Will climate change drive alien invasive plants into areas of high protection value? An improved...

Author's personal copy

Will climate change drive alien invasive plants into areas of highprotection value? An improved model-based regional assessment toprioritise the management of invasions

J.R. Vicente a,*, R.F. Fernandes a, C.F. Randin b, O. Broennimann c, J. Gonçalves a, B. Marcos a,I. Pôças a,d, P. Alves a, A. Guisan c,e, J.P. Honrado a

aCIBIO & Faculdade Ciências do Porto, Universidade Porto, Departamento Biologia, FC4, Rua do Campo Alegre S/N, 4169-007 Porto, Portugalb Institute Botany, University Basel, Schönbeinstrasse 6, CH-4056 SwitzerlandcDépt d’Ecologie et d’Evolution, Univ. Lausanne, Bâtiment de Biologie, CH-1015 Lausanne, SwitzerlanddCentro de Engenharia dos Biossistemas, Tapada da Ajuda 1349-017 Lisboa, Portugale Institute of Earth Sciences, University of Lausanne, Geopolis, 1015 Lausanne, Switzerland

a r t i c l e i n f o

Article history:Received 25 March 2013Received in revised form5 September 2013Accepted 14 September 2013Available online 25 October 2013

Keywords:AcaciaCombined predictive modelsConnectivity analysisConservation valueWoody invasive plants

a b s t r a c t

Species distribution models (SDMs) studies suggest that, without control measures, the distribution ofmany alien invasive plant species (AIS) will increase under climate and land-use changes. Due to limitedresources and large areas colonised by invaders, management and monitoring resources must be pri-oritised. Choices depend on the conservation value of the invaded areas and can be guided by SDMpredictions.

Here, we use a hierarchical SDM framework, complemented by connectivity analysis of AIS distribu-tions, to evaluate current and future conflicts between AIS and high conservation value areas. Weillustrate the framework with three Australian wattle (Acacia) species and patterns of conservation valuein Northern Portugal.

Results show that protected areas will likely suffer higher pressure from all three Acacia species underfuture climatic conditions. Due to this higher predicted conflict in protected areas, management might beprioritised for Acacia dealbata and Acacia melanoxylon. Connectivity of AIS suitable areas inside protectedareas is currently lower than across the full study area, but this would change under future environ-mental conditions.

Coupled SDM and connectivity analysis can support resource prioritisation for anticipation andmonitoring of AIS impacts. However, further tests of this framework over a wide range of regions andorganisms are still required before wide application.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Over the past century, invasion by non-native species hasrapidly increased, threatening ecosystems and economies world-wide (Richardson and Rejmánek, 2011). Invasions can alter theprovision of ecosystem services (Le Maitre et al., 2011;

Theoharides and Dukes, 2007) potentially promoting changes atseveral levels of conservation focus (Marchante et al., 2011). Thecontinuous increase of human mobility acts as a key factor in theprocess of biological invasion by facilitating species migration andcolonisation (Chytrý et al., 2008). Climate change also plays a keyrole in invasion processes through the stimulation of mechanismsthat favour the proliferation of invasive species (Kleinbauer et al.,2010).

Protected areas are pivotal for the conservation of endangeredecosystems, habitats and species since they represent key areas tostudy the efficiency of several barriers (e.g. geographic barriers, andboth environmental and biotic filtering; Richardson et al., 2000) toAIS invasions, and AIS impacts on native species (Py�sek et al., 2002).The Convention on Biological Diversity (CBD, 2010) considers

* Corresponding author. Tel.: þ351220402790; fax: þ351220402799.E-mail addresses: [email protected], [email protected] (J.R. Vicente), rui_

[email protected] (R.F. Fernandes), [email protected] (C.F. Randin),[email protected] (O. Broennimann), [email protected] (J. Gonçalves),[email protected] (B. Marcos), [email protected] (I. Pôças),[email protected] (P. Alves), [email protected] (A. Guisan), [email protected] (J.P. Honrado).

Contents lists available at ScienceDirect

Journal of Environmental Management

journal homepage: www.elsevier .com/locate/ jenvman

0301-4797/$ e see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.jenvman.2013.09.032

Journal of Environmental Management 131 (2013) 185e195


protected areas as the cornerstone of biodiversity conservation buttheir importance and effectiveness vary across regions dependingon human landscape management (Py�sek et al., 2002). Oneexample of its applicability at an European scale is the Natura 2000network that aims at ensuring the protection of important speciesand habitats in Europe (Opdam et al., 2009).

One problem with networks of protected areas is their staticnature when related to species range and ecosystem changes(Pressey et al., 2007). In this context, AIS plants invading a pro-tected area represent a serious concern due to their capacity toreplace native flora and change ecosystem functions and properties(Py�sek et al., 2002). Once AIS are established, eradication is timeconsuming and expensive so preventing future invasions isconsidered the most cost-effective management approach(Genovesi, 2005).

In heterogeneous environments AIS occur in suitable patchesthat may be isolated from each other and connected throughdispersal corridors (e.g. rivers, roads). It has been demonstratedthat habitat connectivity offers preferential routes for the spread ofAIS across landscapes (Minor et al., 2009; Proches et al., 2005). Theanticipation of future invasions is, therefore, commonly regarded asan important conservation task (Bradley and Mustard, 2006;Theoharides and Dukes, 2007).

Species distribution models (SDMs) are widely used in ecology.Over the last decades, they have been increasingly used to identifykey environmental variables driving AIS distribution (explanatorymodels sensu Shmueli, 2010; Guisan and Thuiller, 2005; Vicenteet al., 2010) and to predict the potential distribution of AIS undercurrent conditions and environmental change scenarios (Bradleyand Mustard, 2006; Vicente et al., 2010, 2011). Also, SDMs havebeen particularly useful to predict areas of introduction in theinvaded range (Broennimann et al., 2007). Furthermore, SDMsallow quick and cost-efficient assessments, because they are lesscomplex to calibrate than mechanistic models.

