What passes through the extinction filter? Historical and contemporary patterns of vulnerability of the most extinction-prone bird family (Aves: Rallidae)

Abstract

Patterns of extinction risk can vary across taxa, with species of some groups being particularly vulnerable to extinction. Rails (Aves: Rallidae) represent one of the most extreme yet well-documented cases of mass extinction within a modern vertebrate group. Between 54 and 92% of rail species became extinct following waves of human contact during both the Holocene and the Anthropocene eras, and a third of the extant species are currently threatened or near-threatened. Here, we (1) examine extinction filters through consecutive human contacts with rails, investigating the role of intrinsic life-history traits and (2) investigate the drivers of contemporary vulnerability. During the most recent wave of extinction, we found that body size was an important correlate of rail extinctions, with both smaller and larger bodied species more likely to become extinct. Island endemism and small clutch size were the strongest predictors of contemporary vulnerability. Overall, island endemic rails tend to follow the same trajectory as extinct species, suffering mostly from invasive predators and overhunting, but with different traits targeted contemporarily compared to past extinctions. Moreover, modern anthropogenic threats have created the potential for new intricate pathways – or a contemporary ‘field of bullets’ – making future vulnerability potentially less predictable.

Impact statement

Vulnerability to extinction in rail bird species has shifted over time, from traits like flightlessness and naivety to humans, to slow reproduction in the modern context. While past extinctions primarily occurred on islands, contemporary threats have diversified, making future vulnerability less predictable and highlighting the urgent need for tailored solutions to island conservation.

Introduction

In just over three centuries, the pace of extinction has accelerated far beyond natural background rates (Pimm et al., 1995; Crutzen, 2002; Ceballos et al., 2015), leading experts to consider this sixth major extinction event as a new geological epoch, termed the Anthropocene (Crutzen, 2002; Zalasiewicz et al., 2008). Understanding why and how species become extinct when facing anthropogenic activities is a major question in extinction biology (Diamond, 1989). This provides the potential to better predict future biodiversity loss, with the ultimate goal of providing efficient conservation efforts (Brook and Alroy, 2017). A particularly critical aspect is the need to determine whether past extinction events are provoked randomly: the ‘field of bullets’ hypothesis (Raup, 1991), or linked to species’ life-history traits: the ‘extinction-filter effect’ hypothesis (Pimm et al., 1995; Balmford, 1996; Turvey and Fritz, 2011).

During past extinction events, some taxa were over-represented in the assemblages of extinct species. Within the avifauna, for example, island endemic birds were disproportionately vulnerable during the first human contact in the Holocene (Steadman, 1995) and at a subsequent contact on islands (mostly with European settlers, starting in the 16th century; Pimm et al., 2006; Duncan et al., 2013). Some bird families were found to be systematically more prone to extinction during these events, with many hundreds or thousands of species going extinct due to vulnerabilities induced by specific life-history traits such as island endemism, large body size and flightlessness (Steadman, 2006; Boyer, 2010; Duncan et al., 2013). This suggests that these island-bird extinctions were driven by extinction filters.

Identifying potential shifts in vulnerability between past and contemporary patterns can be crucial to provide reliable context and potential projections for the future. These shifts are expected, as predominant threats have changed throughout time and may therefore target different traits or species (Boyer, 2010; Bromham et al., 2012). Identifying them would answer the question of ‘can past avian extinctions help to forecast extinction risk in birds?’ Historically, overhunting, introduced predators, and to a lesser extent habitat changes were the main anthropogenic threats to birds (Johnson and Stattersfield, 1990; Steadman, 1995; Blackburn et al., 2004; Duncan et al., 2013), while habitat loss created by accelerated land-use changes and climate change are more impactful to species in contemporary times (Bennett et al., 2001; Urban, 2015; Ducatez and Shine, 2017). Contemporary anthropogenic activities also appear to be magnified by an increase in amplitude, intensity and diversity, relative to the past. For example, the IUCN Red List references 38 different types of anthropogenic threats to wildlife (IUCN, 2019).

Most comparative studies on extinction risk investigate contemporary patterns globally and across taxa to extract intrinsic reasons for vulnerability. However, human impacts are not spatially or temporally consistent and species responses to threats can depend on their resistance to previous extinction filters (Diamond, 1984; Steadman, 1999; Biber, 2002; Boyer, 2008; Bromham et al., 2012). Therefore, analysing the evolution of vulnerability within extinction-prone families that went through successive extinction waves is a way to break down patterns of extinction and resistance, and to identify intrinsic causes.

