Introduction

One of the greatest threats to biodiversity is the worldwide spread of invasive species (Hooper et al. 2012). Numbers of both intentional and accidental exotic species introductions have increased exponentially and have reshaped native communities in many parts of the world, particularly islands (Kolar and Lodge 2001). Myers et al. (2000) show that at least 9 of the 25 proposed biodiversity hotspots consist principally or entirely of islands, highlighting their importance for global biodiversity. Islands host over 20% of the world’s biodiversity despite constituting only 5% of the global terrestrial area (Kier et al. 2009). Exotic species introductions have been shown to have the most profound effects on isolated islands, especially those supporting depauperate biotic communities with a high percentage of endemic taxa (Mueller-Dombois 1981). Island ecosystems are highly susceptible to disturbances because they tend to have simplified trophic webs and high rates of endemism (Courchamp et al. 2003). Insular populations are more susceptible to extinction than mainland species (MacArthur and Wilson 1967; Foufopoulos et al. 2011). Reflecting this susceptibility of island taxa, 90% of known reptile and amphibian extinctions, 93% of avian extinctions, and 81% of mammalian extinctions worldwide, have occurred on islands (Honegger 1981; King 1985; Ceballos and Brown 1995). Many of the most detrimental island invasives are human associates, such as rats (Rattus spp.), cats (Felis catus), rabbits (Oryctolagus cuniculus), and escaped livestock (Bowen and Van Vuren 1997; Nogales et al. 2004; Jones et al. 2008). Feral livestock, including pigs (Sus scrofa) and goats (Capra hircus), tend to excessively remove plant biomass, eliminating primary producers from island food webs, harming the native plant communities, causing soil erosion, and damaging island landscapes (Coblentz 1978).

A key factor facilitating the proliferation of introduced mammalian herbivores on islands is increased food availability due to insular vegetation’s lack of defenses. Most plants have evolved some type of anti-herbivore defenses, the extent of which is often proportional to the risk of browsing (Rhoades 1979). Until the arrival of humans, mammalian herbivores were absent from most small island ecosystems (Atkinson 1989). Due to the energetic costs of resistance in plants, many island taxa lack defenses against herbivory such as chemical deterrents, physical armature, or a tolerance to grazing (Coblentz 1978; Vitousek 1988). This lack of resistance can, in turn, lead to pronounced shifts in the diet preferences of herbivores towards less defended endemic flora, giving more heavily defended invasive plants a competitive advantage (Loope and Scowcroft 1985; Van Vuren and Coblentz 1987). In the Mediterranean, plant species vary greatly in their levels of phenols and preferential grazing occurs on species with the lowest phenolic levels (Massei et al. 2000).

Goats are recognized today as the sixth leading threat to species worldwide because of their impact on vegetation (Bellard et al. 2016). In the early twentieth century, Sir Alfred Russell Wallace noted that, “goats are the greatest of all foes to trees” (Wallace 1911). Domesticated in the dry highlands of western Iran 10,000 years ago (Zeder and Hesse 2000), goats are able to utilize many arid habitats unsuitable for other herbivores. This adaptability arises from the combination of a slow metabolism, an efficient digestive system, low water requirements, high reproduction rates, and a generalist diet (Silanikove 2000; Chynoweth et al. 2013). Goats also have the ability to consume tougher, chemically-defended food, making more vegetation available to them (Devendra and McLeroy 1982).

Many experimental studies have been conducted to quantify the effects of goat grazing on island vegetation productivity and species richness. In Hawaii, Mueller-Dombois and Spatz (1972) found that goat exclosures showed a stark increase in vegetation cover. Outside such areas, there were fewer endemic, and more exotic species, and much more exposed soil and rock. Regeneration of the endemic koa tree (Acacia koa), was disrupted in grazed areas, but abundant inside goat exclosures (Spatz and Mueller-Dombois 1973). On the Galápagos, goats have reduced or eliminated 77% of all plant species on the islands (Eckhardt 1972), and compete directly with endemic Galápagos tortoises. Bellingham et al. (2010) found that grazing promoted the growth of dense, unpalatable understories in New Zealand, restricting the succession of native forests. However, it has also been suggested that we should not generalize the negative influences of grazing and focus on individual case studies, as Randriamalala et al. (2016) found no significant grazing impacts on diversity, height, or biomass of vegetation in Madagascar.