Hierarchical approaches integrating environmental factorsacting at different spatial scales may capture dimensions of speciesdistributions that are ignored at a single scale, and thus cancontribute to improve conservation planning (Minor et al., 2009;Pearson et al., 2004; Vicente et al., 2011). One example uses sub-sets of predictors (same extent and grain) classified a priori ac-cording to their scale of influence and to the ecological scales atwhich they are expected to operate (Vicente et al., 2011). Whencompared to models with unclassified predictors, projections ofcombined models contain a higher level of information about therelative importance of regional and local drivers of relevance formanagement (Vicente et al., 2011). Also, ensemble forecastingcalibrated with different SDM techniques represents a significantimprovement when compared to SDM calibrated with one tech-nique (Thuiller et al., 2009).

SDM studies suggest that AIS distribution will likely increase inthe future due to climate and land-use changes (Kleinbauer et al.,2010; Py�sek et al., 2002; Vicente et al., 2010, 2011). Priorities forAIS monitoring and management may however be set quitedifferently depending on the conservation value of the invadedareas (Blossey, 1999; Roura-Pascual et al., 2010). Regarding this,SDM predictions could support prioritisation of control, eradicationand monitoring efforts.

In this paper we build on a proposed combined SDM frame-work (Vicente et al., 2011) coupled with range connectivityanalysis to evaluate the current and future conflicts between AISand protected areas. We illustrate the framework with theexample of three Australian wattle (Acacia) species in a heavilyinvaded area of Northern Portugal. We finally related the pre-dicted invasion risks with various potential drivers acting atdifferent scales.

2. Methods

2.1. Study area and test species

2.1.1. Study areaThe study area is located in the Northwest of the Iberian

Peninsula and corresponds to the North of Portugal (Fig. 1),covering 21,515 km2 in the transition between the Euro-Siberianand the Mediterranean biogeographic regions. Topographicallythis is a rather heterogeneous region, with an elevation rangingbetween 0 and 1545 m, resulting in marked variations of envi-ronmental conditions and land uses, expressed on a diversifiedvegetation cover. Geologically, the area is dominated by granite andschist, and therefore by acid soils. The climate varies fromtemperate Atlantic in the western areas to sub-continental Medi-terranean in the eastern areas. Mean annual rainfall ranges from ca.400 mm in the eastern valleys to over 2500 mm in the westernmountain summits. About 25% of the area is covered by areas ofconservation value, including several integrated areas in the Natura2000 network of EU and national areas with legal protection (Na-tional Parks and Natural Reserves, natural monuments and pro-tected landscapes) that are mostly within Natura 2000 areas. TheNorth of Portugal is therefore fairly covered by nature protectionregimes, but at the same time it is invaded by many AIS in a largediversity of habitat types (e.g. Vicente et al., 2010).

2.1.2. Test speciesTo illustrate our modelling framework, we chose three invasive

woody plants introduced from Australia: Acacia dealbata Link,Acacia melanoxylon R.Br., and Acacia longifolia (Andrews) Willd.Australian Acacia species are important plant invaders emerging inmany parts of the world, modifying vegetation structures andnative species composition and otherwise affecting ecosystemfunctioning (Marchante et al., 2011; Richardson and Rejmánek,2011). These plants were selected as test species on the basis ofthe following criteria: i) being aggressive invasive species in theregion; ii) having partially similar habitat requirements; and iii)exhibiting the same general life strategy and invasiveness traits, e.g.production of large numbers of seeds, germination stimulated byfire, occurrence in dense populations, and exclusion of native spe-cies due to competition for resources (Lorenzo et al., 2010) andallelopathic ability (in A. dealbata and A. melanoxylon; Marchanteet al., 2011).

The three species were reported as naturalised in Portugal in the20th century, having been introduced mainly for ornamental pur-poses and control of soil erosion in mountainous and coastal areas(Almeida and Freitas, 2006), except for A. melanoxylon, which is alsoa forestry species. These three species are now common in forestareas, scrubland, along river margins (A. dealbata, A. melanoxylon)and sand dunes, as is the case of A. longifolia (Marchante et al.,2011).

The study area was first stratified based on the mean annualtemperature (climate), bedrock type (geology) and percentage offorest cover, thus reflecting the major environmental gradientswithin the geographic region. The mean annual temperature andpercentage of forest cover were each split into three classes byidentifying the natural breaks in their distributions in ArcGIS (ESRI,2010). Bedrock types were reclassified qualitatively as graniticrocks, schistose rocks, and all other types that have a limited dis-tribution in the area. The study area was then stratified bycombining these classes to generate 27 strata, of which 23 wererepresented in the area. This stratificationwas performed using theArcGIS Spatial Analyst extension (ESRI, 2010). We then used anequal-stratified sampling design (as recommended by Hirzel andGuisan, 2002) to randomly select plots of 1 km2 (corresponding

J.R. Vicente et al. / Journal of Environmental Management 131 (2013) 185e195186


to the 1 km� 1 km cells of the GIS layers) in each stratum. Samplingunits, cells of 1 km2, were sampled by collecting presences andabsences of the three Acacia species (between March and April2011). A fixed sampling effort (1 h per cell/or time to record the 3taxa presence/2 persons) was established while visiting all habitattypes on a targeted, non-systematic approach. The sampling effortwas distributed according to the relative cover of habitat types ineach landscape mosaic. Occurrence records available from previoussurveys and a limited number of opportunistic records was addedto the dataset. The final occurrence datasets included 1380 records(856 presences, 524 absences) for A. dealbata, 1063 records (82presences, 981 absences) for A. longifolia, and 1106 records (360presences, 746 absences) for A. melanoxylon. These datasets wereindividually submitted to occurrence selection so as to minimisegeographic aggregation (see “Model calibration and evaluation”).

