Bellwether of the Canaries: Anthropogenic influence on the land. snail community of the Canary Islands. Alexander F. Wall, June 13, 2016

Transcription

1 Bellwether of the Canaries: Anthropogenic influence on the land snail community of the Canary Islands Alexander F. Wall, June 13, 2016 B.S., University of Iowa, 2007 B.A., University of Iowa, 2007 A thesis submitted to the Graduate School of the University of Cincinnati in partial fulfillment for the degree of Master of Science in the Department of Geology of the College of Arts and Sciences by July 2016 Name of committee chairs: Dr. Yurena Yanes Dr. Arnold I. Miller Committee member: Dr. Joshua Miller

2 Abstract In many parts of the world, human activities are now commensurate with natural environmental factors in governing the distributions and abundances of species. Natural areas near human-modified landscapes experience pollution, fragmentation, water diversion, introduced species, and other factors that may affect local biodiversity. The land snails of the Canary Islands provide excellent opportunities to diagnose the importance of anthropogenic agents in mediating the diversity and distribution of species. Land snails are particularly sensitive to disturbance and are an integral part of terrestrial ecosystems. This study analyzed the distributions and abundances of land snail subfossils at 60 localities throughout the Canary Islands coastal scrub biome. This was accomplished using data on natural and anthropogenic variables to assess their relative importance in governing land-snail diversity. A total of 34,801 dead shells represent a diverse malacofauna with highly localized endemism. A novel combination of analytical methods was able to distinguish the impact of anthropogenic factors on native land-snail diversity at the biome scale. Regression tree analyses indicated that diversity decreased with increased proximity to agricultural sites and airports the latter taken as a proxy for heavily urbanized areas. Sites varied in biodiversity, e.g. from an average of 3.6 species in sites nearest human-modified areas, to an average of 6.9 species in the most remote sites. Impacts from modified areas affected natural areas at distances from ~8 km to less than 0.1 km away. Protected coastal scrub areas ii

3 near human-modified landscapes may not, therefore, be effectively protecting some members of their constituent communities. iii

4 iv

5 Acknowledgements This research was funded by the following grants: The Geological Society of America Graduate Student Research Grant, The Conchologists of America Academic Grant, and the Society for Sedimentary Geology Student Assistance Grant. Thank you to the government of the Canary Islands for permitting the collection of the samples that are the foundation of this research. I would like to thank my advisors Yurena Yanes and Arnold I. Miller for their advice, support, and assistance in the field, laboratory, and classroom. Together, they and my fellow graduate student Elizabeth Bullard contributed the majority of specimens analyzed in this thesis and provided feedback and fresh ideas for every step in that analysis. Fellow graduate student Elizabeth Hausner helped plan the fieldwork while simultaneously teaching me the GIS skills I would make extensive use of in the production of this thesis. My third committee member, Dr. Joshua Miller introduced me to R and has provided constant support, both technical and theoretical, throughout the project. Special thanks go to Miguel Ibañez and Maria R. Alonso (Universidad de La Laguna) for their assistance with some land snail species identifications. v

6 Table of contents Acknowledgements... ii Table of contents... vi Abstract... ii Introduction... 1 Field methods... 5 Statistical Methods Results Land snails species composition and distribution Environmental factors controlling biodiversity Discussion Conclusions References Supplemental Material... 1 List of Figures Figure 1. Geographical location of the Canary Archipelago... 7 Figure 2. Rarefaction curves of each site Figure 3. Rarefied richness of land snails by site Figure 4. Evenness of snails by site Figure 5. A three dimensional NMDS Figure 6 Regression tree of land snail richness vi

7 Figure 7. Regression tree of land snail Shannon diversity Figure 8. Regression tree of land snail Simpson diversity Table 1. Summary of measured environmental factors... 9 Table 2. Summary of land snail species per island Table 3. Summary of endangered species Supplementary Materials: Figures Figure SM 1. Map of field sites... 1 Figure SM 2. Boxplot showing richness grouped by Otala lactea... 5 Figure SM 3. Richness regression tree first split... 6 Figure SM 4. Richness regression tree second split... 7 Figure SM 5. Richness regression tree pruning parameters... 8 Figure SM 6. Shannon diversity regresiion tree first split... 9 Figure SM 7. Shannon divesity regression tree second split Figure SM 8. Shannon diversity regression tree pruing parameters Figure SM 9. Simspon diversity regression tree first split Figure SM 10. Simpson divesity regression tree second split Figure SM 11. Simpson diversity regression tree pruning parameters vii

8 Figure SM 12. Detailed map of El Hierro land cover and sites with 1 km buffers Figure SM 13. Detailed map of Fuerteventura land cover and sites with 1 km buffers Figure SM 14. Detailed map of Gran Canaria land cover and sites with 1 km buffers Figure SM 15. Detailed map of La Gomera land cover and sites with 1 km buffers Figure SM 16. Detailed map of Lanzarote land cover and sites with 1 km buffers Figure SM 17. Detailed map of La Palma land cover and sites with 1 km buffers Figure SM 18. Detailed map of Tenerife land cover and sites with 1 km buffers Supplementary Materials: Tables Table SM 1. Species counts... 1 Table SM 2. Results of rarefaction Table SM 3 Hill Numbers Example viii

9 Introduction Anthropogenic landscape modification has unintended ecological consequences for organisms in the modified areas (Fischer and Lindenmayer 2007 and sources cited therein). Crops, buildings, infrastructure, etc. replace natural biomes (the coexisting organisms in an area of similar environmental conditions). Yet these anthropogenic landscapes also create a halo of impact around the modified area through pollution, dispersal obstruction, water diversion, introduced species, and other factors (Wood et al. 2000). Quantifying the degree to which anthropogenic activities impact the biodiversity of nearby natural biomes, and the reach of these impacts, is critical to successful conservation initiatives. Discriminating between natural and anthropogenic agents in any ecosystem is not without its challenges. Historical records of biodiversity predating human modification tend to be rare and incomplete, so information on pre-disturbance baselines is rare. As an alternative, substituting space for time has met with success in some studies that compare community structure among areas experiencing varying degrees of disturbance (McKinney 2008 and sources cited therein, McDonnell and Hahs 2008 and sources cited therein, Horsák et al. 2009, Lososová et al. 2011, Horsák et al. 2013, Kolbe et al. 2016). Most of these studies seek to hold fixed the influence of natural factors by limiting the study area, for example, to a single city and its surroundings. As an alternative, an investigation of sites that share key natural attributes but are distributed over a 1

