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

Learn More →

What drives invasibility? A multi‐model inference test and spatial modelling of alien plant species richness patterns in northern Portugal

What drives invasibility? A multi‐model inference test and spatial modelling of alien plant... Biological invasions constitute the second most severe threat to biodiversity around the world, after habitat destruction, and represent a global problem causing biotic homogenisation along with enormous ecological and economical losses ( Theoharides and Dukes 2007 ). Invasion can be defined as the emergence and geographical expansion of a species in an area where it was previously absent ( Vermeij 1996 ). Invasion of (semi‐) natural ecosystems by alien species is known to induce changes in the composition, structure and function of those ecosystems, with important consequences for both the conservation of native biodiversity and the provision of ecosystem services ( Vitousek 1990 , Le Maitre et al. 2004 ). Some of the most negative ecological impacts of invasive alien species are related to competition with native species, which can lead to the invasive species occupying a dominant position in their new environment or even to the replacement of the native residents. Such replacement is one of the key threats to biodiversity and ecosystem function in large parts of the world ( McKinney and Lockwood 1999 , Pauchard and Shea 2006 ). Many pathways for the introduction of alien species are linked to, or caused by, human activities at local, regional and continental scales. Ongoing climate and land‐use changes are expected to boost invasion in a number of habitats ( Theoharides and Dukes 2007 ). Ecosystem invasion by alien species is still increasing and is a major contributor to the loss of biodiversity at the habitat and landscape levels. Control and eradication planning are expensive, and their cost‐effectiveness remains to be evaluated. Therefore, anticipating future invasions is a major task in conservation biology, since it is crucial to forecast accurately where expansions or new invasions will most likely take place ( Theoharides and Dukes 2007 ). Understanding and anticipating invasions can be approached from two perspectives, depending on whether the focus is on traits of species that enable them to invade (‘invasiveness’), or on traits of the receiving communities, habitats or landscapes (‘invasibility’; Richardson and Pysek 2006 ). Here, we focus mostly on invasibility to test whether some habitats or landscapes are more invasible than others (‘differential invasibility’). Invasibility can be assessed at habitat or landscape levels ( Richardson and Pysek 2006 ), and at both local and regional scales ( Pauchard and Shea 2006 ). In this context, habitat distribution models (HDMs) derived from spatially explicit information on habitats and landscapes can be used to predict invasibility by testing or quantifying relationships of invasive species richness with various characteristics of the habitats and/or landscapes. The rationale is that distinct habitats and landscapes will show different susceptibility to invasion, which can then be used to assess invasion risk ( Chytry et al. 2008 ). Common predictors used to model invasive plant species richness and distributions at the regional scale include topography, climate and geology/substrate ( Holmes et al. 2005 , Pino et al. 2005 ). However, human disturbance also plays an important role as a determinant of biological invasions through the introduction and dispersion of new propagules. This may be reflected by distances from human settlements and infrastructure, as well as by disturbance regimes, which in turn can determine landscape composition, fire regime, and landscape fragmentation ( Le Maitre et al. 2004 ). Even though invasibility is thought to be primarily determined by habitat suitability and propagule pressure ( Williams et al. 2005 , Brooks 2007 , Kowarik and Lippe 2007 ), it is also expected to be further mediated by plant strategies, life forms, and the region of origin of invaders (i.e. invasiveness of individual plant species; Pysek and Richardson 2007 ). A commonly cited stabilising mechanism for invasion resistance is an optimal partitioning of available resources ( Chesson 2000 ), and thus maximum niche complementarity, by diverse plant community assemblages. This supports the assumption that the diversity of plant functional groups in receiving communities is a mechanism for resisting invasion ( Theoharides and Dukes 2007 ). Many functional classifications have been proposed for exploring invasion biology, including those based on life strategies, growth forms and reproductive strategies. For plant invasion, the integrative power of the ‘C‐S‐R functional signature’ (i.e. the relative abundance of Competitors, Stress‐tolerants and Ruderals in a given species pool), which is based on Grime's (1977) classification and is related to gradients of stress and disturbance, can be used to indicate levels of resistance, resilience, eutrophication and dereliction ( Hunt et al. 2004 ). In addition, using functional groups instead of individual species can be helpful to understand community dynamics ( Caccianiga et al. 2006 ). One approach to assessing the role of the environment in controlling biological invasions (i.e. invasibility) is to fit statistical models relating alien species distribution and diversity to various environmental predictors that are expected to affect invasibility ( Thuiller et al. 2005 ). These models have been increasingly used to predict the geographic distribution of taxa and biodiversity ( Guisan and Zimmermann 2000 ), or to test hypotheses about which environmental variables (hereafter predictors) determine distributions ( Guisan and Thuiller 2005 , Austin 2007 ). Statistical predictor selection procedures often result in the retention of a single best model with only one set of predictor variables ( Guisan and Zimmermann 2000 ). However, in recent years, modern statistical science has been moving away from traditional methodologies based solely on such null hypothesis testing ( Stephens et al. 2007 ). Instead, it has been suggested that single, best‐fit models should be replaced by information‐theoretic approaches ( Burnham and Anderson 2002 , Lavoué and Droz 2009 ). One of the most commonly used information metrics is the Akaike information criterion (AIC; Akaike 1973 , Reineking and Schöder 2006 ) and its derivatives. AIC is an estimator that quantifies the information lost when a model is fitted to approximate the ‘truth’ (i.e. predefined real distribution; Burnham and Anderson 2002 ). Hence, it is an estimate of the formal strength of evidence (support) for each hypothesis and its related model. By repeating this process for many models of different predictors selected from a fixed pool of predictors (hypotheses), multi‐model selection and inference ranks the statistical support for each of the competing hypotheses, resulting in multiple models that best explain the ecological system. Unfortunately, few examples exist of such a multi‐model information‐theoretic approach implemented in a predictive species and diversity distribution modelling framework ( Dormann et al. 2008 , Gray et al. 2009 , Wisz and Guisan 2009 ). Determining whether single or multiple environmental influences control invasibility and how invasiveness can modulate this response would support the use of universal or specific controls and eradication strategies for different habitats and/or alien plant groups. Here we use an information‐theoretic approach to assess habitat and landscape invasibility at the regional scale by: (1) modelling the spatial and ecological patterns of alien plant invasions in landscape mosaics; and (2) testing competing hypotheses of which environmental predictors control invasibility. We illustrate the approach with a set of 86 alien species in northwestern Portugal, classified into plant functional types according to their C‐S‐R life strategies. We first focus on predictors influencing total alien species richness and expressing invasibility. We then evaluate whether richness of the distinct C‐S‐R groups responds to the same or different groups of environmental predictors in order to further assess the possible influence of species invasiveness. Methods Analytical framework: questions, hypotheses and competing models Analyses were organised according to two major research questions. The first addressed total alien species richness, and the second addressed species richness for distinct alien plant strategies. For each of these two questions, we tested three general invasibility hypotheses using combinations of competing models related to specific hypotheses ( Fig. 1 , Table 1 , 2 ). A total of 13 competing models and related specific hypotheses was developed from combinations of predictor types or sets of predictors of different types (see below, Table 3 ). Competing models and their underlying specific hypotheses and principles for our four response variables are given in Table 2 . Details of the variables included in each model are given in Table 4 . 1 Conceptual framework of the nested approach used to assess the effects of multiple environmental gradients on plant invasion ecology in northwestern Portugal. The primary gradient model was related to climate in all cases (Step 1). Using the spatial predictions of that model, three different areas were selected, ‘full area’; ‘second, third, and fourth quartiles’ (referred to as ‘area above the first quartile’ hereafter) and ‘third and fourth quartiles’ (referred to as ‘area above the second quartile’ hereafter), and then were used to fit models with the primary and secondary gradients as predictors. For the three geographical areas and for each of the four response variables (total and C, S, and R species richness), hypotheses were tested using multi‐model inference (Step 3) to answer our research questions (Step 4). 1 Questions and general hypotheses related to controls of alien invasion (for details on more specific hypotheses, see Table 2 ). Questions General hypotheses Description Question 1. What controls alien species richness? Simple‐type controls Species richness is explained mainly by one or several variables within a group, with an expected prevailing role of climate, landscape composition, and regional corridors promoting dispersal (H 1 to H 7 in Table 2). Multi‐type controls Species richness is determined by combinations of variables selected from the predictor groups, with the prevalence of benign environmental conditions expected to be the most important (H 8 to H 10 in Table 2). Land‐use intensity Areas under intermediate management regimes host the highest numbers of alien species, followed by areas under more intensive management, and then by those under less intensive management (H 11 to H 13 in Table 2). Question 2. Do different plant strategies yield distinct models? Simple‐type controls Patterns of different plant strategies are explained by multiple environmental predictors, with a prevailing role of climate and geology for stress tolerant species (S‐strategists), climate for competitor species (C‐strategists), and landscape composition for ruderal species (R‐strategists) (H 1 to H 7 in Table 2). Multi‐type controls The prevalence of benign environmental conditions constrains S‐strategists and promotes C‐strategists, the presence of regional dispersal corridors mostly promotes R‐strategists, and environmental heterogeneity promotes the presence of all three strategies (H 8 to H 10 in Table 2). Land‐use intensity Areas under intensive management regimes host the highest numbers of R‐ and S‐strategists, followed by areas under intermediate management regimes and then those under less intensive management. C‐strategists exhibit an inverse pattern, with areas under less intensive management hosting the highest numbers of species and areas under intensive management hosting fewer species (H 11 to H 13 in Table 2). 