Rails (Aves: Rallidae) are the most extinction-prone bird family, with 54 to 92% of all species going extinct after their first contact with humans during the mid-Holocene (representing 200 to 2,000 estimated extinct species; Steadman, 1995; Curnutt and Pimm, 2001). They went through a second wave of extinction from the 16th century when European settlers spread worldwide (hereafter ‘Era of Colonialism’). Globally, a third of extant rails are currently threatened or near-threatened (47 species). Island endemic rails are the most threatened rails (Lévêque et al., 2021) and have historically been observed in diverse situations of human contact, resulting in either extinction or coexistence. In the context of island rails, it is evident that their interactions with humans have varied substantially over time, offering a compelling lens through which to explore biogeographical patterns and processes. Some rail species have coexisted with humans since the Pleistocene/Holocene transition, adapting to early human-induced changes, whereas others encountered humans much later, during the mid- to late Holocene or even as recently as the Era of Colonialism (16th to 20th century), in locations such as Saint Helena, Ascension and Tristan da Cunha. This staggered timeline of human arrival and colonisation across different islands has led to a diverse array of impacts on rail populations, from minimal disturbance in some areas to complete extinction in others. By analysing these varied interactions and their outcomes, we can dissect the roles of different extinction filters – such as habitat destruction, introduced predators and over-hunting – and their sequential impact as islands were colonised over time. This biogeographical perspective can both shed light on the historical dynamics that have shaped current rail distributions and yield insights into the broader principles governing species survival and extinction on islands.

Here, we provide a thorough review of correlates to extinction risk and vulnerability throughout different temporal (Era of Colonialism or current time) and spatial scales (globally or on islands only), using the rail family as our exemplar (Figures 1 and 2, see detailed hypotheses and references in Supplementary Table S1). The scientific records for ancient rail extinctions (prior the 16th century) and their life-history traits are largely incomplete from the many hundreds or thousands of species estimated (Steadman, 1995; Curnutt and Pimm, 2001). Therefore, our study focuses on the extinctions that happened from the Era of Colonialism onwards.

Figure 1. Overview of analytical framework: this chart delineates the three distinct parts of our study – past extinction risk, contemporary vulnerability (IUCN status) and contemporary vulnerability (impact from threats). Each section outlines the response variables, the set of predictor variables used, the temporal and spatial scales of analysis and the primary hypotheses tested herein. The analysis uses boosted regression trees with sample sizes indicated for each part. Detailed justification for each hypothesis and trait selection, alongside associated references, can be found in Supplementary Table S1.

Figure 2. Diagram of different rails’ fate (extinction or persistence) over time and their use in the different parts of the analyses. We determined different pathways for rails extinctions: at first contact with humans during (i) the Pleistocene/Holocene or (ii) the Era of Colonialism and (iii) at second or subsequent contact with humans. “(Excluded)” means that the species have been excluded from the analysis and “(ignored)” that the species’ previous state is considered for the analysis of extinction risk. Figure made with BioRender (https://biorender.com/).

The study aims to investigate how the traits of rail species (body size, habitat diversity, migration behaviour, island endemism, island characteristics, flightlessness, naivety to humans and predators and socio-economic status of countries, Table 1) are associated with the following:

1. 1) Extinction or persistence on islands (as all extinctions previously occurred on islands)

2. 2) Contemporary IUCN status (threatened or not), globally and on islands

3. 3) Major threat categories (habitat loss, overhunting and introduced predators), globally and on islands.

Table 1. Explanatory variables used in extinction risk and vulnerability models for rails

Note: See Supplementary Table S1 for references and data sources.

* Human density: Population density is midyear (2017) population divided by land area in square kilometres.

GDP: GDP per capita (PPP) compares GDP on a purchasing power parity basis divided by population as of 1 July for the same year.

Human population growth: Population growth rate compares the average annual percent change in populations, resulting from a surplus (or deficit) of births over deaths and the balance of migrants entering and leaving a country. The rate may be positive or negative.

Methods

Database of rail traits and threats

We compiled information on life-history traits, biogeographic and socio-economic contexts that have been hypothesised to increase extinction risk in birds using the comprehensive Guide to the Rails, Crakes, Gallinules and Coots of the world (Taylor and van Perlo, 1998) and other external sources for information (Table 1; see Supplementary Table S1 for references and data sources). We constructed a database for the 124 species of extant rails (including 33 island endemic species) and 27 recently extinct species (25 officially extinct and 2 considered as extinct in this analysis only, Supplementary Table S2). We extracted their threatened status and impact from threats from the online 2019 version of the International Union for Conservation of Nature (IUCN) Red List database (IUCN, 2019; http://iucnredlist.org). Databases can be found in Appendix 3. Some variables could not be reliably obtained across species and were excluded from the global analysis (in Part 2 [global scale vulnerability with IUCN threatened status]: diet, clutch size and mating system). We considered island endemic species as those restricted to one (single‐island endemics) or a group of islands (multi‐island endemics).