10 much broader area might provide opportunities to evaluate the effects of a range of anthropogenic factors. Anthropogenic-impact studies often compare the biodiversity metrics of a community found in both its natural and modified states (Clergeau 1998, McKinney 2008 and sources cited therein, McDonnell and Hahs 2008 and sources sited therein, Horsák et al. 2009, Lososová et al. 2011, Horsák et al. 2013, Kolbe et al. 2016). For example, Niemelä et al. (2002) found that the richness of carabid beetle communities tended to decrease with increasing urbanization. Conservation efforts, however, often focus on areas that have not been modified by humans and therefore may be better served by studies focused on human factors that "leak" into natural areas. This may be accomplished by selecting sites that maintain the primary constituents and general structure of the natural biome, but vary in their exposure to anthropogenic factors. This should highlight areas where the biome is ostensibly intact yet may contain depauperate constituent communities. Finally, many studies of anthropogenic impacts characterize anthropogenic factors qualitatively; categorizing impacts as low, medium, and high (Clergeau 1998, McKinney 2008 and sources cited therein, McDonnell and Hahs 2008 and sources sited therein, Horsák et al. 2009, Lososová et al. 2011, Horsák et al. 2013). In contrast, Kolbe et al. (2016) used data from a variety of sources to quantify a broad suite of natural and anthropogenic factors. They then used regression trees to tease apart the varying degrees of influence attributable to different factors. 2

11 Here, the ecological fallout of anthropogenic factors is quantified at a geographical scale directly applicable to conservation efforts. Selecting field sites in the semi-arid coastal scrub biome of the Canary Islands controls for the influence of natural factors. A series of estimates of the types and extent of nearby landscape modification were used to assess anthropogenic pressures on each site. Native land-snail diversity was used as a measure of community "health" and, considering the often-observed decline in native snail elsewhere (McMillan et al. 2003, Horsák and Némethová 2008, Lososová et al. 2011, Yanes 2012b, Douglas et al. 2013), it was hypothesized that proximity to humanmodified landscapes has a deleterious effect on the diversity of the native malacofauna of the Canarian coastal scrub. Land snails have characteristics that are particularly advantageous for biodiversity assessments. After death, they often leave behind their shells ( subfossils ), which can be plentiful and are easily collected. Land-snail subfossil assemblages have been shown to be a close match to the living assemblage (Rundell and Cowie 2003, Pearce 2008, Thurman et al. 2008, Cernohorsky et al. 2010, Yanes 2012a, Albano 2014). Plus, multi-year to decadal time-averaging reduces the expression of ephemeral variations (Kidwell 2007, Pearce 2008, Albano 2014, Yanes 2012b). Unlike the multi-millennial ages of marine subfossils in many coastal settings (Kowalewski et al. 1998, Kidwell et al. 2005), Pearce (2008) found that land-snail shells in a temperate forest had a half-life of years and suggested this may be much longer in arid, carbonate-rich settings (like the coastal scrub). A variety of studies have taken advantage of these 3

12 characteristics of subfossil land-snail assemblages, using them in studies of urban-rural gradients (Horsák et al. 2008, Lososová et al. 2011, Horsák et al. 2013), forest succession after logging (Douglas et al. 2013), anthropogenic impacts on live-dead fidelity (Yanes 2012a, b), natural park management (Bros et al. 2016), and heavy metal pollution (Regoli et al. 2006). Their growing reputation as sentinel species for biome health, however, has not been tested on the biome geographic scale In addition to being an expeditious study system, the Canarian malacofauna is understudied. While Pleistocene (pre-human) extinctions have been minimal (Yanes et al., 2011c) and historical extinctions have not been recorded (Núñez and Núñez 2010), continual urbanization appears to be growing threat (Fernández-Palacios and Whittaker 2008). Little has been published on the abundances and distributions of Canarian snails, so data on many species are insufficient to properly assess their conservation statuses (IUCN 2015). This study has the added dimension of providing such information. The coastal scrub biome's relatively consistent climate and vegetation type throughout the seven Canary Islands make it suitable for this study. As a case in point, moisture availability is one of the most important natural factors controlling land-snail distribution (Cook 2001), and the coastal scrub is relatively uniformly semi-arid. Average annual precipitation fluctuates about 100 mm within the biome on any of the Canary Islands, and ranges from 150 to 400 mm among all sites. The coastal scrub also has fairly clear boundaries with adjoining biomes (its upper/inland boundary being thermophilous forest, its outer the ocean), 4

13 abundant land snails with high preservation potential due to generally calciumrich soils (Fernández Caldas et al. 1987), and is the only biome present on all seven islands. Finally and crucially, coastal scrub occurs near agriculture and urban development, as well as in relatively pristine areas. Methods Field methods This study used the European Environment Agency's (EEA) coordination of information on the environment (CORINE) land cover inventory (EEA 2013) to target sampling sites. Human-modified areas were most broadly categorized as agricultural or artificial (the latter including urban, recreational, and industrial areas) following EEA (1995) nomenclature. Natural areas that the EEA identified as "sclerophyllous vegetation" within 5km of the coast were considered coastal scrub in this study. On each island, targeted areas included natural coastal scrub vegetation that was: (1) adjacent to, or surrounded by, agricultural and artificial landscapes; (2) in remote "pristine" areas; or (3) somewhere in between. Maps of these areas are provided in the Supplementary Material (Figures 12 18). Within targeted areas, the presence of typical coastal scrub flora (e.g. Euphorbia, Schizogyne, Launaea, Lycium (Otto et al. 2006)), an elevation below 500 m a.s.l., and accessibility determined the placement of sites. In all, this study sampled 60 sites in the coastal scrub biome throughout the Canary Islands (Fig. 1). In general, more sites were sampled on larger 5

14 islands, but at least six were conducted on each island. The methods prescribed by Cameron and Pokryszko (2005), and Coppolino (2010) were used to collect land-snail specimens. Initial prospecting of sites ensured the presence of snails to maximize their abundance and richness. This practice is common for collecting macro-snails and helps to compensate for their patchy distributions and improves detection of rare species (Szybiak 2009, Coppolino 2010, Bros et al. 2016). Two to four workers marked out 30 m by 30 m plots, then visually searched for and collected any dead snail shells for one hour. Whenever found, living specimens were counted directly in the field. This work, however, focused on data extracted from subfossil assemblages only. Four workers collected dead specimens at each site in Tenerife, Fuerteventura, and Lanzarote; in El Hierro three workers; and two in La Palma, La Gomera, and Gran Canaria. The only exceptions were sites 23 and 24 in Lanzarote, and 45 in La Gomera, which had 3, 2, and 1 collector, respectively. Counts were rarefied to compensate for this discrepancy (see Statistical Methods). 6