2 Specific hypotheses with their ecological rationale (see Table 1 for details of general hypotheses). Specific hypotheses Name Rationale H 1 Climate Minimum temperatures control habitat invasibility by frost‐sensitive alien invaders (Pino et al. 2005), and summer drought stress controls alien invasion in Mediterranean ecosystems ( Godoy et al. 2008 ). H 2 Landscape composition Land cover and land‐use controls alien invasion because they determine suitable habitat availability, and because anthropogenic habitats have been shown to provide suitable conditions for more invasive species ( Chytry et al. 2008 ). Also, more alien invaders can find suitable conditions in landscapes with greater compositional diversity ( Pino et al. 2005 ) H 3 Landscape structure and function Landscape invasibility is controlled by patch shape and size, which determine ecotone density and diversity (Le Maitre et al. 2004, Dufour et al. 2006). The density of the local hydrographic network is related to landscape fragmentation, which provides more opportunities for local survival and dispersal ( Foxcroft et al. 2007 ). Primary productivity is correlated with both native and alien plant species richness ( Williams et al. 2005 ). H 4 Fire regime Fire is a common source of disturbance in Mediterranean areas and influences population dynamics of invasive plants (Keeley et al. 2005). H 5 Geology Different bedrock types support distinct soil properties, which affect alien invasion (Rose and Hermanutz 2004), and also support distinct landscape mosaics in the region, thus providing different sets of habitats for alien invaders. Also, more alien invaders can find suitable conditions in landscapes with greater soil diversity. H 6 Topography The local diversity of terrain morphology controls species richness, since more complex terrain usually provides a wider diversity of habitat types ( Dufour et al. 2006 ). Topographic diversity is also related to local hydrographic networks, thus controlling alien invasion in riparian habitats ( Holmes et al. 2005 ). H 7 Regional dispersal corridors Many invasive species use large corridors (e.g. forests or floodplains along rivers) as preferential dispersal routes (Pysek and Prach 1993, Pauchard and Shea 2006 ). H 8 Benign environmental conditions Benign climate conditions and moist, nutrient rich soils promote species richness (Rose and Hermanutz 2004, Pino et al. 2005, Williams et al. 2005). H 9 Environmental heterogeneity The variability of environmental conditions (land‐use types, topography, soils and landscape configuration) promotes species richness (Dufour et al. 2006). H 10 Total dispersal corridors The presence of corridors controls alien invasion, since many invasive species use corridor networks as dispersal routes (Riffell and Gutzwiller 1996, Parendes and Jones 2000 , Foxcroft et al. 2007 ). H 11 Utilization/replacement areas ‘Utilization’ and ‘Replacement’ land uses (Brooks 2007), with intermediate disturbance regimes, host the highest numbers of alien species. H 12 Removal areas ‘Removal’ land uses (Brooks 2007), with more intensive disturbance regimes, host an intermediate number of alien species. H 13 Conservation areas ‘Conservation’ land uses (Brooks 2007), with less intensive disturbance, host the lowest number of alien species. 3 Predictors used in the models, grouped into environmental types that reflect their ecological meaning, and the corresponding scientific references. Environmental types Predictors References Climate TMN (minimum temperature of the coldest month) Arévalo et al. 2005 SPRE (summer precipitation) Pino et al. 2005 Godoy et al. 2008 Landscape composition pNFo (% cover of natural forest) Pino et al. 2005 pUrb (% cover of urban areas) Chytry et al. 2008 pAFo (% cover of forest stands) SWIlu (local diversity of land cover types) Landscape structure and function MSI (mean shape index – average perimeter‐to‐area ratio for all patches reflecting complexity) Le Maitre et al. 2004 dHNe (density of local hydrographic network) Williams et al. 2005 GPP (mean gross annual primary productivity) Fire regime NFir (total number of fire occurrences) Keeley et al. 2005 Geology pGra (percentage of granite) Rose and Hermanutz 2004 SWIso (local diversity of soil types) Dufour et al. 2006 pFlu (percentage of fluvisols) Topography SWIsl (local variation of slope) Holmes et al. 2005 Regional dispersal corridors disH (distance to main rivers) Pauchard and Shea 2006 4 Predictor variables included in the competing models to test each of the specific hypotheses presented in Table 2 (see Table 3 for codes of predictors). Model TMIN SPRE pNFo pUrb pAFo SWIlu MSI dHne GPP NFir PGra SWIso pFlu SWIsl disH H 1 – Climate H 2 – Landscape composition H 3 – Landscape structure and function H 4 – Fire regimes H 5 – Geology H 6 – Topography H 7 – Regional dispersal corridors H 8 – Benign environmental conditions H 9 – Environmental heterogeneity H 10 – Total dispersal corridors H 11 – Utilization/replacement areas H 12 – Removal areas H 13 – Conservation areas The three general hypotheses are related to: (1) the effects of single‐type invasibility controls; (2) the influence of multi‐type invasibility controls; and (3) the role of land‐use intensity. Land‐use intensity is a recognized determinant of alien invasion, with areas under extensive management being less prone to invasion but more challenging in terms of alien control than areas where intensive land use is the norm ( Brooks 2007 ). Therefore, besides assessing the influence of single‐type and multi‐type controls, we were interested in testing more specifically whether land‐use intensity was relevant to landscape invasibility, particularly when a large set of alien invasive species exhibiting distinct plant strategies was analysed. By comparing the weights of competing models and associated hypotheses expressing distinct land‐use intensities, we expected to gain further insight into the specific responses of alien species and strategies to features of land management regimes. Our study area is ideal for assessing the role of land‐use in plant invasions because it contains land management regimes ranging from wilderness and low‐intensity cattle grazing in the mountains, to urbanised land with well‐developed infrastructures and intensive agriculture at lower altitudes. To formulate general hypotheses related to land‐use intensity (both for species richness and richness of distinct plant strategies), we used Brooks’ (2007) framework, in which four major land‐use types are recognised along a gradient of impacts of land management on ecosystems. We combined Brooks’‘utilization’ and ‘replacement’ areas, since they represent intermediate land‐use intensities when compared to the extreme ‘conservation’ and ‘removal’ management types ( Table 2 ). Nested environmental gradient approach Because most alien invasive plant species in our study area are known to be frost‐sensitive, we expected climate to act as a strong primary gradient determining the spatial pattern of our response variables and masking the effect of other gradients. Therefore, we used a spatially nested approach to assess the relative importance of secondary environmental gradients ( Fig. 1 ). In such an approach, referred to hereafter as the nested environmental gradient approach, predictions from a species richness model calibrated with the environmental predictors related to climate ( Fig. 1 – Step 1) were used to sub‐sample the study area. This sub‐sampling was done by using the quartiles of predictions from the climatic GLM and resulted in areas that are progressively more homogeneous regarding the primary gradient and with greater predicted alien species richness ( Fig. 1 – Step 2). In this way, we focused on those areas that are more problematic in terms of invasion and conservation and that provide a suitable setting to address the effects of multiple environmental gradients on alien species richness in the more invasible areas. These areas were then used as the extent for model selection based on the information criterion to test research hypotheses, allowing secondary gradients to be more easily detected ( Fig. 1 – Step 3). Study area The study area is located in the northwestern Portugal (8°52′–8°2′W, 41°24′–42°9′N; Fig. 2 ). It covers an area of 3462 km 2 size at the transition between the Atlantic and Mediterranean biogeographic regions. Elevation ranges from sea level to 1540 m in the eastern mountains, with valleys of major rivers running from east to west. The annual mean temperature ranges from ca 9°C to 15°C, and the mean total annual precipitation varies between ca 1200 mm in the lowlands to 3000 mm in the eastern mountain summits. 2 Study area (a) and its location in the Iberian Peninsula (b) and in Europe (c). Sampling strategy The study area was first stratified based on the mean annual temperature (climate), bedrock type (geology) and percentage of forest cover, thus reflecting the major environmental gradients within the geographic region. The mean annual temperature and percentage of forest cover were each split into three classes by identifying the natural breaks in their distributions in ArcGIS ( ESRI 2008 ). Bedrock types were reclassified qualitatively as granitic rocks, schistose rocks, and all other types that have a limited distribution in the area ( Table 5 ). The study area was then stratified by combining these classes to generate 27 strata, of which 23 were represented in the area. This stratification was performed using the ArcGIS spatial analyst extension ( ESRI 2008 ). 5 Variables used in the equal‐stratified sampling design. Variable type Variable Breaks Classes Climate Mean annual temperature (°C) Natural breaks 8.1–11.2 11.2–13.3 13.3–15.0 Geology Bedrock type Qualitative breaks granites schist others Landscape Percentage of natural forests Natural breaks 0–39.6 39.6–72.5 72.5–100 We then used an equal‐stratified sampling design ( Hirzel and Guisan 2002 ) to randomly select four plots of 1 km 2 size (corresponding to the 1×1 km cells of the GIS layers; hereafter cells) in each stratum, except in one stratum that was represented by only three cells. All three cells were selected for sampling in this stratum. The 91 selected cells were surveyed between April and May of 2008, and the occurrence of alien plant species was recorded using fixed sampling effort (one hour per cell) while visiting all habitat types on a targeted, non‐systematic approach. The sampling effort was distributed according to the relative cover of habitat types in each landscape mosaic. Response variables Alien plant species richness was estimated by combining all alien plant species occurring in at least one of the 91 surveyed cells. We did not measure or include in our analysis native plant richness. Alien species were then classified according to the C‐S‐R plant strategy classification of Grime (1977) by relating each species to one of the seven strategies (Supplementary material Appendix 1). Because information about the abundance of alien plants was not available to compute a functional signature (as proposed by Hunt et al. 2004 ), we calculated the frequency of each extreme strategy (C, S, and R) in each cell. Species belonging to intermediate strategies were used to compute the frequency of both corresponding extreme strategies (e.g. a species classified as CS was used to estimate the richness of both C‐strategists and S‐strategists because it exhibits traits related to both strategies). Four response variables were thus used for model fitting: total species richness (SR), competitor species richness (SR C ), stress tolerant species richness (SR S ), and ruderal species richness (SR R ). Predictors We selected 15 literature‐based predictors to explain alien species richness ( Table 3 ). These predictors were then grouped into seven environmental types reflecting direct, indirect and disturbance variables (sensu Austin and Heyligers 1989 , Guisan and Zimmermann 2000 ; Table 3 ). Depending on a set of a priori hypotheses, predictors were used either by types or independently in multivariate models ( Table 4 ). To reduce multicolinearity, only predictors with a Spearman correlation <±0.7 and a generalized variance lnflation factor VIF <5 ( Neter et al. 1983 ) were used during modelling. Model calibration and model selection We fit a set of