Rewilding giant tortoises engineers plant communities at local to landscape scales

Trophic rewilding is increasingly being used to promote megafauna reintroductions to island ecosystems, yet ecosystem response to population restoration once megafauna reintroduction occurs remains understudied. In this study of a population of Galapagos giant tortoises reintroduced to an arid island, tortoise exclosures monitored over an 8‐year‐long period revealed that, in response to the presence of tortoises, herbaceous plant cover and numbers of regenerating woody plants decreased, whereas extent of grass cover increased. Vegetation mapping over a 15‐year‐long period across the island indicated a threshold density of 1–2 tortoises per hectare halted incursion of woody plants and triggered a shift in this savannah‐type ecosystem toward more grasses. Restoration of this giant tortoise population has shaped plant communities at both local and landscape scales with cascading effects on many components of biodiversity on the island.


INTRODUCTION
Mega-herbivores have been drastically reduced in most regions of the globe, rendering them functionally extinct in many areas (Morrison et al., 2007). These declines have had significant ecosystem-level consequences due to the loss of many ecosystem functions that mega-herbivores provide (Hyvarinen et al., 2021). Concerted efforts are being made to restore megaherbivores around the world, with "trophic rewilding"-the restoration of large-animal populations to revive top-down interactions and reverse environmental degradation (Bakker & Svenning, 2018;Svenning et al., 2016)-often used as the rationale (Genes & Dirzo, 2022).
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2023 The Authors. Conservation Letters published by Wiley Periodicals LLC.
Islands are a particular focus for megafauna restoration because island faunas have undergone much higher rates of extinction than continental faunas (Wood et al., 2017). Although trophic rewilding is increasingly being used to justify megafauna reintroductions to island ecosystems (Frazier, 2021), our understanding of the effects of herbivore collapse on ecosystems is based largely on studies of large mammals on continents. This is problematic because the megafauna of island ecosystems is dominated by reptiles, and ectothermic (reptile) herbivores function in fundamentally different ways than do endothermic (mammalian) herbivores. Assessments of ecological consequences of restoration of reptile populations for island ecosystems are starting to emerge (e.g., Jones et al., 2022), yet it remains uncertain whether restoration of trophic structure through rewilding of mega-reptiles catalyzes change in island ecosystems (Fraser et al., 2015).
We focus on the Galapagos Islands where 12 species of giant tortoises (Chelonoidis spp.) are currently the target of ongoing restoration programs. Giant tortoises along with Galapagos land iguanas (Conolophus spp.) in different combinations on different islands are the primary herbivores in these insular ecosystems. Most populations of all extant tortoise species have been decimated (Tapia et al., 2021), and at least three species have been driven to extinction. Giant tortoises consume a wide variety of plants (Blake et al., 2015) and are long-distance seed dispersers (Blake et al., 2012). In addition to browsing, giant tortoises trample vegetation, which in combination potentially affects recruitment of woody plants in areas where tortoises occur at ecologically effective densities (>1 per hectare; . We studied tortoise-plant community interactions operating at two spatial scales in a savannah-type ecosystem on Espanola Island, Galapagos, where the population of the island's endemic species of giant tortoises (Chelonoidis hoodensis) has been restored from its nadir of just 14 individuals in the 1960s to some 3000 individuals today (Cayot, 2021). We used exclosures to examine local-scale (ca. 10 2 m 2 ) effects of giant tortoise activity on plant community composition over an 8-year-long period. We also compared change over a 15-year-long period in woody plant cover at the landscape scale (10 7 m 2 ) in areas with varying densities of re-established tortoises. We predicted that presence of reintroduced giant tortoises would reduce woody plant recruitment through browsing and trampling (Figure 1), thereby facilitating a shift to grasses at the local scale in a manner that would be reflected at the landscape scale as reduced dominance of trees in areas where tortoises had become established versus areas where they had not.