Focal taxa included Rallidae (Gruiformes) and followed the IUCN classification. The IUCN still includes the Sarothruridae (flufftails) as members of this family. Species of both families have convergent body plans, leading to taxonomic confusion regarding their placement based on morphology alone (e.g., Livezey, 1998). However, genetic data recovered the sister relationship between flufftails and Heliornithidae (Hackett et al., 2008) and continues to reveal species of “rail” for example, Rallicula forbesi, as flufftails (Garcia-R et al., 2020). Therefore, we excluded known flufftails from our analyses (but see analyses and results including this family in Supplementary material).

Species considered ‘data deficient’ or that have not been recognised by the IUCN (e.g., due to a recent phylogenetic split) were excluded from the analyses (Supplementary Table S2). The metrics for all countries in which a species occurred were averaged. For human density, we used the density of the land where present (i.e., countries, or if present on an island only, would use human density for that island).

Here, we define ‘extinction risk’ as the likelihood of becoming extinct or not for a species, and ‘vulnerability’ as the likelihood of being considered threatened, according to the IUCN Red List (a threatened status being attributed for ‘vulnerable’, ‘endangered’ or ‘critically endangered’ status).

Classification of extinct and extant rail species

We determined different pathways for rail extinctions at either one of three critical time periods: (i) as naïve species experiencing their first contact with humans in the Pleistocene/Holocene, (ii) during the Era of Colonialism or (iii) at second contact with humans after surviving their initial contact during the Pleistocene/Holocene (‘savvy’, Figure 2). We defined ‘contact’ as a period of significant human impact (e.g., introduction of alien species) or settlement. Recent extinctions (e.g., 1970s) resulting from the lasting impacts of the Era of Colonialism (e.g., introduced predators) were analysed as part of the Era of Colonialism. Two species went extinct due to contemporary causes (Figure 2, ‘(iii) Extinct (ignored)’) and were therefore analysed retrospectively as ‘savvy’ during the Era of Colonialism (Supplementary Table S2). Too little information was available from ancient extinctions for a robust classification of life-history traits or extinction drivers; thus, extinctions from (i) were not analysed. We compiled the causes to island rails’ extinctions since the 16th century. To the best of our knowledge, no rail species has been described extinct nor fossil remains of extinct rails dating younger than Pleistocene were found on continental landmasses. The Era of Colonialism mostly concerned European settlers, but some extinctions in New Zealand islands were linked to the impact of Polynesian settlers.

Gallirallus lafresnayanus and Pareudiastes pacificus are two ‘critically endangered’ rail species that have not been seen with certainty since the 19th century and are suspected to be already extinct by some authors (IUCN, 2019); they were considered as extinct for the analyses (Supplementary Table S2). Hypotaenidia owstoni was considered as ‘extinct in the wild’ until 2019 (when its status changed for ‘critically endangered’; IUCN, 2019) and therefore was considered extinct for the analysis.

Statistical analysis

We implemented all modelling in R (version 3.6.3; R Core Team, 2021). We used boosted regression trees (BRTs) to investigate correlates of extinction risk and vulnerability globally and on islands. BRT is a powerful machine-learning approach recognised for its capacity to handle high-dimensional data, capture non-linear relationships implicitly and tolerate collinearity among predictors. BRTs distinctively learn the structure of data, helping to adaptively uncover complex patterns that can be overlooked by traditional methods. Simple Generalized Linear Models (GLM) were initially run (see Supplementary material); however, they fitted the data poorly, therefore BRTs were preferred. Different results between the GLM and BRT analyses could be observed (Supplementary material).

Unlike traditional linear models, BRTs do not require predictors to be orthogonal. This is due to their tree-based structure, where the algorithm selects variables for splitting based on their individual contributions to reducing prediction error, rather than their interrelationships. This process, combined with BRT’s ability to model non-linear relationships and complex interactions through an additive approach of combining multiple trees, substantially mitigates the impact of collinear variables. Consequently, BRTs can effectively handle high-dimensional and correlated data thereby ensuring the reliability of analyses in such cases. This robustness makes it particularly suitable for classification tasks (like whether rails went extinct or not) when faced with a multifaceted array of possible ecological, socio-economic and biological predictor variables. We note that while BRTs inherently accommodate hierarchical dependencies, they do not explicitly account for phylogenetic autocorrelation; however, they are robust even in the presence of incompletely independent data (Jones et al., 2006; Boyer, 2010). Model coefficients are shown as mean ± standard error. The variables ‘Island size’, ‘Human density’ and ‘GDP’ were log₁₀-transformed, and all continuous variables (including the ones log-transformed) were standardised using z-scores. Information on clutch size was scarce, and the data were missing for 44% of the species in the extant island endemic species, as well as for 93% of the extinct species. Therefore, we did not include the clutch size variable in the island models.