competing models and applied multi‐model inference (MMI; Burnham and Anderson 2002 ) to assess how well each model was supported by the data. We used the corrected Akaike information criterion (AIC; Akaike 1973 ) for small sample sizes (AIC c , Shono 2000 ), as recommended when the ratio between n (the number of observations used to fit the model) and K (the number of parameters in the largest model) is <40 ( Shono 2000 , Burnham and Anderson 2002 ). We also limited the maximum number of predictors per model to four because of small sample sizes. For comparing among models we calculated the AIC c difference, where ▵i=AIC c initial – AIC c minimum ( Burnham and Anderson 2002 ). These differences (▵i) were used to derive Akaike weights ( w i ), which represent the probability that a candidate model will be the best approximating and most parsimonious model given the data and set of models. Akaike weights are scaled between zero and one and represent the proportional support for a particular model given all models. Finally, we averaged all competing models weighted by their w i and used the averaged model for spatial prediction. All models were fit using GLMs in R (R 2.9.2, R Development Core Team 2008 ) and associated packages available in CRAN (http://cran.r‐project.org). Species richness was used as the response variable in GLMs with Poisson variance and log link function ( Vincent and Haworth 1983 , Guisan and Zimmermann 2000 ). Second‐order polynomials (linear and quadratic terms) were allowed for each predictor in the GLMs, with the linear term being forced in the model each time the quadratic term was retained (adapted from Burnham and Anderson 2002 , Wisz and Guisan 2009 ). The logistic regression equations were spatially implemented using the raster calculator in the ArcGIS 9.3 spatial analyst extension ( ESRI 2008 ). Results The results of the Akaike weights (w i ) for all the hypotheses related to alien species richness and alien plant strategies richness are summarised in Table 6 . Note that the Akaike weights (w i ) always sum up to 1. 6 Results of information‐theoretic‐based model selection and multi‐model inference: Akaike weights (w i ) for SR (species richness), SR C (C strategy richness), SR S (S strategy richness), and SR R (R strategy richness) for each of the three areas: full area (Full; 91 plots used to fit the model), area above the first quartile (>1st Q; 61 plots used to fit the model) and area above the second quartile (>2nd Q; 40 plots used to fit the model); note that the Akaike weights (w i ) always sum up to 1; for further information see Supplementary material Appendix 2. SR species richness SR C C strategy richness SR S S strategy richness SR R R strategy richness General hypotheses/specific hypotheses Full >1 st Q >2 nd Q Full >1 st Q >2 nd Q Full >1 st Q >2 nd Q Full >1 st Q >2 nd Q Single‐type controls H 1 – climate 0.855 0.833 0.521 0.867 0.817 0.292 0.864 0.262 0.124 0.878 0.857 0.430 H 2 – landscape composition 0.000 0.007 0.004 0.000 0.000 0.001 0.000 0.663 0.039 0.000 0.007 0.001 H 3 – landscape structure and function 0.000 0.000 0.006 0.000 0.000 0.007 0.000 0.000 0.008 0.000 0.000 0.006 H 4 – fire regimes 0.000 0.000 0.077 0.000 0.000 0.038 0.000 0.003 0.092 0.000 0.000 0.220 H 5 – geology 0.000 0.000 0.096 0.000 0.000 0.160 0.000 0.000 0.007 0.000 0.000 0.063 H 6 – topography 0.000 0.000 0.024 0.000 0.000 0.124 0.000 0.004 0.090 0.000 0.000 0.030 H 7 – regional dispersal corridors 0.000 0.000 0.008 0.000 0.000 0.018 0.000 0.000 0.041 0.000 0.000 0.008 Multi‐type controls H 8 – benign environmental conditions 0.145 0.160 0.139 0.133 0.183 0.111 0.136 0.049 0.018 0.122 0.136 0.108 H 9 – environmental heterogeneity 0.000 0.000 0.002 0.000 0.000 0.007 0.000 0.000 0.006 0.000 0.000 0.001 H 10 – total dispersal corridors 0.000 0.000 0.001 0.000 0.000 0.003 0.000 0.000 0.016 0.000 0.000 0.001 Land‐use intensity H 11 – utilization/replacement areas 0.000 0.000 0.002 0.000 0.000 0.009 0.000 0.000 0.022 0.000 0.000 0.001 H 12 – removal areas 0.000 0.000 0.061 0.000 0.000 0.015 0.000 0.018 0.336 0.000 0.000 0.118 H 13 – conservation areas 0.000 0.000 0.059 0.000 0.000 0.213 0.000 0.000 0.200 0.000 0.000 0.014 Hypothesis testing – alien plant richness Single‐type controls The best approximating and most parsimonious specific hypothesis given the data and all competing specific hypotheses was based on climate variables (H 1 , w i =0.855), and reflects the string influence of climate in our study area. When applying the nested gradient approach, specific hypotheses related to climate (H 1 ) and landscape composition (H 2 ) were the best supported for the area above the first quartile; however, the landscape composition specific hypothesis had substantially less adjustment to the data ( w i =0.007, compared to w i =0.833 for climate). The nested approach provided evidence of effects of all the secondary environmental gradients for the area above the second quartile. All seven competing specific hypotheses showed high w i when compared to the full area, and the area above the first quartile. In addition to the hypothesis based on the primary gradient (H 1 , w i =0.521), other specific hypotheses that represent secondary gradients were also important, including geology (H 5 , w i =0.096) and fire regime (H 4 , w i =0.077). Multi‐type controls Comparing the results of the three competing specific hypotheses representing multi‐type controls, i.e. benign environmental conditions (H 8 ), environmental heterogeneity (H 9 ) and dispersal corridors (H 10 ), the model reflecting benign environmental conditions was best for all test areas ( Table 6 ). In contrast, specific hypotheses related to environmental heterogeneity and dispersal corridors presented some weight only when the area above the second quartile was used. Land‐use intensity When competing specific hypotheses reflecting different land‐use intensities (H 11 –H 13 ) were compared, model selection for the whole area and for the area above the first quartile showed very low values of w i for all three specific hypotheses. When the area above the second quartile was used, the best models were those including removal areas (H 12 , w i =0.061) and conservation areas (H 13 , w i =0.059), suggesting that alien species richness responds more to variations within the two extreme land‐use intensities than to variations within areas with intermediate management regimes. Hypothesis testing – alien plant strategies Single‐type controls Among competing hypotheses related to different types of environmental variables (H 1 to H 7 ; Table 3 ), the best approximating and most parsimonious specific hypothesis to explain the species richness of all three strategies was climate (SR C , SR S , and SR R : H 1 w i >0.8; Table 6 ), once again highlighting the role of climate as a primary environmental gradient in the study area. When analyzing the area above the first quartile, climatic (H 1 ) was again selected as the best‐supported specific hypothesis for all three strategies. The landscape composition hypothesis was also selected as best supported for S‐strategists (H 2 w i = 0.663) and as less supported for R‐strategists (H 2 w i = 0.007). For S‐strategists, fire (H 4 , w i = 0.003) and topography (H 6 , w i = 0.004) also showed some importance. When the area above the second quartile was considered, all specific hypotheses were selected for at least one of the three plant strategies, with a prevailing role for climate (H 1 ), fire regime (H 4 ), geology (H 5 ), and topography (H 6 ; Table 6 ). The landscape composition hypothesis (H 2 ) was most relevant for S‐strategist richness. The specific hypothesis expressing regional dispersal corridors (H 7 ) was selected for all three strategies, but with better support for C‐ and S‐strategists. Multi‐type controls Among the three competing specific hypotheses representing multi‐type controls (H 8 –H 10 ) for the whole area, and for the area above the first quartile, only the one expressing benign environmental conditions was selected as best supported (H 8 ; Table 4 ). When the area above the second quartile was considered, all three specific hypotheses were selected for all three strategies, with benign environmental conditions (H 8 ) being the best supported, followed by environmental heterogeneity (H 9 ), and by total dispersal corridors (H 10 ). Land‐use intensity Alien plant strategies exhibited contrasting responses to specific hypotheses expressing land‐use intensity (H 11 –H 13 ; Table 6 ). For all three strategies, specific hypothesis selection for the whole area showed very low values of w i for all three competing hypotheses. In the area above the first quartile, only the specific hypothesis related to removal areas (H 12 ) had any weight, and only for S‐strategists (w i =0.018). Finally, in the area above the second quartile, all three competing specific hypotheses showed some weight for all three strategies. The conservation areas specific hypothesis (H 13 ) was the best fit for C‐strategists (w i =0.213), followed by H 12 (removal areas; w i =0.015) and H 11 (utilization/replacement areas; w i =0.009). For S‐ and R‐strategists, the best specific hypothesis was the one related to removal areas (H 12 ), followed by conservation areas (H 13 ) and finally by utilization/replacement areas (H 11 ; Table 6 ). Spatial predictions from average models Spatial predictions from average models for each combination of response variable and test area (nested gradient approach) are presented in Fig. 3 . These spatial predictions are progressively more detailed regarding patterns of species richness in the more invasible areas as we move from the whole area to the areas above the first and second quartiles. Spatial predictions for C‐ and R‐strategists are very similar, e.g. regarding the location of richness hotspots. On the other hand, predictions for S‐strategists reflect the fact that they are more responsive to the presence of urbanised areas (or ‘removal areas’; Table 6 ), particularly for the area above the first quartile ( Fig. 3 h). 3 Spatial predictions from average models for the four response variables (horizontally: (a) to (c)=species richness; (d–f)=competitor species richness; (g–i)=stress tolerant species richness; (j–i)=ruderal species richness) in the three test areas obtained from the nested gradient approach (vertically: (a), (d), (g) and (j)= full area; (b), (e), (h); (k)=area above the first quartile and (c), (f), (i) and (l)= area above second quartile). Discussion Primary and secondary environmental determinants of alien species richness Climate was the prevailing determinant of alien species richness in our study area, representing a strong primary gradient that masks the effects of other variables at the regional scale. This agrees with other reports of the prevailing influence of climate, particularly frost and low temperatures, on alien invasion ( Walther 2002 , Walther et al. 2007 ). The effects of secondary gradients were identifiable only when the area above the second quartile of the climatic predictions was tested. When this more species‐rich and more homogeneous area was sub‐sampled, all seven specific hypotheses were selected, with those related to climate, fire regime and geology being the best supported. Our general hypothesis that alien species richness is explained by multiple predictor types is therefore only partially supported given that effects of secondary gradients emerged only in the sub‐sampled areas with highest predicted species richness. Our multi‐type controls hypothesis that combinations of factors control alien species richness, was mostly supported when the area above the second quartile was tested ( Table 6 ). The second part of the hypothesis, that benign environmental conditions prevail over environmental heterogeneity and density of dispersal corridors as determinants of total species richness ( Riffell and Gutzwiller 1996 , Williams et al. 2005 , Dufour et al. 2006 ), was fully supported for all three test areas. Becuase this specific hypothesis is the only one including climatic predictors, this further confirmed the dominant effects of climate as the primary