2.2. Environmental predictors and predictor classification

First we selected 45 environmental predictors that, by expertknowledge and according to previously reporting in the scientificliterature, acts as determinants of the ecology and distribution ofthe target species (for more detailed information see Appendix S1in Supporting Information). To prevent multicollinearity, onlypredictors with Spearman’s rho correlation coefficient lower than0.7 (Elith et al., 2006; Wisz and Guisan, 2009) and GeneralisedVariance Inflation Factor VIF <5 (Neter et al., 1983) were

considered. In the case of correlated pairs of predictors, we chosethe one with the most direct ecological impact on plant speciesdistributions (Guisan and Zimmermann, 2000). This yielded a finalset of 24 environmental predictors, grouped into eight environ-mental types that reflect their ecological meaning (climate;dispersal corridors; geology; landscape composition; landscapestructure; fire regimes; phonological metrics; and productivitymetrics) to calibrate the models for each species (Table 1). Climaticpredictors were obtained from the Worldclim database (Hijmanset al., 2005; available at http://worldclim.org/download). Pre-dictors of ecosystem phenology and primary productivity werecomputed from NDVI (Normalised Difference Vegetation Index;Rouse et al., 1973) time series derived from the MODIS satellitesensor. The phenological indices were calculated with the widelyused TIMESAT software (as described in Jönsson and Eklundh,2004). Using this software, per-pixel time series’ upper envelopewas fitted using double-logistic functions, from which the pheno-logical variables were then extracted (for more detailed informa-tion see Appendix S2). Other predictors (dispersal corridors,geology, landscape composition and structure, and fire regimes)were obtained from thematic environmental maps (for more de-tails see Appendix S1).

We classified the variables into two groups considering theirspatial scale of variation and using a statistical classification basedon spatial autocorrelation. Using point pattern statistics (PPSA e

spdep R package; http://cran.r-project.org/web/packages/spdep),

Fig. 1. Location of the study area in Europe (a) and in the Iberian Peninsula (b), digital terrain model (c), and mean annual temperature (d).

J.R. Vicente et al. / Journal of Environmental Management 131 (2013) 185e195 187


we calculated two indices of spatial autocorrelation (Moran’s I andGeary C; Legendre and Legendre, 1998) for all predictors variablesfor increasing neighbourhood distances (more details in Vicenteet al., 2011). Finally, a classification based on fuzzy clustering(fanny, R software cluster package; Maechler et al., 2011) was per-formed based on Euclidean distance matrix of all correlation values(complete linkage method), reflecting the degree of belonging ofeach predictor to each class. The 24 predictors were consistentlyclassified (see Appendix S3) as having local or regional patterns ofvariation and therefore expected influence on species distributions(Vicente et al., 2011; Table 1).

2.3. Analytical framework

We produced spatial projections of current and future conflictsbetween each of the three AIS and conservation value areas (CVA).The analytical design was divided in three major steps, as illus-trated in Fig. 2.

First, we mapped CVA based on the two nature protectionnetworks present in the region: EU Natura 2000 and the nationalpark network. Each map, representing one of the protection net-works, was classified into four classes, from 1 (no protection, i.e.protected areas absent) to 4 (highest protection; Fig. 2 e Step 1). Inthe case of the national network, protection classes were extractedfrom the corresponding management plans elaborated by the na-tional agency for nature conservation (ICNF). For the Natura 2000map we used the following classes: 1-No protection, 2-SpecialProtection Area (SPA; EU Birds Directive), 3-Special Area of Con-servation (SAC; EU Habitats Directive), and 4-Simultaneously SPAand SAC. For each grid cell (1 km2) the protection value was

computed from the percentage of the cell occupied by each pro-tection class (weighted mean). Finally, the two maps were spatiallycombined to obtain a CVA map for the study area.

Second, we developed combined ensemble forecasting models(see Vicente et al., 2011) for the current and future distributions ofA. dealbata, A. longifolia and A. melanoxylon (Fig. 2, Step 2). In thiscombined modelling procedure, separate models were fitted usingeither “regional” or “local” predictors, and a final model was ob-tained by spatially summing the projections from the two partialmodels (Vicente et al., 2011). The combinations of predicted re-sponses can be of four types: suitable regional conditions and localhabitat (A), only suitable local habitat (B), only suitable regionalconditions (C), and unsuitable regional conditions and local habitat(D). All models were produced for the same spatial extent and usingthe same grain size (1 km2).

Finally, we calculated current and future (under climate changescenarios) spatial conflicts between the predicted distribution (AeD) of each test species and CVA (Fig. 2, Step 3).

2.4. Model calibration and evaluation

We applied the combined predictive modelling frameworkdeveloped by Vicente et al. (2011) to predict current and futuredistributions of the three test species using BIOMOD (Thuiller et al.,2009). BIOMOD was applied in the R statistical software (RDevelopment Core Team, 2012) to fit models using the nine avail-able techniques. For the final ensemble forecasting, only modelswith an area under the curve (AUC) > 0.7 were used (see Thuilleret al., 2009 for more details on model parameterisation for thetechniques available in BIOMOD).

We removed aggregation in species occurrence data by selectinga subset of the original occurrences with a minimal distance opti-mising the Clark and Evans (1954) aggregation index R. This index isa measure of clustering or ordering of a point pattern. It is the ratioof the observed mean distance between nearest neighbours in thedata to that expected for a Poisson point process of the same in-tensity. A value of R ¼ 1 indicates that the points are distributedrandomly, whereas R > 1 suggests ordering, while R < 1 suggestsclustering. For each species, we removed from the initial datasetoccurrences geographically less distant to its closest neighbourthan a threshold distance. We proceeded iteratively until the ag-gregation index reached a value R z 1, thus ensuring that thepattern of occurrences was not clustered anymore. To avoid biasesdue to edge effects, we computed the corrected Clark and Evansindex as proposed by Donnelly (1978). The computation was donein R using the spatstat package (available at http://cran.r-project.org/web/packages/spatstat/index.html).

To evaluate the performance of model predictions, we used thearea under the curve (AUC) of a receiver operating characteristic(ROC) plot (Fielding and Bell, 1997) on an independent evaluationdataset not used to calibrate the models. The AUC was obtained asan output for each model. BIOMOD uses a repeated split-sampleprocedure, keeping 20% of the initial data out of the calibrationfor subsequent validation of the predictions. The number of repe-titions was set to 50.

Model predictions were finally stacked into a single ensemblemodel using a weighted approach, available for generatingensemble models in BIOMOD. This approach ranks models by theirAUC evaluation score using a decay measure expressing the relativeimportance of model weights, a decay of 1 being equivalent to acommittee averaging (i.e. same weights given to all elements). Forour ensemble models, we chose the decay of 1.6, as recommendedby Thuiller et al. (2009). Spatial projections of models werereclassified into presenceeabsence using a ROC-optimisedthreshold maximising jointly the percentage of presences and

Table 1Environmental predictors used to calibrate predictive models, with a shortdescription (for more detailed information about predictors see Appendices S1 andS2), the corresponding type of environmental factor (“ecological type”), and thespatial scale of variation (and influence).