Part 1: Past extinction risk on islands

We investigated the role of naivety to humans, island size, flightlessness and body size as correlates of historical extinction risk for island rails during the Era of Colonialism. The model for past island extinctions took the form:

• • ISLAND_Extinctions ~ body size + flightlessness + island size + naivety to humans

Part 2: Contemporary vulnerability (IUCN threatened status)

For the correlates of contemporary vulnerability in rails, we tested how life-history traits (island endemism, island characteristics, body size, habitat diversity and migration behaviour) and socio-economic attributes of the countries where present (GDP, human population growth and human density), were associated with the likelihood of a rail species being threatened or not (Table 1). Models were run at two spatial scales: i) globally (all modern rail species) and ii) on islands (island endemic rails).

• • GLOBAL_Vulnerability-_IUCN ~ migratory behaviour + body size + clutch size + habitat diversity + island endemism + GDP + population growth + human density_countries

• • ISLAND_Vulnerability-_IUCN ~ body size + habitat diversity* + flightlessness* + island size + island isolation* + human density_islands

Where * indicates predictor variables that were excluded from the island model after preliminary results, to avoid overfitting with too many marginal predictors (their importance was below 0.4 in initial models). Remaining parameters were body size, island size, and human density.

We modelled vulnerability as a binomial dependent variable: ‘non-threatened’ species (category including ‘least concern’ and ‘near-threatened’ IUCN Red List status) versus ‘threatened’ species (‘vulnerable’, ‘endangered’ and ‘critically endangered’ status; IUCN, 2019). Four species that were classified as ‘threatened’ by the IUCN were done so on the basis of potential future threats. As this was outside the scope of this study, we attributed them with a (current) ‘non-threatened’ status for the analysis (Supplementary Table S2).

Part 3: Contemporary vulnerability (impact from threats: Habitat loss, overhunting and introduced sp.)

We assessed how habitat loss, overhunting and introduced species can disproportionally impact species with particular life-history traits, at both global and island scales. We used the ‘Threat Impact Scoring System (IUCN – CMP Unified Classification of Direct Threats, version 3.2)’ proposed by the IUCN (2019) (https://www.iucnredlist.org/resources/threat-classification-scheme) to extract whether a species was impacted by either habitat loss, overhunting or introduced species. Current and past impacts were included. Species suffering from habitat loss would be classified as such if their habitats were impacted through ‘1. Residential & Commercial Development’, ‘2. Agriculture’, ‘3. Energy Production & Mining’ or ‘7. Natural System Modifications’ under the IUCN categories of threats, encompassing all types of habitat alteration (following Green, 1996). Species were classified as suffering from introduced species and overhunting if they were recognised by the IUCN as suffering from a threat of, respectively, ‘8.1 Invasive non-native/alien species/diseases’ and ‘5.1 Hunting & collecting terrestrial animals’.

Models used for the threat of habitat loss, globally and on islands, respectively, took the forms:

• • GLOBAL_Vulnerability-_{HABITAT LOSS} ~ habitat diversity + artificial habitat + island endemic

• • ISLAND_Vulnerability-_{HABITAT LOSS} ~ habitat diversity + artificial habitat** + island size + flightlessness

Models used for the threat of overhunting, globally and on islands, respectively:

• • GLOBAL_Vulnerability-_OVERHUNTING ~ body size + clutch size + island endemism

• • ISLAND_Vulnerability-_OVERHUNTING ~ body size + island size + flightlessness

Models used for the threat of introduced species, globally and on islands, respectively:

• • GLOBAL_Vulnerability-_{INTRODUCED SPECIES} ~ body size + clutch size + island endemism

• • ISLAND_Vulnerability-_{INTRODUCED SPECIES} ~ body size + island size + flightlessness** + naivety to predators

Where ** indicates predictor variables that were excluded from the island models after preliminary analyses to avoid overfitting with too many marginal predictors (their importance was below 0.6 in initial models).

Modelling process

Step 1: Selection of the optimum parameters for the BRT models

For each of the aforementioned models, we performed a grid search to estimate which boosted regression model’s parameters would maximise the out-of-bag true skill statistic (TSS) score (see results in Table 2). TSS evaluates how well a predicted outcome can distinguish between positive and negative instances, taking into account all components of the confusion matrix (Allouche et al., 2006; Rahmati et al., 2019). The TSS score was calculated using the predict function (package caret, thresholded at 0.5). We used the gbm.step function (package dismo), using deviance as the loss function to estimate the optimum number of trees in order to avoid overfitting (Elith et al., 2008). Models were run using cross-validation with five folds and a maximum of 10,000 trees. The minimum number of observations per node (n.minobsinnode) was kept at 10 during the creation of the model but could decrease to 5 for predictions because of small datasets (Table 2).