15 Figure 1. Geographical location of the Canary Archipelago. Black dots and numbers depict field sites where dead shells were collected. Shells were commonly found on the soil surface, between or under rocks, or amongst plant litter. Only shells that were complete enough for positive identification and that included the apex were counted. Species were identified based on comparison with specimens from the mollusk collection in the Malacology Laboratory of the University of Cincinnati, combined with the most recent literature (Ibáñez et al. 2006, Yanes et al. 2009a, Yanes et al. 2011a, Yanes et al. 2011b, Castro et al. 2012, Santana et al. 2013, Alonso et al. 2013, Alonso and Ibáñez 2015a, Alonso and Ibáñez 2015b, Alonso and Ibáñez 2015c), and the assistance of local experts Miguel Ibáñez and Maria R. Alonso from the Universidad de La Laguna. Compiled data on 39 different factors for each sampling location (Table 1) allowed characterization of the environments in which snails were living. Five of these were characteristics for each island (e.g., island area). Island ruggedness (or roughness) was calculated as the ratio of true surface area and the 7

16 planimetric area, such that a 1 km 2 patch with a steep slope or convoluted surface shape has a higher value than one nearer a flat plane (Hoechstetter et al. 2006). Many of the factors were derived from the EEA CORINE data set (EEA 2013). As described above, areas were categorized as agricultural, artificial, and natural landscapes. Following Kolbe et al. 2016, some of the environmental variables were reported as areas within 1 km buffers around sites. Measurements of length, distance, and area were calculated in log 10 km to normalize distributions. The length of edges shared between the coastal scrub and the other land cover types indicated landscape heterogeneity. For example, a site adjacent to a large plantation may have the same nearby agricultural area (Area Ag ) as a site surrounded many small fields, but the surrounding scrub vegetation may share much less of its length of borders with agriculture (Edge ScrubAg ) in the former case. Also included are simple distances to agricultural and artificial landscapes, as well as specific features that may be particularly important, including garbage dumps, recreational areas, airports, and industrial areas. Roads were categorized according to EEA (1995) designations. Impact is a qualitative description as to whether a site were impacted by agricultural, artificial, both, or neither types of anthropogenic factor based on observations in the field (e.g. nearby buildings, litter, goats, etc.). Sites were categorized as protected if they were within the limits of a governmentdesignated natural area with restrictions on landscape modification. The remaining measured factors are described in Table 1. Topographical data were taken from the EEA Elevation map of Europe (EEA 2004). Climatic and 8

17 population data were taken from the WorldClim global climate data set (Hijmans et al. 2006). All geographic analyses were completed using ArcGIS Desktop release 10.3 (ESRI 2014). Table 1. Summary of measured environmental factors and the abbreviations used to encode them for analysis. Natural factor code Anthropogenic factor code Island area, log 10 km 2 IsArea 1 Population density as of 2000, the most recent data available Age of island, millions IsAge 1 Total length of roads within of years 1km buffer, log 10 km Number of biomes on IsHabDiv 1 Distance to the nearest island small road, log 10 km Island ruggedness, IsRug 1 Distance to the nearest 3D/2D area ratio highway, log 10 km Island's shortest distance to the African continent, log 10 km Annual precipitation mm, 50 year average Maximum annual temperature ºC, 50 year average Minimum annual temperature ºC, 50 year average AfricaDist 1 AnnualPrecip MaxAnnualT MinAnnualT Distance to the nearest freeway, log 10 km Distance to nearest road of any kind, log 10 km Distance to the nearest agricultural area, log 10 km Distance to nearest artificial surface, log 10 km Elevation in meters Elevation Distance to the nearest airport, log 10 km Slope in degrees Slope Distance to the nearest urban area, log 10 km Direction of slope Northness Distance to the nearest exposition (N = 1, S = dump or landfill, log 10 km -1) Direction of slope exposition (E = 1, W = -1) Coastal scrub area within 1km of site, log 10 km 2 Edges of coastal scrub within 1km buffer, log 10 km Coastal scrub edge bordering natural features, log 10 km Distance to the nearest stream, Eastness Area Scrub 2 Edge ScrubTot 2 Edge ScrubNat 2 Dist Stream Distance to the nearest recreational area, log 10 km Distance to the nearest industrial site, log 10 km Human modified area of any kind within 1km buffer, log 10 km 2 Agricultural area within 1km buffer, log 10 km 2 Artificial area within 1km buffer, log 10 km 2 HumPop RoadDens Dist SmRd Dist MedRd Dist BigRd Dist AnyRd Dist Ag Dist Art Dist Air Dist Urb Dist Dump Dist Golf Dist Ind Area Mod Area Ag Area Art 9

18 log 10 km A qualitative assignment of agricultural or urban influence Impact 3 1 Measurements of islands parameters, not individual sites 2 Measurements could result from natural or anthropogenic factors 3 Categorical variables Coastal scrub edge bordering agriculture, log 10 km Edges shared between artificial and agricultural surfaces, log 10 km Coastal scrub edge bordering artificial surfaces, log 10 km Edges shared between artificial and natural surfaces, log 10 km Coastal scrub edge bordering any humanmodified surface, log 10 km Whether site is part of a gov t designated natural area Edge ScrubAg Edge ArtAg Edge ScrubArt Edge ArtNat Edge ScrubMod Protected 3 Statistical Methods Hill numbers are used to report diversity statistics. As discussed in greater detail in Chao et al. 2014, the generalized equation for Hill numbers when q 1 is (Equation 1) where S is the number of species, p i is the proportional abundance of species i, and q the order of diversity. Values of q are chosen by the investigator to vary emphasis on common species. When q = 0, 0 D is simply sample richness. As q increases, common species are given increasing weight. When q = 1, (Equation 2) 10

19 This is identical to e raised to the power of the Shannon-Wiener Index; that is, ln( 1 D) is the Shannon-Wiener Index and similarly weights species proportionally to their abundances. 2 D is the inverse of Simpson's index and also weights common species more heavily. Following Chao et al. (2014), These three commonly used Hill numbers are referred to as richness ( 0 D), Shannon diversity ( 1 D), and Simpson diversity ( 2 D). Hill numbers hold several advantages over their index counterparts. They are reported with a unit: species equivalents sometimes simply called true diversity which is the number of equally abundant species required to achieve the same diversity measurement (Chao et al. 2014). More importantly, Hill numbers adhere to the doubling property: when two samples with identical diversity measurements, but no shared species, are combined into a single sample, the diversity measurement of the third sample should be double that of either of the original two (Hill 1973). This makes comparison between samples and diversity measurements more straightforward (see example in Supplementary Material, Table 3). Most measures of diversity, including Hill numbers, are sensitive to sampling effort, and, as noted earlier, my samples were collected with some variability in number of collectors (from 1 to 4). To mitigate this, samples were rarefied after Chao and Jost (2012). Rarefaction curves were created by finding a sample's expected richness at many subsample sizes. These demonstrated how richness increased with increased subsample size for each sample. Sample completenesses were reduced to that of the least complete sample (sample 33 11

