Modelling individual and collective species responses

Biological Conservation 167 (2013) 283–291

Contents lists available at ScienceDirect

Biological Conservation journal homepage: www.elsevier.com/locate/biocon

Modelling individual and collective species responses to climate change within Small Island States Shobha S. Maharaj a,b,⇑, Mark New c,d,e a

Department of Plant Sciences, University of Oxford, South Parks Road, Oxford OX1 3RB, United Kingdom Department of Life Sciences, University of the West Indies, St. Augustine Campus, Trinidad and Tobago c School of Geography and Environment, University of Oxford, South Parks Road, Oxford OX1 3QY, United Kingdom d African Climate and Development Initiative, University of Cape Town, Private Bag X3, Rondebosch 7701, South Africa e Department of Environmental and Geographical Science, University of Cape, University of Cape Town, Private Bag X3, Rondebosch 7701, South Africa b

a r t i c l e

i n f o

Article history: Received 16 August 2012 Received in revised form 30 July 2013 Accepted 19 August 2013

Keywords: Small Island States Climate change Species distribution modelling Terrestrial biodiversity

a b s t r a c t Very little empirical work has been done to assess the potential impacts of climate change upon terrestrial biodiversity within small islands, many of which contribute to global species diversity due to high levels of endemicity. This study illustrates projections of not only individual but also the ‘collective’ response of a group of high conservation value tree species to climate change within the Caribbean small island of Trinidad. The species distribution modelling algorithm MaxEnt was used to construct models of the realised present environmental space occupied by these species based on present day climate and other environmental factors. These models were then used to estimate present and future (2050; SRES A2) distributions of these species across Trinidad. Both present and future model output were incorporated to create change maps which illustrate projected expansions, contractions and areas of stable environmental space for each species. Individual change maps were combined to create a ‘collective’ change map portraying projected changes in the environmental space of this species group as a whole. Most individual species and the collective group response were projected to lose more than 50% of present environmental space, with the latter being limited to the southern edge of the island. Our results suggest that small islands may experience an eventual disappearance of endemics and other valuable species under SRES A2 conditions, which may serve to further depreciate global terrestrial species diversity. Application of this ‘collective’ response may be particularly useful for planning within the limited geographic spaces available for conservation within small islands. Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction 1 The term ‘biodiversity hotspot’ was coined during the late 1980s to highlight regions across the globe containing large numbers of endemics within relatively small geographic regions that were under significant threats of habitat destruction (Myers, 1988, 1990). These biodiversity hotspots along with other areas of natural habitat continue to be threatened by a range of human activities (de Chazal and Rounsevell, 2009), in addition to which, climate change has been postulated by some to become an additional major driver of global species diversity loss (e.g. Thomas et al., 2004). Already linked to events such as recent species extinctions (Pounds et al., 2006) and

⇑ Corresponding author. Present address: Department of Life Sciences, St. Augustine Campus, University of the West Indies, Trinidad and Tobago. Tel.: +1 (868) 487 4477; fax: +1 (868) 663 5241. E-mail addresses: [email protected] (S.S. Maharaj), mark.new@ acdi.uct.ac.za (M. New). 1 SD modelling: species distribution modelling. GCM: Global Climate Model. RCM: Regional Climate Model. PRECIS: Providing REgional Climates for Impacts Studies. 0006-3207/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.biocon.2013.08.027

phenological changes (Walther et al., 2002), the disassembly of species within ecological communities have already been reported, with species moving pole-ward and upward in elevation (Hickling et al., 2006; Lenoir et al., 2008). Of the 25 hotspots originally suggested by Myers et al. (2000), nine were comprised mainly of islands, and almost all tropical islands were reported to belong to one hotspot or another. However, the species diversity within many of these islands, and in particular, small islands, are considered to be particularly vulnerable to climate change because of limited geographic space in conjunction with high human population pressure for limited resources (Heller and Zavaleta, 2009). The Caribbean is such an example: consisting mainly of Small Island States (SIS), the region was ranked as the third richest hotspot in terms of the percentage of global plant and vertebrate endemics (Myers et al., 2000). Stretching over just 0.15% of the earth’s surface, this region has been estimated to host approximately 7000 endemic vascular plant species within just 11.3% of its remaining primary vegetation (Mittermeier et al., 2005; Myers