METHODS
Espanola Island (1.38 • S, 89.68 • W) is the southeasternmost of the Galapagos Islands ( Figure 1). The island is small (60 km 2 ), low (maximum elevation = 206 m), and arid (10−600 mm annual precipitation and 23.8 • C average annual temperature). The island's primary plant community is an association of grasses and herbaceous plants (Galactia striata, Galactia tenuiflora, Phaseolus mollis, Rhynchosia minima, and Paspalum spp.) interspersed with woody plants (primarily Cordia lutea and Prosopis juliflora), now heavily skewed to the latter over much of the island due to both the prolonged absence of tortoises and the occurrence over nearly a century of large numbers of feral goats, removed in 1978, whose overgrazing resulted in the proliferation of woody plants at the expense of the grasses ). An arboreal cactus (Opuntia megasperma), also reduced by feral goats, is limited to a few thousand individuals mostly in the island's central region, a 1250-ha area where repatriates from the 50-year-long ex situ giant tortoise breeding program have been released (Cayot, 2021). The endemic Espanola giant tortoise is the only resident large-bodied herbivore; no species of land iguana occurs on the island.

Local-scale effects of tortoises on plant community
To measure the response of the plant community at the site level to the presence of giant tortoises, twenty-five 6 × 6 m square-shaped plots each centered upon an adult Opuntia cactus were established in 2014 near "el Caco" (Figure 2). Herbivore exclosures, which amount to experimental manipulations of herbivore densities, represent a powerful means to isolate the effects of herbivores on plant communities. To establish exclosures, at random 12 plots were selected to be fenced to exclude tortoises (chain-link fences; Figure 1) and 13 were left unfenced to serve as "controls" accessible to tortoises. Plots were subdivided into 2 × 2 m subplots upon which measures of plant community were made starting in May 2013 (immediately prior to the construction of the exclosures), and then repeated during May-August (at the end of each rainy season when plant growth had peaked) in 2013, 2014, 2015, 2017, 2019, and 2021. Measures of vegetation on each subplot included extent of herbaceous plant and grass cover (proportion of area, visually estimated) and censuses of all woody plant stems identified to species, cactus seedlings, fallen cactus fruits and pads, and tortoise droppings.
To test effects of the tortoise exclusion treatment on the plant communities, we used linear mixed-effects models (Zuur et al., 2009) as implemented in the lme4 package (Bates et al., 2015). Because the sampling design corresponded to Before-After-Control-Impact (BACI; Underwood, 1993), the variables Plot (with the factors Control F I G U R E 1 Impacts of rewilding giant tortoises on vegetation on Espanola Island, Galapagos (a). Predictions of tortoise impacts to arid zone savannah-type vegetation of Galapagos (b) in which "current" situation represents prevalence of woody plants due to lack of native herbivores and legacy of previous infestation of goats, now removed . A male Chelonoidis hoodensis Espanola giant tortoise beneath to a mature Opuntia megasperma cactus (c). One of the 6 × 6 m tortoise exclosures ("treatments") used in the study (d).

F I G U R E 2
Sample grid consisting of 1650, 1-ha cells on Espanola Island, Galapagos where effect of tortoises occurring at different densities on change in woody cover was measured at the landscape scale over a 15-year-long period. Darkened cells indicate areas where tortoises occurred. The area where tortoise exclusion plots and their controls are located is indicated by the red dot.