Predictors Description Ecological type Classification(scale)

MTCM Minimum temperature coldestmonth

Climate Regional

TAR Temperature annual rangePWM Precipitation wettest monthPS Precipitation seasonalityDensRiN Density local hydrographic

networkDispersal corridors Local

DensRoN Density local road networkpCambi Percentage cambisoils GeologypGran Percentage granitespAnnC Percentage cover annual crops Landscape

compositionpBlFor Percentage cover broad-leafforests

pCoFor Percentage cover artificialstands

pMixFor Percentage cover mix forestspPioMo Percentage cover pioneer

mosaicspUrb Percentage cover urban areasMPAR Mean perimeterearea ratio Landscape structureMSI Mean shape indexNumFir Number fires Fire regimesSOS Time of the start of growing

season (GS)Phenological metrics

MOS Time of the mid of GSEOS Time of the end of GSLOS Length of GSINT NDVI integral during GS Productivity metricsAMP Amplitude of NDVI values

during GSMAX Maximum NDVI during GS



absences correctly predicted (i.e. the probability wheresensitivity ¼ specificity; Liu et al., 2005).

2.5. Spatial projections under current and future climate scenarios

Spatially combined scenario projections were mapped over thefull geographical extent of the study area, using ArcGIS (ESRI, 2010).The foremost intention of using scenarios in this study was toillustrate the greater informative power of the combined modellingapproach compared to projections from a single full model (Vicenteet al., 2011). We chose two plausible and accessible socio-economicscenarios from the Intergovernmental Panel on Climate Change e

IPCC (Nakicenovic and Swart, 2000) driven by a general circulationmodel (HadCM3): A1B (very rapid economic growth withincreasing globalisation, with a balanced requirement of differentenergy sources (fossil and non-fossil) and high increases in

temperature, ranging between 1.4 and 6.4 �C), and B2 (develop-ment following sustainable locally oriented pathways, with lowerrates of global population growth and yielding moderate increasesin temperature, between 1.4 and 3.8 �C). Future scenarios onlyincluded changes in regional climatic processes. A1B scenario focuson economic growth in a homogeneous world (globalisation) whilethe B2 scenario has a more environmental focus (ecologicallyfriendly approach) on local and regional levels in a heterogeneousworld (regionalisation; Nakicenovic and Swart, 2000).

2.6. Spatial conflict between CVA and invasion

To identify spatial conflicts between CVA and invasion, we usedthe spatial projections derived from combinedmodels, which allowa refined detection of present and future conflict areas (Vicenteet al., 2011), since they are able to predict combinations (Fig. 3aef):

Fig. 2. Conceptual and analytical framework to forecast current and future spatial conflicts between AIS and CVA. The CVA was mapped combining Natura 2000 sites and NaturalParks each classified in four classes ranging from no protection to highest protection (Step 1). Current and future models were calibrated and projected for the three AIS (Acaciadealbata, Acacia longifolia, and Acacia melanoxylon) with the combined modelling approach, that resulted in four predicted responses: (a) suitable regional conditions and localhabitat, (b) only suitable local habitat, (c) only suitable regional conditions, and (d) unsuitable regional conditions and local habitat (Step 2). Finally spatial conflicts between thepredicted distribution of the species (aed) and the CVA were calculated (Step 3).



a) Of highest concern, where the invasive species has suitableregional conditions and local habitats are available (type A),and CVA is high or medium;

b) Of probable conflict with low conservation relevance, wherethe invasive species has suitable regional conditions andlocal habitats are available (type A), but CVA is low;

c) Of conflict possible but uncertain, with conservation rele-vance, where the invasive species has only suitable regionalconditions or local habitats available (type B or C), and theCVA is high or medium;

d) Of conflict possible but uncertain, with low conservationrelevance, where the invasive species has only suitableregional conditions or local habitats available (type B or C),and CVA is low;

e) Of lowest concern, where the invasive species is predicted tobe absent (D); and

f) Without conflict (no CVA), even with the invasive speciespredicted to be present (A, B or C).

2.7. Connectivity of invasive species distributions and CVA

Temporal changes in spatial connectivity of predicted suitableareas for each of the three species across the test area were in firstcalculated using the connectivity index developed by Randin et al.(2009). Here we considered values of species presence: proba-bility of 1 for the response type A, probability of 0.5 for responsetypes B and C, and probability of 0 for response type D. The con-nectivity index attains a maximum value of 1 when all cells sur-rounding a focal suitable cell are also suitable. Then, changes inthe spatial relationship between CVA and the connectivity ofpredicted suitable areas were quantified using the raster calcu-lator in ArcGIS 9.3 (ESRI, 2010). We analysed the trends of con-nectivity for each species in space (full area and inside CVA) andtime (2020 and 2050). Finally a Spearman rank correlation testwas used to identify significant relations between connectivityand protection value.

3. Results

3.1. Invasive species under current and future climates

A. dealbata and A. longifolia were predicted to follow oppositetemporal dynamics, with tendencies for both species more stronglypronounced in climatic scenario A1B (Table 2). Areas with suitableregional conditions and suitable local habitats (response class A)and areas with only suitable regional conditions (C) were predictedto increase in the future for A. dealbata and to decrease forA. longifolia. Conversely, areas with local suitable habitats only (B)and areas with unsuitable regional conditions and local habitats (D)were predicted to decrease for A. dealbata and to increase forA. longifolia (see also Appendix S4).

The occupancy of A. melanoxylon was predicted differentlydepending on the scenario considered. A decrease was predictedfor both time periods (2020 and 2050) for response class A underthe A1B scenario. Under B2, an increase of response class A for theperiod 2020 was predicted whereas a decrease was predicted for2050. For response class B, an increase was predicted along timeconsidering the A1B climatic scenario, while considering the B2climatic scenario a decrease was predicted for 2020 and then anincrease for 2050. Under the A1B scenario, response class C waspredicted to increase for 2020 but to decrease for 2050, whereasconsidering the B2 climatic scenario an increase was predictedalong time. Finally, response class D was predicted to decreaseunder the A1B climatic scenario by 2020 and to increase by 2050,whereas for the B2 climatic scenario a consistent decrease waspredicted.