20 with 73 individuals) so diversity measurements could be compared (see Supplementary Material, Table 2). Richnesses at sites including introduced species were compared to those with only natives using Welch's t-tests (Welch 1947). This test was favored over the Student's t-test due to unequal sample sizes and variances. Native diversity was compared between the group of sites with an introduced species to the group without it, to detect signs of invasiveness (broadly defined as introduced species having a detrimental effect on native biodiversity). To directly investigate differences in species composition between sites, an NMDS (non-metric multidimensional scaling) was constructed. Samples were transformed using Wisconsin double standardization, that is, species maximum transformed and then sample total transformed. Samples were then plotted using Bray-Curtis similarity (Faith et al. 1987) such that sites nearer each other had more similar species compositions. Regression trees of the first three Hill numbers richness, Shannon diversity, and Simpson diversity were the primary means of investigation for this study. A regression tree is a machine learning technique that creates a hierarchical ranking of variables that best explain the distribution of a response variable. In this case, the explanatory variables were environmental factors, and the response variable was a measure of diversity ( q D). Each explanatory variable was used to recursively split sites into two groups until the total variance of the response variable is minimized. The variable that minimizes the variance is considered to have the most explanatory power and the process is repeated for 12

21 each of the two subgroups. Explanatory variables are recycled such that one may be used to make multiple splits. The process is repeated until the terminal subgroups include few samples and further splits do not improve the model. To avoid overfitting, a standard pruning procedure using the one standard error rule (Breiman et al. 1984), is then applied to remove splits that don't sufficiently improve the model. This reduces the size of the tree to the most parsimonious sub-tree whose error is no more than one standard error greater than the error of the best sub-tree (Breiman et al. 1984, Lemon et al. 2003, Hothorn et al. 2015). All statistical analyses were completed using RStudio vers (RStudio Team, 2015) and R vers (R Development Core Team 2016). NMDS was conducted using functions available in the vegan package (Oksanen et al. 2016). Regression trees were produced using the party package (Hothorn et al. 2015). Results Land snails species composition and distribution 13

22 14

23 Figure 2. Rarefaction curves of each site, color-coded by island. Inset map indicates color-coding, dots mark actual sample sizes and include the site number. 60 sites distributed among the seven Canary Islands yielded total of 34,801 subfossil land-snail specimens. Sites varied in richness from 1 to 14 species and in abundance from 34 to 2,770 individuals. Rarefaction of species richness (Figure 2) illustrated the great variety of richness and compositions encountered. For example, site 30 (Lanzarote) had a single species that was highly abundant, while site 6 (Tenerife) included fewer individuals than most sites, yet had 13 species and appears to be nowhere near its asymptote (i.e. greater sampling effort would have likely yielded many more species). Western islands' sites generally overlapped, while Lanzarote and Fuerteventura had relatively depressed curves with low richness, despite large sample sizes. The number of species encountered on each island and their biogeographical characters are summarized in Table 2; a complete species list and locality information are presented in the Supplementary Material, Table 1. 15

24 Table 2. Summary of the number of large (>5mm in length) land snail species recovered from the coastal scrub biome on each island. The actual sums of values in each column are greater than the Totals depicted at the bottom of each column because some species are found on multiple islands. The native column includes only non-endemic natives. Island # species # endemic species # native cosmopolitan (non-endemic) species # introduced species El Hierro La Palma La Gomera Tenerife Gran Canaria Fuerteventura Lanzarote Total macrosnail species (i.e., those larger than 5mm in maximum shell length) from the coastal scrub biome, including 72 Canary Islands endemics, 8 non-endemic natives, and 4 introduced species (Table 2) were recovered. Subsequent analysis excluded six semislug species (terrestrial gastropods unable to retract fully into their shell) and six microsnail species (those with a longest dimension of <5 mm as adults). Field methods were designed to maximize efficient detection of macrosnails, but did not include bulk sampling of soils, which has been suggested as the most effective method for detecting microsnails (Coppolino 2010). Furthermore, many microsnail species are burrowing, so their presence in mostly surficial samples may have more to do with soil processes than ecology. Similarly, the fragility of microsnail and semislug shells likely lead to much lower persistence in the assemblage and, therefore, greater taphonomic bias (Cadée 1999, Ménez 2002, Pearce 2008). 16

25 Their abundances and locality information, however, are included in Supplementary Material, Table 1. The great majority of macrosnail species encountered were highly endemic. 69 species were single-island endemics; 27 of these were found only at one site and 23 only at two. Eight species listed as endangered or critically endangered by the IUCN (IUCN 2015) were encountered (Table 3). These included Monilearia (syn. Xerotricha) arguinaguinensis, a critically endangered species occasionally thought to be extinct (Fernández-Palacios and Whittaker 2008) and whose shells made up ~60% of the site 60 sample. This further emphasizes the need for intensive and holistic land snail surveys in the Canaries, as the preservation statuses of most species are unknown or seem inaccurate. Table 3. Summary of endemic land snail species considered endangered by the IUCN encountered as subfossil in this study. IUCN Endangered species Island Site # Abundance of shells Canariella huttereri El Hierro 33, 34 1, 76 Canarivitrina falciferia La Gomera 47 1 Hemicycla plicaria Tenerife Monilearia arguinaguinensis Gran Canaria Monilearia granostriata Fuerteventura 17 3 Napaeus isletae Gran Canaria Theba grasseti Gran Canaria 58, 59 37, 15 Xerotricha pavida La Palma 38, 40, 42 19, 18, 18 Several cosmopolitan native species could be found on multiple islands. Theba geminata, though almost certainly an endemic species complex (Greve et al. 2010, Greve et al. 2012), was encountered on three islands: Fuerteventura, Lanzarote, and Gran Canaria. 16,157 subfossils (nearly as many as all other species combined) of this group were recovered from 20 sites. Where it occurred, T. geminata made up more than 50% of individuals in all but three 17

26 sites. Monilearia persimilis was collected on Gran Canaria, La Palma, El Hierro, and Tenerife. Non-endemic native Rumina decollata (an introduced species in the Great Britain and North America) was present in samples from all islands except Tenerife and La Palma. R. decollata is omnivorous and typically made up ~10% of samples when present, but were nearly 97% of La Gomera site 48's sample, yielding hundreds of individuals. Finally, Caracollina lenticula, a highly cosmopolitan native, was the only species found on all seven islands, occurring in all but eight samples. Four introduced macrosnail species were collected. Cernuella virgata and Cornu aspersum were each found on single islands: two sites on Gran Canaria and four sites on La Palma, respectively. Xerotricha conspurcata was found at one site each on Tenerife and Gran Canaria. Otala lactea occurred on all islands except La Palma, though at no more than 1/3 of sites on any one island (Supplementary Material, Table 1). It has also been reported as extinct on La Gomera (Núñez and Núñez 2010) where, in this study, it was represented only by a single, relatively weathered shell. While this large snail was relatively widely distributed, it made up at most 6% of individuals found at a given site except at El Hierro site 33. There, more than 50% of individuals were O. lactea. There is no published evidence that human-introduced snail species in the Canaries have become invasive, generally defined as causing damage to the native biodiversity. Neither do the results presented here seem to indicate a threat from these species, with the possible exception of O. lactea at site 33, given its overwhelming abundance there. Yet, Welch's t-tests comparing sites 18

