods, forest openings and trails. Distributional data and climate data The general distribution data is based on maps published by Meusel . (1965 ), Jäger (1970 ), and de Bolós & Vigo (1990 ) and was completely revised using a great amount of recently published data, especially for the compilation of the newly colonized regions and localities in North America. The main advantage of the mapping method used is that it allows the possibility to draw on very heterogeneous data sources, ranging from simple presence–absence indications for large regions to detailed dot maps showing single locations. The approach is useful for finding distributional gaps, outposts, exclaves and sectors where occurrence is continuous (for methodology, see also Hoffmann & Welk, 1999 ; Hoffmann, 2001 ). Global climate data (monthly means of precipitation and temperature) were obtained from the Potsdam Institute of Climate Impact Research (CLIMATE database version 2.1; W. Cramer, Potsdam, personal communication). The distribution map was digitized, and subsequently transformed into a grid of the same resolution as the climate data. Calculations were performed using the program Arc/Info® ( ESRI, 1992 ). Climate analysis The investigations of Bartlein . (1986 ) and Huntley . (1995 ) showed that the distributional range of a species may be considered on a global scale as a function of the endogenous ecological constitution of the species and the climate. Clearly, plant distribution is influenced by and interacts with the environment in a very complex way. However, on large spatial scales it is sufficient and suitable to use statistical models, which summarize the effect of the interaction between climate and plant distribution using a smaller number of parameters. The study by Huntley . (1995 ) substantiates evidence that distributional ranges may be modelled using only climatic factors. The method of the climatic response surfaces ( Bartlein ., 1986 ) has been modified to obtain the position of the species in the world's climate system (see Hoffmann, 2001, 2002 ). Intervals of the climate data used are defined as follows: 0.1 K for temperature and 1 mm for precipitation. Frequency of occurrences was recorded by counting the number of occupied grid cells of the species within the defined interval of temperature and precipitation, respectively. The data is presented in frequency diagrams. Frequency diagrams These diagrams show the number of grid cells (percentage) occupied by a species, along the chosen climatic gradient ( Fig. 1 ). Despite the great diversity of individual shapes, some common characteristics can be observed in the curves (see also Hoffmann, 2000, 2001 ). 1 Frequency diagrams of Alliaria petiolata . The x‐axes show the climatic range of the respective mean monthly precipitation (1a, mm) and temperature (1b, °C) within which A . petiolata occurs. The y‐axes show the percentage of grid cells occupied by the species. For reasons of clarity, precipitation values above 200.5 mm are omitted from the graphs. Shape and slope of a curve are the most important characteristics of the diagrams. The point where the slope changes from steep to flat may indicate the position of a range limiting factor (critical level) or a geographical barrier (mountains, oceans and deserts), whereas a gentle slope may point to the fact that this climatic factor is of less importance for the limitation of the range. Isolated records in the tails of the graph may belong to various categories, e.g. relicts, outposts, occurrences in unusual habitats, temporary occurrences due to synanthropic dispersal, or errors due to inaccurate climate data. Climatic modelling To obtain a climate‐based model of the spatial distribution of A. petiolata , each of the 24 climate variables (monthly means of precipitation and temperature) are analysed. The first step is to count and visualize the number of populated grid cells along climatic gradients (from minimum to maximum values) in frequency diagrams for the general native distribution range. Shape and ascent of the resulting 24 frequency diagrams are examined to find out the strongest correlations between range limits and the regarded climatic variables. This could have been done using mathematical methods, but because the distributional data used to map the range borders analysed varies strongly in accuracy and relevance, a manual method based on personal experience was chosen. To improve the fit of spatial models the removal of observations that are beyond the last positive observation by greater than 1% of the sample is recommended ( Austin ., 1995 ). For the modelling of general distributions a further step is necessary. To increase predictive power, the input data of the model (24 climate grid‐layers) is gradually optimized by omission of observations that are beyond the range of the examined critical levels (see above). All grid cells within the chosen interval were selected in the corresponding monthly climate grid and assigned a value of ‘1’. Cells from outside the chosen interval were labelled with a value of ‘0’. Finally, the resulting 24 ‘clipped’ binary grids were added. This overlay of binary maps results in a cumulative map showing the spatial pattern of the number of climatically supportive months — the climatic model. The resulting model fit was analysed by examining the number of observations (populated grid cells) correctly predicted by the model variant (C), as well as the proportion of omission and commission errors, which respectively predict the species to be absent when it is present, and vice versa (A, B). This examination was done using rigid calculations of similarity (Jaccard‐Index; J = C × 100/A × B percentage) between the native range and the spatial pattern that was created using climate variables for Eurasia. The application of this index allows exclusion of the vast number of grid cells, which remain vacant in both the model and the actual area of the plant in northern Eurasia (D), and thus computes only the exact congruence between the two patterns, regardless of the size of the surrounding modelling arena. In this way threshold values for the regarded climate intervals that minimized the omission and commission errors were chosen manually. Results Distribution Map The native range of A. petiolata is shown in Fig. 2(A) . The TNC‐Element Stewardship Abstract by Nuzzo (2000 ), which is otherwise well investigated, describes the native range incorrectly as extending eastwards from England to Czechoslovakia (cited also in the TNC‐weed alert by Morisawa, 2000 ). The species behaves as an apophyte in most parts of its native distribution area (apophytes are elements of the natural vegetation that benefit from human influences like disturbance or eutrophication). Alliaria petiolata benefits especially from the effects of increasing alkalinity and nutrient content of soils and is considered an expanding and abundant species almost everywhere within its native distribution area. 2 Native, neophytic, and climatically modelled range of Alliaria petiolata . (A) Distribution range of A. petiolata in Eurasia (native, archeophytic) and North America (introduced, neophytic). Open circles indicate geographically imprecise records of occurrence. (B) Climatically modelled range and long‐term probability zones for invasion in North America. The legend indicates the percentage of similarity between the climate of the modelled native range and the climate in North America based on the number of supportive (climatically suitable) months (zone I: 100%, zone II: > 96%, zone III: > 92%, zone IV: > 88%, zone V: ≤ 88%). According to Jäger (1970 ) the range of garlic mustard belongs to the Eupatorium ‐type of Mediterranean–Middle European plant distribution areas. The species of this type (e.g. Eupatorium cannabinum , Iris pseudacorus , Ranunculus ficaria , Crataegus monogyna , Rumex obtusifolius and Prunus spinosa ) are elements of the temperate broadleaved forest zone and prefer relatively moist habitats (floodplain forests, riverbanks, margins of lakes and ponds). There are naturalized occurrences of most of the above‐mentioned species in North America and they are considered locally as invasive or potentially invasive plants. The altitudinal distribution increases from north (400 m in south Norway, 350 m in the British Isles) to south (900–1600 m in Iraq, 1100–2500 m in Tadzhikistan, and 2200–3100 m in Nepal). The neophytic North American range is also shown in Fig. 2(A) . The new distribution area has grown exponentially since introduction, and by 2000 the species had spread to 34 US states and 4 Canadian provinces. For this reason all data concerning distribution has to be considered as preliminary until the species has occupied its full potential range. Alliaria petiolata is most widespread in the Midwestern and north‐eastern United States, in south‐western Ontario, and in the St. Lawrence Valley. Infrequent collections are reported from mountain states (Colorado, Utah), and sub‐boreal regions (Gaspé/Quebec). In the Pacific Northwest, A. petiolata is established only in Portland (Oregon), in several locations around Seattle (Washington) and in Victoria and Vancouver (British Columbia, first report from 1948). North America is the main area for new occurrences for A. petiolata . Only one locality has been known of since 