3.2. Spatial conflict between CVA and invasive species under currentand future climates

The three test species were predicted to follow distinct dy-namics of conflict with CVA under both scenarios A1B and B2(Fig. 4). All spatial combinations expressing conflict (types aed)between A. dealbata and CVA were forecasted to increase for both2020 and 2050 (particularly under scenario A1B). Areas classified as

Fig. 3. Results of the possible spatial conflicts combinations between the AIS pre-dictions and the CVA map. (a) Combinations of highest concern, (b) probable conflictwith low conservation relevance, (c) conflict possible but uncertain, with conservationrelevance, (d) conflict possible but uncertain, (e) with low conservation relevance;areas of lowest concern, and (f) areas without conflict.

Table 2Number of km2 predicted for each of the four model classes; (A e suitable regionalconditions and local habitat; B e only suitable local habitat; C e only suitableregional conditions; D e unsuitable regional conditions and local habitat) for theyears 2000, 2020 and 2050, for each climatic scenario: A1B and B2.

Year Modelclass

Climaticscenario

Acacia dealbata(km2)

Acacia longifolia(km2)

Acacia melanoxylon(km2)

2000 A 6390 1243 5145B 1757 1247 1478C 4419 1711 1306D 8826 17,191 13,463

2020 A A1B 7781 58 4613A B2 7178 1206 5380B A1B 443 2408 1997B B2 1046 1260 1230C A1B 8237 18 1545C B2 5711 1474 1574D A1B 4931 18,908 13,237D B2 7457 17,452 13,208

2050 A A1B 7985 0 2622A B2 7806 377 4725B A1B 239 2466 3988B B2 418 2089 1885C A1B 10,645 0 1265C B2 8677 432 1928D A1B 2523 18,926 13,517D B2 4491 18,494 12,854



Fig. 4. Possible spatial conflicts between the AIS predictions and the CVA map for the three test species: Acacia dealbata, Acacia longifolia, and Acacia melanoxylon. For each species, and for each data (2000, 2020A1B, 2020B2, 2050A1B,2050B2) are presented the results of the number of km2 of spatial conflict between the AIS and the protection value for each conflict combination (see Fig. 3): (a) Highest concern (response type A and protection value between 6 and16), (b) Probable conflict with low conservation relevance (response type A and protection value between 1 and 6), (c) Conflict possible but uncertain with medium/high conservation relevance (response type B or C and protectionvalue between 6 and 16), (d) Conflict possible but uncertain with low conservation relevance (response type B or C and protection value between 1 and 6), (e) Lowest concern (response type D), and (f) no conflict but invader present(response type A, B or C and protection value 1).

J.R.Vicente

etal./

Journalof

Environmental

Managem

ent131

(2013)185

e195

191


of lowest concern (e) were predicted to decrease in time and moreseverely under A1B (Fig. 4).

The spatial conflict between CVA and A. longifoliawas forecastedto experience little changes, only affecting some types and mostlybetween years 2000 and 2020 (Fig. 4). Nonetheless, areas ofprobable conflict with low conservation relevance (b) were gener-ally predicted to decrease. The areas of lowest concern (e) and thosewithout conflict (f) were also predicted to decrease along time,particularly under the A1B scenario.

For A. melanoxylon, conflicts with CVA show different trends intime and across scenarios (Fig. 4). A general increase was fore-casted for areas of highest concern (a), of conflict possible butuncertainwith conservation relevance (c), and without conflict (f).On the other hand, there was a tendency of decrease for areas ofprobable conflict with low conservation relevance (b) as well asfor those of lowest concern (e). The areas of conflict possible butuncertain with low conservation relevance (d) and those withoutconflict (type f) were predicted to increase in both periods under

Fig. 6. Relation between connectivity of distributions of AIS and protection value areas along time and across climate scenarios. (a) Empirical distribution of connectivity values forthe whole study area; (b) Empirical distribution of connectivity values for areas with protection value; (c) Spearman rank correlation between connectivity of species predictions forprotected-value areas (protection value > 1) and protection value for the three considered dates. Ad e Acacia dealbata.

Fig. 5. Results of the spatial conflicts combinations between the AIS predictions and the CVA map for the three test species: Acacia dealbata (Ad); Acacia longifolia (Al), and Acaciamelanoxylon (Am) for 2000 and future climatic scenarios (A1B and B2) for 2020 and 2050.



both climatic scenarios (but particularly under A1B; see AppendixS4).

Considering the spatial patterns of conflict and their dynamicsin the region (Fig. 5), themain conflicts will potentially take place inthe western part of the study area, particularly along the westernlimits of protected areas, where high protection value coincideswith suitable conditions for invasive species. A. dealbata was fore-casted to expand significantly in both protected and non-protectedareas. On the contrary, only minor conflict dynamics was forecastedfor both A. melanoxylon and A. longifolia, as these species are pre-dicted to remain absent from large parts of the region under bothcurrent and future climate conditions (Fig. 5).

3.3. Relations between CVA and connectivity of invasive speciesdistributions under current and future climates

Current connectivity in the whole study area was predictedhigher for A. melanoxylon than for A. dealbata and for A. longifolia(Fig. 6a). Connectivity was higher in the whole area than inside theprotection value areas, for all three species and for all time periods(Fig. 6a vs. b). No significant increase in connectivity was forecastedunder future climatic conditions. We have only detected minorincreases for A. dealbata for the full area by 2020 under both sce-narios, and for areas with protection value by 2020 and 2050 underthe A1B scenario (Fig. 6a and b).

Correlations between species distribution connectivity andprotection value areas were negative when considering all datesand all scenarios, meaning that high connectivity values are mostlyoutside areas with protection value. Inside protected areas, thiscorrelation was lowest for A. dealbata under current climate con-ditions (Fig. 6c), but it was predicted to increase slightly in thefuture. Such correlation was predicted to decrease in the future forA. longifolia and A. melanoxylon, with some differences betweenclimatic scenarios (Fig. 6c). These correlations were also tested forthe full area and showed similar results.