284

S.S. Maharaj, M. New / Biological Conservation 167 (2013) 283–291

et al., 2000). A more recent assessment of the vulnerability of global biodiversity hotspots to climate change ranked this region among those hotspots that are ‘especially vulnerable’, with extinctions (based on species–area relationships) projected to exceed 3000 plant species and 200 vertebrate species (Malcolm et al., 2006). Apart from the traditional drivers of habitat loss such as the expansion of industry, agriculture and tourism, the limited geographic space and steep climate gradients of many Caribbean SIS leave terrestrial species with very little room to shift or adapt to changing climatic conditions. Additionally, a range of other limitations such as finite natural, technological and economic resources, in conjunction with increasing human population densities (ECLAC, 2009; Pulwarty and Hutchinson, 2009), impinge on the adaptive abilities of these nations to adequately respond to challenges such as climate change and its associated loss of species diversity. The updating of the biodiversity conservation and adaptation strategies of many SIS, including those of the Caribbean, to incorporate consideration of the impacts of climate change upon terrestrial biodiversity would require the forecasting of potential species distributions within these islands in response to future climate change scenarios. Species distribution (SD) modelling, one of the most recognised and well used tools for producing such projections (Franklin, 2009). SD models are correlative models which first analyse and identify environmental conditions important to the current distribution of a given species, and then use this information to project suitable areas for the species’ survival under future climate conditions. However, SD modelling of terrestrial species within many SIS, including those of the Caribbean, has rarely been undertaken, mainly because of a lack of adequate data. Firstly, while for many islands within this region, SD data are available from several sources such as historical surveys, natural history collections and other opportunistic sources of species observations, such data are biased within geographic space and are typically strongly skewed towards more accessible areas (e.g. Schulman et al., 2007). In recent years, several papers have cautioned against the use of such non-randomly distributed location data, as this can dramatically lower the predictive ability of a SD model due to unrealistic bias among environmental parameters during its generation (Elith et al., 2011; Phillips et al., 2009). Additionally, the distribution data for many species within these islands consist of very small numbers (0.85 or < 0.85 as highly correlated, we selected the more ecologically meaningful variable within pairs of highly correlated variables for the analyses (Rissler and Apodaca, 2007). The final list of predictor variables used in these analyses is indicated within Table A2. 2.5. Removal of poor SD models and top contributing predictor variables We used the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) of the test data to evaluate the predictive accuracy of the mean model output of each species’ potential present SH. As outlined by Phillips et al. (2006), for presence-only algorithms such as MaxEnt, AUC values express the probability that a randomly selected occurrence point and a randomly selected background point will be correctly ordered by the classifier. Models having an AUC < 0.7 were considered to be poorly fitted and not accurate enough for further use (Swets, 1988), and were removed from further analyses (Table A1). These eliminated species were all ubiquitous in distribution, while those with AUC P 0.7 were range-restricted in their distribution. For those species where the MaxEnt model was considered accurate enough, the results of jackknife tests (Baldwin, 2009), which were also carried out while each model was being built by MaxEnt, were used to determine the variables which contributed the most to explaining the species’ potential SH. 2.6. Construction of presence/absence maps The MaxEnt output for the 11 species that passed the AUC test were expressed in a logistic format. These consisted of two (present and future) probability SH grid maps, within which the value of each pixel (ranging from 0 to 1) illustrates habitat suitability within that given area and time period. In order to define the potential SH of these species across the island, these SH grid maps were converted to ‘presence/absence’ binary maps (e.g. Fig. A1, Supporting information) based on a fixed sensitivity approach which selected a threshold of occurrence at a 5% omission level (analogous to setting a fixed sensitivity of 0.95). All pixels with probabilities less than the selected threshold of occurrence were given the value ‘0’ (absent) while all pixels with values more than

286


or equal to the threshold of occurrence were labelled as ‘1’ (present). This maximised the number of observed presences categorised as ‘present’, but imposed standardisation to accommodate large differences in the number of observed localities among the eleven species being modelled.

climates – within which the value of each pixel represented the total number of species for which habitat was projected to be suitable within that area of the island (Fig. 1a and b). These total count grids were then differenced to discern projected changes in the number of species with suitable habitat across the island from the present to future conditions (Fig. 1c).

2.7. Construction of change maps 2.9. Creation of a collective change map We combined presence/absence grids of each species for both the present and future climates to create ‘change maps’ which illustrate the modelled changes in each species’ SH range from its present SH to its projected SH during 2050 under an SRES A2 future. The following steps were involved: a. For each pixel within the presence/absence grids, a binary chronological sequence of pixel values was developed by combining corresponding values from the present and future grids in the format: (present future). For example, if at a given pixel, species A was projected to be absent during the present climate but present during the future climate scenario, this would be depicted by the sequence (0 1). b. There were four possible chronological sequences and corresponding decimal values which could occur: (i) (0 0) = 0 = suitable conditions do not occur at present and may not occur in the future. (ii) (0 1) = 1 = suitable conditions do not occur at present but could occur in future (expansion). (iii) (1 0) = 2 = suitable conditions occur at present but not in the future (contraction). (iv) (1 1) = 3 = suitable conditions occur both at present and in the future (stable). c. A final grid which expresses the change status of a given species was then constructed by labelling each pixel with the above decimal values of these binary combinations (e.g. Fig. A2, Supporting information). These were referred to as ‘change maps’ (Fig. A3, Supporting information) which depicted the projected change in a species’ SH across the island. 2.8. Construction of total maps The binary presence/absence grids of all 11 species were summed to yield a ‘total count’ grid for the present and the future