Beyond vegetation, invasive goats can have profound impacts on other island taxa including breeding seabird populations (Pafilis et al. 2013). Because islands constitute relatively closed terrestrial systems, breeding seabird populations can provide critically important allochthonous marine nutrient subsidies to the simple local food webs (Sánchez-Piñero and Polis 2000). Seabird-derived guano, food scraps, and carrion can serve as fertilizers for plants, which in turn can boost arthropod and other primary consumer populations (Sánchez-Piñero and Polis 2000). Indeed, nesting seabirds have been found to increase the limiting soil nutrients nitrogen (N) and phosphorus (P) (Wait et al. 2005), which in turn support dense insular primary producer populations (Kolb et al. 2010). Seabirds can also increase CaCO3 content and lower C:N ratios (García et al. 2002). The ratio of C to N is a crucial measurement for organic matter decomposition (Parnas 1975). Thus, one aim of this study is to investigate whether the presence of goats is associated with decreased nesting seabird populations and how goats may affect arthropods as well as soil nutrients. Given the importance of seabirds for nutrient cycling in island systems, factors that reduce their numbers can have cascading impacts (Towns et al. 2009). Overgrazing has also been observed to affect physical soil properties and is associated with soil loss and desertification of many landscapes. In Australia, Bayne et al. (2004) analyzed sediment fluxes in areas with varying grazing intensities and found that there was less vegetation in areas with more goats, therefore promoting soil erosion and subsequently increasing the risk of further plant loss. Important parameters affecting erosion susceptibility are soil organic matter content and grain size distribution (Wischmeier et al. 1971).

Given the perceived impacts of feral goats on island ecosystems, conservation organizations have attempted to eradicate goats from numerous island sites worldwide—though with variable success. To date, approximately 164 goat eradications have succeeded (DIISE 2015) thanks to advances in technology and improved field techniques including the use of Global Positioning Systems (GPS), aerial hunting, and the use of ‘Judas goats’ (Campbell and Donlan 2005). Results of these eradications have been variable and difficult to predict—island ecosystems may recover on their own, require some restoration or reintroduction, or even become more damaged due to ecological destabilization (Courchamp et al. 2003; Schweizer et al. 2016). Few studies have implemented monitoring programs to evaluate the post-eradication recovery of these systems and predicting the success of eradication attempts has been difficult.

Successful goat eradications can create surprising and unintended conservation problems. Adequate assessments of the consequences of eradication are hindered because ecological relationships between island organisms are often poorly understood prior to eradication (Zavaleta et al. 2001). For example, goat eradications on the Galápagos Islands led to a decline, rather than a recovery, of the endangered Galápagos Hawk (Buteo galapagoensis), because the species had come to depend on goat-altered habitats (Rivera-Parra et al. 2012). Similarly, on the Bonin Islands, unbeknownst to conservation managers, goat grazing kept newly arrived invasive plants under control. Once the goats were eradicated, these exotic plants overran the landscape, impairing the native plant communities (Mack and Lonsdale 2002; Osawa et al. 2016). In the Mediterranean, goat grazing has had profound ecological impacts and whole habitats have been shaped by historical livestock management systems (San Miguel-Ayanz et al. 2010). Therefore, to prevent unintended consequences of goat eradications, it is critical to understand the effects of this important species before any eradication efforts are implemented (Zavaleta et al. 2001).

Goats and their products have always been an integral part of Mediterranean island economies (Hadjigeorgiou et al. 2002); most inhabited islands have been traditionally grazed year-round by roving herds of goats. With the increased availability of boat motors and associated reliability of access, shepherds have expanded their grazing area to previously inaccessible islets. Such relatively small islets are often of great conservation importance, either as refugia for endemic taxa that have been extirpated on larger islands, or as breeding sites for significant seabird populations. Typically, Mediterranean shepherds will release herds of goats onto such islands after the onset of the winter growing season, coming back to collect them after they feed on the spring vegetation flush. Such free-ranging goat herds will typically graze with limited, if any, supplemental feeding (Pafilis et al. 2013). Because islands are generally overstocked, even seasonal presence can have severe impacts on local plant communities. However, in the last 20 years, because of a shift away from traditional livestock husbandry and increased conservation concerns, goat releases have been discontinued on several islands. This sets the stage for an investigation of the potential recovery of local ecosystems after goats have been removed.

In contrast to studies that focus on exclosure plots, we test for effects on an island-wide scale (Greuter 1995). Islands represent spatially discrete entities making them reliable and realistic study systems (Wardle 2002). The present dearth of large-scale experiments has been attributed to a combination of logistic difficulties, moral challenges, and concerns about pseudoreplication (Soulé 1990; Donlan et al. 2002; Oksanen 2001). However, as ecosystem-level approaches become more prevalent in conservation, so will the need for large-scale experimental manipulations (Zavaleta et al. 2001). Here we conduct whole island manipulations and examine the effects of goats and their removal on multiple trophic levels. In particular, we quantify vegetation characteristics, physical and chemical soil traits, arthropod characteristics, and seabird populations on each island. We also provide secondary analyses of the interactions between these variables to further elucidate the ecosystem-level processes mediating effects of goat grazing on island species communities. We hypothesize (1) that goat grazing will have pronounced effects on vegetation characteristics, arthropod community structure, soil, as well as seabird presence and (2) that these changes will become reversed following goat removal.