4. Discussion

Assessing current and future conflicts and impacts of AIS onnative biodiversity and ecosystems has become a matter of debatein the scientific community (Davis et al., 2005), as well as, a priorityfor managers. Here we proposed a robust and efficient frameworkto identify areas of possible conflict between AIS and areas of na-ture protection value. Our framework combines the statisticalrobustness and informative projections of combined predictivemodelling with connectivity analysis of invasive species distribu-tions. Using this new framework we predicted the current patternsof conflict for threewoody invasive plants in Northern Portugal andtheir future dynamics under alternative climate change scenarios.Overall our framework yielded valuable information on the pat-terns of invasion expressed as a gradient of concern, providinguseful information for management discussions while rising chal-lenging new research questions in the field of plant invasions. Af-terwards, we discussed the dynamics of conflict betweenprotection value and the test invaders, the effectiveness of com-bined predictive models (CPMs) in the prediction of conflicts, andthe added value of connectivity analysis for assessing AIS distri-butions in space (full area vs. protected-value areas) and time (2000vs. 2020 vs. 2050).

4.1. Current and future conflicts between protection value andinvaders in the test region

Predictions of current conflicts between the three Acacia speciesand the protection value areas confirmed that themost problematic

species nowadays in the North of Portugal is A. dealbata, as it hasalready spread over most of the study area, including large areaswith protection value. Although all three test species were pre-dicted to conflict with protected areas mostly along their borderinglowlands, A. dealbata is already invading the core of some protectedareas (Lorenzo et al., 2010). Since the main areas of introduction ofAIS in this region are located along the coastline (Vicente et al.,2010), and due to the spatial configuration of the dispersal corri-dors present in the region (e.g. highways and rivers), conflicts withprotected areas are currently mostly taking place in the westernpart of the region. This general pattern is consistent with pre-dictions of invasion theory regarding the importance of introduc-tion sites and dispersal corridors (Theoharides and, Dukes, 2007).

Predictions of future conflicts suggest that invasion manage-ment in Northern Portugal should be prioritised towards A. deal-bata and A. melanoxylon, as they are predicted to have a larger areaof conflict with conservation areas. Nonetheless, surveillanceshould be conducted in the future for A. longifolia, even if its dis-tribution was predicted to remain stable, since future changes inenvironmental conditions could potentially lead to a shift in itsinvasive behaviour (Marchante et al., 2011). Overall, the mostproblematic species in the future will still be A. dealbata, probablypromoted by its longer residence time (Lorenzo et al., 2010) andalso due to its ability to colonise different habitat types.

4.2. Effectiveness of CPMs for predicting conflicts between plantinvaders and protection value

The higher thematic resolution and the more informativecharacter of combined models compared to standard SMDs canmake the former particularly valuable for multiple applications inconservation, management and monitoring (Vicente et al., 2011).Discriminating local from regional effects of environmental factorson species distribution is important in many conservation andmanagement issues, especially when related to invasive species.For example, it is possible to locally influence environmental con-ditions through planning and management (e.g. through land-usepolicies), and thereby constrain invasive species expansion to-wards areas with high protection value (e.g. promoting forestmanagement regimes that reduces ecosystem invasibility, insideand around protection areas). However, it is not possible to rapidlycontrol regional effects (e.g. climate) with local management ini-tiatives. Therefore multi-scale modelling approaches are alwaysmore informative and realistic due to the different levels of driversthat shape species distributions (Pearson et al., 2004; Vicente et al.,2011).

Conflict maps obtained with combined models are also valuablefor multiple applications, since they allow the discrimination of agradient of “invasion concern” (see Vicente et al., 2011) related tothe spatial conflict between AIS and protected areas. This way, aprioritisation of the areas where actions are and will be requiredcan be based on response types of highest concern related to po-tential conflicts. This is of the utmost importance when facingfinancial resource limitations for land management and invasioncontrol (Genovesi, 2005). Also, conflict maps for both current andfuture conditions can support the development of monitoringnetworks optimised for early signalling of potential impacts of alieninvasion on valuable elements of conservation areas. By targetingmonitoring efforts for areas where the highest concern conflicts areexpected to occur, early signals of impacts may be detected and thecost-efficiency of the monitoring programs can be improved(Blossey, 1999).

We also showed that the added value by CPMs could beparticularly valuable in the study of biological invasions to predictwhere and when invasions will occur. Predictions for current



distributions using CPMs allow the identification of areas where themore suitable conditions make landscapes more prone to invasion(i.e. increased invasibility) and also in which cases it will coincidewith CVA (Kleinbauer et al., 2010). Also, combined models identifynon-invaded areas where species find suitable regional conditions,which may become invaded if local landscape management createssuitable local conditions for the species, admitting there are nosignificant dispersal constraints. Overall, this additional amount ofinformation provided by combined models has obvious positiveimplications for land management aimed at preventing new bio-logical invasions (Py�sek et al., 2002; Roura-Pascual et al., 2010).

4.3. CPMs, connectivity of invader distributions, and protectionvalue

Connectivity analysis is useful in ecological research and spe-cifically in the study of invasive species, as they allow identifying ina spatially-explicit way where AIS are and will potentially be mostproblematically distributed (Minor et al., 2009). Surfaces with highconnectivity represent potential dispersal corridors, if suitableenvironmental conditions remain available over time (Procheset al., 2005). Moreover, areas with high connectivity will be theones where management may be less effective, as local controlmeasures may be counteracted by immigration and propagulepressure from neighbouring invaded areas.

In this study, we used a connectivity index based on the nearestneighbours (Calabrese and Fagan, 2004). This index, althoughsimple to obtain, is sometimes considered a crude connectivitymetric (Moilanen and Nieminen, 2002). However, from CPM pro-jections it was possible to weight the potential suitable pixels forthe Acacia species by the availability of both regional conditionsand suitable local habitat. The two intermediate response types(with either local habitat available or regional climatic conditionssuitable) provided additional information on the nature of thefactors constraining invasive species. Sites with only regional con-ditions or local habitats are often predicted as absences by tradi-tional SDM projections. However they can be potential corridors forspecies dispersal. This level of information is kept when applyingCPMs.