Table 2. The combinations of the parameters and model settings used to derive the best model and the optimum number of boosted regression trees.

Note: N is the total number of species used in each analysis.

Step 2: Variable importance and fluctuations in the models

We used the gbm function (package gbm) with the optimum number of trees and parameters identified in step 1 (Table 3). Relationships between the response variable and the predictors were analysed by producing partial-dependence plots.

Table 3. Optimum parameters and model performance for the boosted regression trees.

Note: Results presented for Part 2 (islands) have a reduced number of predictors after an initial model selection. TSS is the out-of-bag true skill statistic score.

Results

Part 1: Past extinction risk

During the Era of Colonialism, there were 27% naïve species (18/67), and 73% savvy species (of which, 37% (25/67) had met and survived earlier contact/settlement with arriving sailors and 36% (24/67) had coevolved with indigenous people since the Pleistocene/Holocene (Figure 3). At least 62% of the rails were flightless (39 species), and of those, only 36% were naïve to humans. All extinctions took place on islands.

Figure 3. Global distribution of island rails since the Era of Colonialism (i.e., 16th century onwards). Symbols illustrate their fate of extinction (cross: extinct, vertical cross: extinct for contemporary reasons (i.e., extant in the analyses), plain: extant). Colours illustrate rails’ state of naivety to humans at the time of contact (blue: naïve, pink: not naïve). The Inaccessible Rail (Atlantisia rogersi) was excluded since it did not have a substantial contact with humans. Projection information: WGS84, centred on 150°E.

Body size was the most influential predictor in the extinction risk model (explaining 90% of the relative influence; TSS = 0.5, n=67, Table 3, Figure 4), with smaller (⩽24 cm) and larger body sizes (⩾35 cm) being the most extinction-prone. Flightless rails were more prone to extinction than flying ones but the relative importance of flightlessness in the model was only 6% (Figure 4). Small island size and naivety had very little relative influence in the model (Figure 4, 14 species naïve species went extinct).

Figure 4. Extinction risk in island rails: the relative influence (left) and partial dependence plots (right) of predictor variables for the boosted regression tree model on extinction risk. Y is the probability of becoming extinct. All continuous variables were standardised using z-scores.

Part 2: Contemporary vulnerability (IUCN status)

Globally, 23% of all extant rails are currently considered as threatened. The body size of modern rails ranges from 12 to 63 cm (mean 27.4 ± 10.1 SD). On average, they live in two to three habitats per species globally (2.6 ± 1.5 SD) and in 2.4 different habitats (± 1.2 SD) on islands. Then, 29% of all rail species are island endemic, and of these, 51% are threatened. Of the island endemic rails, 46% are flightless and of these, 47% are threatened. Then, 50% of the island flying rails are threatened.

Globally

Human density was the main predictor of contemporary global rail vulnerability, with >36% relative variable importance (Figure 5). The relationship between rail vulnerability and human density was complex, with little consistency in its direction of influence, although the overall trend was lower impacts at higher human densities. This result is likely to be indicative of a random effect arising from differences between countries (see Discussion). Small clutch size (≤3 eggs), being an island endemic, and having a smaller body size (<28 cm) were all also associated with being more at risk of being threatened (Figure 5).

Figure 5. Global vulnerability: the relative influence (left) and partial dependence plots for the four main predictor variables (right) of the boosted regression tree model on rails’ global vulnerability. See Supplementary Figure S1 for partial dependence plots for all predictor variables. Y is the probability of being threatened. All continuous variables were standardised using z-scores.

On islands

Our island model had a lower predictive performance (TSS=0.28, Table 3), suggesting relationships with predictor variables were weak. Human density was the only predictor to influence the model (100% relative influence).

Part 3: Contemporary vulnerability (impact from threats)

Globally

We analysed the impact of the three main anthropogenic threats (habitat loss, overhunting and introduced species) to rail vulnerability. Habitat loss was the most common anthropogenic threat associated with rail vulnerability (Table 4), but this was not predicted well by any life-history (intrinsic) traits we examined (TSS = 0.21, Table 3). We found that vulnerability to overhunting and the threat of introduced species were both predicted by island endemism (Figure 6, Supplementary Figure S2). Overhunting was also predicted by clutch size (most strongly, accounting for over 40% of the model influence, with rails with clutches smaller than three eggs or larger than eight eggs being the most vulnerable to overhunting), and increasing body size (Figure 6). To the threat of introduced species, large-bodied rails were also more vulnerable (Supplementary Figure S2).

Table 4. Proportion of rail species impacted by the three main threatening processes currently or in the recent past

Figure 6. Global vulnerability to overhunting: relative influence (left) and partial dependence plots (right) of predictor variables for the boosted regression tree model on rails’ vulnerability to overhunting globally. Y is the probability of being threatened by overhunting. All continuous variables were standardised using z-scores.