Materials and methods

Study area

All fieldwork occurred in May–July 2015 on the Cycladic islands (central Aegean Sea, Greece). The climate is typical for the Mediterranean region with warm, dry summers and mild, modestly wet winters (Gikas and Tchobanoglous 2009). The islands are mainly composed of limestone and flysch substrates with shallow to no soil profiles. Located within the Mediterranean biodiversity hotspot, they represent a hyperdiverse landscape covered by summer-deciduous phrygana, semi-arid coastal heaths and sclerophyllous maquis vegetation types (Médail and Quézel 1999). “Islet specialist” plant species—taxa found mostly on small islands—play a major role in the study ecosystems (Bergmeier et al. 2003). All of the study islands have relatively simple food webs in which the top predators are lizards (Podarcis erhardii, Hemidactylus turcicus, and Mediodactylus kotschyi). Sand boas (Eryx jaculus) are present only on the two largest islands in this study (Drionissi and Gramvoussa). All of our study islands are uninhabited and ecologically similar to each other, with no other known introduced herbivores such as rabbits; they were selected to fall into one of three categories: islands currently being grazed by goats (Aspronissi, Fidussa, Agrilou, and Venetiko); islands which have never been grazed by goats (Turlos, Preza, Agia Kali, Drionissi, Grambonissi, North Varvaronissi, and South Varvaronissi); and those with previously extant but now removed goat herds (Mikros Ambelas, Petalidi, Kisiri, Psalida, and Gramvoussa) (Fig. 1). We considered an island to be grazed if goats had been brought onto the island for more than one spring season. From interviews with local shepherds, we determined that all islands considered eradicated have been devoid of goats for at least 10 years. Because the islands are relatively small (0.79–79.25 ha; see Appendix A in ESM), and the habitat open, we were able to census goats visually.

Fig. 1
figure 1

Map of Greece and Cycladic study areas in the Aegean. Study islands: Agia Kali (AK), Agrilou (AG), Aspronissi (AS), Drionissi (DR), Fidussa (FI), Grambonissi (GB), Gramvoussa (GV), Kisiri (KS), Mikros Ambelas (MA), North Varvaronissi (NV), Petalidi (PE), Preza (PR), Psalida (PS), South Varvaronissi (SV), Turlos (TU), and Venetiko (VE), all located near the larger islands of Amorgos (AM), Ios (IO), Irakleia (IR), Naxos (NA), Paros (PA), and Schinoussa (SC)

Full size image

Quantification of vegetation condition

To assess vegetation characteristics we established four 50 m transects, one in each cardinal direction. On the smallest islands (Kisiri, Mikros Ambelas, and North Varvaronissi), which could not accommodate this design, three transects were used. We measured the presence of vegetation along the entire length of each transect and averaged the values for a measure of percent vegetation cover. We also applied this method to assess percent bedrock and bare soil. To determine vegetation height, we measured the height of plants every 2 m along each transect and averaged the values for each island. We sampled vegetation biomass in five randomly placed 80 cm × 80 cm squares. In these squares, all vegetation was clipped to ground level; all plant matter was collected and dried until no further weight loss was observed, and then weighed. Biomass data were averaged and expressed as a single g/m2 island value. Plant communities were determined from 80 cm × 80 cm quadrats placed every 5 m along each established transect; in each quadrat we recorded the identity of all plants present. Utilizing the program EstimateS (Colwell 2013), we generated a sample-based incidence rarefaction curve (Gotelli and Colwell 2001; Colwell et al. 2004). A bias-corrected form of the Chao2 asymptotic estimator was used to estimate the actual number of plant species on the island (Chao 1987; Colwell and Coddington 1994, Gotelli and Colwell 2011):

$$S_{Chao2} = S_{obs} + \left( {\frac{m - 1}{m}} \right)\frac{{q_{1} \left( {q_{1} - 1} \right)}}{{2\left( {q_{2} + 1} \right)}}$$

where S Chao2 = the estimated number of species, S obs  = the observed number of species, m = the total number of samples, q 1 = the number of unique species, and q 2 = the number of duplicate species.

It is necessary to correct for area effects when making inter-island comparisons because the number of species increases with island size (MacArthur and Wilson 1967). This species-area relationship can be defined by the power law S = CAz, where C and z are coefficients, A is the area, and S is the number of species. To estimate species density C (an area-independent metric of species richness given by the equation C = S/Az), we obtained a data-based estimate of the coefficient z by plotting the species-area relationship for our study system and extracting the exponent of the fitted curve (Rosenzweig et al. 2011). Non-metric multidimensional scaling (NMDS) was used to analyze the changing species composition between study sites (Marín et al. 2016). In order to construct a species composition distance matrix, we employed the Bray–Curtis similarity index. Surface fitting was used to interpret the effects of different environmental variables on our species composition.