Finally, since conflict maps result from the spatial overlay ofCPM projections with protection value areas, our results were alsomore detailed in terms of the spatio-temporal range dynamics ofthe modelled species. In fact, a critical result is that the areas ofconflict of higher concern or with high conservation relevance arepredicted to increase over the next decades for two of our threetested species (A. dealbata and A. melanoxylon). This shows howCPMs can support prioritisation towards improved conservationstrategies for anticipating and mitigating the impacts of invasiveplants and climate change in protected areas or other sensitiveareas.

In much of our study area, connectivity of suitable areas for thetest invaders inside protected areas is currently lower than acrossthe full study area, since higher protection value areas coincidegeographically with areas of high elevation and also due to therelatively short residence time and scarcity of dispersal corridors(e.g. roads). Nevertheless, because of the changes in environmentalconditions predicted for the future, this connectivity may change asthe potential distribution of the species will also change (seeFigs. 4e6). This is particularly expected for A. dealbata since con-nectivity was predicted to moderately increase in the future. Underfuture conditions, relationships between protected areas and thespecies potential distributions might shift towards more negativevalues, suggesting an increase of connectivity inside protectedareas with the lowest protection value (which would balance thecurrent higher connectivity of non-protected areas). Overall, our

connectivity analysis showed that protected areas would likelysuffer from a higher pressure from the three Acacia species in thefuture.

5. Conclusions

Predictions from our new modelling framework generated newworking hypothesis that could be used as research directions oninvasion ecology, conservation planning and ecosystem restoration(Le Maitre et al., 2011; Pressey et al., 2007). However, the mainoutput is related to the application of the framework and results toimprove regional and local management strategies. In this regard,this study of the potential conflicts between AIS and nature pro-tection value has shown that:

� Taxonomically and/or functionally related invasive species mayhave distinct levels of conflict with protected areas in receivingregions, due e.g. to different ecological requirements, residencetimes and introduction pathways;

� Detailed predictions of future conflicts highlight that resourcesfor conservation and for invasion management in the regionshould be prioritised for some species and areas;

� CPMs can support resource prioritisation for efficiently antici-pating and mitigating the impacts of invasive plants and climatechange on protected areas;

� Connectivity analysis, both inside and around protected areas,can further contribute to prioritise management actionsconsidering regional dispersal corridors and the probability ofrecolonisation after local control;

� Future changes in environmental conditions may induce shiftsin AIS distributions and increase their connectivity; this mayresult in higher pressure from invasive species and reducedsuccess of management actions in the future;

� The application of this framework over a wide range of situa-tions, tailored as necessary to assess specific managementchallenges raised by invaders’ traits and region-specific net-works of protected areas, will verify its adequacy to supportadaptive and efficient control of invasions in the wider contextof conservation planning and management.

Acknowledgements

This study was financially supported by FCT (Portuguese ScienceFoundation) through PhD grant SFRH/BD/40668/2007 to J.R.Vicente.

This study benefited support from FCT through project PTDC/AGR-AAM/104819/2008 (ECOSENSING).

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.jenvman.2013.09.032.

References

Almeida, J.D., Freitas, H., 2006. Exotic naturalized flora of Continental Portugal e areassessment. Bot. Complut. 30, 117e130.

Blossey, B., 1999. Before, during and after: the need for long-term monitoring ininvasive plant species management. Biol. Invasions 1, 301e311.

Bradley, B.A., Mustard, J.F., 2006. Characterizing the landscape dynamics of aninvasive plant and risk of invasion using remote sensing. Ecol. Appl. 16,1132e1147.

Broennimann, O., Treier, U., Müller-Schärer, H., Thuiller, W., Peterson, A., Guisan, A.,2007. Evidence of climatic niche shift during biological invasion. Ecol. Lett. 10,701e709.

Calabrese, J.M., Fagan, W.F., 2004. A comparison shoppers’ guide to connectivitymetrics: trading off between data requirements and information content. Front.Ecol. Environ. 2, 529e536.



Chytrý, M., Maskell, L., Pino, J., Py�sek, P., Vilà, M., Font, X., Smart, S., 2008. Habitatinvasions by alien plants: a quantitative comparison among mediterranean,subcontinental and oceanic regions of Europe. J. Appl. Ecol. 45, 448e458.

Clark, P.J., Evans, F.C., 1954. Distance to nearest neighbor as a measure of spatialrelationships in populations. Ecology 35, 445e453.

Davis, M.A., Thompson, K., Grime, J.P., 2005. Invasibility: the local mechanismdriving community assembly and species diversity. Ecography 28, 696e704.

Donnelly, K., 1978. Simulations to Determine the Variance and Edge-effect of TotalNearest Neighbor Distance. SImulation Methods in Archaeology. CambridgeUniversity Press, London, pp. 91e95.

Elith, J., Graham, C.H., Anderson, R.P., Ferrier, M., Dudík, S., Guisan, A., Hijmans, R.J.,Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A.,Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., McC Overton, J., Peterson, A.T.,Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberón, J.,Williams, S., Wisz, M.S., Zimmermann, N.E., 2006. Novel methods improve pre-diction of species’ distributions from occurrence data. Ecography 29, 129e151.

ESRI, 2010. ArcGIS 9.3. Environmental Systems Research Institute Inc.Fielding, A.H., Bell, J., 1997. A review of methods for the assessment of prediction

errors in conservation presence/absence models. Environ. Conserv. 24, 38e49.Genovesi, P., 2005. Eradications of invasive alien species in Europe: a review. Biol.

Invasions 7, 127e133.Guisan, A., Zimmermann, N.E., 2000. Predictive habitat distribution models in

ecology. Ecol. Modell. 135, 147e186.Guisan, A., Thuiller, W., 2005. Predicting species distribution: offering more than

simple habitat models. Ecol. Lett. 8, 993e1009.Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high reso-

lution interpolated climate surfaces for global land areas. Int. J. Climatol. 25,1965e1978.

Hirzel, A., Guisan, A., 2002. Which is the optimal sampling strategy for habitatsuitability modelling? Ecol. Modell. 157, 331e341.