On islands

The proportion of species impacted by habitat loss, overhunting and introduced species was higher on islands than globally (Table 4). While there was no pattern found for habitat loss globally, the signal on islands was much stronger (TSS=0.88; Table 3). Island size was the main factor leading to increased vulnerability from habitat loss on islands, with species living on islands between 4,000 km² and 163,790 km² (approximately log₁₀ 3.6–5.2 area units) having up to 96% chance of being impacted by habitat loss (Supplementary Figure S3; 33% of the island rails). Smaller habitat diversity also increases the risk of habitat loss.

Island size was also the most important factor for predicting overhunting, with rails on larger islands having more chance to being over-hunted. Large rails and flightless rails were also more likely to be over-hunted (Supplementary Figure S4).

Vulnerability to introduced predators was predominantly explained by the naivety to mammal predators (i.e., absence of native mammals; Supplementary Figure S5). Island size and body size and had little influence on the vulnerability to introduced species (Supplementary Figure S5).

Discussion

Part 1. Extinction risk

As hypothesised (Figure 1, Supplementary Table S1), our study revealed that rail extinction events did not occur at random, and that during the Era of Colonialism the extinction-filter targeted island endemic rails only. While we posited that naïve, flightless and large-bodied rails would be more likely to go extinct, we found that both smaller and larger rails were more likely to go extinct during human occupation. Flightlessness or naivety were not found to play a role in these extinctions during this period. The other predictors displayed complex or no clear relationships to extinction vulnerability, as explored in detail below.

Holocene extinctions

Holocene extinctions (not modelled explicitly herein) help in understanding the context of extinctions happening during the Era of Colonialism (focus of this study). During the first human-initiated extinction wave during the Holocene, all extinct island birds were naturally naïve to people and most of them had no anti-predatory behaviour as the oceanic islands they evolved on had no mammalian predators. Local studies on island-groups (e.g., Hawaii, New Zealand and Pacific islands) found that flightlessness and body size were the main extinction drivers during this first extinction wave, both for rails (Steadman, 1995; Curnutt and Pimm, 2001; Duncan et al., 2013; Alcover et al., 2015) and other birds (Boyer, 2010; Sayol et al., 2020). Nevertheless, focusing on the case of New Zealand’s birds, Duncan et al. (2002) concluded that flightlessness was not an important factor during the first bird extinction wave (whereas body size and naivety were), because naïve birds were easy to hunt regardless of their flight ability, since they lacked anti-predator behaviour.

Recent extinctions

Body size

In the rails’ second extinction wave (during the Era of Colonialism, the focus of this study), we found that body size was the key factor of extinction risk (with species <24 cm and >35 cm being most prone to extinction). The observation that intermediate-sized rails were most resilient is challenging to explain but might be an example of a ‘critical weight range’ that was large enough to avoid predation by rodents (at small sizes) and small enough to avoid being targeted by people (at large sizes), or having a sufficiently high reproductive rate (which allometrically, is correlated negatively with body size) to compensate for increased depredation. Indeed, studies have previously found that while the extinction-filter at first contact with humans drove large naïve birds extinct, the effects of body size on the surviving birds were more complex during the next extinction wave: studies mostly found either no effect (Bromham et al., 2012) or non-linear and complex effects (Boyer, 2008, 2010). In cases where body size played a role (e.g., Hawaii), two pathways to extinction were proposed (Boyer, 2010): smaller species became extinct due to the predation of their eggs or chicks by small introduced predators such as rodents (Holdaway, 1999), while larger species went extinct due to the predation by larger introduced predators (e.g., dog, pig) and human hunting, which mostly targeted larger prey. Indeed, in support of this general hypothesis, 62% of the recent rail extinctions were linked to overhunting, and to introduced predators for 69%, including rats for 62% of them (Lévêque et al., 2021).

Naivety to humans

Most studies working on modern extinctions (Era of Colonialism) compare small island groups through the different waves of extinctions (e.g., Boyer, 2008, 2010; Bromham et al., 2012), and consequently do not account for the role of naivety to humans – excluding places with long-term cohabitation with people (e.g., Indonesia, Madagascar) or remote human-free islands (e.g., Tristan da Cunha, Galápagos Islands and Mascarene Islands). We found that rails that had survived initial contact or coevolved with humans were not less vulnerable to the second wave of extinction than naïve ones, indicating that previous human colonisation had not pre-selected resilient rail species like some have suggested for birds (Biber, 2002).