regional environmental gradient related to alien species richness. The strong influence of climate on alien invasion has been frequently reported and has received much attention due to its implications for climate change ecology ( Pino et al. 2005 , Walther et al. 2007 , Godoy et al. 2008 ). Moreover, other variables such as land cover or land‐use are known to be correlated with climate at different scales ( Thuiller et al. 2004 , Randin et al. 2009 ). In our study area, extreme climatic conditions (namely low winter temperatures in mountains) seem to inhibit alien species richness, since most alien plant species that are invasive in the region are frost‐sensitive due to their sub‐tropical and/or lowland origins. For this reason, mountain landscapes in the area are today mostly devoid of alien invaders. However, they may become prone to invasion if climate warming reduces the number of frost days in the future, particularly if landscape composition also changes (driven by land‐use change) towards facilitating invasions. Because most mountains within our study area are included in nature reserves (including the Peneda‐Gerês National Park), our results have evident implications for conservation planning in order to prevent new invasions in areas with the greatest conservation value. Sub‐sampling our study area based on areas of highest predicted species richness eliminated mountain areas, which are, as discussed above, mostly devoid of alien invaders due to cold climate conditions and isolation from other mountain ranges by wide lowland areas ( Pauchard et al. 2009 ). This allowed us to focus on areas of greater conservation and management concerns. In these areas, a larger diversity of alien species, potentially occurring in a higher diversity of habitat types, raises more complex challenges regarding not only species control and eradication but also landscape management aimed at reducing new invasion likelihoods. For these areas, in the predictions above the first quartile, climate was joined by landscape composition as a major secondary gradient, supporting previous reports on the importance of landscape traits in determining alien invasion ( Pino et al. 2005 , Chytry et al. 2008 ). A large number of gradients emerged only for the area above the second quartile, confirming the additional importance of secondary controls, such as geology ( Rose and Hermanutz 2004 ), fire regime ( Keeley et al. 2005 ), regional dispersal corridors ( Pysek and Prach 1993 , Pauchard and Shea 2006 ), and topography ( Dufour et al. 2006 ) in explaining alien invasions. Responses of plant strategies to environmental determinants of landscape invasibility Functional traits of individual species act as controls of community assembly because they influence recruitment from the regional species pool. Species richness for distinct plant functional types has been shown to respond differently when modelled against a common set of environmental gradients ( Steinmann et al. 2009 ). We confirmed this for alien invaders by showing that all three tested plant strategies are explained by multiple predictor types, and that different sets of predictors are relevant for distinct strategies. When the variation of the dominant primary gradient (climate) was narrowed, all models were selected for the three strategies, with a particular emphasis on geology (C‐strategists), topography (C‐ and S‐strategists), and fire regime (S‐ and R‐strategists). The prevailing role played by climate for all three strategies supported our working hypothesis of single‐type controls for C‐ and S‐strategists but contradicted it for R‐strategists, for which a prevailing role of climate was a priori not expected but observed. This may be related to the fact that most ruderal species in the study area have tropical and subtropical origins. Also, the CR secondary strategy is most common among species with R traits in the area; because these species are capable of colonising a wider local diversity of habitats than pure R‐strategists, they are less dependent on disturbance regimes and may thus become more responsive to climate ( Pysek et al. 1995 ). This prevalence of climate as a control for all three strategies was further confirmed by the dominant role played by benign environmental conditions among multi‐type controls. Landscape composition was also important for S‐ and R‐strategists, given these species tend to occur in specific habitats, either related to regular disturbance (e.g. crop fields; R‐strategists) or characterised by chronic stress (e.g. rocky environments; S‐strategists; Pysek et al. 1995 ). For C‐strategists, environmental heterogeneity and benign environmental conditions were the most important controls, confirming the ability of C‐strategists to occur in a wide range of habitats if no limiting factors constrain primary production ( Pysek et al. 1995 ). In the more homogeneous test area above the second quartile, variations in benign conditions and in the presence of dispersal corridors were the most important controls for S‐strategists, whereas R‐strategists mostly responded to benign environmental conditions (prevalent in lowlands with higher values of primary production) and were only marginally influenced by environmental heterogeneity and availability of dispersal corridors. Human disturbance, land‐use intensity, and alien invasion Human disturbance has been reported as a major determinant of alien invasion ( Le Maitre et al. 2004 , Brooks 2007 ). In our study, land‐use intensity was an important determinant of landscape invasibility when the variation of the primary climatic gradient was narrowed ( Table 6 ). Models representing removal areas and conservation areas fit the data better than the model representing intermediate land‐use intensity, showing that alien species richness, regardless of plant strategy, responds more to extreme land management regimes than to variations in the area characterised by intermediate human disturbance. This may be explained by the fact that intermediate disturbance regimes favour native species richness ( Huston 1994 ), with fewer niches becoming thus available for alien invaders (‘biotic resistance’; Theoharides and Dukes 2007 ). Overall, the highest alien species richness values coincide with areas subject to removal uses, suggesting that land‐use intensification facilitates alien invasion and highlighting the importance of urban areas and other heavily managed habitats in promoting alien species invasions ( Brooks 2007 ). Alien plant strategies exhibited a relationship to land‐use intensity when climatic variation was narrowed through the nested sub‐sampling approach. Alien species richness for all three plant strategies was controlled by the existence of conservation areas, but S‐ and R‐strategists responded more strongly to heavily managed areas (here represented by urban areas; Pysek et al. 1995 ). Positive effects of land‐use intensity on alien species richness were found for all three plant strategies. In fact, the areas of highest species richness for all three strategies tend to coincide with removal areas, where human disturbance often causes the decline or even extinction of native species, with alien invaders capitalising on the newly available resources ( Le Maitre et al. 2004 ). At the other extreme, conservation areas usually host fewer alien species, independently of strategy type, which may be due to ecosystem stability related to infrequent disturbance ( Pokorny et al. 2005 ). Limitations and further prospects Interactions among variables have been a major topic of discussion in ecological modelling ( Austin 2007 ), but they can be difficult to implement in a multi‐model inference framework. Only interactions among variables included in a given model can be evaluated, which could have consequences on the support provided to models since interactions may occur between variables that are never shared in a same model. This would create confounding effects difficult to handle when testing among competing hypotheses. We argue that, for testing our hypotheses, interactions are not formally needed, but could improve the fit of some models and the support of some hypotheses. Yet, the trends observed are real and the exclusion of interactions does not diminish their importance. Nonetheless, the development of an adequate framework for integrating them should be a research priority in the near future. Testing spatial autocorrelation in model residuals could also reveal additional insights ( Austin 2007 , Dormann et al. 2007 ). To be addressed properly, spatial autocorrelation analyses should be conducted for all competing models, which could increase the complexity of multi‐model analyses and hypotheses testing. In addition, such tests would rather address quite different hypotheses and should thus support future developments of the framework presented here. A third issue for future development of the framework is model validation. Cross validation could be a valid alternative to using information‐theoretic approaches but has three major disadvantages: first, in algorithm comparison tests ( Burnham and Anderson 2002 ), it was concluded that AICc was selecting the most parsimonious model when compared to cross validation; second, when using either larger sample sizes or large number of variables, the cross‐validation approach becomes much more computer‐intensive than the information‐theoretic approach; and third, the cross validation approach does not allow framing a hypothesis testing framework as MMI does (i.e. there is no need to relate groups of variables to hypotheses in cross validation, as we did here). Nonetheless, in future developments AICc weights may be complemented with cross‐validated or (rather) bootstrapped goodness‐of‐fit measures (e.g. D 2 or AUC) of each competing model, to assess its stability to data resampling and associated uncertainty. Conclusions Our results confirm that in our study area: 1) climate is the main determinant of alien invasions, and can thus be considered a primary environmental gradient determining landscape invasibility. Effects of secondary gradients are detected only when the study area is sub‐sampled according to predictions based on primary gradients. 2) Once secondary environmental gradients are revealed, multiple types of predictors are involved in determining patterns of alien species richness and thus landscape invasibility. Some types of predictors (e.g. landscape composition, topography, fire regime) prevail over others in explaining these observed patterns. 3) Alien species richness responds most to extreme land management regimes, regardless of plant strategy, suggesting that intermediate disturbance may control alien invasion by promoting native species richness and thus biotic resistance to invasion processes. Overall, land‐use intensification facilitates alien invasion, whereas conservation areas usually host few alien species, highlighting the importance of ecosystem stability in preventing alien invasions. 4) Distinct alien plant strategies show different responses to environmental gradients, particularly when the variation in the dominant primary gradient is narrowed by sub‐sampling the study area. The differential responses of distinct plant strategies to environmental predictors in different subareas suggest using distinct control and eradication approaches for different areas and alien plant groups. Supplementary material (Appendix E6380 at < www.oikos.ekol.lu.se/appendix > appendix. Appendix 1–2. Acknowledgements This study was financially supported by FCT (Portuguese Science Foundation) through PhD grant SFRH/BD/40668/2007 to JV. AG received support from the Swiss NCCR for ‘Plant survival in natural and agroecological landscapes’ (< www.unine.ch/nccr >). This paper is a contribution of the 3rd International Riederalp Workshop: ‘The Utility of Species Distribution Models as Tools for Assessing Impacts of Global Change’. We would like to thank Nicklaus Zimmermann, Catherine Graham, Peter Pearman, Thomas Edwards and Jens‐Christian Svenning, guest editors of this special issue in Ecography. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Ecography Wiley