Number of species 0 4 1 5 2 6 3 7

8 9 10 11

Number of species 0 4 8 1 5 9 2 6 3 7

A ‘collective’ change map was constructed based upon a threshold number of species. This map combines the present and future projections of all individual species within a given group (in this case, all 11 species) to highlight areas within an island which may be suitable for the future survival of at least this threshold number of species. Our means of selecting a threshold was based on the wellestablished observation that tropical flora are characterised by high levels of alpha diversity, within which species distributions tend to follow a hollow curve, with few frequent and relatively many infrequent species (e.g. Wright, 2002). We therefore constructed frequency distribution curves of the number of species versus number of projected pixels from the present and future total count grids (Fig. 2a); from which, the number of projected pixels across the island was used as a frequency surrogate, to identify an suitable threshold number of species. We defined this threshold as the number of species which was representative of the ratio of the area projected for future collective species presence (stable and expanding) versus future areas of collective species contraction for this particular group of species. Choosing too few species would likely result in a wide fluctuation in the number of these pixels, while too many species would result in diminishing returns as the number of species progressively decreases. Hence the tangent gradients of these curves (Fig. 2a) were used as a guide to select candidate thresholds (CTs) of 2, 3, 4 and 5 species for the construction of ‘collective’ change maps (e.g. Fig. 3c). Each of these CTs was used to construct a ‘collective’ presence/ absence map for (i) the present and (ii) the future climates based on the same method used to create the individual species presence/absence maps (Section 2.6). Using the method described in Section 2.7 to create the individual change maps, the (binary) grids for these collective species presence/absence maps of the present and future climates were

Decrease of 9 to 11 species Decrease of 5 to 8 species Decrease of 1 to 4 species No change Increase of 1 to 4 species

Species occurence 0

30 km

Fig. 1. Total count grids (Present, Future and [Future – Present]). The relatively high numbers of species modelled to exist at present across the northern and western parts of the island are projected to drastically decrease under future climate conditions; during which, the southern part of the island is projected to contain the highest numbers of these species (up to a maximum of 4 species per pixel).

287

Number of pixels

2500

(a)

2000

present future

1500 1000 500 0

0

1

2

3

4

5

6

7

8

9 10 11

Number of species present

Area collective species presence / Area collective species contraction


y x

0.8

0.45

(b)

0.6 y x

0.4

0.24 y x

0.2 0

2

3

0.09

4

5

Threshold number of species

Fig. 2. Relationships analysed to determine optimal threshold for collective change map. The optimal threshold represents the number of species for which the ratio (area projected for future collective species presence versus future areas of collective species contraction) is maximised. (a) The tangent gradients of the frequency distribution curves (number of species versus number of projected pixels) from the present and future total count grids are used to identify potential optimal thresholds of 2, 3, 4 and 5 species. (b) The maximum gradient within the curve (number of pixels projecting future collective species presence versus number of pixels projecting future collective species contraction) versus potential optimal threshold was then used to identify the optimal threshold (3 species).

Collective species absent Collective species present

Collective species absent Collective species present

Collective species ... ... absence ... expansion ... contraction ... presence stable

0

30 km

Fig. 3. Combination of collective presence/absence maps to produce collective change map. Combination of present and future collective presence/absence maps to produce a collective change map (based on an optimal threshold of 3 species). A drastic decrease in the area of collective species presence is projected to occur within the northern and central regions of the island from present to future climate conditions. Areas of future collective presence projected to occur mainly within a narrow, tapering zone that extends from the south-western and south-central edge of the island to its central region.

combined to create a ‘collective species change map’ for each of the CTs. From these collective change maps (CTs = 2, 3, 4 and 5 species), we plotted the relationship between the ratio (projected areas of future collective species presence versus future collective species contraction) and the CT number of species (Fig. 2b). A threshold of 3 species was chosen as it was seen as the ‘transition’ point between a wide fluctuation in the area of collective species presence (between CTs of 2 and 3 species) and diminishing returns in the area of collective presence where the CT was more than 3 species. This technique offers a compromise in defining the area of collective species presence between the selection of a large number of species (e.g. P2 species) with highly variable distribution data, and a smaller number of species (P4 species) with the potential for fairly constant distribution data. We however, emphasise that the validity of this technique is likely to vary depending of the planning objectives involved and hence does not constitute the only way for deriving a threshold number of species to illustrate the collective response of a given group of species. For example, a conservation planner whose objective is to define an area of suitable habitat for species with specialised habitat requirements may opt to define the area of collective species presence by simply using a smaller number of species as a threshold – which gives a more conservative, less varied area of collective species presence.

3. Results 3.1. Most frequent top contributing predictor variables Among the top contributing predictor variables, BIO17 (precipitation of driest quarter) proved to be the most frequent among the eleven species models that were generated. This was followed by BIO13 (precipitation of wettest month) and a tie for the third most frequent place between BIO14 (precipitation of driest month) and Elevation. Additionally, despite just two temperature variables being included in the predictor variable list, BIO9m4 (mean temperature of driest consecutive 4-month period) also featured to be an important predictor in 3 of the 11 species modelled (Fig. A4, Supporting information).