1893 in New Zealand ( Webb ., 1988 ). Cavers . (1979 ) reported newly established occurrences for Sri Lanka, but this could not be confirmed by recent floristic literature ( Philcox, 1995 ). The inclusion of North Africa and India into the list of invaded regions in the ‘Weed Alert’ of The Nature Conservancy ( Morisawa, 2000 ) is another error because North India and coastal regions of North Africa are parts of the native range of the species. Frequency diagrams The diagrams show the percentage of occupied grid cells of A. petiolata along the chosen climatic gradients ( Fig. 1 ). For reasons of clarity, precipitation values above 200.5 mm are not represented in the graphs. The omitted data range contains only single frequency values, generally lower than 0.02%. Several diagrams reflect the above‐mentioned characteristics of the curves. Graphs with one distinct peak are shown, e.g. for precipitation variables. Flat graphs with several smaller peaks can also be observed, e.g. for winter temperatures (December to February). In addition, very steep ascents of the graphs are to be found for precipitation in the spring months (March to May). The model The modelled west Eurasian range of A. petiolata is shown in Fig. 2(B) . The percentage agreement between the two patterns in the modelling arena ‘Eurasia North of the tropic of cancer’ is 99.6%. The similarity between the observed and modelled range is approximately 60% according to the Jaccard Index. It would of course be possible to compute Indices like Cohen's Kappa, or Chi‐Square. However, in taking the vast number of vacant grid cells into account, they are less rigid and thus lead to similarity values over 98%. These values may underline the excellent fit of the distribution area modelled using this rather labour‐intensive method, but are less appropriate for critical model fit assessment. The short‐term probability of occurrence is influenced by the environment in a very complex way. It depends on factors like seed size, seed production, seed type, or human vectoring and may be accelerated by effects of ‘ecological release’. However, in a first approximation the number of supportive months may be considered as a measure of the long‐term probability of a plants’ occurrence. Therefore, the potential area for invasion is divided into five probability zones. These zones are based on the degree of congruence (zone I: 24 supportive months [100%], zone II: 23 months [> 96%], zone III: 22 months [> 92%], zone IV: 21 months [> 88%], zone V: less than 21 months [≤ 88%]) between the climates in North America and the climatically modelled native distribution area of the species. A compact core area (zone I, 100% congruence) is conspicuous in the north‐eastern United States and adjacent Canada. It ranges from Prince Edward Island in the north‐east to Minnesota in the north‐west, from here to Iowa in the south‐west and through Illinois and Kentucky to western North Carolina in the south‐east. The main part of this area is congruent with the Appalachian floristic province ( Takhtajan, 1986 ). The northern limit of the core area reaches only slightly into the southern parts of the boreal forest climate and vegetation (Ontario, Quebec). A small western part of the area reaches into regions characterized as the oak–savanna ecotone with the eastern deciduous forest (Iowa). The south‐eastern Coastal Plain with a warm temperate‐subtropical climate (permanently humid and with hot summers) and the treeless prairies and plains of the west are excluded from the completely homologous climatic region. A second sizeable potential distribution area of zone I is shown in the Pacific Northwest region. It ranges from south‐western British Columbia along the coast to south‐eastern Oregon and through the northern tip of Idaho to north‐west Montana. On a map of this scale, Washington State belongs nearly entirely to this second climatically supportive region of the potential A. petiolata ‐distribution area. Small isolated parts of this western distribution range are localized in central Alberta, southern Idaho and adjacent Utah. The main part of this area belongs to the Rocky Mountain Region of Takhtajan (1986 ). The Vancouverian floristic province, which is the range of Pacific Coast conifer forests (broadleaved trees are more conspicuous than in the Rocky Mountain ranges), is the most suitable area for A. petiolata . It is characterized by an oceanic cool temperate forest climate ( Bruillet & Whetstone, 1993 ). The Intermountain grassland with its dry steppe climate and cold winters along with the Mediterranean woodlands and scrublands of California are excluded from zone I. Table 1 lists US states and Canadian provinces, which could be affected by occurrences of A. petiolata , according to the climate model prediction. The table contains only zone‐regions with a climatic congruence or long‐term invasion probability of 100%. 1 Regions 100% climatically congruent with the native range of Alliaria petiolata State/Province Subdivision (county, district, part) I. North American Atlantic Region Ontario Easternmost Rainy River, not N of line Sudbury — Timiskaming Quebec Not N of line Témiscamingue — Montmagny Prince Edward Island Complete except NE‐tip New Brunswick Not Madawaska, Restigouche Nova Scotia Not NE of line Colchester — Halifax Maine Complete New Hampshire Complete Vermont Complete Massachusetts Complete Rhode Island Complete Connecticut Complete New York Complete New Jersey Not along the south coast (Salem to Cape May) Maryland Not E of line Frederick‐Carroll Pennsylvania Not York, Lancaster, Chester Ohio Complete Indiana Not SW of line Vigo — Dearborn Michigan Complete Wisconsin Not N of line Vernon — Florence Minnesota Fillmoore, Winona, Houston Iowa Not NW of line Howard — Pottawattamie, not Appanoose to Lee Illinois Not S of line Henderson — Edgar Kentucky Not W of line Mason — Harlan West Virginia Complete Virginia Not E of line Loudoun — Grayson North Carolina Not E of line Transylvania — Stokes Tennessee Not W of line Folk — Johnson II. Rocky Mountain Region British Columbia Victoria, Capital, Cowichan Valley, Vancouver, Sunshine Coast Alberta Calgary Washington Complete except Whatcom, NW‐Clallam Oregon Not SE of line Coos — Hood River Idaho Not S of Nez Perce — Clearwater, but local in Lincoln and Jerome Montana Mineral, Sanders, Lincoln Utah Local in central Box Elder Regions of lower climatic congruence are shown in the map ( Fig. 2B ). The area of zone II builds a narrow stripe surrounding the core area. Zone III is larger and exceeds the compact Atlantic sector of the potential range more distinctly, especially in the north‐east (Gaspé Peninsula), west (Minnesota) and in the south (Kansas, Missouri, Illinois). The western, Pacific part of zone III consists mainly of three adjacent areas (the region surrounding zone II, the western slope of the Rocky Mountains in Alberta and adjacent regions, and a compact area ranging from northern Utah to southern Idaho). Isolated localities are shown in central British Columbia (Nechako and Fraser River Valleys) and northern California (Klamath and North Coast Ranges). Zone IV is much larger than the climatic core area. It comprises Newfoundland and an isolated area south of Lake Winnipeg. Large parts of the Prairie Province, including most of South Dakota and Nebraska connect the compact Atlantic Region with the more scattered occurrence of climatically similar areas of the Rocky Mountain Region. Here, zone IV ranges from northern California and Colorado in the south to British Columbia and Alberta in the north. Zone V contains the remaining regions of North America, which are supposedly unfavourable for successful establishment of A. petiolata . Comparison of current and potential range in North America Evaluation of the quality of the model must primarily be based on occurrences reported from beyond the predicted zones (omission errors). Due to the continuing expansion of the relatively ‘young’ distribution area, it cannot be assumed that invasion is complete. However, large gaps found within areas which are almost completely invaded must be included in a critical analysis of the quality of the model. At a first glance, the comparison of the current and potential range shows one main result. The spread of A. petiolata in eastern North America has reached almost every climatically suitable region, at least with isolated occurrences. Repeated introduction ( Meekins ., 2001 ) and human‐facilitated dispersal enable the species to ‘fill’ its potential distribution area very quickly. But within this potential range, garlic mustard has not yet invaded every possible habitat, and thus, the process of invasion is far from over. A large proportion of the current occurrences ( c. 70%) are situated in zone I; 6%, 7%, 14% and 3% of the currently known localities are situated within zones II, III, IV and V, respectively. Table 2 provides an overview of the 30% of discovered occurrences situated outside of the predicted core area (zone I). ‘Climatic outliers’ are reported more frequently from the southern section of the invaded area, compared to the isolated, locally restricted occurrences in the northern part of the area and the southern Rocky Mountains. In addition to isolated outposts in Oklahoma, Arkansas, Georgia and North Carolina, much more numerous occurrences are to be found particularly in Kansas and Tennessee, and probably in Illinois and Missouri. 