27 with and without introduced species found those including O. lactea tend to have significantly higher native-species richness than those without (see Supplementary Material, Figure 2). The presence of other introduced species was not significantly correlated with richness of native species. Site 33 follows this trend with a rarefied richness of native species of 5.0 (this site happens to be the least complete site, so was effectively not rarefied), somewhat above El Hierro's average of 4.6. The site's native species exhibit evenness and abundances typical of the other sites on the island. This suggests O. lactea's abundance has not been detrimental to native biodiversity. It is, however, the only site on the island lacking Caracollina lenticula. The 432 individuals belonging to seven introduced species (357 individuals of introduced macrosnail species and 75 individuals of three introduced microsnails) were removed from all subsequent analyses. There is evidence that introduced species have quite different distribution patterns than native species. For example, less disturbed systems are often less susceptible to colonization by introduced species (Elton 1958, Davis et al. 2000, Levine 2000, Stachowicz et al. 2002). Introduced species are also more likely than native species to be synanthropic and may exhibit a higher diversity in anthropogenically influenced areas (McKinney 2008, Horsák et al. 2013), quite the opposite of this study's hypothesis for native species. Finally, the community structure of samples from Fuerteventura and Lanzarote (the two easternmost islands) were markedly different from that of the rest of the Canary Islands. Samples from these two islands had significantly 19

28 lower diversities and evennesses compared with all other islands, despite typically higher abundances (Figs 2, 3, 4). Furthermore, NMDS revealed that samples from Fuerteventura and Lanzarote had essentially indistinguishable faunal compositions, but were quite distinct from all other islands (Fig. 5). Santos et al. (2010) found that these islands deviated from the island species-area relationships of the other Canary Islands and excluded them from many of their analyses. Finally, the dominant genus at all sites on Fuerteventura and Lanzarote was Theba, which includes many cryptic species. This made accurate species identification, and therefore diversity measurement, difficult (Greve et al. 2012). For these reasons these two islands were excluded from the remaining analyses. Figure 3. Rarefied richness of land snails at each site from the coastal scrub biome across the Canary Islands, grouped by island. 20

29 Figure 4. Evenness, measured as PIE (the probability of interspecific encounter, (Hurlbert 1971)) of coastal scrub land snail communities across the Canary Islands, grouped by island. 21

30 Figure 5. A three dimensional NMDS representing Bray-Curtis similarities of native land snail species abundance between sites: points nearer each other in the ordination space have more similar species compositions. Smaller symbols denote being farther from the viewer. Only species occurring at two or more sites were included. Each island's samples tend to group together, exhibiting natural variability between islands and exhibiting a distribution in ordination space similar to their geographical distributions. Sites from Lanzarote (purple) and Fuerteventura (yellow) are most distinct from those of other islands, forming an exclusive cluster centered in the third octant (bottom left, or -axis 1, -axis 2, and +axis 3). Stress: 0.14 A total of 14,211 native macrosnails were collected from 42 sites on the five westernmost islands. After rarefaction, subsequent analyses were conducted on a total of 2,314 subfossil specimens from these sites. A summary subsample sizes and completenesses used may be found in Table 2 of the Supplemental Material. Environmental factors controlling biodiversity 22

31 Regression trees analyzed the relationships between the suite of 39 environmental factors and three biodiversity measurements across (Figures 6, 7, and 8). Their constructions are detailed in Supplementary Material, Figures In every case, anthropogenic factors best explained the distribution of diversity measurements. Total area within 1 km of the site used for agriculture (Area Ag ) had the greatest explanatory power for richness. A more species-rich group was produced that included sites with less than km 2 of nearby agriculture and a less species-rich group that included more Area Ag. The latter group could be further split by the total length of edges the coastal scrub shared with noncoastal-scrub natural areas within 1km of the site (Edge ScrubNat ). The 18 sites with less than 0.27 km Edge ScrubNat shared much of their edges with modified areas, so this may reflect a largely anthropogenic signal. Figure 6. Regression tree of land snail richness per site. Nodes representing a split are in labeled rectangles with a numbered square at the top, terminal nodes 23

32 are represented by boxplots. The mean richness for all 42 sites is 4.8 species equivalents. The mean richness for each node is: Node 1, 4.8; Node 2, 4.3; Node 3, 3.6; Node 4, 5.0; Node 5, 6.9. Boxplot midlines represent the median. The distance to the nearest airport (Dist Air ) was the most important factor in determining the primary splits for both the Shannon (Figure 7) and Simpson (Figure 8) diversity trees. For both trees, nine less-biodiverse sites were found nearer than 8.33 km to airports. The group of 33 sites farther from airports could be further split by distance to the nearest agricultural area (Dist Ag ). Again, lessdiverse sites were found nearer the human-modified landscape. Unlike the primary splits, the second splits in these trees were not identical. Dist Ag split the Shannon diversity subgroup at 0.38 km and the Simpson diversity subgroup at just 0.07 km. Figure 7. Regression tree of land snail Shannon diversity. The primary split is at 8.33 km from an airport. The secondary split is at 0.38 km from an agricultural area. The mean Shannon diversity for each node is: Node 1, 3.2; Node 2, 2.5; Node 3, 3.6; Node 4, 3.2; Node 5, 4.7. Boxplot midlines represent the median. 24

33 Figure 8. Regression tree of land snail Simpson diversity. Like the Shannon diversity regression tree, the primary split is at 8.33 km from an airport. The secondary split is at just 0.07 km from an agricultural area. The mean Simpson diversity values are: Node 1, 2.7; Node 2, 2.2; Node 3, 3.0; Node 4, 2.3; Node 5, 3.4. Boxplot midlines represent the median. Discussion The coastal scrub of the five western Canary Islands includes 60 endemic species, nearly all of which are highly localized. Cameron et al. (1996) argued that the deeply dissected topography of Porto Santo (Madeira, Portugal), likely played an important role in spurring speciation by creating barriers between populations there. It is easy to imagine a similar effect for the five high, rugged western-most Canary Islands. Studies of Pacific Island systems have noted a marked homogenization of malacofauna with increasing anthropogenic impact (Cowie 2001a, Cowie 2001b). This was not observed in the Canarian coastal scrub. Invasive species are implicated as being the primary causes of homogenization, so the lack of obvious invaders in the Canaries may explain why its malacofauna is not homogenized. 25