3.2. ‘Winners’, ‘losers’ and ‘decreasing shifters’ Change maps (Fig. A3) of these 11 species were placed into broad categories of ‘winners’, ‘losers’ and ‘decreasing shifters’ (Walmsley et al., 2007) based upon projected future changes in species range (inclusive of new areas of expansion) compared to potential present SH range under the SRES A2 emissions scenario:

288


a. ‘Winners’ = Future SH range P 100% of present range. b. ‘Losers’ = Future SH range < 50% of present range. c. ‘Decreasing shifters’ = Future SH range is 50% 6 present range 6 99% (from which projected areas of expansion > 15% of present range). The majority (7) of the 11 species were categorised as losers, while just three species Brosimum alicastrum, Virola surinamensis and Ilex arimensis were categorised as decreasing shifters. Finally, the lone winner, Mora excelsa, had a 22.7% increase in range size compared to that of its potential present SH. These change maps indicated a projected loss in the SH range of all species (except M. excelsa), with the proportion and location of these losses varying widely. Additionally, in contrast to the generally large and contiguous areas of projected species disappearance, the proportions of stable SH ranges projected for all species were generally much smaller and less contiguous. Although some areas of range expansion occurred for all species, except Sterculia puriens var. glabrescens and Calophyllum lucidum, these were generally smaller than areas of stable and contracting range, and varied in terms of both location and contiguity.

species contraction in Fig. 3c; the remaining zones of (i) no change and (ii) increasing species within Fig. 1c were not similar to the areas of projected stable and/or expanding collective species presence in Fig. 3c. 3.6. RCM sensitivity analysis As our climate change projection data were derived from a single RCM under the SRES A2 emission scenario, we also explored the sensitivity of projected responses of these 11 species to climate deviations from this scenario (Table A3, Supporting information). Details of these analyses are included within Table A4 (Supporting information). The effects of changes in temperature and precipitation upon species, expressed as the anticipated change in the area of a species’ SH compared to the original SRES A2 scenario, indicated that the size of the majority of forecasted ‘‘future species SH’’ increased under drier, cooler conditions than those of the ‘‘standard’’ SRES A2 scenario. In contrast, the majority of species were projected to experience a decrease in SH under warmer or wetter conditions (Table A4).

3.3. Total maps 4. Discussion Comparison of the total grids (present versus future) revealed that areas of highest densities for these 11 species were projected to both shift and decrease (in density levels) from the northern edge and north-central region to the southern edge and southcentral region of the island (Fig. 1a and b). Differencing the present and future total count grid revealed a change in density of 11 to +4 species in the future (Fig. 1c). The greatest loss of species was projected for the northern edge and north central region of island, with the severity of loss diminishing in a southward direction. Additionally, both (i) no change in the number of species and (ii) increases ranging from 1 to a maximum of 4 species in the future were projected for the western part of the island, with increases in species number being particularly concentrated within its south-western and south-central regions. 3.4. Collective presence/absence maps The collective species presence/absence maps (Section 2.9) indicated drastic contractions of the ‘collective’ species suitable habitat range across the island from the present to future climates (e.g. Fig. 3a and b). 3.5. Collective change map The collective change map (threshold = 3 species) illustrates the anticipated changes in the ‘collective’ species presence from the present to 2050 under an SRES A2 scenario (Fig. 3c). It indicates that the collective species’ range is projected to contract to less than half its present size. Additionally, relatively smaller areas of stable collective species presence are projected to occur mainly along the central and western regions of the island’s southern edge, tapering in a northward direction into the central part of the island. While areas of collective species expansion were much smaller than the projected areas of stable collective species presence, in all cases, the former were located immediately adjacent to the latter. Smaller pockets of both stable and expanding collective species presence were also projected to occur within the south-eastern and north-eastern corners of the island as well as being scattered across the western part of the island’s northern coast. Finally, a comparison of Figs. 1c and 3c revealed that while a combination of the zones of decreasing species in the former appeared to be a rough approximation of the projected collective

4.1. Collective trends Like many SD modelling studies reported in the literature, none of the species modelled in this study were projected to maintain entirely stable SHs under the SRES A2 scenario (Fig. A3); with large contractions in potential SH range being projected for both the majority of individual species and for the collective response. Such projections across Trinidad (and likely other SIS), which has limited areas of forest available for conservation within already small, fixed geographic spaces, may imply the possible disappearance of many endemic and other high conservation value species as climate continues to change. Because of the high endemicity associated with tropical SIS (Myers et al., 2000), such a phenomenon may be likely to lead to an increased depreciation of global terrestrial species diversity, with the degree of this decline increasing if climate change becomes more severe than that projected for the SRES A2 scenario (Hannah, 2008). Further, the pattern of collective species range response in Trinidad is contrary to trends that have been reported in the literature which describe the movement of species either pole-ward or upward in altitude – in search of pre-existing climatic conditions as climate changes (but see Ogawa-Onishi et al., 2010). This is perhaps because, from the RCM sensitivity study, the majority of these species appear to favour drier, cooler conditions (Table A4) resulting from combinations of temperature, precipitation and elevation levels which did not necessarily occur in poleward directions or at higher altitudes. Furthermore, these species’ affinity for drier, cooler conditions is not anomalous, despite tropical forests being generally portrayed as consisting of a collection of plant species which thrive in humid, moist conditions. This generalisation applies only to some types of tropical forests – such as tropical rainforest; there are however other types of tropical forests which do exist outside such humid, moist conditions (Richards, 1996). Indeed consultation with local experts from Trinidad confirmed that many tree species within the island require drier conditions during important phases of their reproductive cycles and are known to be limited by too much moisture availability (Baksh-Comeau, 2010). This is further supported by the early works of Marshall (1934) and Beard (1946) which described the island’s vegetation in terms of a series of associations and formations that were directly resulting from moisture availability.