2 Recent Alliaria occurrences (dots) outside zone I (zone of complete climate suitability) State/Province Zone II III IV V Notes I. North American Atlantic Region Quebec • One record from Gaspé, not confirmed since 1891 Minnesota • Reported from Clay and Dakota county North Dakota • Reported only from Cass county South Dakota • Reported only from Brooking county Nebraska • Geographically imprecise record of occurrence Kansas • • • Reported from moist riparian woods in 13 counties Missouri • • Underrepresented, occurs in many counties Oklahoma • Reported from Delaware and Adair county Arkansas • • Benton, Washington, Franklin (and Pulaski?) county Illinois • • • At least 41 counties Indiana • Reported from a few counties in the south Kentucky • • • Union and Logan county in zone IV Tennessee • • • At least 13 counties in zone IV Georgia • Geographically imprecise record of occurrence North Carolina • Reported only from Rockingham county Virginia • • Reported from many counties Maryland • Densely populated area II. Rocky Mountain Region Colorado • Reported from one location in 1952, 1958, 1966 Utah • Reported from one location in 1971, 1983, 1984 All reported occurrences in the Pacific North‐west are situated within ‘zone I’ as predicted by the climatic model. However, this part of the range comprises only a few localities, and thus the occupation of the western subsection has to be considered as being in an initial phase of invasion. Discussion The data For identifying and predicting species’ habitat preferences and requirements in relation to climatic variables, the quality of spatial data about climate and distribution is crucial. On a global scale, floristic knowledge differs greatly from region to region (see Jäger's map in Frodin 2001 ). In some regions data is sparse in terms of time (short normal periods) and space (density of weather stations). The validity of the interpolated climate data is somewhat limited, especially in these regions. Another problem in modelling plants distribution areas is the use of different types and numbers of climatic parameters. It might be argued that it is the extreme values in the year that are most critical to plants. Indeed, many modellers in this field use metrics, such as minimum monthly temperatures or, on an annual scale, minimum temperatures of the coldest month. Extreme values are, however, not always particularly robust measures. The use of mean values in place of extreme values can be justified particularly with oscillations of the plants distribution areas, especially as climate is likewise subjected to oscillations. Mean values of many years of production and reproduction are crucial for the fitness of the populations. The fact that modelling with both extreme values and mean values produces good results is due to the close relationship (parallelism) between monthly extremes and mean values. Finally, on large spatial scales, range boundaries result from the complete life cycle of the plants, which depend on the climate over the whole year and thus includes the possibility of regeneration as well as any temporary damage. Another problem is the possible life cycle changes of the species due to genetic differentiation in isolated populations or adaptation to new combinations of environmental factors. Thompson . (1999 ) calculated a series of ‘bioclimatic’ variables that were proposed as more direct controls over plant distributions than temperature and precipitation. However, on large spatial scales (e.g. global and Holarctic), these indices might not reflect the important differences between the climatic rhythms of the eastern and western sides of the main landmasses ( Jäger, 1995 ); a situation that favours the use of the basic climatic parameters. Artificially limiting the numbers of climatic parameters may result in poorly fitting models, such as in Beerling . (1995 ). The FloraMap™ system ( CIAT, 1999 ), which is designed for mapping large‐scale climate probabilities mainly for the tropic regions of the world, uses monthly average diurnal temperature ranges in addition to rainfall and temperature data (altogether 36 climate variables). For a spatially more explicit large scale modelling of plant ranges, high quality distribution maps and a wide array of basic climatic parameters are necessary. Frequency diagrams Comparing the data of Baskin & Baskin (1992 ), Byers & Quinn (1998) ; Cavers . (1979 ), Grime . (1988 ), Meekins (2000), Nuzzo (1993) ; Rejmánek (2000) and Trimbur (1973) with the diagrams ( Fig. 1 ), it appears that life cycle data are reflected to a large extent. At the time of germination (late February until beginning of April in northern localities) there must be an average minimum monthly precipitation of 25 mm at temperatures between 0 °C and 12 °C. In the plant's native distribution range, optimal mean values are 5–10 °C and 30–60 mm. In May (main time of flowering), precipitation must reach an average minimum of 30 mm, combined with average monthly temperatures ranging from 8 °C to 18 °C. For a successful establishment of the young plants, the elongation of the flowering stems, the development of flowers and seed formation, a frost‐free time of at least 4 months is probably necessary. The development of the plants is clearly temperature related; long‐term studies have documented that A. petiolata flowers open earlier with increasing temperature ( Fitter ., 1995 ). In July and August, the young rosettes of A. petiolata expose minimal leaf size. Adult plants are in the life cycle phase of seed ripening at this time and therefore may need a certain minimum temperature, whereas rainfall is of lesser importance. In the plant's native distribution range, mean values of the July temperature (between 16 °C and 23 °C) appear to be most favourable for the species. Autumn precipitation does not seem to be of great importance for the distribution of A. petiolata . The relevant diagrams ( Fig. 1A ) show a modest increase at lower amounts of precipitation. However, only a few occurrences are observed at precipitation values of less than 20 mm for September and of less than 30 mm for October The diagrams of temperature ( Fig. 1B ) indicate that optimal temperatures have to be well above the freezing in these months. It is possible that these temperatures are necessary for the further growth of the plants and the formation of new rosette leaves. During the winter, A. petiolata may require temperatures low enough to break seed dormancy through cold stratification. If daytime temperatures are above freezing the rosettes can continue growth and biomass production. Below a sufficiently thick snow cover, the rosettes can survive even periods of severe frost. The diagrams for winter temperature ( Fig. 1B ) show a relatively wide range of values (26–28 K for December, January and February), but mean temperatures below 5 °C seem to be necessary. This value is not necessarily the prerequisite for breaking seed dormancy; however, these minimal values are the expression of temperature conditions that enable this physiological process. On the other hand, these low temperatures may prevent the development of competing evergreen vegetation. In summary, it can be said that the native distribution range of A. petiolata is roughly approximated by climatic factors, such as moisture (≥ 500 mm annual rainfall), sufficient warmth during the main time of development (the 9 °C May isotherm), sufficient cold winters (the 6 °C January isotherm), and a sufficient time span for plant development (symbolized by the isoline of > 120 frost‐free days). The climatic model The climatic model of the distribution range of A. petiolata describes both the natural and the adventitious, neophytic range of the species with a large degree agreement. This may confirm the assumption that, on a global scale, climate is the main range‐limiting factor for this species. However, the agreement of the model and the observed native range is not complete, especially in mountainous regions. These mismatches are caused by differences between the altitudinal distribution of the plant and the interpolated elevation of the underlying climate grids. Large distribution ranges, partly limited by seacoasts, are generally reducing the possibilities of climatic modelling. For example, it is impossible to characterize the potential western range boundary in Europe with climatic variables alone. Additionally, for relatively large parts of its range (e.g. Turkey) only very sparse floristic data are available. Thus, the agreement with the model produced from climatic data could possibly be higher here. In the north‐eastern part of the area (Byelorussia and adjacent north‐west Russia) edaphic factors probably mask the effect of climate. Here, unsuitable soil conditions (wet and acidic, see above) seem to produce a large gap in an otherwise climatically suitable region. The very detailed distributional data for north‐western Europe (Denmark and Scotland) emphasis the importance of soil factors, which modify or override the influences of climate in these areas. In North America, the reasons for the differences between the current distribution range and that of the climatic model are clear. The large, uninvaded areas north of the Great