One explanation is that the second wave of extinction might have operated via different mechanisms to the first (Holocene), for which savvy rails did not have adequate defences. For example, a new wave of introduced predators (e.g., dog, pig and cat) exerted different predatory pressures to humans and rats, and the introduction of new competitor species (e.g., goat) and diseases (Milberg and Tyrberg, 1993; Loehle and Eschenbach, 2012) that can contribute to the accelerated demise of endemic species (Wood et al., 2017; Kouvari and van der Geer, 2018). Moreover, each island had different types of contact with people (varying in intensity, duration, type of settlement, introduction of alien species etc.) which could influence the species’ responses (Wood et al., 2017), whose impacts could be buffered by the local biodiversity composition. More research on identifying the intrinsic mechanisms that determine resistance at first contact for island endemics would provide interesting insights on the mechanisms of extinctions.

Parts 2 and 3: Contemporary pattern of vulnerability

Our hypotheses were that sedentary, habitat specialist, island endemic, large-bodied, slow breeders and flightless species would be more threatened. Developing or human-dense countries, and small, isolated and human-dense islands would support more threatened species (IUCN status). We found that human density, small clutch size, island endemism and small body size were the four main predictors for contemporary global rail vulnerability. Human density was the main predictor; however, the signal was complex and likely to play a random effect in the model. We did not find a pattern of overall vulnerability for contemporary island endemic rails.

For the threats of habitat loss, overhunting and introduced predators, we hypothesised that introduced predators and overhunting would have a greater impact on large, island endemic and flightless species, as well as species living on small islands, naïve to mammal predators and with smaller clutch size. Habitat loss was hypothesised to impact more habitat specialists, island endemic and flightless species, as well as species living on small islands and unable to exploit disturbed habitats. We found that vulnerability to overhunting and introduced species was predicted by island endemism. Overhunting was also predicted by both smaller and larger clutch size, and larger body size. As to the threat of introduced species, large-bodied rails were also more vulnerable. Habitat loss was not predicted by any life-history trait we examined. On islands, large island size was a factor of vulnerability to overhunting and habitat loss. Large and flightless rails were also more likely to be over-hunted. Vulnerability to introduced predators was predominantly explained by the naivety to mammal predators while island size and body size had little influence.

As most bird extinctions and all rail extinctions occurred on islands, our models found that island endemic rails comprised most of the threatened rail species. This also held true when considering island endemism as a predictor of vulnerability to introduced predators and overhunting. Island endemic birds are globally threatened, and this trend is increasing – they represented 39% of all threatened birds in 1990 (Johnson and Stattersfield, 1990), and 47% 20 years later (BirdLife International, 2017). In Rallidae, 67% of threatened species are island endemic (Lévêque et al., 2021).

Owens and Bennett (2000) compared different bird lineages and found that rails had various routes to vulnerability, from habitat loss, human persecution (overhunting) and introduced predators, because they were typically more ecologically specialised, with larger bodies and slower breeding rates compared with other families. Our analyses showed that some of these life-history traits were indeed increasing rails’ vulnerability to these external threats. Globally, clutch size and body size had little influence on vulnerability to introduced species, but rail species with smaller and larger clutch sizes and larger bodies were more vulnerable to overhunting. For island endemics, large-bodied and flightless species, as well as those living on large islands, were more threatened by overhunting, while naivety to predators was the main factor to the threat of introduced predators. Interestingly, while smaller rails are at more risk to be threatened globally, we found that larger rails are at more risk to the threat of overhunting (both globally and at the island scale), validating our initial hypotheses. This pattern is also found overall in birds (Ripple et al., 2017). This suggests overall that body size is a complex driver of vulnerability that can also interact with other ecological traits in ways that are not yet fully understood (Bennett and Owens, 1997; Boyer, 2010).

Similarly, Bennett et al. (2001) have suggested that smaller-sized birds, when specialised and fast breeding like rails, would be more likely to be threatened largely due to habitat loss. However, looking broadly within the rail family, we did not find any link between small body size and any other intrinsic trait to the threat of habitat loss. Nevertheless, the main IUCN justification for the threatened status of small-bodied rails in this study (nine species) was that they generally had a very restricted range, often in wetlands that were fragmented and with ongoing loss. Most of these species were continental, found in South America and Asia, and half were from the genus Laterallus. While no traits were found to increase rail vulnerability to habitat loss globally, at the island scale, rails living on large islands were found to be more at risk of becoming threatened due to habitat loss, and for species with smaller habitat diversity.

Clutch size and body size

We also found that at the global scale, rails with small clutches (≤3 eggs) and smaller bodies (<20 cm) were also the most likely to have a threatened IUCN status (i.e., “VU”, “EN” or “CR”). Small clutch size, a measure of slow reproductive rate, is known to increase vulnerability due to the slow regeneration of population and validates our hypotheses at the global scale (Bennett and Owens, 1997; Lee and Jetz, 2011; Garcia-R and Di Marco, 2020). Interestingly, rail species with either larger clutches (>8 eggs) or clutches with fewer eggs (<3 eggs) are the most vulnerable to the threat of overhunting. This suggests that species that produce more eggs tend to be targeted by overhunting because of the abundance of eggs they provide, while slow-breeding species would struggle to recover from exploitation (Owens and Bennett, 2000).