A key limitation of the application of SD modelling analyses to SIS within the Caribbean is a paucity of data, both in terms of species occurrences and the generally coarse resolution of available climate change data. Ideally, models developed from the regional (versus in-island) distribution of a given species, then projected onto the geographic area of Trinidad would be more robust – as these would be based on a more complete environmental domain. Such an endeavour would require a region wide vegetation inventory conducted within a stratified random sampling structure for each island of the Caribbean (and also across northern South America). Additionally a region-wide collation of historical survey data would be needed. However, the vast scopes of both these items make them unavailable to date. Additionally, Worldclim data were used for present climatology due to unavailability of more detailed climate data derived specifically for this region. Despite just three points within Trinidad used in the Worldclim interpolation, these did vary in elevation range and hence should have captured this climate–elevation relationship reasonably well. This is particularly true in the case of temperature variables as the associated lapse rates are known to be relatively constant for geographic domains of this size. However, errors may have been larger for rainfall as these elevation relationships vary with aspect and position relative to prevailing weather conditions. Also, compared to GCM data at c. 250 km 250 km – which do not even ‘recognise’ small islands as land, this study made use of the highest resolution RCM projection data (625 km2) that currently exists for the western Caribbean. These 625 km2 data provided 12 points across the island, adding regional variation to the projection compared to alternative GCM data (1 point). These 12 points were able to show contrasting changes between the northern and southern regions of the island, and also ‘recognised’ regions of higher elevation within the island’s interior. Furthermore, in line with concerns within the literature that representation of climate at fine resolutions could lead to misleading interpretation of results (Wiens and Bachelet, 2010), our interpolation process only projected the 625 km2 response in the RCM onto the higher resolution baseline (1 km2 – present climate), and hence did not include higher resolution local responses that are additional to those simulated at 625 km2. Additionally, adherence to the prevailing recommendation that interpretation of SD modelling forecasts be limited to general patterns (rather than details), served to minimise misinterpretation (Araújo et al., 2005; Brook et al., 2009). However, this limitation of climate data to resolutions of 625 km2 or similar can also be viewed as a fundamental limitation to the applicability of SD modelling analyses conducted at the SIS scale, as the small geographic spaces of these islands may require insight into species response at a finer scale for purposes such as conservation planning. Similar to other regions across the globe, climate scientists across the Caribbean echo the sentiment that there is a need for more accurate climate data that detail regional variation of climate and the complex topography of many of these SIS (Charlery and Nurse, 2010). They cite the lack of historical climate observations as one of the main contributors to this paucity. This has been reflected in the limited number of good quality precipitation and temperature-based predictor variables that can currently be used for SD modelling analyses within most of the Caribbean (Maharaj, 2012). However, within many islands of the Caribbean such as Trinidad, observation data for Tmax and Tmin is particularly scarce, as was evident in this study, with 7 out of the 9 ‘least correlated’ Bioclim predictor variables used for these analyses being precipitation-based instead. Indeed, our use of relatively more robust Tmean-based parameters but exclusion of those derived from Tmax and Tmin imply that while first order temperature responses were captured by the mean temperature data, second order responses due to subtler variations in Tmax and Tmin parame-