Lakes and the valley of the St. Lorenz Stream may be attributed to the prevalence in these areas of shallow soils derived primarily from ancient acidic bedrock, which are associated predominantly with coniferous forest vegetation. Although some vigorous Alliaria populations are reported from habitats upon the Canadian Shield ( IPCP, 1999 ), it should be assumed that A. petiolata reaches its climatically defined northern boundary only locally within areas of human settlement (parks, hedges, edges of road and gardens). Its limited occurrence in New Brunswick is also within an area of calcareous bedrock. More extensive distributions of A. petiolata can only be expected in the Mixedwood Plain Ecozone ( Barbour & Christensen, 1993 ), a biogeographical region dominated by limestone bedrock, where soils and water bodies tend to be alkaline. Certainly, probability zones II–IV will remain devoid of A. petiolata in Canada. This prediction is strongly supported by the long‐term, relatively limited distribution of invading A. petiolata in Canadian compared with the situation in the adjacent United States. The second remarkably sparsely populated area is the Pacific Northwest section of the predicted range. At this point in time, A. petiolata may be at an early stage of invasion (the ‘lag‐phase’, see Jäger, 1995 ), as the first report dates back to 1948. The area of distribution predicted by the climatological model depends on the interpolated climate surfaces and does not satisfactorily reflect the great regional variation caused by the very diverse topography. Beside possible occurrences in ruderal habitats, cultivated land and other man‐made vegetation types, suitable growth conditions seem in fact to be restricted to small parts of the predicted area. In the forests of the Pacific, or Humid Transition area, dense coniferous overstoreys above rich ericaceous shrub, herb and moss canopies offer very limited opportunities for A. petiolata to invade natural habitats of the region. Favourable growth conditions may occur along river valleys of lowlands and in humid grasslands. The Cascade Mountains form a sharp and efficient climatic barrier across the predicted range. For the arid eastern part, with predominantly sagebrush, bunchgrass and yellow pine vegetation, suitable growth conditions are hardly encountered. Here, A. petiolata may invade habitats with extrazonal vegetation, like river valleys, moist places, shady, north facing slopes and several mountainous habitat types. However, more probable are occurrences in urbane areas and disturbed, nutrient‐rich places around human settlements. The next conspicuous difference between the modelled and the observed distribution are the relatively numerous Alliaria occurrences in probability zones III–V reported from Kansas. The localities are concentrated in the north‐eastern part of the state. The linearly arranged dots (see Fig. 2A ) indicate the association with riverside forest strips of the wooded floodplains along Missouri and Kansas Rivers, and their tributaries. These are extrazonal vegetation types whose distribution is predominantly influenced by topography, geology and soil characteristics, rather than by climatic variables. The same interpretation may apply to the occurrences reported from Tennessee. According to Baskin & Baskin (1992 ), possible life cycle changes (see above) were not detected in this region. Further discussion about differences between the modelled and the actual distribution seem unnecessary, in so much as areas predicted by the model to be suitable for Alliaria but which are uninhabited might either not have been reached by diaspores, or were predicted falsely as suitable due to the locally restricted predictive power of climatic variables. Another, and more speculative, explanation for the absence of the plant from climatically suitable areas as predicted by the model might be due to the origin of the introduced diaspores. It is possible that A. petiolata is locally adapted to particular climate conditions in its natural range. From the entire source population of A. petiolata constituting its natural distribution range, a random sample has been transported by humans to North America (see Meekins ., 2001 ). This sample may not be able to spread with equal speed to all regions that are predicted as climatically suitable. In summary, it can be stated that not all of the predicted areas will necessarily be invaded by Alliaria . The probability zones show only the spatial framework within which — if further requirements and preconditions are met (such as the presence of human vectors) — Alliaria has a high (climatically defined) likelihood of inhabiting in the long term. The presence of non‐acidic soils, deciduous wood vegetation and mesic habitat conditions seems to be a particularly important combination of additional prerequisites. If they are fulfilled, habitats with extrazonal vegetation and man‐made habitats can become invaded locally, even if they are in zones III and IV. However, wide‐ranging invasions of natural, zonal vegetation types are not predicted to occur here. Usually, predictions are easy to make but difficult to test. Pyšek (2001 ) states that most of the predictions in invasion ecology were made too recently to know whether they are accurate or not. The research on well known, non‐indigenous European species in North America, and vice versa, provides us with opportunities for long‐term field tests because many of the species had have enough time to reach even the remote parts of their potential distribution ranges on their ‘new’ continents. With a review of the results of investigations on a large number of species with different life history strategies, life forms and native range types, our understanding of the different capacities of climatic range models for predicting invasiveness could be improved. Acknowledgments This research was supported by the Deutsche Forschungsgemeinschaft. The manuscript was improved by the helpful suggestions of Karsten Wesche, Linda Scott and two anonymous referees. The authors would like to express thanks to W. Cramer (Potsdam Institute of Climate Impact Research) for providing the climate data. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Diversity and Distributions Wiley

Present and potential distribution of invasive garlic mustard ( Alliaria petiolata ) in North America

Loading next page...
 
/lp/wiley/present-and-potential-distribution-of-invasive-garlic-mustard-alliaria-rIPxum7Kmk

References (44)

Publisher
Wiley
Copyright
Copyright © 2002 Wiley Subscription Services, Inc., A Wiley Company
ISSN
1366-9516
eISSN
1472-4642
DOI
10.1046/j.1472-4642.2002.00144.x
Publisher site
See Article on Publisher Site

Abstract

Introduction Occurrences of non‐indigenous plant species have become components of the vegetation of most regions of the world as a result of increasing species exchange between continents and vegetation transformation by man. At a global scale, the problem of invasive alien plants has increased greatly over the past decades ( Perrings ., 2000 ). Procedures for assessing the potential invasiveness of non‐indigenous plants are urgently needed for objective and scientifically based quarantine regulations for preventing introduction or further spread, or for making informed decisions about deliberate introductions. Recent systems for predicting invasiveness of plants have been developed for parts of the non‐indigenous floras of North America ( Reichard & Hamilton, 1997 ), Australia ( Pheloung ., 1999 ), and South Africa ( Tucker & Richardson, 1995 ). They are based on the analysis of a number of biological and ecological characters of the species in question (life history, habitat characteristics, invasiveness elsewhere and biogeography). However, according to ( Cronk & Fuller, 1995 ) environmental factors, such as climate, and particularly seasonality of climate, are of special importance because the relationship between climate and the distribution of plants is well documented ( Grace, 1987 ; Woodward, 1987 ). Climate may be considered as setting the broad limits for plant distribution, while other factors such as geology, soils and competition will determine the presence or absence of a species in a particular area and on a finer regional or local scale. Accepting that climate usually limits the range of a species, analysis of the climatic preferences of a species can be used to predict areas where a species might be expected to occur. Previously, some climatic models have been developed to define the potential distribution of different plant forms ( Box, 1981 ), of forest trees for optimized cultivation ( Booth, 1991 ; Booth & Jones, 1998 ), of possible locations of species ( Box ., 1993 ; Skov & Borchsenius, 1997 ; Skov, 2000 ) and of weeds in newly colonized areas ( Howden, 1985 ; Panetta & Dodd, 1987 ; Panetta & Mitchell, 1991 ). The cited prediction systems use comparisons of weather station data with simple distribution data of where the species is recorded to construct climate profiles for the species. Inadequate sampling, especially where climatic gradients are steep or complex, is a possible problem inherent to these correlative models. A set of localities, or even considerable parts of the known range, often do not contain data representative of the full climatic tolerance