What drives invasibility? A multi‐model inference test and spatial modelling of alien plant species richness patterns in northern Portugal

Download PDF
12 pages

Loading next page...
 
/lp/wiley/what-drives-invasibility-a-multi-model-inference-test-and-spatial-GovFp7heES

References (56)

Publisher
Wiley
Copyright
© 2010 The Authors
ISSN
0906-7590
eISSN
1600-0587
DOI
10.1111/j.1600-0587.2010.6380.x
Publisher site
See Article on Publisher Site

Abstract

Biological invasions constitute the second most severe threat to biodiversity around the world, after habitat destruction, and represent a global problem causing biotic homogenisation along with enormous ecological and economical losses ( Theoharides and Dukes 2007 ). Invasion can be defined as the emergence and geographical expansion of a species in an area where it was previously absent ( Vermeij 1996 ). Invasion of (semi‐) natural ecosystems by alien species is known to induce changes in the composition, structure and function of those ecosystems, with important consequences for both the conservation of native biodiversity and the provision of ecosystem services ( Vitousek 1990 , Le Maitre et al. 2004 ). Some of the most negative ecological impacts of invasive alien species are related to competition with native species, which can lead to the invasive species occupying a dominant position in their new environment or even to the replacement of the native residents. Such replacement is one of the key threats to biodiversity and ecosystem function in large parts of the world ( McKinney and Lockwood 1999 , Pauchard and Shea 2006 ). Many pathways for the introduction of alien species are linked to, or caused by, human activities at local, regional and continental scales. Ongoing climate and land‐use changes are expected to boost invasion in a number of habitats ( Theoharides and Dukes 2007 ). Ecosystem invasion by alien species is still increasing and is a major contributor to the loss of biodiversity at the habitat and landscape levels. Control and eradication planning are expensive, and their cost‐effectiveness remains to be evaluated. Therefore, anticipating future invasions is a major task in conservation biology, since it is crucial to forecast accurately where expansions or new invasions will most likely take place ( Theoharides and Dukes 2007 ). Understanding and anticipating invasions can be approached from two perspectives, depending on whether the focus is on traits of species that enable them to invade (‘invasiveness’), or on traits of the receiving communities, habitats or landscapes (‘invasibility’; Richardson and Pysek 2006 ). Here, we focus mostly on invasibility to test whether some habitats or landscapes are more invasible than others (‘differential invasibility’). Invasibility can be assessed at habitat or landscape levels ( Richardson and Pysek 2006 ), and at both local and regional scales ( Pauchard and Shea 2006 ). In this context, habitat distribution models (HDMs) derived from spatially explicit information on habitats and landscapes can be used to predict invasibility by testing or quantifying relationships of invasive species richness with various characteristics of the habitats and/or landscapes. The rationale is that distinct habitats and landscapes will show different susceptibility to invasion, which can then be used to assess invasion risk ( Chytry et al. 2008 ). Common predictors used to model invasive plant species richness and distributions at the regional scale include topography, climate and geology/substrate ( Holmes et al. 2005 , Pino et al. 2005 ). However, human disturbance also plays an important role as a determinant of biological invasions through the introduction and dispersion of new propagules. This may be reflected by distances from human settlements and infrastructure, as well as by disturbance regimes, which in turn can determine landscape composition, fire regime, and landscape fragmentation ( Le Maitre et al. 2004 ). Even though invasibility is thought to be primarily determined by habitat suitability and propagule pressure ( Williams et al. 2005 , Brooks 2007 , Kowarik and Lippe 2007 ), it is also expected to be further mediated by plant strategies, life forms, and the region of origin of invaders (i.e. invasiveness of individual plant species; Pysek and Richardson 2007 ). A commonly cited stabilising mechanism for invasion resistance is an optimal partitioning of available resources ( Chesson 2000 ), and thus maximum niche complementarity, by diverse plant community assemblages. This supports the assumption that the diversity of plant functional groups in receiving communities is a mechanism for resisting invasion ( Theoharides and Dukes 2007 ). Many functional classifications have been proposed for exploring invasion biology, including those based on life strategies, growth forms and reproductive strategies. For plant invasion, the integrative power of the ‘C‐S‐R functional signature’ (i.e. the relative abundance of Competitors, Stress‐tolerants and Ruderals in a given species pool), which is based on Grime's (1977) classification and is related to gradients of stress and disturbance, can be used to indicate levels of resistance, resilience, eutrophication and dereliction ( Hunt et al. 2004 ). In addition, using functional groups instead of individual species can be helpful to understand community dynamics ( Caccianiga et al. 2006 ). One approach to assessing the role of the environment in controlling biological invasions (i.e. invasibility) is to fit statistical models relating alien species distribution and diversity to various environmental predictors that are expected to affect invasibility ( Thuiller et al. 2005 ). These models have been increasingly used to predict the geographic distribution of taxa and biodiversity ( Guisan and Zimmermann 2000 ), or to test hypotheses about which environmental variables (hereafter predictors) determine distributions ( Guisan and Thuiller 2005 , Austin 2007 ). Statistical predictor selection procedures often result in the retention of a single best model with only one set of predictor variables ( Guisan and Zimmermann 2000 ). However, in recent years, modern statistical science has been moving away from traditional methodologies based solely on such null hypothesis testing ( Stephens et al. 2007 ). Instead, it has been suggested that single, best‐fit models should be replaced by information‐theoretic approaches ( Burnham and Anderson 2002 , Lavoué and Droz 2009 ). One of the most commonly used information metrics is the Akaike information criterion (AIC; Akaike 1973 , Reineking and Schöder 2006 ) and its derivatives. AIC is an estimator that quantifies the information lost when a model is fitted to approximate the ‘truth’ (i.e. predefined real distribution; Burnham and Anderson 2002 ). Hence, it is an estimate of the formal strength of evidence (support) for each hypothesis and its related model. By repeating this process for many models of different predictors selected from a fixed pool of predictors (hypotheses), multi‐model selection and inference ranks the statistical support for each of the competing hypotheses, resulting in multiple models that best explain the ecological system. Unfortunately, few examples exist of such a multi‐model information‐theoretic approach implemented in a predictive species and diversity distribution modelling framework ( Dormann et al. 2008 , Gray et al. 2009 , Wisz and Guisan 2009 ). Determining whether single or multiple environmental influences control invasibility and how invasiveness can modulate this response would support the use of universal or specific controls and eradication strategies for different habitats and/or alien plant groups. Here we use an information‐theoretic approach to assess habitat and landscape invasibility at the regional scale by: (1) modelling the spatial and ecological patterns of alien plant invasions in landscape mosaics; and (2) testing competing hypotheses of which environmental predictors control invasibility. We illustrate the approach with a set of 86 alien species in northwestern Portugal, classified into plant functional types according to their C‐S‐R life strategies. We first focus on predictors influencing total alien species richness and expressing invasibility. We then evaluate whether richness of the distinct C‐S‐R groups responds to the same or different groups of environmental predictors in order to further assess the possible influence of species invasiveness. Methods Analytical framework: questions, hypotheses and competing models Analyses were organised according to two major research questions. The first addressed total alien species richness, and the second addressed species richness for distinct alien plant strategies. For each of these two questions, we tested three general invasibility hypotheses using combinations of competing models related to specific hypotheses ( Fig. 1 , Table 1 , 2 ). A total of 13 competing models and related specific hypotheses was developed from combinations of predictor types or sets of predictors of different types (see below, Table 3 ). Competing models and their underlying specific hypotheses and principles for our four response variables are given in Table 2 . Details of the variables included in each model are given in Table 4 . 1 Conceptual framework of the nested approach used to assess the effects of multiple environmental gradients on plant invasion ecology in northwestern Portugal. The primary gradient model was related to climate in all cases (Step 1). Using the spatial predictions of that model, three different areas were selected, ‘full area’; ‘second, third, and fourth quartiles’ (referred to as ‘area above the first quartile’ hereafter) and ‘third and fourth quartiles’ (referred to as ‘area above the second quartile’ hereafter), and then were used to fit models with the primary and secondary gradients as predictors. For the three geographical areas and for each of the four response variables (total and C, S, and R species richness), hypotheses were tested using multi‐model inference (Step 3) to answer our research questions (Step 4). 1 Questions and general hypotheses related to controls of alien invasion (for details on more specific hypotheses, see Table 2 ). Questions General hypotheses Description Question 1. What controls alien species richness? Simple‐type controls Species richness is explained mainly by one or several variables within a group, with an expected prevailing role of climate, landscape composition, and regional corridors promoting dispersal (H 1 to H 7 in Table 2). Multi‐type controls Species richness is determined by combinations of variables selected from the predictor groups, with the prevalence of benign environmental conditions expected to be the most important (H 8 to H 10 in Table 2). Land‐use intensity Areas under intermediate management regimes host the highest numbers of alien species, followed by areas under more intensive management, and then by those under less intensive management (H 11 to H 13 in Table 2). Question 2. Do different plant strategies yield distinct models? Simple‐type controls Patterns of different plant strategies are explained by multiple environmental predictors, with a prevailing role of climate and geology for stress tolerant species (S‐strategists), climate for competitor species (C‐strategists), and landscape composition for ruderal species (R‐strategists) (H 1 to H 7 in Table 2). Multi‐type controls The prevalence of benign environmental conditions constrains S‐strategists and promotes C‐strategists, the presence of regional dispersal corridors mostly promotes R‐strategists, and environmental heterogeneity promotes the presence of all three strategies (H 8 to H 10 in Table 2). Land‐use intensity Areas under intensive management regimes host the highest numbers of R‐ and S‐strategists, followed by areas under intermediate management regimes and then those under less intensive management. C‐strategists exhibit an inverse pattern, with areas under less intensive management hosting the highest numbers of species and areas under intensive management hosting fewer species (H 11 to H 13 in Table 2). 