289

ters may have been missed. There is also a pervading lack of knowledge on the effect of temperature upon the distribution of these species within the Caribbean, and hence it is unsurprising that no temperature-based predictor variables were suggested as ecologically meaningful by regional experts. Indeed this paucity of both sound temperature-based climate data as well as literature or expert opinion on the mechanistic impact of temperature upon many floral species within these islands may not be co-incidental; with the latter perhaps being symptomatic of the former. However, the results of this study’s RCM sensitivity analyses indicate that these species are sensitive to changes in temperature, hence further exploration of this issue is suggested. 4.2. Value of collective response It has also been well established within a large and increasing literature that species are responding on an individual basis to an already changing climate (Elith and Leathwick, 2009), with the majority of SD modelling literature reporting species response from both an individual or multi-species perspective. However, to the authors’ knowledge, there are no reports within the SD modelling literature which report a ‘collective’ species response (i.e. the combined response of multiple species). Existing approaches have combined model projections from more than one species via ‘stacking’ by summing binary presence/absence maps of individual species to describe species richness across a given study area (Dubuis et al., 2011; Mateo et al., 2012). The collective change map takes this concept of stacking a step further by allowing for the combination of threshold-based stacked models from two scenarios into one layer – illustrating projected change highlighted as areas of stable, expanding or contracting species presence. This collective species response approach used in this study should not be confused with attempts to model community response to climate change such as those discussed by Ferrier et al. (2002) and Ferrier and Guisan (2006). While the initial stages of the construction of the collective change map may be similar to the ‘predict first, assemble later’ strategy described by Ferrier and Guisan (2006), this study does not attempt to derive its collective response via ecological assembly rules (Keddy, 1992) to the models of individual species. While biotic interactions may be important to the future SH of some species, such interactions are generally considered to be (i) dynamic and prone to constant change in space and time and (ii) more important when modelling at the micro-scale, while climate factors tend to dominate at the macro-scale. Hence, this collective change map also does not integrate community-level ordinations, aggregations or classifications, nor does it imply that the areas of collective presence are partially based upon species interactions. Instead, by assuming additive properties of the individual species presence/absence maps, it provides an easy to understand and implement method of illustrating a collective (or thresholdbased majority) response of the components within any given set of species to novel conditions such as climate change. Further, construction of these collective maps is flexible depending on the planning objectives involved and can be applied sans an optimal threshold of species; for example, such maps can be constructed by replacing the optimal threshold with a fixed number of species determined on an a priori basis. With both palaeoecological records (Graham and Grimm, 1990) and modelled simulations of future range shifts indicating that species are likely to respond individually (instead of as community) to climate change, and future species assemblages are likely to be substantially different from those of the present (Huntley et al., 2006; Thuiller et al., 2011), we believe that this technique has the potential to be of considerable applied value. For example, its application can provide conservation practitioners with a means of assessing the placement, potential biodi-

290


versity value and hence, the effectiveness of present/future protected area networks – based on suitable environmental space for a maximised collection of high conservation value or endemic species (not necessarily belonging to the same present-day community) rather than for individual species. Such information could provide a basis upon which further planning considerations can be integrated; for example some projected areas of SH may not necessarily be occupied, hence additional costs would be required for the establishment of missing plant species. As species continue to shift with the changing climate, such application may be particularly useful for conservation planning within many SIS - which have very limited choices of land areas that can be used for the designation of protected areas as a result of deforestation driven by fierce competition from alternative uses such as agriculture and housing. Additionally, the simplicity of this technique allows for its applications to other areas within and outside of conservation planning which require processing and collation of sequential spatial data. 5. Conclusion We believe that despite the above mentioned limitations of applying SD modelling to SIS, such work is necessary in order to understand, and perhaps enable better planning and management of species diversity at global, regional and local scales in response to an already changing climate. The application of SD modelling results such as those presented in this work to real-world conservation issues such as the assessment of protected area effectiveness may be a small step in this direction. As evidence from elsewhere suggests, it can no longer be assumed a priori that the existing terrestrial protected area networks within SIS will continue to be effective if changes in climate induce species loss and shifts (Hannah, 2008; Pressey et al., 2007). Hence, we are in the process of considering climate change along with other factors such as protected areas and their buffers to further inform sustained conservation of important species within these islands. Acknowledgements The authors are grateful to Dr. John Charlery (University of the West Indies, Cavehill Campus, Barbados) for provision of postprocessed downscaled climate data used these analyses; Dr. Stephen Harris (Department of Plant Sciences, University of Oxford) for generously editing and providing very constructive suggestions on the structuring of the paper’s background; Prof. John Agard (University of the West Indies, St. Augustine Campus, Trinidad) and Dr. Nick Brown (Linacre College, University of Oxford) for their advice and support. We also acknowledge the support received at a writeshop organised by the Stockholm Environment Institute and the University of the West Indies, funded by the UN International Strategy for Disaster Reduction. Finally, we thank Linacre College at the University of Oxford for its financial support of this work. Appendix A. Supplementary Material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.biocon.2013. 08.027. References Araújo, M.B., Pearson, R.G., Thuiller, W., Erhard, M., 2005. Validation of speciesclimate impact models under climate change. Glob. Change Biol. 11, 1504– 1513. Baksh-Comeau, Y.S., 2010. Interview by author. Baksh-Comeau, Y.S., Hawthorne, W.D., Harris, S.A., Maharaj, S.S., Filer, D.L., Unpublished. The vascular flora of Trinidad and Tobago: a checklist with conservation status.