of the species. Recently, some of these model systems became available as user‐friendly computer programs (Climate™: developed from concepts contained in the Bioclim™ Prediction System, Climex™, or floramap ™© 1999 International Centre for Tropical Agriculture [CIAT]). For Bioclim™ see Busby (1991 ), for Climex™ see Sutherst & Maywald (1985 ). The main advantage of the bioclimatic modelling systems mentioned above is that they are easy to handle and thus they allow decision makers or land managers to obtain acceptable models for considerable numbers of plant species in a reasonable time frame. However, once a plant is recognized as behaving aggressively or invasively, decision makers, conservationists or land managers may be interested in assessments which are geographically more accurate for both their selected species and the region for which they are responsible, even if the method is more labour consuming. For this special purpose, we would like to introduce a method for determining the spatially different likelihood of long‐term establishment of introduced species. In order to achieve more detailed results, more time and data are necessarily required. A case study for Alliaria petiolata (M. Bieb.) Cavara & Grande in North America is used to present this bioclimatic modelling method, which is based not only on climatic tolerances at a set of localities but also on frequency distributions along climatic gradients in the general native distribution area. Species and methods Long‐term probability zones of potential distribution in North America were assessed for Alliaria petiolata (M. Bieb.) Cavara & Grande (garlic mustard). Alliaria petiolata is a tall, short‐lived herb in the Brassicaceae, native to western Eurasia. Used as a culinary herb and because of its perceived medicinal value, garlic mustard was probably introduced to North America by early European settlers and has become one of the most rapidly expanding invasive plants of woodland habitats in eastern North America. It invaded and now dominates the forest ground layer in many regions from New England through the Midwest, and from southern Ontario to Tennessee. Its spread through forests of the eastern and Midwestern United States and Canada has caused great concern ( Blossey ., 2001 ), and thus garlic mustard is an example of a species that may justify the use of greater time and data intensive methods to determine its potential distribution, like the approach we present here. In Europe, A. petiolata is most common in habitats of relatively high air humidity. Slightly shaded places beside rivers and at roadsides and tracks are reported as optimal habitats from regions with a humid maritime climate (British Isles, Grime ., 1988 ). In more arid regions (Sicily and Greece) the species occurs mainly in shaded habitats and mountainous areas where it exhibits a marked bias towards north facing slopes ( de Halácsy, 1901 ). In North America, A. petiolata most frequently occurs in moist, shaded soils of river floodplains, forests, roadsides, edges of woods, forest openings and trails. Distributional data and climate data The general distribution data is based on maps published by Meusel . (1965 ), Jäger (1970 ), and de Bolós & Vigo (1990 ) and was completely revised using a great amount of recently published data, especially for the compilation of the newly colonized regions and localities in North America. The main advantage of the mapping method used is that it allows the possibility to draw on very heterogeneous data sources, ranging from simple presence–absence indications for large regions to detailed dot maps showing single locations. The approach is useful for finding distributional gaps, outposts, exclaves and sectors where occurrence is continuous (for methodology, see also Hoffmann & Welk, 1999 ; Hoffmann, 2001 ). Global climate data (monthly means of precipitation and temperature) were obtained from the Potsdam Institute of Climate Impact Research (CLIMATE database version 2.1; W. Cramer, Potsdam, personal communication). The distribution map was digitized, and subsequently transformed into a grid of the same resolution as the climate data. Calculations were performed using the program Arc/Info® ( ESRI, 1992 ). Climate analysis The investigations of Bartlein . (1986 ) and Huntley . (1995 ) showed that the distributional range of a species may be considered on a global scale as a function of the endogenous ecological constitution of the species and the climate. Clearly, plant distribution is influenced by and interacts with the environment in a very complex way. However, on large spatial scales it is sufficient and suitable to use statistical models, which summarize the effect of the interaction between climate and plant distribution using a smaller number of parameters. The study by Huntley . (1995 ) substantiates evidence that distributional ranges may be modelled using only climatic factors. The method of the climatic response surfaces ( Bartlein ., 1986 ) has been modified to obtain the position of the species in the world's climate system (see Hoffmann, 2001, 2002 ). Intervals of the climate data used are defined as follows: 0.1 K for temperature and 1 mm for precipitation. Frequency of occurrences was recorded by counting the number of occupied grid cells of the species within the defined interval of temperature and precipitation, respectively. The data is presented in frequency diagrams. Frequency diagrams These diagrams show the number of grid cells (percentage) occupied by a species, along the chosen climatic gradient ( Fig. 1 ). Despite the great diversity of individual shapes, some common characteristics can be observed in the curves (see also Hoffmann, 2000, 2001 ). 1 Frequency diagrams of Alliaria petiolata . The x‐axes show the climatic range of the respective mean monthly precipitation (1a, mm) and temperature (1b, °C) within which A . petiolata occurs. The y‐axes show the percentage of grid cells occupied by the species. For reasons of clarity, precipitation values above 200.5 mm are omitted from the graphs. Shape and slope of a curve are the most important characteristics of the diagrams. The point where the slope changes from steep to flat may indicate the position of a range limiting factor (critical level) or a geographical barrier (mountains, oceans and deserts), whereas a gentle slope may point to the fact that this climatic factor is of less importance for the limitation of the range. Isolated records in the tails of the graph may belong to various categories, e.g. relicts, outposts, occurrences in unusual habitats, temporary occurrences due to synanthropic dispersal, or errors due to inaccurate climate data. Climatic modelling To obtain a climate‐based model of the spatial distribution of A. petiolata , each of the 24 climate variables (monthly means of precipitation and temperature) are analysed. The first step is to count and visualize the number of populated grid cells along climatic gradients (from minimum to maximum values) in frequency diagrams for the general native distribution range. Shape and ascent of the resulting 24 frequency diagrams are examined to find out the strongest correlations between range limits and the regarded climatic variables. This could have been done using mathematical methods, but because the distributional data used to map the range borders analysed varies strongly in accuracy and relevance, a manual method based on personal experience was chosen. To improve the fit of spatial models the removal of observations that are beyond the last positive observation by greater than 1% of the sample is recommended ( Austin ., 1995 ). For the modelling of general distributions a further step is necessary. To increase predictive power, the input data of the model (24 climate grid‐layers) is gradually optimized by omission of observations that are beyond the range of the examined critical levels (see above). All grid cells within the chosen interval were selected in the corresponding monthly climate grid and assigned a value of ‘1’. Cells from outside the chosen interval were labelled with a value of ‘0’. Finally, the resulting 24 ‘clipped’ binary grids were added. This overlay of binary maps results in a cumulative map showing the spatial pattern of the number of climatically supportive months — the climatic model. The resulting model fit was analysed by examining the number of observations (populated grid cells) correctly predicted by the model variant (C), as well as the proportion of omission and commission errors, which respectively predict the species to be absent when it is present, and vice versa (A, B). This examination was done using rigid calculations of similarity (Jaccard‐Index; J = C × 100/A × B percentage) between the native range and the spatial pattern that was created using climate variables for Eurasia. The application of this index allows exclusion of the vast number of grid cells, which remain vacant in both the model and the actual area of the plant in northern Eurasia (D), and thus computes only the exact congruence between the two patterns, regardless of the size of the surrounding modelling arena. In this way threshold values for the regarded climate intervals that minimized the omission and commission errors were chosen manually. Results Distribution Map The native range of A. petiolata