34 Site 33, nearest the airport on El Hierro was the only site with a majority of nonnative individuals. The abundances and proportions of native species, however, were not noticeably unusual relative to other sites on the island. Site 38, on the outskirts of the town of La Playa de Santiago in La Gomera, was also unusual in having a great abundance and near monoculture of the native cosmopolitan species Rumina decollata. At the 14 other sites where the species was present, it made up only ~7% of individuals on average. Fuerteventura and Lanzarote, the easternmost islands, have a land snail fauna distinct from the other five islands. This was immediately apparent in the field as most sites had an incredible abundance of individuals from the genus Theba a minor constituent in the few western-island sites where it is present (Supplementary Material, Table 1). Their markedly different malacofauna may ultimately be explained by their topographies. The two islands are much flatter than their western neighbors, with a maximum elevation of 807 m a.s.l. They also effectively have only a scrub plant biome, in contrast to the other islands' multiple plant biomes, due to a small altitudinal gradient precluding interception of the trade winds. Whittaker et al. (2008) described a hump-shaped progression of diversity on volcanic islands, with diversity peaking concomitant with topographic complexity, followed by parallel senescence driven by erosion and eventual submergence of the island. These two 20 Myr old islands quite ancient for volcanic islands appear to be the stage of decline. Regression tree analyses found that anthropogenic factors had greater power for explaining biodiversity at sites than did natural factors. While many 26

35 climatic natural factors were constrained by the choice of field area, the influence of island biogeographical factors on local (alpha) diversity was not. For example, the number of unique biomes varied among islands and could affect overall species density in the coastal scrub, as other biomes could serve to spur speciation (Whittaker et al. 2001). Nevertheless, these factors were not found to be as important as anthropogenic ones in the five western islands. It should be noted that many of the factors used were not independent of one another. For example, the distance to the nearest airport is, for obvious reasons, strongly correlated with road density. Regression trees are robust to highly correlated explanatory factors, yet it bears mentioning that biodiversity has a component of stochastic variability. One of a set of multiple highly-correlated variables, may therefore better explain diversity patterns simply by chance. In this sense, the environmental parameters explored in this study are not likely the direct causes of observed differences. Rather, they are proxies for the true, direct causes. Considering proximity to an airport as an example, factors associated with the presence of an airport, including a host of concomitant factors related to urbanization, increased exposure to pollution, synanthropic predators, impeded dispersal, and/or other parameters are probably the true causes of limited diversity. While the actual values reported at nodes indicate breaking points in the data rather than definitive demarcations of "pristine" and "impacted" areas, results like those observed here can still be very informative for conservation purposes by confirming these factors do have a significant impact. 27

36 Proximity to agriculture was also found to be an important factor for all diversity measures. The regression tree for richness identified the total area used for agriculture within a 1 km radius of the site as the best explanatory variable, but the split was at just km 2. Of the sites with less agriculture, three sites had no agriculture within 1 km, and the other six included agriculture just at the periphery of the 1 km radius (as opposed to a very small field nearer the site). These nine sites had a mean rarefied richness of 6.9 species equivalents, more than half again of those sites with more nearby agriculture, which had a mean of 4.3 species equivalents. Agriculture was also an important factor for Shannon and Simpson diversities, dictating the second split of each tree for those 26 sites more distant from airports, though their regression trees selected the distance to the nearest agricultural area (Dist Ag ) rather than Area Ag, the most important factor for richness. An interesting pattern is diagnosed by comparing the effects of agriculture on the three measures of diversity. Richness, which weights rare and common species equally, significantly decreased when more than a few hundred square meters of agriculture were present within 1 km. [As a caveat, Area Ag and Dist Ag are not strictly interchangeable, but they are highly correlated and those lower- Area Ag sites in the richness tree (Fig. 6, Node 5) were all greater than 0.60 km from any agriculture.] The Shannon diversity regression tree's split for agriculture was at 0.38 km away and the Simpson diversity regression tree's split for the same split was just 0.07 km. This shift from greater to smaller distance is likely a result of the way these three measures of diversity weight rare species. It 28

37 suggests that rare species are affected by agricultural landscapes at a greater distance than are more common species. Again, the exact values for the distances may be less informative than the pattern itself. The total length of coastal scrub edge bordering natural features (Edge ScrubNat ), in the richness regression tree is responsible for splitting the subgroup of sites with more agriculture within 1 km. The 18 sites with low Edge ScrubNat all had edges shared with modified areas. On the other hand, Edge ScrubNat also varied among more "pristine" sites, between those more contiguous with surrounding coastal scrub vegetation and those in close contact with other natural landscapes. The regression analysis, however, grouped these possible subsets of pristine sites together, in contrast to the more modified group. Edge ScrubNat therefore effectively reflects an anthropogenic signal. Overall, the methods used here permitted assessment of the effect of anthropogenic factors in a geographic context, making them especially useful for conservation practices. Regression tree analysis of multiple diversity measurements and of many environmental factors may also help tailor conservation efforts to different species. For example, species that are locally abundant may be resilient to agricultural factors at a closer proximity than rarer species. This study presents the first quantitative evaluation of the interplay of natural and anthropogenic factors on land snail diversity in the Canary Archipelago. The results, however, may have some limitations associated with the limited number of sampling sites available. Future studies in the region will 29

38 incorporate additional sites and plant-biomes, likely allowing for the construction of larger regression trees, and permitting further insight into the complex relationships natural and anthropogenic factors exhibit in this eclectic island system. Diversity of the native land snail fauna of the Canarian coastal scrub was negatively affected by close proximity to human-modified landscapes. This, coupled with the apparent lack of extinctions in the biota, indicates the ranges of its constituent species are shifting. These species, many of which are known only from single valleys (IUCN 2015), are losing habitat overtly to landscape modification, but also cryptically to anthropogenic factors that extend far into natural areas. Many of the few species with sufficient population data are in decline and some seem to be endangered (IUCN 2015). This study's results may inform efforts to keep this unique malacofauna intact. In particular, the seven Canarian Rural Parks and four UNESCO Biosphere Reserves are protected areas comprising a mosaic of natural biomes and agricultural areas (Fernández- Palacios and Whittaker 2008) that may not sufficiently serve species very sensitive to those anthropogenic factors. The Canary Islands have been targeted for conservation efforts, especially after being described part of the Mediterranean biodiversity hyper-hotspot (Myers et al. 2000). Yet land snails as a group have many vulnerable members worldwide in need of conservation, especially in island systems, while remaining woefully under-studied (Lydeard et. al 2004, Fernández-Palacios and Whittaker 2008, Régnier et al. 2009). Régnier et al. (2009) estimated that only 3% of 30