2 Specific hypotheses with their ecological rationale (see Table 1 for details of general hypotheses). Specific hypotheses Name Rationale H 1 Climate Minimum temperatures control habitat invasibility by frost‐sensitive alien invaders (Pino et al. 2005), and summer drought stress controls alien invasion in Mediterranean ecosystems ( Godoy et al. 2008 ). H 2 Landscape composition Land cover and land‐use controls alien invasion because they determine suitable habitat availability, and because anthropogenic habitats have been shown to provide suitable conditions for more invasive species ( Chytry et al. 2008 ). Also, more alien invaders can find suitable conditions in landscapes with greater compositional diversity ( Pino et al. 2005 ) H 3 Landscape structure and function Landscape invasibility is controlled by patch shape and size, which determine ecotone density and diversity (Le Maitre et al. 2004, Dufour et al. 2006). The density of the local hydrographic network is related to landscape fragmentation, which provides more opportunities for local survival and dispersal ( Foxcroft et al. 2007 ). Primary productivity is correlated with both native and alien plant species richness ( Williams et al. 2005 ). H 4 Fire regime Fire is a common source of disturbance in Mediterranean areas and influences population dynamics of invasive plants (Keeley et al. 2005). H 5 Geology Different bedrock types support distinct soil properties, which affect alien invasion (Rose and Hermanutz 2004), and also support distinct landscape mosaics in the region, thus providing different sets of habitats for alien invaders. Also, more alien invaders can find suitable conditions in landscapes with greater soil diversity. H 6 Topography The local diversity of terrain morphology controls species richness, since more complex terrain usually provides a wider diversity of habitat types ( Dufour et al. 2006 ). Topographic diversity is also related to local hydrographic networks, thus controlling alien invasion in riparian habitats ( Holmes et al. 2005 ). H 7 Regional dispersal corridors Many invasive species use large corridors (e.g. forests or floodplains along rivers) as preferential dispersal routes (Pysek and Prach 1993, Pauchard and Shea 2006 ). H 8 Benign environmental conditions Benign climate conditions and moist, nutrient rich soils promote species richness (Rose and Hermanutz 2004, Pino et al. 2005, Williams et al. 2005). H 9 Environmental heterogeneity The variability of environmental conditions (land‐use types, topography, soils and landscape configuration) promotes species richness (Dufour et al. 2006). H 10 Total dispersal corridors The presence of corridors controls alien invasion, since many invasive species use corridor networks as dispersal routes (Riffell and Gutzwiller 1996, Parendes and Jones 2000 , Foxcroft et al. 2007 ). H 11 Utilization/replacement areas ‘Utilization’ and ‘Replacement’ land uses (Brooks 2007), with intermediate disturbance regimes, host the highest numbers of alien species. H 12 Removal areas ‘Removal’ land uses (Brooks 2007), with more intensive disturbance regimes, host an intermediate number of alien species. H 13 Conservation areas ‘Conservation’ land uses (Brooks 2007), with less intensive disturbance, host the lowest number of alien species. 3 Predictors used in the models, grouped into environmental types that reflect their ecological meaning, and the corresponding scientific references. Environmental types Predictors References Climate TMN (minimum temperature of the coldest month) Arévalo et al. 2005 SPRE (summer precipitation) Pino et al. 2005 Godoy et al. 2008 Landscape composition pNFo (% cover of natural forest) Pino et al. 2005 pUrb (% cover of urban areas) Chytry et al. 2008 pAFo (% cover of forest stands) SWIlu (local diversity of land cover types) Landscape structure and function MSI (mean shape index – average perimeter‐to‐area ratio for all patches reflecting complexity) Le Maitre et al. 2004 dHNe (density of local hydrographic network) Williams et al. 2005 GPP (mean gross annual primary productivity) Fire regime NFir (total number of fire occurrences) Keeley et al. 2005 Geology pGra (percentage of granite) Rose and Hermanutz 2004 SWIso (local diversity of soil types) Dufour et al. 2006 pFlu (percentage of fluvisols) Topography SWIsl (local variation of slope) Holmes et al. 2005 Regional dispersal corridors disH (distance to main rivers) Pauchard and Shea 2006 4 Predictor variables included in the competing models to test each of the specific hypotheses presented in Table 2 (see Table 3 for codes of predictors). Model TMIN SPRE pNFo pUrb pAFo SWIlu MSI dHne GPP NFir PGra SWIso pFlu SWIsl disH H 1 – Climate H 2 – Landscape composition H 3 – Landscape structure and function H 4 – Fire regimes H 5 – Geology H 6 – Topography H 7 – Regional dispersal corridors H 8 – Benign environmental conditions H 9 – Environmental heterogeneity H 10 – Total dispersal corridors H 11 – Utilization/replacement areas H 12 – Removal areas H 13 – Conservation areas The three general hypotheses are related to: (1) the effects of single‐type invasibility controls; (2) the influence of multi‐type invasibility controls; and (3) the role of land‐use intensity. Land‐use intensity is a recognized determinant of alien invasion, with areas under extensive management being less prone to invasion but more challenging in terms of alien control than areas where intensive land use is the norm ( Brooks 2007 ). Therefore, besides assessing the influence of single‐type and multi‐type controls, we were interested in testing more specifically whether land‐use intensity was relevant to landscape invasibility, particularly when a large set of alien invasive species exhibiting distinct plant strategies was analysed. By comparing the weights of competing models and associated hypotheses expressing distinct land‐use intensities, we expected to gain further insight into the specific responses of alien species and strategies to features of land management regimes. Our study area is ideal for assessing the role of land‐use in plant invasions because it contains land management regimes ranging from wilderness and low‐intensity cattle grazing in the mountains, to urbanised land with well‐developed infrastructures and intensive agriculture at lower altitudes. To formulate general hypotheses related to land‐use intensity (both for species richness and richness of distinct plant strategies), we used Brooks’ (2007) framework, in which four major land‐use types are recognised along a gradient of impacts of land management on ecosystems. We combined Brooks’‘utilization’ and ‘replacement’ areas, since they represent intermediate land‐use intensities when compared to the extreme ‘conservation’ and ‘removal’ management types ( Table 2 ). Nested environmental gradient approach Because most alien invasive plant species in our study area are known to be frost‐sensitive, we expected climate to act as a strong primary gradient determining the spatial pattern of our response variables and masking the effect of other gradients. Therefore, we used a spatially nested approach to assess the relative importance of secondary environmental gradients ( Fig. 1 ). In such an approach, referred to hereafter as the nested environmental gradient approach, predictions from a species richness model calibrated with the environmental predictors related to climate ( Fig. 1 – Step 1) were used to sub‐sample the study area. This sub‐sampling was done by using the quartiles of predictions from the climatic GLM and resulted in areas that are progressively more homogeneous regarding the primary gradient and with greater predicted alien species richness ( Fig. 1 – Step 2). In this way, we focused on those areas that are more problematic in terms of invasion and conservation and that provide a suitable setting to address the effects of multiple environmental gradients on alien species richness in the more invasible areas. These areas were then used as the extent for model selection based on the information criterion to test research hypotheses, allowing secondary gradients to be more easily detected ( Fig. 1 – Step 3). Study area The study area is located in the northwestern Portugal (8°52′–8°2′W, 41°24′–42°9′N; Fig. 2 ). It covers an area of 3462 km 2 size at the transition between the Atlantic and Mediterranean biogeographic regions. Elevation ranges from sea level to 1540 m in the eastern mountains, with valleys of major rivers running from east to west. The annual mean temperature ranges from ca 9°C to 15°C, and the mean total annual precipitation varies between ca 1200 mm in the lowlands to 3000 mm in the eastern mountain summits. 2 Study area (a) and its location in the Iberian Peninsula (b) and in Europe (c). Sampling strategy The study area was first stratified based on the mean annual temperature (climate), bedrock type (geology) and percentage of forest cover, thus reflecting the major environmental gradients within the geographic region. The mean annual temperature and percentage of forest cover were each split into three classes by identifying the natural breaks in their distributions in ArcGIS ( ESRI 2008 ). Bedrock types were reclassified qualitatively as granitic rocks, schistose rocks, and all other types that have a limited distribution in the area ( Table 5 ). The study area was then stratified by combining these classes to generate 27 strata, of which 23 were represented in the area. This stratification was performed using the ArcGIS spatial analyst extension ( ESRI 2008 ). 5 Variables used in the equal‐stratified sampling design. Variable type Variable Breaks Classes Climate Mean annual temperature (°C) Natural breaks 8.1–11.2 11.2–13.3 13.3–15.0 Geology Bedrock type Qualitative breaks granites schist others Landscape Percentage of natural forests Natural breaks 0–39.6 39.6–72.5 72.5–100 We then used an equal‐stratified sampling design ( Hirzel and Guisan 2002 ) to randomly select four plots of 1 km 2 size (corresponding to the 1×1 km cells of the GIS layers; hereafter cells) in each stratum, except in one stratum that was represented by only three cells. All three cells were selected for sampling in this stratum. The 91 selected cells were surveyed between April and May of 2008, and the occurrence of alien plant species was recorded using fixed sampling effort (one hour per cell) while visiting all habitat types on a targeted, non‐systematic approach. The sampling effort was distributed according to the relative cover of habitat types in each landscape mosaic. Response variables Alien plant species richness was estimated by combining all alien plant species occurring in at least one of the 91 surveyed cells. We did not measure or include in our analysis native plant richness. Alien species were then classified according to the C‐S‐R plant strategy classification of Grime (1977) by relating each species to one of the seven strategies (Supplementary material Appendix 1). Because information about the abundance of alien plants was not available to compute a functional signature (as proposed by Hunt et al. 2004 ), we calculated the frequency of each extreme strategy (C, S, and R) in each cell. Species belonging to intermediate strategies were used to compute the frequency of both corresponding extreme strategies (e.g. a species classified as CS was used to estimate the richness of both C‐strategists and S‐strategists because it exhibits traits related to both strategies). Four response variables were thus used for model fitting: total species richness (SR), competitor species richness (SR C ), stress tolerant species richness (SR S ), and ruderal species richness (SR R ). Predictors We selected 15 literature‐based predictors to explain alien species richness ( Table 3 ). These predictors were then grouped into seven environmental types reflecting direct, indirect and disturbance variables (sensu Austin and Heyligers 1989 , Guisan and Zimmermann 2000 ; Table 3 ). Depending on a set of a priori hypotheses, predictors were used either by types or independently in multivariate models ( Table 4 ). To reduce multicolinearity, only predictors with a Spearman correlation <±0.7 and a generalized variance lnflation factor VIF <5 ( Neter et al. 1983 ) were used during