is shown in Fig. 2(A) . The TNC‐Element Stewardship Abstract by Nuzzo (2000 ), which is otherwise well investigated, describes the native range incorrectly as extending eastwards from England to Czechoslovakia (cited also in the TNC‐weed alert by Morisawa, 2000 ). The species behaves as an apophyte in most parts of its native distribution area (apophytes are elements of the natural vegetation that benefit from human influences like disturbance or eutrophication). Alliaria petiolata benefits especially from the effects of increasing alkalinity and nutrient content of soils and is considered an expanding and abundant species almost everywhere within its native distribution area. 2 Native, neophytic, and climatically modelled range of Alliaria petiolata . (A) Distribution range of A. petiolata in Eurasia (native, archeophytic) and North America (introduced, neophytic). Open circles indicate geographically imprecise records of occurrence. (B) Climatically modelled range and long‐term probability zones for invasion in North America. The legend indicates the percentage of similarity between the climate of the modelled native range and the climate in North America based on the number of supportive (climatically suitable) months (zone I: 100%, zone II: > 96%, zone III: > 92%, zone IV: > 88%, zone V: ≤ 88%). According to Jäger (1970 ) the range of garlic mustard belongs to the Eupatorium ‐type of Mediterranean–Middle European plant distribution areas. The species of this type (e.g. Eupatorium cannabinum , Iris pseudacorus , Ranunculus ficaria , Crataegus monogyna , Rumex obtusifolius and Prunus spinosa ) are elements of the temperate broadleaved forest zone and prefer relatively moist habitats (floodplain forests, riverbanks, margins of lakes and ponds). There are naturalized occurrences of most of the above‐mentioned species in North America and they are considered locally as invasive or potentially invasive plants. The altitudinal distribution increases from north (400 m in south Norway, 350 m in the British Isles) to south (900–1600 m in Iraq, 1100–2500 m in Tadzhikistan, and 2200–3100 m in Nepal). The neophytic North American range is also shown in Fig. 2(A) . The new distribution area has grown exponentially since introduction, and by 2000 the species had spread to 34 US states and 4 Canadian provinces. For this reason all data concerning distribution has to be considered as preliminary until the species has occupied its full potential range. Alliaria petiolata is most widespread in the Midwestern and north‐eastern United States, in south‐western Ontario, and in the St. Lawrence Valley. Infrequent collections are reported from mountain states (Colorado, Utah), and sub‐boreal regions (Gaspé/Quebec). In the Pacific Northwest, A. petiolata is established only in Portland (Oregon), in several locations around Seattle (Washington) and in Victoria and Vancouver (British Columbia, first report from 1948). North America is the main area for new occurrences for A. petiolata . Only one locality has been known of since 1893 in New Zealand ( Webb ., 1988 ). Cavers . (1979 ) reported newly established occurrences for Sri Lanka, but this could not be confirmed by recent floristic literature ( Philcox, 1995 ). The inclusion of North Africa and India into the list of invaded regions in the ‘Weed Alert’ of The Nature Conservancy ( Morisawa, 2000 ) is another error because North India and coastal regions of North Africa are parts of the native range of the species. Frequency diagrams The diagrams show the percentage of occupied grid cells of A. petiolata along the chosen climatic gradients ( Fig. 1 ). For reasons of clarity, precipitation values above 200.5 mm are not represented in the graphs. The omitted data range contains only single frequency values, generally lower than 0.02%. Several diagrams reflect the above‐mentioned characteristics of the curves. Graphs with one distinct peak are shown, e.g. for precipitation variables. Flat graphs with several smaller peaks can also be observed, e.g. for winter temperatures (December to February). In addition, very steep ascents of the graphs are to be found for precipitation in the spring months (March to May). The model The modelled west Eurasian range of A. petiolata is shown in Fig. 2(B) . The percentage agreement between the two patterns in the modelling arena ‘Eurasia North of the tropic of cancer’ is 99.6%. The similarity between the observed and modelled range is approximately 60% according to the Jaccard Index. It would of course be possible to compute Indices like Cohen's Kappa, or Chi‐Square. However, in taking the vast number of vacant grid cells into account, they are less rigid and thus lead to similarity values over 98%. These values may underline the excellent fit of the distribution area modelled using this rather labour‐intensive method, but are less appropriate for critical model fit assessment. The short‐term probability of occurrence is influenced by the environment in a very complex way. It depends on factors like seed size, seed production, seed type, or human vectoring and may be accelerated by effects of ‘ecological release’. However, in a first approximation the number of supportive months may be considered as a measure of the long‐term probability of a plants’ occurrence. Therefore, the potential area for invasion is divided into five probability zones. These zones are based on the degree of congruence (zone I: 24 supportive months [100%], zone II: 23 months [> 96%], zone III: 22 months [> 92%], zone IV: 21 months [> 88%], zone V: less than 21 months [≤ 88%]) between the climates in North America and the climatically modelled native distribution area of the species. A compact core area (zone I, 100% congruence) is conspicuous in the north‐eastern United States and adjacent Canada. It ranges from Prince Edward Island in the north‐east to Minnesota in the north‐west, from here to Iowa in the south‐west and through Illinois and Kentucky to western North Carolina in the south‐east. The main part of this area is congruent with the Appalachian floristic province ( Takhtajan, 1986 ). The northern limit of the core area reaches only slightly into the southern parts of the boreal forest climate and vegetation (Ontario, Quebec). A small western part of the area reaches into regions characterized as the oak–savanna ecotone with the eastern deciduous forest (Iowa). The south‐eastern Coastal Plain with a warm temperate‐subtropical climate (permanently humid and with hot summers) and the treeless prairies and plains of the west are excluded from the completely homologous climatic region. A second sizeable potential distribution area of zone I is shown in the Pacific Northwest region. It ranges from south‐western British Columbia along the coast to south‐eastern Oregon and through the northern tip of Idaho to north‐west Montana. On a map of this scale, Washington State belongs nearly entirely to this second climatically supportive region of the potential A. petiolata ‐distribution area. Small isolated parts of this western distribution range are localized in central Alberta, southern Idaho and adjacent Utah. The main part of this area belongs to the Rocky Mountain Region of Takhtajan (1986 ). The Vancouverian floristic province, which is the range of Pacific Coast conifer forests (broadleaved trees are more conspicuous than in the Rocky Mountain ranges), is the most suitable area for A. petiolata . It is characterized by an oceanic cool temperate forest climate ( Bruillet & Whetstone, 1993 ). The Intermountain grassland with its dry steppe climate and cold winters along with the Mediterranean woodlands and scrublands of California are excluded from zone I. Table 1 lists US states and Canadian provinces, which could be affected by occurrences of A. petiolata , according to the climate model prediction. The table contains only zone‐regions with a climatic congruence or long‐term invasion probability of 100%. 