Landscape-scale effects of tortoise activity on extent of woody vegetation
We contrasted woody plant cover on the island over a 15year-long period among areas hosting varying densities of tortoises. To do so, spatial extent of woody vegetation on 1650 square-shaped 1-ha grid cells in the central part of the island (Figure 2), where most tortoises occur (Cayot, 2021), was determined by delineating crowns of woody plants (primarily Cordia lutea) evident in cloud-free satellite imagery (Quickbird, December 29, 2006, 0.6 m resolution; WorldView-1, July 7, 2020, 0.5 m resolution). Woody vegetation was categorized using a machine-learning algorithm and tree-based classifier (RandomForest) as implemented in Weka 3.6.5 (Hall et al., 2009) developed for each set of imagery. Classification accuracy was determined by cross-tabulating the predicted identity (woody vegetation vs. grasses/herbaceous plants) of 500 randomly located points with their actual identity as determined through inspection of the aerial imagery and field validation.
Tortoise populations were surveyed in 2000,2002,2003,2007,2010,2019, and 2021 by groups of park rangers methodically searching for tortoises throughout the sampled area who noted the geographic location (accurate to 2 m) of each tortoise encountered (Gibbs & Goldspiel, 2021); these encounters were summed by survey per grid cell to estimate tortoise density on a given survey, which was then averaged across surveys for a given grid cell to provide an index of resident tortoise density on each plot. The difference in woody plant extent in 2020 versus 2006 was then calculated and its magnitude contrasted across discrete levels of averaged tortoise density (0, 1, 2, and 3 tortoises/ha). An analysis of covariance (ANCOVA) was used to examine the contribution of woody cover extent in 2006 to that in 2020 (both asin(sqrt(x)) transformed prior to analysis) as mediated by tortoise density.

Local-scale effects of tortoises on plant community
Tests using linear mixed-effects models of effects of tortoise exclusion on the plant communities ( Figure 3) coupled with visualization of those effects (Figure 4) indicated that droppings of tortoises disintegrated inside the exclosures within 2 years, with no new droppings appearing, indicating efficacy of the treatment to exclude tortoises. Tortoises rapidly and dramatically reduced availability of Opuntia fruits and cladodes, with 3-10× higher numbers of fruits and cladodes accumulating on plots from which tortoises were excluded. Extent of herbaceous plant cover and numbers of regenerating woody plants decreased in the presence of tortoises, whereas extent of grass cover increased. Too few Opuntia seedlings emerged in plots of either treatment over the study period to permit analysis of change in cactus regeneration.

3.2
Landscape-scale effects of tortoise activity on extent of woody vegetation Of the validation points, 93.1% and 95.7% of those dominated by woody vegetation were correctly classified as such during 2006 and 2020, respectively, whereas 89.4% and 92.3% of those lacking woody vegetation were correctly classified for 2006 and 2020, respectively, suggesting that the machine-learning classifier of woody plants in aerial imagery performed adequately. Woody vegetation dominated the landscape of the central portion of the island in 2020, with 78% of the 1650, 1-ha grid cells supporting ≥95% cover by woody vegetation . Magnitude of change in woody cover between 2006 and 2020 was positive on the 85% of grid cells where no tortoises occurred, suggesting ongoing incursion of woody plants over much of the island, but was negative and increasingly so on grid cells with increasing tortoise density ( Figure 5). The ANCOVA indicated that the effect of resident tortoise density was mostly additive (Table 1) albeit with a significant interaction, such that both the slope and intercept of vegetation change between 2006 and 2020 differed, an interaction mostly driven by the highest density of tortoises ( Figure 6).

DISCUSSION
Our study provided an opportunity to ask questions about the role of giant tortoises as reptile mega-herbivores in structuring the plant communities on islands, as well as to explore the ramifications of restoring reptile populations on islands to promote ecosystem recovery (Hansen et al., 2010). The study provided evidence that giant tortoises reintroduced to Espanola Island do engineer the structure of its terrestrial plant communities, with giant tortoises potentially mediating relationships between trees and grasses in these savannah-type ecosystems. Recruitment of woody vegetation and herbaceous plants was reduced in areas that tortoises could access. These interactions at the local scale were manifested at the landscape scale in that areas with no tortoises became increasingly dominated by woody cover over the 15-year-long monitoring period, whereas those where tortoises had become established lost woody cover. Increasing population density of these mega-herbivores produced what appeared to be largely additive effects on reducing ingrowth of woody plants in the plant community studied, albeit with the highest tortoise densities observed catalyzing a reversal in the plant community from a trend toward increasing domination by woody plants toward a tree-grass savannah-type association.