Baldwin, R.A., 2009. Use of maximum entropy modeling in wildlife research. Entropy 11, 854–866. Beard, J.S., 1946. The Natural Vegetation of Trinidad. Clarendon Press, Oxford. Brook, B.W., Akçakaya, H.R., Keith, D.A., Mace, G.M., Pearson, R.G., Araújo, M.B., 2009. Integrating bioclimate with population models to improve forecasts of species extinctions under climate change. Biol. Lett. 5, 723–725. Campbell, J.D., Taylor, M.A., Stephenson, T.S., Watson, R.A., Whyte, F.S., 2011. Future climate of the Caribbean from a regional climate model. Int. J. Climatol. 31, 1866–1878. Charlery, J., Nurse, L., 2010. Areal downscaling of global climate models: an approach that avoids data remodeling. Clim. Res. 43, 241–249. de Chazal, J., Rounsevell, M.D.A., 2009. Land-use and climate change within assessments of biodiversity change: a review. Global Environmental Change 19, 306–315. Dubuis, A., Pottier, J., Rion, V., Pellissier, L., Theurillat, J.-P., Guisan, A., 2011. Predicting spatial patterns of plant species richness: a comparison of direct macroecological and species stacking modelling approaches. Divers. Distrib. 17, 1122–1131. ECLAC, 2009. Economic Commission for Latin America and the Caribbean. Elith, J., Graham, C.H., 2009. Do they? How do they? WHY do they differ? On finding reasons for differing performances of species distribution models. Ecography 32, 66–77. Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation and prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697. Elith, J., Phillips, S.J., Hastie, T., Dudík, M., Chee, Y.E., Yates, C.J., 2011. A statistical explanation of MaxEnt for ecologists. Divers. Distrib. 17, 43–57. Ferrier, S., Guisan, A., 2006. Spatial modelling of biodiversity at the community level. J. Appl. Ecol. 43, 393–404. Ferrier, S., Watson, G., Pearce, J., Drielsma, M., 2002. Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species-level modelling. Biodivers. Conserv. 11, 2275–2307. Fowler, H.J., Blenkinsop, S., Tebaldi, C., 2007. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. Int. J. Climatol. 27, 1547–1578. Franklin, J.F., 2009. Mapping Species Distributions: Spatial Inference and Prediction. Cambridge University Press, Cambridge, UK. Government of Trinidad and Tobago, 1972a. 1:25,000 Soil Map Series. Government of Trinidad and Tobago, 1972b. 1:25,000 Topographic Map Series, Sheet Numbers 3–76. Government of Trinidad and Tobago, 1972c. Land Capability Survey of Trinidad and Tobago. Graham, R.W., Grimm, E.C., 1990. Effects of global climate change on the patterns of terrestrial biological communities. Trends Ecol. Evol. 5, 289–292. Hannah, L., 2008. Protected areas and climate change. Ann. N. Y. Acad. Sci. 1134, 201. Hawthorne, W.D., Abu-Juam, M., 1995. Forest Protection in Ghana: With Particular Reference to Vegetation and Plant Species. IUCN in Collaboration with ODA and the Forest Department Republic of Ghana, Gland, Switzerland and Cambridge, UK. Heller, N.E., Zavaleta, E.S., 2009. Biodiversity management in the face of climate change: a review of 22 years of recommendations. Biol. Conserv. 142, 14–32. Helmer, E.H., Ruzycki, T.S., Benner, J., Voggesser, S.M., Scobie, B.P., Park, C., Fanning, D.W., Ramnarine, S., 2012. Detailed maps of tropical forest types are within reach: forest tree communities for Trinidad and Tobago mapped with multiseason Landsat and multiseason fine-resolution imagery. For. Ecol. Manage. 279, 147–166. Hickling, R., Roy, D.B., Hill, J.K., Fox, R., Thomas, C.D., 2006. The distributions of a wide range of taxonomic groups are expanding polewards. Glob. Change Biol. 12, 450–455. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25, 1965–1978. Huntley, B., Collingham, Y.C., Green, R.E., Hilton, G.M., Rahbek, C., Willis, S.G., 2006. Potential impacts of climatic change upon geographical distributions of birds. IBIS 148, 8–28. Keddy, P.A., 1992. Assembly and response rules: two goals for predictive community ecology. J. Veg. Sci. 3, 157–164. Kennedy, H., 2004. Data in Three Dimensions: A Guide to ArcGIS 3D Analyst. OnWord Press, New York. Lenoir, J., Gégout, J.C., Marquet, P.A., de Ruffray, P., Brisse, H., 2008. A significant upward shift in plant species optimum elevation during the 20th century. Science 320, 1768–1771. Maharaj, S.S., 2012. The impact of climate change on the small island developing states of the Caribbean. Plant Sciences. University of Oxford, Oxford. Malcolm, J.R., Liu, C., Neilson, R.P., Hansen, L.J., Hannah, L., 2006. Global warming and extinctions of endemic species from biodiversity hotspots. Conserv. Biol. 20, 538–548. Marshall, R.C., 1934. Trees of Trinidad and Tobago. Government Printing Office, Port-of-Spain, Trinidad. Mateo, R.G., Felicísimo, Á.M., Pottier, J., Guisan, A., Muñoz, J., 2012. Do stacked species distribution models reflect altitudinal diversity patterns? PLoS ONE 7, e32586. Mittermeier, R.A., Gil, P.R., Hoffman, M., Pilgrim, J., Brooks, T., Mittermeier, C.G., Lamoreux, J., da Fonseca, G.A.B., 2005. Hotspots revisited: earth’s biologically richest and most endangered terrestrial ecoregions. Conservation International.