39 mollusks are described worldwide while they constitute more than half of known historical extinctions. Evaluating the health of land snail communities provides valuable information on an important group, can inform conservation efforts of vulnerable species, and may prove a logistically expeditious way of testing for and calibrating for anthropogenic disturbance. Conclusions 1. The land snail fauna of the Canarian coastal scrub includes a large majority of endemic species. Of 82 macrosnail species encountered, 72 were endemic to the Canaries and 69 were endemic to a single island. Unlike many oceanic archipelagoes, the Canary Islands host no known invasive land snails and its native malacofauna have experienced no known historical extinctions. Yet, their often-small ranges, the decreasing area of coastal scrub, and the results presented here suggest they are susceptible to recent human modification. 2. The results of this study indicate that the effects of anthropogenic factors on biodiversity in natural areas can be quantified when natural factors are sufficiently constrained. A straightforward way to accomplish this is to limit sites to a single plant biome as conducted here. 3. Using a combination of analytical methods, including regression tree analysis, it was found that close proximity to anthropogenic modifications impacted the diversity of natural areas covered by coastal scrub. Impacts related to airports were associated with reduced diversity in sites up to ~8 km away from them. Agriculture was associated with low diversity in sites less than 1 31

40 km away. These results may provide guidelines for the establishment or expansion of protected areas for this unique biome. 4. This study adds to the growing number of investigations that recommend assessments of land snail diversity as a way to measure biome "health." Time-averaged subfossil snail assemblages provide dependable representations of typical living assemblages and avoid possible biases, for example, by moderating the influence of ephemeral population variations. A change in biodiversity as depicted in subfossils can be taken, therefore, as the result of a more sustained phenomenon. 32

41 Supplemental Material Figure SM 1. Map of field sites on the Canarian archipelago for use with Supplementary Material Table 1, below. Table SM 1. All species are endemic to the Canary Islands except a are merely native to the Canary Islands, and b are introduced. Microsnails, species with a maximum dimension of less than 5 mm as adults, are indicated with c. Semislugs are indicated with d. Counts are ordered to correlate with the site numbers. (Ibáñez et al. 2006, Yanes et al. 2009a, Núñez and Núñez 2010, Yanes et al. 2011a, Yanes et al. 2011b, Castro et al. 2012, Santana et al. 2013, Alonso et al. 2013, Alonso and Ibáñez 2015a, Alonso and Ibáñez 2015b, Alonso and Ibáñez 2015c) species total counts sites Canariella bimbachensis a , 1, , 35, 36 Canariella discobolus Canariella fortunata , 25, 98 6, 9, 10 Canariella hispidula 79 45, 34 11, 12 Canariella huttereri 77 1, 76 33, 34 Canariella multigranosa Canariella plutonia 273 3, 2, 2, 3, 173, 90 15, 16, 19, 20, 22, 27 Canariella tenuicostulata , 54, 8 45, 46, 47 Canarivitrina falcifera d Canarivitrina taburientensis d 3 1:2 37, 39 1

42 Caracollina lenticula a , 60, 184, 63, 292, 98, 53, 78, 70, 208, 147, 78, 1, 8, 3, 2, 3, 8, 11, 1, 5, 15, 110, 5, 245, 117, 154, 369, 103, 136, 202, 58, 90, 24, 78, 66, 18, 192, 7, 1, 1, 31, 9, 27, 1, 28, 59, 10, 7, 6, 18, 46 Cecilioides acicula b,c Cernuella virgata b 66 2, 64 58, 59 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 18, 19, 20, 23, 24, 25, 27, 29, 31, 32, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60 Cornu aspersum b 32 12, 2, 10, 8 38, 39, 40, 43 Ferussacia attenuata c Ferussacia folliculus b,c 33 4, 1, 6, 15, 1, 5, 1 10, 22, 27, 38, 39, 43, 58 Gibbulinella dealbata 20 2, 18 52, 58 Gibbulinella dewinteri 54 10, 6, 11, 4, 10, 5, 8 6, 10, 37, 38, 39, 41, 44 Gibbulinella cf. macrogira 3 2, 1 45, 46 Granopupa granum b,c 41 13, 8, 1, 5, 9, 3, 2 1, 2, 3, 4, 5, 22, 58 Hemicycla berkeleyii 53 40, 13 56, 57 Hemicycla bethencourtiana 91 14, 77 5, 12 Hemicycla bidentalis 98 5, 93 6, 10 Hemicycla consobrina 110 6, 4, 8, 26, 66 1, 2, 3, 4, 5 Hemicycla ethelema 77 54, 23 54, 60 Hemicycla eurythyra Hemicycla fritschi 95 68, 17, 2, 6, 2 45, 46, 47, 49, 50 Hemicycla fuenterroquensis 12 4, 1, 7 38, 41, 42 Hemicycla glasiana , 91 52, 59 Hemicycla gomerensis Hemicycla guamartemes Hemicycla laurijona Hemicycla maugeana , 25, 101, 57, 82 32, 33, 34, 35, 36 Hemicycla paivanopsis 216 1, 46, , 49, 50 Hemicycla aff.paivanopsis , 110, 82, 1 45, 46, 47, 51 Hemicycla plicaria Hemicycla pouchet Hemicycla psathyra , 51, 77, 32 52, 53, 55, 59 2

43 Hemicycla quadricincta 56 10, 46 48, 51 Hemicycla sarcostoma Hemicycla sp Insulivitrina canariensis d 16 4, 7, 1, 4 31, 32, 34, 35 Insulivitrina lamarkii d Insulivitrina nogalesi d 2 1, 1 54, 55 Insulivitrina solemi d 10 5, 5 40, 44 Monilearia arguineguinensis Monilearia caementitia 57 55, 2 52, 59 Monilearia granostriata Monilearia monilifera , 1, 2, 2, 4, 6, 345, 309, 5, 4, 25, 597, 3 Moniliaria multipunctata 11 8, 2, 1 13, 16, 18 Monilearia persimilis , 25, 116, 3, 34, 44, 27, 33, 50, 172, 277, 243, 121, 65, 139, 307, 50, 26, 16 Monilearia phalerata , 80, 58, 28, 246, 28, 85, 1, 449, 8, 73, 23, 13, 69, 29 Monilearia aff. woodwardia Napaeus baeticatus 43 11, 32 6, 10 Napaeus bertheloti 9 8, 1 45, 46 Napaeus encastus Napaeus gruereanus Napaeus interpunctatus 11 9, 2 52, 53 Napaeus isletae Napaeus moquinianus Napaeus rupicola 7 3, 2, 2 45, 49, 50 Napaeus servus Napaeus sp Napaeus subsimplex 57 10, 46, 1 31, 32, 36 Napaeus variatus 70 1, 59, 3, 7 5, 6, 9, 12 Napeus subgracilior 30 14, 14, 2 38, 39, 44 Obelus discogranulatus a , 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 27, 29 6, 7, 8, 9, 31, 32, 33, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 54, 55 1, 2, 3, 4, 5, 7, 8, 9, 10, 12, 53, 56, 57, 58, 59 3