F I G U R E 3
Temporal trends in vegetation on plots from which giant tortoises were excluded versus on control plots where they had access over an 8-year-long period on Espanola Island, Galapagos. Exclosures were established after the first year of measurement in 2013. Box plots indicate minimum, first quartile, median, third quartile, and maximum for each response variable at each treatment level.

TA B L E 1
Outcomes of an analyses of covariance examining the effect on woody cover extent in 2006 of that in 2020, tortoise density (0, 1, 2, and 3 tortoises/ha), and their interaction, on each of 1650, 1-ha grid cells over a 15-year-long period on Espanola Island, Galapagos. Tortoises exerted substantial impacts particularly on cactus, a keystone species for the entire vertebrate community. Tortoises consumed virtually every cladode (pad) that fell to the substrate. This would effectively eliminate vegetative reproduction in cactus; however, there was no obvious recruitment of Opuntia near adult plants in the areas from which tortoises were excluded by asexual means (rooted pads or rooted fruit pericarp). Birds (finches, mockingbirds, and doves) consume and disturb pads, effectively preventing their rooting, which may explain the lack of vegetative reproduction in cacti from areas where tortoises were excluded, despite large volumes of cactus cladodes being deposited over the 8-year study ( Figure 3). Similarly, tortoises consumed almost all cactus fruits deposited, which would greatly expand the scope for seed dispersal away from the parent plant, where bird predation on seeds is most intense (Grant & Grant, 1981). We expect that seed dispersal by tortoises provides a critical vector for sexual reproduction in cacti , resulting in more cactus recruits in the larger landscape, whereas intense consumption of cactus parts by tortoises renders asexual reproduction by cactus inviable. Similarly giant tortoises (Geochelone gigantea) introduced to an island ecosystem in the Indian Ocean played an important role through seed dispersal in regenerating a palmrich communities while having only modest effects F I G U R E 4 Effect size estimates associated with tortoise exclusion (treatment) on the plant communities over an 8-year-long period on Espanola Island, Galapagos, as derived from linear mixed-effects models for each response variable (see Section 2). Effects sizes represent the number of standard deviations the coefficient differs from zero (blue = positive effects, red = negative effects). Points represent effect means, lines 95% confidence intervals, and asterisks the significance level in relation to difference of an effect size from zero (*≤0.05, **≤0.01, ***≤0.001).