S.S. Maharaj, M. New / Biological Conservation 167 (2013) 283–291 Myers, N., 1988. Threatened biotas: ‘‘Hot spots’’ in tropical forests. Environmentalist 8, 187–208. Myers, N., 1990. The biodiversity challenge: expanded hot-spots analysis. Environmentalist 10, 243–256. Myers, N., Mittermeier, R.A., Mittermeier, C.G., da Fonseca, G.A.B., Kent, J., 2000. Biodiversity hotspots for conservation priorities. Nature 403, 853–858. Ogawa-Onishi, Y., Berry, P.M., Tanaka, N., 2010. Assessing the potential impacts of climate change and their conservation implications in Japan: a case study of conifers. Biol. Conserv. 143, 1728–1736. Phillips, S.J., Anderson, R.P., Schapire, R.E., 2006. Maximum entropy modeling of species geographic distributions. Ecol. Model. 190, 231–259. Phillips, S.J., Dudík, M., 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31, 161–175. Phillips, S.J., Dudík, M., Elith, J., Graham, C.H., Lehmann, A., Leathwick, J., Ferrier, S., 2009. Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecol. Appl. 19, 181–197. Pounds, A.J., Bustamante, M.R., Coloma, L.A., Consuegra, J.A., Fogden, M.P.L., Foster, P.N., La Marca, E., Masters, K.L., Merino-Viteri, A., Puschendorf, R., Ron, S.R., Sánchez-Azofeifa, G.A., Still, C.J., Young, B.E., 2006. Widespread amphibian extinctions from epidemic disease driven by global warming. Nature 439, 161– 167. Pressey, R.L., Cabeza, M., Watts, M.E., Cowling, R.M., Wilson, K.A., 2007. Conservation planning in a changing world. Trends Ecol. Evol. 22, 583–592. Pulwarty, R.S., Hutchinson, N., 2009. Vulnerability and Capacity Assessment: A Guidance Manual for the Conduct and Mainstreaming of Climate Change Vulnerability and Capacity Assessments in the Caribbean Region. Belmopan, Belize. Richards, P.W., 1996. The Tropical Rainforest. An Ecological Study, second ed. Cambridge University Press, Cambridge, UK. Rissler, L.J., Apodaca, J.J., 2007. Adding more ecology into species delimitation: ecological niche models and phylogeography help define cryptic species in the black salamander (Aneides flavipunctatus). Syst. Biol. 56, 924–942. Schulman, L., Toivonen, T., Ruokolainen, K., 2007. Analysing botanical collecting effort in Amazonia and correcting for it in species range estimation. J. Biogeogr. 34, 1388–1399. Spak, S., Holloway, T., Lynn, B., Goldberg, R., 2007. A comparison of statistical and dynamical downscaling for surface temperature in North America. J. Geophys. Res. 112, D08101. Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science, 1285– 1293.

291

Taylor, M.A., Centella, A., Charlery, J., Borrajero, I., Bezanilla, A., Campbell, J.D., et al., 2007. Glimpses of the Future: A Briefing from the PRECIS Caribbean Climate Change Project. Caribbean Community Climate Change Centre, Belmopan, Belize, p. 24. Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M., Beaumont, L.J., Collingham, Y.C., Erasmus, B.F.N., de Siqueira, M.F., Grainger, A., Hannah, L., Hughes, L., Huntley, B., van Jaarsveld, A.S., Midgley, G.F., Miles, L., Ortega-Huerta, M.A., Townsend Peterson, A., Phillips, O.L., Williams, S.E., 2004. Extinction risk from climate change. Nature 427, 145–148. Thuiller, W., Lavergne, S., Roquet, C., Boulangeat, I., Lafourcade, B., Araujo, M.B., 2011. Consequences of climate change on the tree of life in Europe. Nature 470, 531–534. Walmsley, C.A., Smithers, R.J., Berry, P.M., Harley, M., Stevenson, M.J., Catchpole, R., 2007. MONARCH – Modelling Natural Resource Responses to Climate Change: A Synthesis for Biodiversity Conservation. Oxford. Walther, G.-R., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T.J.C., Fromentin, J.-M., Hoegh-Guldberg, O., Bairlein, F., 2002. Ecological responses to recent climate change. Nature 416, 389–395. Wiens, J.A., Bachelet, D., 2010. Matching the multiple scales of conservation with the multiple scales of climate change. Conserv. Biol. 24, 51–62. Wright, J.S., 2002. Plant Diversity in Tropical Forests: A Review of Mechanisms of Species Coexistence. Oecologia 130, 1–14.

Glossary AUC: Area Under the Curve Bioclim: Bioclimatic Predictor Variable GCM: Global Climate Model P: Precipitation PRECIS: Providing REgional Climates for Impacts Studies RBS: Rapid Botanical Survey RCM: Regional Climate Model ROC: Receiver Operating Characteristic SIS: Small Island State SD: Species Distribution Tmax: Mean Monthly Maximum Temperature Tmean: Mean Monthly Temperature Tmin: Mean Monthly Minimum Temperature