modelling. Model calibration and model selection We fit a set of competing models and applied multi‐model inference (MMI; Burnham and Anderson 2002 ) to assess how well each model was supported by the data. We used the corrected Akaike information criterion (AIC; Akaike 1973 ) for small sample sizes (AIC c , Shono 2000 ), as recommended when the ratio between n (the number of observations used to fit the model) and K (the number of parameters in the largest model) is <40 ( Shono 2000 , Burnham and Anderson 2002 ). We also limited the maximum number of predictors per model to four because of small sample sizes. For comparing among models we calculated the AIC c difference, where ▵i=AIC c initial – AIC c minimum ( Burnham and Anderson 2002 ). These differences (▵i) were used to derive Akaike weights ( w i ), which represent the probability that a candidate model will be the best approximating and most parsimonious model given the data and set of models. Akaike weights are scaled between zero and one and represent the proportional support for a particular model given all models. Finally, we averaged all competing models weighted by their w i and used the averaged model for spatial prediction. All models were fit using GLMs in R (R 2.9.2, R Development Core Team 2008 ) and associated packages available in CRAN (http://cran.r‐project.org). Species richness was used as the response variable in GLMs with Poisson variance and log link function ( Vincent and Haworth 1983 , Guisan and Zimmermann 2000 ). Second‐order polynomials (linear and quadratic terms) were allowed for each predictor in the GLMs, with the linear term being forced in the model each time the quadratic term was retained (adapted from Burnham and Anderson 2002 , Wisz and Guisan 2009 ). The logistic regression equations were spatially implemented using the raster calculator in the ArcGIS 9.3 spatial analyst extension ( ESRI 2008 ). Results The results of the Akaike weights (w i ) for all the hypotheses related to alien species richness and alien plant strategies richness are summarised in Table 6 . Note that the Akaike weights (w i ) always sum up to 1. 6 Results of information‐theoretic‐based model selection and multi‐model inference: Akaike weights (w i ) for SR (species richness), SR C (C strategy richness), SR S (S strategy richness), and SR R (R strategy richness) for each of the three areas: full area (Full; 91 plots used to fit the model), area above the first quartile (>1st Q; 61 plots used to fit the model) and area above the second quartile (>2nd Q; 40 plots used to fit the model); note that the Akaike weights (w i ) always sum up to 1; for further information see Supplementary material Appendix 2. SR species richness SR C C strategy richness SR S S strategy richness SR R R strategy richness General hypotheses/specific hypotheses Full >1 st Q >2 nd Q Full >1 st Q >2 nd Q Full >1 st Q >2 nd Q Full >1 st Q >2 nd Q Single‐type controls H 1 – climate 0.855 0.833 0.521 0.867 0.817 0.292 0.864 0.262 0.124 0.878 0.857 0.430 H 2 – landscape composition 0.000 0.007 0.004 0.000 0.000 0.001 0.000 0.663 0.039 0.000 0.007 0.001 H 3 – landscape structure and function 0.000 0.000 0.006 0.000 0.000 0.007 0.000 0.000 0.008 0.000 0.000 0.006 H 4 – fire regimes 0.000 0.000 0.077 0.000 0.000 0.038 0.000 0.003 0.092 0.000 0.000 0.220 H 5 – geology 0.000 0.000 0.096 0.000 0.000 0.160 0.000 0.000 0.007 0.000 0.000 0.063 H 6 – topography 0.000 0.000 0.024 0.000 0.000 0.124 0.000 0.004 0.090 0.000 0.000 0.030 H 7 – regional dispersal corridors 0.000 0.000 0.008 0.000 0.000 0.018 0.000 0.000 0.041 0.000 0.000 0.008 Multi‐type controls H 8 – benign environmental conditions 0.145 0.160 0.139 0.133 0.183 0.111 0.136 0.049 0.018 0.122 0.136 0.108 H 9 – environmental heterogeneity 0.000 0.000 0.002 0.000 0.000 0.007 0.000 0.000 0.006 0.000 0.000 0.001 H 10 – total dispersal corridors 0.000 0.000 0.001 0.000 0.000 0.003 0.000 0.000 0.016 0.000 0.000 0.001 Land‐use intensity H 11 – utilization/replacement areas 0.000 0.000 0.002 0.000 0.000 0.009 0.000 0.000 0.022 0.000 0.000 0.001 H 12 – removal areas 0.000 0.000 0.061 0.000 0.000 0.015 0.000 0.018 0.336 0.000 0.000 0.118 H 13 – conservation areas 0.000 0.000 0.059 0.000 0.000 0.213 0.000 0.000 0.200 0.000 0.000 0.014 Hypothesis testing – alien plant richness Single‐type controls The best approximating and most parsimonious specific hypothesis given the data and all competing specific hypotheses was based on climate variables (H 1 , w i =0.855), and reflects the string influence of climate in our study area. When applying the nested gradient approach, specific hypotheses related to climate (H 1 ) and landscape composition (H 2 ) were the best supported for the area above the first quartile; however, the landscape composition specific hypothesis had substantially less adjustment to the data ( w i =0.007, compared to w i =0.833 for climate). The nested approach provided evidence of effects of all the secondary environmental gradients for the area above the second quartile. All seven competing specific hypotheses showed high w i when compared to the full area, and the area above the first quartile. In addition to the hypothesis based on the primary gradient (H 1 , w i =0.521), other specific hypotheses that represent secondary gradients were also important, including geology (H 5 , w i =0.096) and fire regime (H 4 , w i =0.077). Multi‐type controls Comparing the results of the three competing specific hypotheses representing multi‐type controls, i.e. benign environmental conditions (H 8 ), environmental heterogeneity (H 9 ) and dispersal corridors (H 10 ), the model reflecting benign environmental conditions was best for all test areas ( Table 6 ). In contrast, specific hypotheses related to environmental heterogeneity and dispersal corridors presented some weight only when the area above the second quartile was used. Land‐use intensity When competing specific hypotheses reflecting different land‐use intensities (H 11 –H 13 ) were compared, model selection for the whole area and for the area above the first quartile showed very low values of w i for all three specific hypotheses. When the area above the second quartile was used, the best models were those including removal areas (H 12 , w i =0.061) and conservation areas (H 13 , w i =0.059), suggesting that alien species richness responds more to variations within the two extreme land‐use intensities than to variations within areas with intermediate management regimes. Hypothesis testing – alien plant strategies Single‐type controls Among competing hypotheses related to different types of environmental variables (H 1 to H 7 ; Table 3 ), the best approximating and most parsimonious specific hypothesis to explain the species richness of all three strategies was climate (SR C , SR S , and SR R : H 1 w i >0.8; Table 6 ), once again highlighting the role of climate as a primary environmental gradient in the study area. When analyzing the area above the first quartile, climatic (H 1 ) was again selected as the best‐supported specific hypothesis for all three strategies. The landscape composition hypothesis was also selected as best supported for S‐strategists (H 2 w i = 0.663) and as less supported for R‐strategists (H 2 w i = 0.007). For S‐strategists, fire (H 4 , w i = 0.003) and topography (H 6 , w i = 0.004) also showed some importance. When the area above the second quartile was considered, all specific hypotheses were selected for at least one of the three plant strategies, with a prevailing role for climate (H 1 ), fire regime (H 4 ), geology (H 5 ), and topography (H 6 ; Table 6 ). The landscape composition hypothesis (H 2 ) was most relevant for S‐strategist richness. The specific hypothesis expressing regional dispersal corridors (H 7 ) was selected for all three strategies, but with better support for C‐ and S‐strategists. Multi‐type controls Among the three competing specific hypotheses representing multi‐type controls (H 8 –H 10 ) for the whole area, and for the area above the first quartile, only the one expressing benign environmental conditions was selected as best supported (H 8 ; Table 4 ). When the area above the second quartile was considered, all three specific hypotheses were selected for all three strategies, with benign environmental conditions (H 8 ) being the best supported, followed by environmental heterogeneity (H 9 ), and by total dispersal corridors (H 10 ). Land‐use intensity Alien plant strategies exhibited contrasting responses to specific hypotheses expressing land‐use intensity (H 11 –H 13 ; Table 6 ). For all three strategies, specific hypothesis selection for the whole area showed very low values of w i for all three competing hypotheses. In the area above the first quartile, only the specific hypothesis related to removal areas (H 12 ) had any weight, and only for S‐strategists (w i =0.018). Finally, in the area above the second quartile, all three competing specific hypotheses showed some weight for all three strategies. The conservation areas specific hypothesis (H 13 ) was the best fit for C‐strategists (w i =0.213), followed by H 12 (removal areas; w i =0.015) and H 11 (utilization/replacement areas; w i =0.009). For S‐ and R‐strategists, the best specific hypothesis was the one related to removal areas (H 12 ), followed by conservation areas (H 13 ) and finally by utilization/replacement areas (H 11 ; Table 6 ). Spatial predictions from average models Spatial predictions from average models for each combination of response variable and test area (nested gradient approach) are presented in Fig. 3 . These spatial predictions are progressively more detailed regarding patterns of species richness in the more invasible areas as we move from the whole area to the areas above the first and second quartiles. Spatial predictions for C‐ and R‐strategists are very similar, e.g. regarding the location of richness hotspots. On the other hand, predictions for S‐strategists reflect the fact that they are more responsive to the presence of urbanised areas (or ‘removal areas’; Table 6 ), particularly for the area above the first quartile ( Fig. 3 h). 3 Spatial predictions from average models for the four response variables (horizontally: (a) to (c)=species richness; (d–f)=competitor species richness; (g–i)=stress tolerant species richness; (j–i)=ruderal species richness) in the three test areas obtained from the nested gradient approach (vertically: (a), (d), (g) and (j)= full area; (b), (e), (h); (k)=area above the first quartile and (c), (f), (i) and (l)= area above second quartile). Discussion Primary and secondary environmental determinants of alien species richness Climate was the prevailing determinant of alien species richness in our study area, representing a strong primary gradient that masks the effects of other variables at the regional scale. This agrees with other reports of the prevailing influence of climate, particularly frost and low temperatures, on alien invasion ( Walther 2002 , Walther et al. 2007 ). The effects of secondary gradients were identifiable only when the area above the second quartile of the climatic predictions was tested. When this more species‐rich and more homogeneous area was sub‐sampled, all seven specific hypotheses were selected, with those related to climate, fire regime and geology being the best supported. Our general hypothesis that alien species richness is explained by multiple predictor types is therefore only partially supported given that effects of secondary gradients emerged only in the sub‐sampled areas with highest predicted species richness. Our multi‐type controls hypothesis that combinations of factors control alien species richness, was mostly supported when the area above the second quartile was tested ( Table 6 ). The second part of the hypothesis, that benign environmental conditions prevail over environmental heterogeneity and density of dispersal corridors as determinants of total species richness ( Riffell and Gutzwiller 1996 , Williams et al. 2005 , Dufour et al. 2006 ), was fully supported for all three test areas. Becuase this specific hypothesis is the only one including climatic predictors, this further confirmed the dominant effects of climate