F I G U R E 5
Relationship between tortoise density and change in woody vegetation from 2006 to 2020 on Espanola Island, Galapagos, with internal box plots depicting minimum, first quartile, median, third quartile, and maximum for ∆woody plant change at each level of tortoise density. F I G U R E 6 Linear relationships between extent of woody cover on each grid cell in 2006 versus 2020 at different tortoise densities (with 95% confidence intervals); dotted line indicates parity of change such that grid cells above the line gained woody vegetation, whereas those below lost woody vegetation between 2006 and 2020. Marginal histograms indicate distribution of woody cover values for 2006 (top margin) and 2020 (right margin). on communities dominated by herbaceous plants ( Moorhouse-Gann et al., 2022).
A limitation of our study could be the representativeness of sites chosen for the exclosure experiments. All exclosures were centered upon adult cacti-areas that attract tortoises searching for the food and the shade the cactus provides, especially during the extended dry season that occurs on Espanola Island. Such areas were chosen to ensure homogeneity among study plots and to focus on sites where tortoise activity was frequent in order to maximize the scope for detecting interactions among tortoises, cactus, and plant communities, should they occur. Tortoise activity diminishes in areas farther from adult cacti (Gibbs et al., 2008) such that there is likely to be a time lag to see comparable effects farther from cacti, areas that comprise much of the island.
Identifying timescales of progress in trophic rewilding is important, because it is easier to mobilize support for efforts that yield outcomes quickly (Bakker & Svenning, 2018). The changes observed over this relatively long-term study represented relatively modest effect sizes for all response variables measured (except on cactus fruit and cladode availability). Galapagos ecosystems are strongly affected by el Niño/la Niña (ENSO) events, which produce a short-term increase in rainfall associated with el Niño (Tye & Aldáz, 1999). No ENSO event occurred during the course of our study. Extending monitoring through the next strong ENSO event will be important to determine if it enhances or abates effects of tortoises on this ecosystem.
The incipient changes to plant communities that giant tortoise rewilding on Espanola Island has catalyzed may ultimately have larger carryover effects on terrestrial organisms related to changes in habitat structure and composition. Tortoises on Espanola Island are creating mosaics of vegetation and maintaining the savannahlike plant communities of arid zones intermediate along the grassland-woodland gradient, changes that will likely affect many other species. Some endemic plant species on this and other arid Galapagos islands (e.g., Acalypha abingdonii) appear restricted to herb-grass meadows, an association that giant tortoises appear to maintain by preventing shrub overdominance (Hamann, 1993). Recovery of the island's tree cactus (Opuntia megasperma) appears to be underway following tortoise introduction due to the seed dispersal services provided by tortoises (Gibbs et al., 2008); given the importance of cactus for much of the island's animal community (Grant & Grant, 1981), Opuntia recovery is likely to have ramifications for faunal diversity. Espanola Island is also effectively the only nesting area on the planet of the endemic albatross (Phoebastria irrorata), a species that appears to rely on giant tortoises to maintain areas free of woody plants that albatross require for landing and taking off to access nest sites (Anderson et al., 2002). More generally, in terrestrial ecosystems, species diversity is strongly related to habitat heterogeneity, an outcome enhanced by tortoise rewilding, in both Galapagos (birds: Geladi et al., 2021;invertebrates: Desender et al., 1999) and elsewhere (birds: Roth, 1976;invertebrates: Law et al., 1988).
Island ecosystems are among those most impacted by humans. Human exploitation on islands has largely destroyed populations of large-bodied native herbivores that probably had large effects on the structure and function of island ecosystems. Because many species of mega-herbivores that once inhabited islands have been driven to extinction, including giant tortoises across much of the globe (Falcon & Hansen, 2018), the process is irreversible, except in the case of using "analog" extant species to replace ecosystem services of extinct forms (e.g., Tapia et al., 2022). Otherwise, current rewilding efforts must focus on species that are extant. Here, we provide strong evidence that rewilding of large herbivores on islands can restore some key ecosystem services. Herbivores clearly reduced the abundance of shrubs and higher densities of herbivores were more efficient in doing so. Secondary, positive impacts of tortoises on other species are also plausible. A fuller elaboration of tortoise impacts is needed as effects on other ecosystem services, such as nutrient cycling and carbon sequestration, remain unexamined.
Many proposals to re-instate key ecological functions through mega-fauna rewilding are pending or underway (Frazier, 2021). One of the primary shortcomings identified for promoting trophic rewilding as a justification for population restoration of large herbivores has been insufficient knowledge about the possible outcomes of rewilding efforts (Rubenstein & Rubenstein, 2016). We have documented that population restoration of a mega-herbivore can alter a plant community and generate cascading effects for other biodiversity components within it. This study suggests that trophic rewilding represents a valid rationale for reintroduction of these and similar mega-reptiles for restoration of island ecosystems. Given the increasing scale and ambition of rewilding projects involving large-bodied herbivores on islands many of which are predicated on restoration of ecosystem services those herbivores might have once provided, more such studies examining other herbivores on other islands in the context of a broader range of ecosystem services are needed.

A C K N O W L E D G M E N T S
Support for the study was provided by the Galapagos Conservancy, NASA Award 20-ECOF20-0019, and the BAND Foundation. Permission to access the study site and perform the fieldwork was provided by the Galapagos National Park Directorate , whose park guards also collected much of the field data.

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The authors declare no conflicts of interest.

D ATA AVA I L A B I L I T Y S TAT E M E N T
Data used in this study are available from the Knowledge Network for Biocomplexity (urn:uuid:205527dd-0e24-424d-9720-240c78dd6916).