1 Regions 100% climatically congruent with the native range of Alliaria petiolata State/Province Subdivision (county, district, part) I. North American Atlantic Region Ontario Easternmost Rainy River, not N of line Sudbury — Timiskaming Quebec Not N of line Témiscamingue — Montmagny Prince Edward Island Complete except NE‐tip New Brunswick Not Madawaska, Restigouche Nova Scotia Not NE of line Colchester — Halifax Maine Complete New Hampshire Complete Vermont Complete Massachusetts Complete Rhode Island Complete Connecticut Complete New York Complete New Jersey Not along the south coast (Salem to Cape May) Maryland Not E of line Frederick‐Carroll Pennsylvania Not York, Lancaster, Chester Ohio Complete Indiana Not SW of line Vigo — Dearborn Michigan Complete Wisconsin Not N of line Vernon — Florence Minnesota Fillmoore, Winona, Houston Iowa Not NW of line Howard — Pottawattamie, not Appanoose to Lee Illinois Not S of line Henderson — Edgar Kentucky Not W of line Mason — Harlan West Virginia Complete Virginia Not E of line Loudoun — Grayson North Carolina Not E of line Transylvania — Stokes Tennessee Not W of line Folk — Johnson II. Rocky Mountain Region British Columbia Victoria, Capital, Cowichan Valley, Vancouver, Sunshine Coast Alberta Calgary Washington Complete except Whatcom, NW‐Clallam Oregon Not SE of line Coos — Hood River Idaho Not S of Nez Perce — Clearwater, but local in Lincoln and Jerome Montana Mineral, Sanders, Lincoln Utah Local in central Box Elder Regions of lower climatic congruence are shown in the map ( Fig. 2B ). The area of zone II builds a narrow stripe surrounding the core area. Zone III is larger and exceeds the compact Atlantic sector of the potential range more distinctly, especially in the north‐east (Gaspé Peninsula), west (Minnesota) and in the south (Kansas, Missouri, Illinois). The western, Pacific part of zone III consists mainly of three adjacent areas (the region surrounding zone II, the western slope of the Rocky Mountains in Alberta and adjacent regions, and a compact area ranging from northern Utah to southern Idaho). Isolated localities are shown in central British Columbia (Nechako and Fraser River Valleys) and northern California (Klamath and North Coast Ranges). Zone IV is much larger than the climatic core area. It comprises Newfoundland and an isolated area south of Lake Winnipeg. Large parts of the Prairie Province, including most of South Dakota and Nebraska connect the compact Atlantic Region with the more scattered occurrence of climatically similar areas of the Rocky Mountain Region. Here, zone IV ranges from northern California and Colorado in the south to British Columbia and Alberta in the north. Zone V contains the remaining regions of North America, which are supposedly unfavourable for successful establishment of A. petiolata . Comparison of current and potential range in North America Evaluation of the quality of the model must primarily be based on occurrences reported from beyond the predicted zones (omission errors). Due to the continuing expansion of the relatively ‘young’ distribution area, it cannot be assumed that invasion is complete. However, large gaps found within areas which are almost completely invaded must be included in a critical analysis of the quality of the model. At a first glance, the comparison of the current and potential range shows one main result. The spread of A. petiolata in eastern North America has reached almost every climatically suitable region, at least with isolated occurrences. Repeated introduction ( Meekins ., 2001 ) and human‐facilitated dispersal enable the species to ‘fill’ its potential distribution area very quickly. But within this potential range, garlic mustard has not yet invaded every possible habitat, and thus, the process of invasion is far from over. A large proportion of the current occurrences ( c. 70%) are situated in zone I; 6%, 7%, 14% and 3% of the currently known localities are situated within zones II, III, IV and V, respectively. Table 2 provides an overview of the 30% of discovered occurrences situated outside of the predicted core area (zone I). ‘Climatic outliers’ are reported more frequently from the southern section of the invaded area, compared to the isolated, locally restricted occurrences in the northern part of the area and the southern Rocky Mountains. In addition to isolated outposts in Oklahoma, Arkansas, Georgia and North Carolina, much more numerous occurrences are to be found particularly in Kansas and Tennessee, and probably in Illinois and Missouri. 2 Recent Alliaria occurrences (dots) outside zone I (zone of complete climate suitability) State/Province Zone II III IV V Notes I. North American Atlantic Region Quebec • One record from Gaspé, not confirmed since 1891 Minnesota • Reported from Clay and Dakota county North Dakota • Reported only from Cass county South Dakota • Reported only from Brooking county Nebraska • Geographically imprecise record of occurrence Kansas • • • Reported from moist riparian woods in 13 counties Missouri • • Underrepresented, occurs in many counties Oklahoma • Reported from Delaware and Adair county Arkansas • • Benton, Washington, Franklin (and Pulaski?) county Illinois • • • At least 41 counties Indiana • Reported from a few counties in the south Kentucky • • • Union and Logan county in zone IV Tennessee • • • At least 13 counties in zone IV Georgia • Geographically imprecise record of occurrence North Carolina • Reported only from Rockingham county Virginia • • Reported from many counties Maryland • Densely populated area II. Rocky Mountain Region Colorado • Reported from one location in 1952, 1958, 1966 Utah • Reported from one location in 1971, 1983, 1984 All reported occurrences in the Pacific North‐west are situated within ‘zone I’ as predicted by the climatic model. However, this part of the range comprises only a few localities, and thus the occupation of the western subsection has to be considered as being in an initial phase of invasion. Discussion The data For identifying and predicting species’ habitat preferences and requirements in relation to climatic variables, the quality of spatial data about climate and distribution is crucial. On a global scale, floristic knowledge differs greatly from region to region (see Jäger's map in Frodin 2001 ). In some regions data is sparse in terms of time (short normal periods) and space (density of weather stations). The validity of the interpolated climate data is somewhat limited, especially in these regions. Another problem in modelling plants distribution areas is the use of different types and numbers of climatic parameters. It might be argued that it is the extreme values in the year that are most critical to plants. Indeed, many modellers in this field use metrics, such as minimum monthly temperatures or, on an annual scale, minimum temperatures of the coldest month. Extreme values are, however, not always particularly robust measures. The use of mean values in place of extreme values can be justified particularly with oscillations of the plants distribution areas, especially as climate is likewise subjected to oscillations. Mean values of many years of production and reproduction are crucial for the fitness of the populations. The fact that modelling with both extreme values and mean values produces good results is due to the close relationship (parallelism) between monthly extremes and mean values. Finally, on large spatial scales, range boundaries result from the complete life cycle of the plants, which depend on the climate over the whole year and thus includes the possibility of regeneration as well as any temporary damage. Another problem is the possible life cycle changes of the species due to genetic differentiation in isolated populations or adaptation to new combinations of environmental factors. Thompson . (1999 ) calculated a series of ‘bioclimatic’ variables that were proposed as more direct controls over plant distributions than temperature and precipitation. However, on large spatial scales (e.g. global and Holarctic), these indices might not reflect the important differences between the climatic rhythms of the eastern and western sides of the main landmasses ( Jäger, 1995 ); a situation that favours the use of the basic climatic parameters. Artificially limiting the numbers of climatic parameters may result in poorly fitting models, such as in Beerling . (1995 ). The FloraMap™ system ( CIAT, 1999 ), which is designed for mapping large‐scale climate probabilities mainly for the tropic regions of the world, uses monthly average diurnal temperature ranges in addition to rainfall and temperature data (altogether 36 climate variables). For a spatially more explicit large scale modelling of plant ranges, high quality distribution maps and a wide array of basic climatic parameters are necessary. Frequency diagrams Comparing the data of Baskin & Baskin (1992 ), Byers & Quinn (1998) ; Cavers . (1979 ), Grime . (1988 ), Meekins (2000), Nuzzo (1993) ; Rejmánek (2000) and Trimbur (1973) with the diagrams ( Fig. 1 ), it appears that life cycle data are reflected to a large extent. At the time of germination (late February until beginning of April in northern localities) there must be an average minimum monthly precipitation of 25 mm at temperatures between 0 °C and 12 °C. In the plant's native distribution range, optimal mean values are 5–10 °C and 30–60 mm. In May (main time of flowering), precipitation must reach an average minimum of 30 mm, combined with average monthly temperatures ranging from 8 °C to 18 °C. For a successful establishment of the young plants, the elongation of the flowering stems, the development of flowers and seed formation, a frost‐free time of at least 4 months is probably necessary. The development of the plants is clearly temperature related; long‐term studies have documented that A. petiolata flowers open earlier with increasing temperature ( Fitter ., 1995 ). In July and August, the young rosettes of A. petiolata expose minimal leaf size. Adult plants are in the life cycle phase of seed ripening at this time and therefore may need a certain minimum temperature, whereas rainfall is of lesser importance. In the plant's native distribution range, mean values of the July temperature (between 16 °C and 23 °C) appear to be most favourable for the species. Autumn precipitation does not seem to be of great importance for the distribution of A. petiolata . The relevant diagrams ( Fig. 1A ) show a modest increase at lower amounts of precipitation. However, only a few occurrences are observed at precipitation values of less than 20 mm for September and of less than 30 mm for October The diagrams of temperature ( Fig. 1B ) indicate that optimal temperatures