Quantification of arthropod characteristics

Five pitfall traps, used to sample epigeal arthropods, were installed on each island at randomly chosen locations near each directional transect. The traps were constructed by sinking plastic cups (7 cm in diameter and 11 cm deep), filled 2/3 with ethylene glycol, flush into the ground (Woodcock 2005; Schmidt et al. 2006). Traps were placed under a large, raised flat stone in a manner that protected them from livestock trampling but still allowed free invertebrate access. After approximately 2 weeks (on average 17.19 ± 4.59 days), the samples were collected, rinsed, identified to morphospecies, sorted, counted, dried, and weighed.

Abundance of each species and total number of observed species were recorded for each island. The number of arthropod individuals was counted for each trap and then averaged for each island and divided by the number of collection days. Each sample was dried under a heat lamp until no further weight reductions were observed and then the weight was recorded. Arthropod biomass was averaged for each island and divided by the number of days collected. The program EstimateS (Colwell 2013) was used to construct a sample-based abundance rarefaction curve for each island (Gotelli and Colwell 2001; Colwell et al. 2004). A bias-corrected form of the Chao1 asymptotic estimator was used to estimate the total number of arthropod species on the island (Chao 1987; Colwell and Coddington 1994, Gotelli and Colwell 2011):

$$S_{Chao1} = S_{obs} + \frac{{f_{1} \left( {f_{1} - 1} \right)}}{{2\left( {f_{2} + 1} \right)}}$$

where S Chao1 = the number of estimated species, S obs  = the observed number of species, f 1 = the number of singleton species, and f 2 = the number of doubleton species.

Since the number of species is intrinsically linked to the size of an island (MacArthur and Wilson 1967), we accounted for area by calculating the coefficient C of the species-area relationship, where C = S/Az (Rosenzweig et al. 2011), and thus obtained comparable metrics. We also used NMDS with surface fitting to analyze arthropod species composition between islands of different grazing statuses and the environmental variables that affect that composition.

Soil classification—geochemistry

Five 1 kg soil samples were collected from each island. Samples were gathered from the four cardinal directions to minimize the effects of aspect on soil characteristics. 80 samples were analyzed for their grain size distribution, total carbon (C), total organic carbon (OC), total nitrogen (N) and total phosphorus (P). Grain size analysis was made using a Malvern Mastersizer 2000. Moment measures were calculated using GRADISTAT V.4 software (Blott and Pye 2001) and based on USDA classification. For C and N, we used a Carlo Erba EA1108 CHNS-O Elemental analyzer. OC content was estimated using the wet oxidation—titration method of Wakley and Black (1934), which has been validated and compared with LECO by Gaudette et al. (1974), and with the combustion-infrared method (Avramidis et al. 2015). We divided the OC by total N to determine a C:N. Traditionally, a conversion factor of 1.72 has been used to convert OC to organic matter based on the assumption that organic matter is 58% C (Nelson and Sommers 1982). We calculated P based on a persulfate digestion method according to APHA 4500-P (2005). CaCO3 was measured using a digital hand-held soil calcimeter (FOGII/Version 2/2014; BD INVENTIONS). More specifically, CaCO3 (%) calculations were based on the measurement of emitted CO2, a method modified from Müller and Gastner (1971). To determine the degree of erosion, soil depth measurements were taken from 30 random locations on each island using a graded metal stake that was sunk into the ground until it encountered bedrock; values were averaged to obtain an island-wide value of soil depth.

Quantification of seabird populations

Relative abundances of nesting seabirds were obtained during extended single island visits coinciding with the peak of the seabird nesting season (Hildén et al. 1995). This reduces any inter-measurement variation resulting from the presence of prospectors on some islands and not others. These estimates are meant to be comparative approximations of resident bird populations and are not true population censuses. We conducted our surveys during similar conditions (sunny with light wind), and during the same time of the day (mid-morning hours). All adult seabirds observed at the island were counted following procedures outlined in Walsh et al. (1995). Breeding seabirds were flushed upon our arrival on the islands and were counted in the air using binoculars by two independent observers for 15 min. Observation sites were chosen so as to encompass the whole island. The focus seabird species was the yellow-legged gull (Larus michahellis). Cliff nesting seabirds were almost completely absent from the islands and were not included in the counts. To our knowledge, none of the study islands harbored any nocturnal seabird colonies (Puffinus yelkouan and Calonectris diomedea). After estimates were calculated, the values were then divided by island area to account for differences in island size which determined our relative seabird abundances (birds/km2).