Jönsson, P., Eklundh, L., 2004. TIMESAT e a program for analyzing time-series ofsatellite sensor data. Comput. Geosci. 30, 833e845.

Kleinbauer, I., Dullinger, S., Peterseil, J., Essl, F., 2010. Climate change might drive theinvasive tree Robinia pseudacacia into nature reserves and endangered habitats.Biol. Conserv. 143, 382e390.

Legendre, P., Legendre, L., 1998. Numerical Ecology e Developments in Environ-mental Modelling, second ed. Elsevier, Oxford.

Le Maitre, D.C., Gaertner, M., Marchante, E., Ens, E.-J., Holmes, P.M., Pauchard, A.,O’Farrell, P.J., Rogers, A.M., Blanchard, R., Blignaut, J., Richardson, D.M., 2011.Impacts of invasive Australian acacias: implications for management andrestoration. Divers. Distrib. 17, 1015e1029.

Liu, C.R., Berry, P.M., Dawson, T.P., Pearson, R.G., 2005. Selecting thresholds ofoccurrence in the prediction of species distributions. Ecography 28, 385e393.

Lorenzo, P., González, L., Reigosa, M.J., 2010. The genus Acacia as invader: thecharacteristic case of Acacia dealbata link in Europe. Ann. For. Sci. 67, 101.

Maechler, M., Rousseeuw, P., Struyf, A., Hubert, M., 2011. Cluster Analysis Basics andExtensions. R Package Version 1.14.1.

Marchante, H., Freitas, H., Hoffmann, J.H., 2011. Post-clearing recovery of coastaldunes invaded by Acacia longifolia: is duration of invasion relevant for man-agement success? J. Appl. Ecol. 48, 1295e1304.

Minor, E.S., Tessel, S.M., Engelhardt, K.A.M., Lookingbill, T.R., 2009. The role oflandscape connectivity in assembling exotic plant communities: a networkanalysis. Ecology 90, 1802e1809.

Moilanen, A., Nieminen, M., 2002. Simple connectivity measures in spatial ecology.Ecology 83, 1131e1145.

Nakicenovic, N., Swart, R., 2000. Special Report on Emissions Scenarios. A SpecialReport of Working Group III of the Intergovernmental Panel on Climate Change.Cambridge University Press, Cambridge, UK.

Neter, J., Kutner, M.H., Nachtsheim, C.J., Wasserman, W., 1983. Applied LinearRegression Models. Richard D D Irwin, Homewood.

Opdam, P.F.M., Broekmeyer, M.E.A., Kistenkas, F.H., 2009. Identifying uncertaintiesin judging the significance of human impacts on Natura 2000 sites. Environ. Sci.Pol. 12, 912e921.

Pearson, R.G., Dawson, T.E., Liu, C., 2004. Modelling species distributions inBritain: a hierarchical integration of climate and land-cover data. Ecography27, 285e298.

Pressey, R.L., Cabeza, M., Watts, M.E., Cowling, R.M., Wilson, K.A., 2007. Conserva-tion planning in a changing world. Trends Ecol. Evol. 22, 583e592.

Proches, S., Wilson, J.R.U., Veldtman, R., Kalwij, J.M., Richardson, D.M., Chown, S.L.,2005. Landscape corridors: possible dangers? Science 310, 779e783.

Py�sek, P., Jarosík, V., Kucera, T., 2002. Patterns of invasion in temperate nature re-serves. Biol. Conserv. 104, 13e24.

Randin, C.F., Engler, R., Normand, S., Zappa, M., Zimmermann, N.E., Pearman, P.B.,Vittoz, P., Thuiller, W., Guisan, A., 2009. Climate change and plant distribu-tion: local models predict high-elevation persistence. Global Change Biol. 15,1557e1569.

R Development Core Team, 2012. R: A Language and Environment for StatisticalComputing.

Richardson, D.M., Rejmánek, M., 2011. Trees and shrubs as invasive alien species e aglobal review. Divers. Distrib. 17, 788e809.

Richardson, D.M., Pysek, P., Rejmanek, M., Barbour, M.G., Panetta, F.D., West, C.J.,2000. Naturalization and invasion of alien plants: concepts and definitions.Divers. Distrib. 6, 93e107.

Roura-Pascual, N., Krug, R.M., Richardson, D.M., Hui, C., 2010. Spatially-explicitsensitivity analysis for conservation management: exploring the influence ofdecisions in invasive alien plant management. Divers. Distrib. 16, 426e438.

Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1973. Monitoring vegetationsystems in the Great Plains with ERTS. In: Third ERTS symposium, pp.309e317.

Secretariat of the Convention on Biological Diversity (CBD), 2010. Global Biodiver-sity Outlook 3. Secretariat of the Convention on Biological Diversity (CBD),Montreal, 94 pp.

Shmueli, G., 2010. To explain or to predict? Stat. Sci. 25, 289e310.Theoharides, K.A., Dukes, J.S., 2007. Plant invasion across space and time: factors

affecting nonindigenous species success during four stages of invasion. NewPhytol. 176, 256e273.

Thuiller, W., Lafourcade, B., Engler, R., Araújo, M.B., 2009. BIOMOD e a plat-form for ensemble forecasting of species distributions. Ecography 32, 369e373.

Vicente, J., Alves, P., Randin, C., Guisan, A., Honrado, J., 2010. What drives invasi-bility? A multi-model inference test and spatial modelling of alien plant speciesrichness patterns in Northern Portugal. Ecography 33, 1081e1092.

Vicente, J., Randin, C., Gonçalves, J., Metzger, M., Lomba, A., Honrado, J., Guisan, A.,2011. Where will conflicts between alien and rare species occur after climateand land-use change? A test with a novel combined modelling approach. Biol.Invasions 13, 1209e1227.

Wisz, M., Guisan, A., 2009. Do pseudo-absence selection strategies affect geographicpredictions of species? A virtual species approach. BMC Ecol. 9 online http://www.biomedcentral.com/content/pdf/1472-6785-9-8.pdf.


Will climate change drive alien invasive plants into areas of high protection value? An improved...

Documents

Transcript of Will climate change drive alien invasive plants into areas of high protection value? An improved...