have to be well above the freezing in these months. It is possible that these temperatures are necessary for the further growth of the plants and the formation of new rosette leaves. During the winter, A. petiolata may require temperatures low enough to break seed dormancy through cold stratification. If daytime temperatures are above freezing the rosettes can continue growth and biomass production. Below a sufficiently thick snow cover, the rosettes can survive even periods of severe frost. The diagrams for winter temperature ( Fig. 1B ) show a relatively wide range of values (26–28 K for December, January and February), but mean temperatures below 5 °C seem to be necessary. This value is not necessarily the prerequisite for breaking seed dormancy; however, these minimal values are the expression of temperature conditions that enable this physiological process. On the other hand, these low temperatures may prevent the development of competing evergreen vegetation. In summary, it can be said that the native distribution range of A. petiolata is roughly approximated by climatic factors, such as moisture (≥ 500 mm annual rainfall), sufficient warmth during the main time of development (the 9 °C May isotherm), sufficient cold winters (the 6 °C January isotherm), and a sufficient time span for plant development (symbolized by the isoline of > 120 frost‐free days). The climatic model The climatic model of the distribution range of A. petiolata describes both the natural and the adventitious, neophytic range of the species with a large degree agreement. This may confirm the assumption that, on a global scale, climate is the main range‐limiting factor for this species. However, the agreement of the model and the observed native range is not complete, especially in mountainous regions. These mismatches are caused by differences between the altitudinal distribution of the plant and the interpolated elevation of the underlying climate grids. Large distribution ranges, partly limited by seacoasts, are generally reducing the possibilities of climatic modelling. For example, it is impossible to characterize the potential western range boundary in Europe with climatic variables alone. Additionally, for relatively large parts of its range (e.g. Turkey) only very sparse floristic data are available. Thus, the agreement with the model produced from climatic data could possibly be higher here. In the north‐eastern part of the area (Byelorussia and adjacent north‐west Russia) edaphic factors probably mask the effect of climate. Here, unsuitable soil conditions (wet and acidic, see above) seem to produce a large gap in an otherwise climatically suitable region. The very detailed distributional data for north‐western Europe (Denmark and Scotland) emphasis the importance of soil factors, which modify or override the influences of climate in these areas. In North America, the reasons for the differences between the current distribution range and that of the climatic model are clear. The large, uninvaded areas north of the Great Lakes and the valley of the St. Lorenz Stream may be attributed to the prevalence in these areas of shallow soils derived primarily from ancient acidic bedrock, which are associated predominantly with coniferous forest vegetation. Although some vigorous Alliaria populations are reported from habitats upon the Canadian Shield ( IPCP, 1999 ), it should be assumed that A. petiolata reaches its climatically defined northern boundary only locally within areas of human settlement (parks, hedges, edges of road and gardens). Its limited occurrence in New Brunswick is also within an area of calcareous bedrock. More extensive distributions of A. petiolata can only be expected in the Mixedwood Plain Ecozone ( Barbour & Christensen, 1993 ), a biogeographical region dominated by limestone bedrock, where soils and water bodies tend to be alkaline. Certainly, probability zones II–IV will remain devoid of A. petiolata in Canada. This prediction is strongly supported by the long‐term, relatively limited distribution of invading A. petiolata in Canadian compared with the situation in the adjacent United States. The second remarkably sparsely populated area is the Pacific Northwest section of the predicted range. At this point in time, A. petiolata may be at an early stage of invasion (the ‘lag‐phase’, see Jäger, 1995 ), as the first report dates back to 1948. The area of distribution predicted by the climatological model depends on the interpolated climate surfaces and does not satisfactorily reflect the great regional variation caused by the very diverse topography. Beside possible occurrences in ruderal habitats, cultivated land and other man‐made vegetation types, suitable growth conditions seem in fact to be restricted to small parts of the predicted area. In the forests of the Pacific, or Humid Transition area, dense coniferous overstoreys above rich ericaceous shrub, herb and moss canopies offer very limited opportunities for A. petiolata to invade natural habitats of the region. Favourable growth conditions may occur along river valleys of lowlands and in humid grasslands. The Cascade Mountains form a sharp and efficient climatic barrier across the predicted range. For the arid eastern part, with predominantly sagebrush, bunchgrass and yellow pine vegetation, suitable growth conditions are hardly encountered. Here, A. petiolata may invade habitats with extrazonal vegetation, like river valleys, moist places, shady, north facing slopes and several mountainous habitat types. However, more probable are occurrences in urbane areas and disturbed, nutrient‐rich places around human settlements. The next conspicuous difference between the modelled and the observed distribution are the relatively numerous Alliaria occurrences in probability zones III–V reported from Kansas. The localities are concentrated in the north‐eastern part of the state. The linearly arranged dots (see Fig. 2A ) indicate the association with riverside forest strips of the wooded floodplains along Missouri and Kansas Rivers, and their tributaries. These are extrazonal vegetation types whose distribution is predominantly influenced by topography, geology and soil characteristics, rather than by climatic variables. The same interpretation may apply to the occurrences reported from Tennessee. According to Baskin & Baskin (1992 ), possible life cycle changes (see above) were not detected in this region. Further discussion about differences between the modelled and the actual distribution seem unnecessary, in so much as areas predicted by the model to be suitable for Alliaria but which are uninhabited might either not have been reached by diaspores, or were predicted falsely as suitable due to the locally restricted predictive power of climatic variables. Another, and more speculative, explanation for the absence of the plant from climatically suitable areas as predicted by the model might be due to the origin of the introduced diaspores. It is possible that A. petiolata is locally adapted to particular climate conditions in its natural range. From the entire source population of A. petiolata constituting its natural distribution range, a random sample has been transported by humans to North America (see Meekins ., 2001 ). This sample may not be able to spread with equal speed to all regions that are predicted as climatically suitable. In summary, it can be stated that not all of the predicted areas will necessarily be invaded by Alliaria . The probability zones show only the spatial framework within which — if further requirements and preconditions are met (such as the presence of human vectors) — Alliaria has a high (climatically defined) likelihood of inhabiting in the long term. The presence of non‐acidic soils, deciduous wood vegetation and mesic habitat conditions seems to be a particularly important combination of additional prerequisites. If they are fulfilled, habitats with extrazonal vegetation and man‐made habitats can become invaded locally, even if they are in zones III and IV. However, wide‐ranging invasions of natural, zonal vegetation types are not predicted to occur here. Usually, predictions are easy to make but difficult to test. Pyšek (2001 ) states that most of the predictions in invasion ecology were made too recently to know whether they are accurate or not. The research on well known, non‐indigenous European species in North America, and vice versa, provides us with opportunities for long‐term field tests because many of the species had have enough time to reach even the remote parts of their potential distribution ranges on their ‘new’ continents. With a review of the results of investigations on a large number of species with different life history strategies, life forms and native range types, our understanding of the different capacities of climatic range models for predicting invasiveness could be improved. Acknowledgments This research was supported by the Deutsche Forschungsgemeinschaft. The manuscript was improved by the helpful suggestions of Karsten Wesche, Linda Scott and two anonymous referees. The authors would like to express thanks to W. Cramer (Potsdam Institute of Climate Impact Research) for providing the climate data.

Journal

Diversity and DistributionsWiley

Published: Jul 1, 2002

There are no references for this article.