Statistical analysis

We compared all observed variables across the three island types—no goats, goats, and goats removed. Assumptions of normality were tested using Shapiro-Wilkes tests; variables that failed the test were either log-transformed (arthropod biomass, number of arthropods, C:N), square root-transformed (seabird density), arcsine-square root-transformed (% CaCO3, % Arachnida, % Diptera, % Annuals, and % Perennials), or Box-Cox-transformed using λ = − 0.5 (% N and % P). For all tests, n = 16. We utilized one-way ANOVA tests followed by post hoc Tukey comparisons to test for differences between the groups. If normality criteria could not be met, we used Kruskal–Wallis tests instead. To quantify vegetation and arthropod species diversity, we used a Shannon–Wiener Diversity Index (SWDI) that was developed to take into account both species richness and evenness (Maurer and McGill 2011). Our secondary analyses investigated the relationships between different variables using linear mixed model regressions (LMM), treating grazing status as a random effect. To assess the significance of differences in community compositions following NMDS for vegetation and arthropods, we used Analysis of Similarities (ANOSIM). All analyses were run in RStudio Team (2015).

Results

Effects of goats

Vegetation

Herbivory by goats has pronounced effects on island plant communities. We identified 119 distinct plant species from the 16 study islands (Appendix B in ESM). Grazing status of an island significantly affects the estimated plant species density (p = 0.00676, F 2,13 = 7.521, ANOVA). Islands with goats removed have significantly fewer species relative to both grazed and ungrazed islands; they have lost 45.4% of their estimated taxa numbers compared to ungrazed islands and 57.9% compared to grazed islands (Fig. 2a). Typical species found on ungrazed islands include Medicago arborea, Matthiola sinuata, and Atriplex halimus. Common plants associated with grazed islands include Plantago coronopus, Tordylium apulum, and Phleum arenarium. The SWDI for plants (p = 0.00412, F 2,13 = 8.628, ANOVA) follows a similar pattern: islands with removed goats have significantly lower SWDI values relative to both grazed and ungrazed islands (Fig. 2b). Percent vegetation cover is significantly lower (p = 0.004, F 2,13 = 8.438, ANOVA) in both grazed (25.96%) and goat-removed (38.56%) islands relative to the ungrazed sample. Plant biomass also decreases significantly (p = 0.00267, F 2,13 = 9.679, ANOVA), with grazed islands experiencing a decline to an average of 66.4% of the baseline (Fig. 2c). Mean plant height (p = 0.013, F 2,13 = 6.174, ANOVA) is on average 63.64% shorter on grazed islands and 64.85% shorter on goat-removed islands relative to ungrazed islands (Fig. 2d). The average percentage of annuals and perennials on the islands changed significantly with grazing regimes (p = 0.0058, F 2,13 = 7.857, ANOVA). Annual plant species increase from 24 to 50.2% of the community with the introduction of goats (subsequently, perennial species decreased from 76 to 49.8% of the community). Plant species composition differed significantly among the islands of different grazing status (p = 0.005, R = 0.357, ANOSIM). The NMDS plot (Fig. 3a) shows a complete separation between the plant communities of islands with goats and islands where goat have been removed. The plot also reveals the relative dissimilarity between the islands without goats and the other two grazing statuses. The surfaces fit onto the NMDS represent environmental variables that were shown to significantly affect community composition similarities: seabird density (p = 0.009), nitrogen content (p = 0.005), phosphorus content (p = 0.004), and soil depth (p = 0.053).

Fig. 2
figure 2

a Comparison of the means of the estimated plant species density. b Comparison of the means of vegetation SWDI. c Comparison of the means of the average aboveground dry vegetation biomass (g/m2). d Grazing effects on average plant height (cm). e Grazing effects on average soil depth (cm). f Grazing effects on seabird population density. Error bars represent standard error

Full size image
Fig. 3
figure 3

a NMDS plot with surface fitting for plant community composition. b NMDS plot with surface fitting for arthropod community composition. Island abbreviations are the same as in Fig. 1

Full size image

Arthropods

We collected a total of 7054 individuals from 118 arthropod taxa. Hymenoptera (particularly ants), Isopoda, Coleoptera, and Arachnida were the most abundant taxa, collectively comprising 84.2% of the collected arthropods (Hymenoptera 29.2%; Isopoda 24.3%; Coleoptera 22.7%; Arachnida 8%). Grazing status does not affect significantly estimated arthropod species density (p = 0.489, F 2,13 = 0.757, ANOVA), average arthropod biomass/trap/day (p = 0.611, F 2,13 = 0.511, ANOVA), or average number of arthropods/trap/day (p = 0.561, F 2,13 = 0.604, ANOVA). However, we documented a marginally significant positive effect of goat grazing on the arthropod SWDI (p = 0.064, F 2,13 = 3.417, ANOVA). Within individual arthropod groups, only the % Dipterans is significantly affected by grazing (p = 0.049, F 2,13 = 3.98, ANOVA) where we see a 109.25% increase in areas with goats relative to ungrazed islands. Grazing has a marginally significant effect on % Hymenopterans (p = 0.063, F 2,13 = 3.442, ANOVA) with goat presence resulting in a 53.2% decrease relative to ungrazed islands. The percentages of Arachnids (p = 0.557, F 2,13 = 0.613, ANOVA), Coleopterans (p = 0.146, F 2,13 = 2.235, ANOVA), Isopods (p = 0.281, χ 2 (2, n = 16) = 2.541, Kruskal–Wallis), and Hemipterans (p = 0.92, χ 2 (2, n = 16) = 0.167, Kruskal–Wallis) are not affected by grazing. Arthropod species assemblages appear to not be significantly affected by grazing status (p = 0.159, R = 0.105, ANOSIM). The NMDS plot (Fig. 3b) shows high overlap between the three statuses. The surface fit onto the NMDS represents soil depth (p = 0.01), an environmental variable that was shown to significantly affect arthropod community composition similarities.