44 Obelus mirandae a 485 8, 126, 86, 38, 195, 32 45, 46, 47, 49, 50, 51 Obelus cf. mirandae Obelus moderatus a Otala lactea b 245 1, 68, 1, 1, 1, 1, 39, 99, 1, 2, 16, 15 Pomatias canariensis , 58 58, 59 Pomatias cf. canariensis 88 5, 83 31, 32 Pomatias laevigatus , 3, 119, 15 5, 6, 10, 12 9, 10, 15, 19, 22, 25, 27, 33, 46, 53, 58, 59 Pomatias cf. laevigatus , 4, 1, 23, 6, 13 45, 46, 48, 52, 54, 55 Pomatias cf. lanzarotensis 35 6, 2, 1, 2, 4, 20 15, 16, 17, 19, 22, 50 Pomatias lanzarotensis Pupoides coenopictus a,c 11 10, 1 1, 2 Retinella lenis 7 5, 2 40, 43 Retinella hierroensis , 98 33, 34 Retinella rochebruni 33 15, 1, 17 45, 46, 48 Rumina decollata a , 24, 60, 1, 21, 6, 5, 220, 12, 8, 2, 3, 3, 833, 43, 102 Theba arinagae Theba geminata a , 201, 77, 60, 1958, 900, 546, 838, 1222, 1039, 585, 712, 699, 50, 2261, 1554, 1810, 123, 461, 95, 403 Theba cf. clausoinflata Theba grasseti 52 37, 15 58, 59 Theba impugnata , 14, 15, 17, 19, 20, 21, 22, 26, 29, 33, 46, 47, 48, 58, 59 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 28, 29, 30, 52, 53, 58, 59 Vitrea contracta a,c 12 3, 1, 5, 1, 1, 1 6, 10, 32, 36, 41, 44 Xerotricha adoptata 66 62, 4 46, 50 Xerotricha conspurcata b 13 1, 12 6, 59 Xerotricha lancerottensis 16 8, 8 19, 20 Xerotricha pavida 55 19, 18, 18 38, 40, 42 Xerotricha orbignii 167 8, 23, 1, 27, 1, 7, 48, 1, 9, 1, 23, 18 Xerotricha cf. orbignii 176 3, , 58 4, 5, 6, 7, 8, 10, 11, 12, 31, 32, 35, 36 4

45 Figure SM 2. Boxplot showing rarefied richness of native land snail species in five westernmost Canary Islands grouped by the presence of the introduced species Otala lactea. Welch's t-test found sites including O. lactea had significantly higher richness, P = El Hierro site 33 is the only site where O. lactea is a dominant species and is marked by a blue triangle. 5

46 Figure SM 3. The regression tree process iteratively splits the data into two groups at all possible values of the explanatory variable until total variance in the response variable (richness) is minimized for both sides. The explanatory variable that best does this for richness in all 42 sites in my data set is SiteAreaAg. The groups are divided by the solid line, the mean values for each group are marked with a dotted line and are 6.9 on the left and 4.3 on the right. (Note site 59 is on the left and 2 on the right.) 6

47 Figure SM 4. The second split for richness regression tree. Groups are divided by the solid line, their average values marked with dotted lines. A group of 18 sites with less than km of landscape edges between coastal scrub vegetation and other natural areas (EdgeVegNat) has a mean richness of 3.6. (All but one of these 18 sites actually have an EdgeVegNat value of 0 km.) The second group with a greater length of EdgeVegNat has a mean of 5.0 species. 7

48 Figure SM 5. Land snail richness regression tree (see Fig 6.) cross validation (Xval) relative error and complexity parameter (cp). To avoid over-fitting, tree is pruned the to largest tree that s relative error is greater than one standard error from the tree with lowest complexity, indicated by the dotted line. This tree were pruned to a cp of 0.13 and 3 terminal nodes. 8

49 Figure SM 6. The first split for Shannon diversity regression tree. Groups are divided by the solid line, their average values marked with dotted lines. One group less than 8.33 kilometers from the nearest airport has an average Shannon diversity of 2.5 species equivalents, the other more than 8.33 km away has an average Shannon diversity of 3.6 species equivalents. 9

50 Figure SM 7. The second split for Shannon diversity regression tree. Groups are divided by the solid line, their average values marked with dotted lines. A group less than 0.38 kilometers from the nearest airport has an average Shannon diversity of 3.2 species equivalents, the group more than 0.38 km away has an average Shannon diversity of 4.7 species equivalents. 10

51 Figure SM 8. Shannon diversity regression tree (see Fig X.) cross validation (Xval) relative error and complexity parameter (cp). To avoid over-fitting, tree is pruned to the largest tree that s relative error is greater than one standard error from the tree with lowest complexity, indicated by the dotted line. This tree were pruned to a cp of 0.12 and 3 terminal nodes. 11

52 Figure SM 9. The first split for Simpson diversity regression tree. Groups are divided by the solid line, their average values marked with dotted lines. One group less than 8.33 kilometers from the nearest airport has an average Shannon diversity of 2.2 species equivalents, the other more than 8.33 km away has an average Shannon diversity of 3.0 species equivalents. 12

53 Figure SM 10. The second split for Shannon diversity regression tree. Groups are divided by the solid line, their average values marked with dotted lines. One group less than 0.07 kilometers from the nearest airport has an average Shannon diversity of 2.3 species equivalents, the other more than 8.33 km away has an average Shannon diversity of 3.4 species equivalents. Figure SM 11. Simpson diversity regression tree (see Fig X.) cross validation (Xval) relative error and complexity parameter (cp). To avoid over-fitting, tree is pruned to the largest tree that s relative error is greater than one standard error from the tree with lowest complexity, indicated by the dotted line. This tree were pruned to a cp of 0.13 and 3 terminal nodes. 13

54 Figure SM 12. Detailed map of El Hierro land cover and sites with 1 km buffers. 14

55 Figure SM 13. Detailed map of Fuerteventura land cover and sites with 1 km buffers. 15

56 Figure SM 14. Detailed map of Gran Canaria land cover and sites with 1 km buffers. 16

57 Figure SM 15. Detailed map of La Gomera land cover and sites with 1 km buffers. 17

58 Figure SM 16. Detailed map of Lanzarote land cover and sites with 1 km buffers. 18

59 Figure SM 17. Detailed map of La Palma land cover and sites with 1 km buffers. 19

60 Figure SM 18. Detailed map of Tenerife land cover and sites with 1 km buffers. Table SM 2. Results of rarefaction, samples were reduced to a subsample such that their completeness is nearest , the completeness of the least complete sample: site 33. Site subsample size completeness Richness Shannon Simpson