as the primary regional environmental gradient related to alien species richness. The strong influence of climate on alien invasion has been frequently reported and has received much attention due to its implications for climate change ecology ( Pino et al. 2005 , Walther et al. 2007 , Godoy et al. 2008 ). Moreover, other variables such as land cover or land‐use are known to be correlated with climate at different scales ( Thuiller et al. 2004 , Randin et al. 2009 ). In our study area, extreme climatic conditions (namely low winter temperatures in mountains) seem to inhibit alien species richness, since most alien plant species that are invasive in the region are frost‐sensitive due to their sub‐tropical and/or lowland origins. For this reason, mountain landscapes in the area are today mostly devoid of alien invaders. However, they may become prone to invasion if climate warming reduces the number of frost days in the future, particularly if landscape composition also changes (driven by land‐use change) towards facilitating invasions. Because most mountains within our study area are included in nature reserves (including the Peneda‐Gerês National Park), our results have evident implications for conservation planning in order to prevent new invasions in areas with the greatest conservation value. Sub‐sampling our study area based on areas of highest predicted species richness eliminated mountain areas, which are, as discussed above, mostly devoid of alien invaders due to cold climate conditions and isolation from other mountain ranges by wide lowland areas ( Pauchard et al. 2009 ). This allowed us to focus on areas of greater conservation and management concerns. In these areas, a larger diversity of alien species, potentially occurring in a higher diversity of habitat types, raises more complex challenges regarding not only species control and eradication but also landscape management aimed at reducing new invasion likelihoods. For these areas, in the predictions above the first quartile, climate was joined by landscape composition as a major secondary gradient, supporting previous reports on the importance of landscape traits in determining alien invasion ( Pino et al. 2005 , Chytry et al. 2008 ). A large number of gradients emerged only for the area above the second quartile, confirming the additional importance of secondary controls, such as geology ( Rose and Hermanutz 2004 ), fire regime ( Keeley et al. 2005 ), regional dispersal corridors ( Pysek and Prach 1993 , Pauchard and Shea 2006 ), and topography ( Dufour et al. 2006 ) in explaining alien invasions. Responses of plant strategies to environmental determinants of landscape invasibility Functional traits of individual species act as controls of community assembly because they influence recruitment from the regional species pool. Species richness for distinct plant functional types has been shown to respond differently when modelled against a common set of environmental gradients ( Steinmann et al. 2009 ). We confirmed this for alien invaders by showing that all three tested plant strategies are explained by multiple predictor types, and that different sets of predictors are relevant for distinct strategies. When the variation of the dominant primary gradient (climate) was narrowed, all models were selected for the three strategies, with a particular emphasis on geology (C‐strategists), topography (C‐ and S‐strategists), and fire regime (S‐ and R‐strategists). The prevailing role played by climate for all three strategies supported our working hypothesis of single‐type controls for C‐ and S‐strategists but contradicted it for R‐strategists, for which a prevailing role of climate was a priori not expected but observed. This may be related to the fact that most ruderal species in the study area have tropical and subtropical origins. Also, the CR secondary strategy is most common among species with R traits in the area; because these species are capable of colonising a wider local diversity of habitats than pure R‐strategists, they are less dependent on disturbance regimes and may thus become more responsive to climate ( Pysek et al. 1995 ). This prevalence of climate as a control for all three strategies was further confirmed by the dominant role played by benign environmental conditions among multi‐type controls. Landscape composition was also important for S‐ and R‐strategists, given these species tend to occur in specific habitats, either related to regular disturbance (e.g. crop fields; R‐strategists) or characterised by chronic stress (e.g. rocky environments; S‐strategists; Pysek et al. 1995 ). For C‐strategists, environmental heterogeneity and benign environmental conditions were the most important controls, confirming the ability of C‐strategists to occur in a wide range of habitats if no limiting factors constrain primary production ( Pysek et al. 1995 ). In the more homogeneous test area above the second quartile, variations in benign conditions and in the presence of dispersal corridors were the most important controls for S‐strategists, whereas R‐strategists mostly responded to benign environmental conditions (prevalent in lowlands with higher values of primary production) and were only marginally influenced by environmental heterogeneity and availability of dispersal corridors. Human disturbance, land‐use intensity, and alien invasion Human disturbance has been reported as a major determinant of alien invasion ( Le Maitre et al. 2004 , Brooks 2007 ). In our study, land‐use intensity was an important determinant of landscape invasibility when the variation of the primary climatic gradient was narrowed ( Table 6 ). Models representing removal areas and conservation areas fit the data better than the model representing intermediate land‐use intensity, showing that alien species richness, regardless of plant strategy, responds more to extreme land management regimes than to variations in the area characterised by intermediate human disturbance. This may be explained by the fact that intermediate disturbance regimes favour native species richness ( Huston 1994 ), with fewer niches becoming thus available for alien invaders (‘biotic resistance’; Theoharides and Dukes 2007 ). Overall, the highest alien species richness values coincide with areas subject to removal uses, suggesting that land‐use intensification facilitates alien invasion and highlighting the importance of urban areas and other heavily managed habitats in promoting alien species invasions ( Brooks 2007 ). Alien plant strategies exhibited a relationship to land‐use intensity when climatic variation was narrowed through the nested sub‐sampling approach. Alien species richness for all three plant strategies was controlled by the existence of conservation areas, but S‐ and R‐strategists responded more strongly to heavily managed areas (here represented by urban areas; Pysek et al. 1995 ). Positive effects of land‐use intensity on alien species richness were found for all three plant strategies. In fact, the areas of highest species richness for all three strategies tend to coincide with removal areas, where human disturbance often causes the decline or even extinction of native species, with alien invaders capitalising on the newly available resources ( Le Maitre et al. 2004 ). At the other extreme, conservation areas usually host fewer alien species, independently of strategy type, which may be due to ecosystem stability related to infrequent disturbance ( Pokorny et al. 2005 ). Limitations and further prospects Interactions among variables have been a major topic of discussion in ecological modelling ( Austin 2007 ), but they can be difficult to implement in a multi‐model inference framework. Only interactions among variables included in a given model can be evaluated, which could have consequences on the support provided to models since interactions may occur between variables that are never shared in a same model. This would create confounding effects difficult to handle when testing among competing hypotheses. We argue that, for testing our hypotheses, interactions are not formally needed, but could improve the fit of some models and the support of some hypotheses. Yet, the trends observed are real and the exclusion of interactions does not diminish their importance. Nonetheless, the development of an adequate framework for integrating them should be a research priority in the near future. Testing spatial autocorrelation in model residuals could also reveal additional insights ( Austin 2007 , Dormann et al. 2007 ). To be addressed properly, spatial autocorrelation analyses should be conducted for all competing models, which could increase the complexity of multi‐model analyses and hypotheses testing. In addition, such tests would rather address quite different hypotheses and should thus support future developments of the framework presented here. A third issue for future development of the framework is model validation. Cross validation could be a valid alternative to using information‐theoretic approaches but has three major disadvantages: first, in algorithm comparison tests ( Burnham and Anderson 2002 ), it was concluded that AICc was selecting the most parsimonious model when compared to cross validation; second, when using either larger sample sizes or large number of variables, the cross‐validation approach becomes much more computer‐intensive than the information‐theoretic approach; and third, the cross validation approach does not allow framing a hypothesis testing framework as MMI does (i.e. there is no need to relate groups of variables to hypotheses in cross validation, as we did here). Nonetheless, in future developments AICc weights may be complemented with cross‐validated or (rather) bootstrapped goodness‐of‐fit measures (e.g. D 2 or AUC) of each competing model, to assess its stability to data resampling and associated uncertainty. Conclusions Our results confirm that in our study area: 1) climate is the main determinant of alien invasions, and can thus be considered a primary environmental gradient determining landscape invasibility. Effects of secondary gradients are detected only when the study area is sub‐sampled according to predictions based on primary gradients. 2) Once secondary environmental gradients are revealed, multiple types of predictors are involved in determining patterns of alien species richness and thus landscape invasibility. Some types of predictors (e.g. landscape composition, topography, fire regime) prevail over others in explaining these observed patterns. 3) Alien species richness responds most to extreme land management regimes, regardless of plant strategy, suggesting that intermediate disturbance may control alien invasion by promoting native species richness and thus biotic resistance to invasion processes. Overall, land‐use intensification facilitates alien invasion, whereas conservation areas usually host few alien species, highlighting the importance of ecosystem stability in preventing alien invasions. 4) Distinct alien plant strategies show different responses to environmental gradients, particularly when the variation in the dominant primary gradient is narrowed by sub‐sampling the study area. The differential responses of distinct plant strategies to environmental predictors in different subareas suggest using distinct control and eradication approaches for different areas and alien plant groups. Supplementary material (Appendix E6380 at < www.oikos.ekol.lu.se/appendix > appendix. Appendix 1–2. Acknowledgements This study was financially supported by FCT (Portuguese Science Foundation) through PhD grant SFRH/BD/40668/2007 to JV. AG received support from the Swiss NCCR for ‘Plant survival in natural and agroecological landscapes’ (< www.unine.ch/nccr >). This paper is a contribution of the 3rd International Riederalp Workshop: ‘The Utility of Species Distribution Models as Tools for Assessing Impacts of Global Change’. We would like to thank Nicklaus Zimmermann, Catherine Graham, Peter Pearman, Thomas Edwards and Jens‐Christian Svenning, guest editors of this special issue in Ecography.

Journal

Ecography – Wiley

Published: Dec 1, 2010

There are no references for this article.