Soil

Grazing also is significantly associated with erosion as evidenced by a decrease of 44.41% of average soil depth (p = 0.000983, F 2,13 = 12.36, ANOVA) (Fig. 2e) and an increase of 82.45% in the average amount of exposed bedrock (p = 0.0158, F 2,13 = 5.804, ANOVA) on grazed relative to ungrazed islands. More importantly, this process of soil loss continued even after goats had been removed, resulting in a 146.07% increase in average percent rock and a 68.31% decrease in average soil depth on islands with removed goats as compared to ungrazed islands (see Fig. 4).

Fig. 4
figure 4

Changes in typical vegetation cover from an island without goats, with goats, and removed goats

Full size image

While grazing by goats was associated with declining amounts of soil, we found little evidence for effects on soil structure or chemistry. 87.8% of our soil samples were classified as a silt loam while 4.9% were classified as sandy loam, 3.7% silt, and 1.2% sand, loamy sand, and loam, respectively. Grazing status does not significantly affect average % bare ground (p = 0.525, F 2,13 = 0.678, ANOVA), average % organic matter (p = 0.591, F 2,13 = 0.549, ANOVA), average % CaCO3 (p = 0.548, F 2,13 = 0.63, ANOVA), average % N (p = 0.65, F 2,13 = 0.445, ANOVA), average % P (p = 0.845, F 2,13 = 0.171, ANOVA), C:N ratio (p = 0.836, F 2,13 = 0.182, ANOVA), average % sand (p = 0.757, F 2,13 = 0.285, ANOVA), average % silt (p = 0.762, χ 2 (2,n = 16) = 0.545, Kruskal–Wallis), or average % clay (p = 0.464, F 2,13 = 0.814, ANOVA).

Seabirds

On grazed islands, we witnessed a trend towards lower seabird populations relative to ungrazed islands; however this effect does not quite reach statistical significance (p = 0.154, F 2,13 = 2.168, ANOVA) (see Fig. 2f).

Secondary analyses

Impacts of soil erosion

Erosion caused by goats has significant implications for vegetation characteristics. The increased occurrence of bedrock, resultant from goat grazing, is associated with reduced average percent vegetation cover (p < 0.00001, t = − 8.721, LMM), shorter average plant height (p < 0.00001, t = − 4.721, LMM), and marginally diminished plant biomass (p = 0.053, t = − 1.932, LMM). Unexpectedly, we document a significant positive relationship between amounts of exposed bedrock and the SWDI for arthropods (p = 0.012, t = 2.524, LMM), as well as estimated arthropod species density (p = 0.064, t = 1.85, LMM).

Similarly, decreased soil depth is associated with reduced vegetation cover (p = 0.00016, t = 3.767, LMM), diminished average vegetation height (p = 0.0015, t = 3.18, LMM), and marginally fewer arthropods collected per trap (p = 0.077 t = 1.77, LMM). In addition, lower C:N ratios are correlated with higher estimated arthropod species density (p = 0.019, t = − 2.351, LMM). Interestingly, higher N & P values are associated with lower plant SWDI values (p = 0.007, t = − 2.695, LMM & p = 0.0006, t = − 3.423, LMM, respectively). Higher N content is also marginally positively associated with arthropod SWDI values (p = 0.057, t = 1.902, LMM) as higher P values are similarly associated with estimated arthropod species densities (p = 0.054, t = 1.93, LMM). Lastly, we document marginally significant positive relationships between organic matter content and plant height (p = 0.059, t = 1.889, LMM).

Impacts on arthropods

A marginally significant positive relationship was found between both estimated plant species density and average plant height and average arthropod biomass (p = 0.068, t = 1.826, LMM & p = 0.051, t = 1.952, LMM, respectively). A marginally negative relationship was also found between average vegetation biomass and estimated arthropod species density (p = 0.082, t = − 1.74, LMM).

Seabird effects

The vast majority of nesting seabird species found on the study sites were yellow-legged gulls (Larus michahellis) with occasional European shags (Phalacrocorax aristotelis). Relative abundance of these seabirds is associated with significantly increased levels of N (p = 0.00016, t = 3.774, LMM) and P (p < 0.00001, t = 7.941, LMM) in the soil, presumably representing marine subsidies (Fig. 5a, b). A significant negative relationship was found between seabird densities and the vegetation SWDI (p = 0.00006, t = − 3.995, LMM). No significant relationships were detected between seabird densities and estimated plant species densities (p = 0.732, t = − 0.342, LMM), average % vegetation cover (p = 0.616, t = − 0.502, LMM), average vegetation biomass (p = 0.84, t = − 0.202, LMM), average plant height (p = 0.609, t = 0.512, LMM), average soil depth (p = 0.781, t = − 0.279, LMM), average % organic matter (p = 0.197, t = 1.289, LMM), average % CaCO3 (p = 0.136, t = − 1.493, LMM), C:N ratio (p = 0.135, t = − 1.497, LMM), estimated arthropod species density (p = 0.286, t = 1.068, LMM), arthropod SWDI (p = 0.193, t = 1.302, LMM), average arthropod biomass/trap/day (p = 0.67, t = − 0.426, LMM), or average number of arthropods/trap/day (p = 0.524, t = 0.638, LMM).

Fig. 5
figure 5

a Linear mixed model showing the correlation between soil nitrogen content and seabird density. b Linear mixed model showing the correlation between soil phosphorus content and seabird density. Note the axes have been Box-Cox transformed using λ = − 0.5

Full size image

Discussion

Effects on vegetation

The presence of goats has wide-ranging impacts on vegetation and soil characteristics and has therefore direct implications for island conservation programs. Trampling of soil and removal of vegetation dislodges and triggers soil movement, which sets in motion a positive feedback loop of desertification that persists even after goats have been removed (López-Bermúdez 1990). Independently of direct goat biomass consumption, loss of soil leads to reduced plant cover, less plant biomass, and overall shorter plants, thus further diminishing the ability of vegetation to stabilize soil. Once goats are removed, vegetation cover and plant biomass rebound partially, but remain restricted to the small pockets of soil persisting in rock depressions. While vegetation cover and biomass partially recover, plant growth forms remain prostrate due to prior browsing which diminishes the heterogeneity of the vertical plant structure (Noy-Meir et al. 1989; Dennis et al. 1998). Overgrazing has the ability to even change the hydrology of ecosystems and increase stream flows through reduced infiltration and increased surface runoff (Van Vuren et al. 2001).

Despite these reductions in plant presence, we did not document a significant change in the estimated plant species density or in plant SWDI between islands with, or without, goat grazing. This is in general agreement with other studies done in the Mediterranean (Alrababah et al. 2007). It has been proposed that grazing intensity is not expected to change the diversity of semi-arid grasslands with a long history of grazing (Milchunas et al. 1988; Osem et al. 2002). However, this lack of change in species diversity masks important dissimilarities in plant community composition between islands (Fig. 3a) (Cingolani et al. 2005). While plant assemblages on islands without goats are characterized by relatively rare, native perennial species such as Medicago arborea, Matthiola sinuata, and Atriplex halimus, islands with goats present are characterized by more annual, generalist, low-lying plant species, usually with sticky seeds, such as Plantago coronopus, Tordylium apulum, and Phleum arenarium. Panitsa et al. (2006) suggested that the increase in plant species associated with goat grazing can be traced to the introduction of seeds in the fur or feces of an animal. The most effective animal seed transport mechanisms are usually seen in annual plant species (Shmida and Ellner 1983). A 32-year study by Snogerup and Snogerup (2004) noticed that the main changes in islet flora concerned soil space opened up by grazing animals and erosion, and also classified roughly 100 taxa that were probably transported by grazing animals. Indeed, high grazing pressure can actually increase plant species richness (Hobbs and Huenneke 1992), especially in nutrient-rich ecosystems, such as seabird islands (Proulx and Mazumder 1998). Fernández-Lugo et al. (2009) similarly found that species richness failed to recover after goat grazing but also noticed significant species turnover related to higher soil magnesium content. Our observations indicate that once goats are removed from an island, the island begins to lose some of these introduced species, yet the native species do not return, presumably due to a lack of seed sources, ultimately resulting in an overall decline in alpha diversity. Mulder et al. (2009) found that if there is an expansion of non-native plant species early in the recovery process, islands could have very different successional trajectories than uninvaded islands, leading to two alternative stable states. Depending on the goals of an island conservation project, the actual identities of the plant community may be more important than overall levels of species richness, and re-introduction of focal plant species may be warranted.

Effects on arthropods

We did not observe a consistent effect of goat grazing on arthropod communities (Fig. 3b). In our study, arthropod numbers did not rise with increased vegetation cover or reduced grazing, perhaps because any gains were passed on to lizards which constitute the next trophic level up. However, both goat grazing and declining vegetation cover were associated with increases in insect SWDI, potentially because more microhabitats were available. Other studies have also found that there are groups of arthropods that prefer increased bare ground associated with higher stocking rates of herbivores (García et al. 2009). Specific arthropod guilds have been reported to be differentially influenced by vegetation covers (Gardner et al. 1997; Woodcock et al. 2005). When arthropods were analyzed by order, we only detected a significant increase in Dipterans with grazing, possibly because many Dipteran taxa utilize goat dung to breed (Khan et al. 2012).

General arthropod community characteristics also appear unaffected by direct seabird densities, in general agreement with previous research (Orgeas et al. 2003; Towns et al. 2009), which documented an effect of seabird numbers on arthropod community composition, but no effect on arthropod species richness or biomass. However, we did record a positive association between arthropod species density and elevated soil P levels and between arthropod SWDI and soil N levels, paralleling similar observations in the past (Anderson and Polis 1998; Sayer et al. 2010). Kaspari et al. (2008) observed that P often tends to be the main limiting nutrient for decomposition in most ecosystems and fertilization experiments have shown that P addition stimulates cellulose-decomposing microbes resulting in greater arthropod biomass.

Soil effects

In our study, soil chemical characteristics do not seem to be influenced by grazing. Instead, it is primarily the amount, and not the chemical properties of the soil, that are affected in this system. We find that pronounced losses of what are already shallow, erosion-prone soils, are the main symptom of goat grazing, as well as the driver for further degradation, and result in sparse, structurally simplified habitats. Our observations suggest that erosion associated with goat presence is not entirely caused by exposure due to the removal of plant cover; additionally the breaking of the fragile soil crust by goat hooves allows for the notoriously strong local winds to carry soil particles out to the sea. Other studies in Mediterranean ecosystems have seen varying and contrasting influences of grazing on soil nutrients (Yates et al. 2000; Sánchez-Marañón et al. 2002; Peco et al. 2006; Fernández-Lugo et al. 2009). Increasing N & P, primarily driven by seabird density, were found to lower plant SWDI, which is suggested to be caused by a few competitive plant species increasing in abundance while the majority decline (Tilman 1987; Farrer et al. 2013).

Effects of seabirds

Seabirds appear to shape the study island communities through bottom-up trophic interactions. Seabirds are thought to prefer nesting on smaller islands (Vidal et al. 2001), an observation that was confirmed in this study. Previous studies in the region (Pafilis et al. 2013) have detected a significant inverse relationship between seabird density and grazing. Our results, although failing to meet statistical significance, are consistent with a similar negative trend between goats and resident seabird populations reported nearby (Pafilis et al. 2013). The reduction in seabird abundance and the subsequent nutrient transport has been shown to have cascading effects on below-ground organisms and the ecosystem processes they drive (Fukami et al. 2006). We also detected a strongly significant increase in soil N and P levels with higher seabird densities. Allochthonous nutrient inputs from marine subsidies are critically important for these isolated terrestrial islands (Sánchez-Piñero and Polis 2000). These nutrients may support higher plant biomass and more nitrophilous plant species (Polis et al. 2002). However, these nutrients were only found to stimulate primary productivity during wet periods (Polis et al. 2002), which may explain why—given the generally arid conditions in the study region—we do not see an effect of soil nutrients on vegetation characteristics in our study.

Seabirds can reduce plant vegetation by trampling and burrowing activities or by creating locally toxic conditions through extremely high nutrient levels (Hata et al. 2014). These negative effects may cancel out some of the positive effects in our study. Seabirds can affect island arthropods in at least 2 ways—first, by providing carcasses and food scraps for scavengers such as Tenebrionidae or Dermestidae beetles (Polis and Hurd 1996), and second, by increasing herbivorous or detritivorous and eventually even predatory arthropods, via increased plant productivity (Anderson and Polis 1999; Polis et al. 2002). Lastly, gulls can also act as important seed dispersers between islands; we regularly found large Pistacia lentiscus seeds among the investigated guano. Given this prominent role that seabirds play in Mediterranean island ecosystems, it is important to protect, and even restore seabird colonies on islands from which goats were eradicated.

Conclusion

This study demonstrates that goat herbivory has severe and long-term effects on many, but not all aspects of island ecosystems. Goats set in motion a positive feedback loop of desertification, which persists even after their removal. The grazing and associated soil erosion contribute to changing plant communities, lower plant biomass, less vegetation cover, and shorter plants. There is also some support for an impact on seabirds, an important source of N and P in our system, although this relationship fails to meet statistical significance. Subsequent studies may benefit from including grazing intensity in the analyses. This study suggests that goat removal programs should be accompanied by restoration of the native vegetation ideally before there is extensive loss of soils. Full restoration of small island ecosystems will be slow given the very long periods needed for soil regeneration.