The effect of small body size in the vulnerability of rails contradicts most studies on birds, where threatened species were mostly large-bodied (Gaston and Blackburn, 1995; Bennett and Owens, 1997; Lee and Jetz, 2011; Ripple et al., 2017), while some studies report no relationship (Morrow and Pitcher, 2003; Chichorro et al., 2019). As the last extinction wave targeted large or small (but not medium-sized) rails, we can suppose that all large-bodied vulnerable rails went extinct, while there are still vulnerable smaller-bodied rails that are still facing the ongoing threat of extinction. With the new diversity of threats impacting rails (Lévêque et al., 2021), different processes can now affect rails. For example, Ripple et al. (2017) found that the lightest-bodied birds were mostly impacted by agriculture, and agriculture is one of the three predominant threats to the rails globally (Lévêque et al., 2021).

Socio-economic status of countries

Human density was the main predictor of global vulnerability for rails, but the complex and apparently counterintuitive relationship (overall lower vulnerability of rails with higher human density) makes it challenging to interpret the exact role. Our interpretation is that human density is acting like a random effect in the decision tree (i.e., it captures a variety of unmeasured tapering effects associated with the idiosyncratic environment and history of any given island), rather than being an actual predictor. However, it could not be modelled formally as a random effect because there was almost always only one rail species per island.

High GDP of the countries where the species inhabit also played a role in increasing the threatening processes, most likely through the encroachment of natural habitats and direct exploitation due to economic development (Czech, 2000). This pattern has also been found in parrots (Olah et al., 2016). Human population growth was another socio-economic attribute that was found as influencing vulnerability, but to a much lesser extent. The predictor’s negative relationship contradicted our original hypothesis that high human population growth would lead species to be more vulnerable. We suspect this to be linked to some species’ particular occurrence: human population growth is particularly high for species occurring in African countries, where rails have high habitat diversity and have long coevolved with high rates of human pressure. Alternatively, the hunting or poaching of predators in those regions could also act to ‘release’ rail populations from natural mortality pressures.

Results from the literature are variable, and while Davies et al. (2006) found a negative effect of GDP on vulnerability of birds globally, indicating that areas of high economic development support fewer threatened bird species, McKee et al. (2013) found that overall higher human density and GDP is linked to an increase in the number of threatened species. Our conflicting results with socio-economic predictors (GDP, human population growth and human density) suggest that the relationship between human development and threatening processes in rails is complex and would require more research to disentangle.

Vulnerability on islands

Interestingly, we did not find a pattern of overall vulnerability for contemporary island endemic rails. To explain the randomness observed, we propose different pathways for vulnerability. Most threatened rails, being island endemics, often take the same trajectory as extinct species, suffering mostly from invasive predators (Lévêque et al., 2021) and also overhunting. As extinction-filters selected savvy rails throughout the different extinction waves, different island traits could contribute to the elevated vulnerability, such as small ranges and population sizes, low genetic diversity, inbreeding and so forth (Frankham, 1998; Purvis et al., 2000; Frankham, 2005). Furthermore, as the breadth of anthropogenic threats has intensified and become more complex, threatened island rails are impacted by more threats than previously (four threat types on average and up to nine different threats, e.g., mining, recreational activities, dam management, Lévêque et al., 2021). The consequence of this was to either create new and intricate pathways to vulnerability or result in a contemporary ‘field of bullets’ (Raup, 1991) where intense and large-scale disturbances make vulnerability unpredictable (largely stochastic or happenchance). Similar results have been found in other taxa (Duncan and Lockwood, 2001). The increased diversity of impacts leading to habitat loss on islands might also create extinction debts on islands (Triantis et al., 2010; Otto et al., 2017).

On islands – Flightlessness and predator naivety

While flightlessness was strongly associated with ancient extinctions (during the first wave in the mid-Holocene), it diminished in importance for the more recent extinction wave (Era of Colonialism). Following this continuity, flightless rails are no more threatened than flying ones in contemporary times. This highlights the role of humans in the selection of resilient species. Vulnerable flightless rails were initially easy prey to the first human arrivals on their islands, with those flightless species surviving that first wave being resilient in other ways (e.g., behaviourally adaptive or preferring habitats like wetlands that were less accessible). A global study on the roles of flightlessness and naivety for bird extinctions (beyond just Rallidae), via different extinction waves at a global scale, would help shed light on the drivers of the more recent extinctions. With only 19 flightless rails remaining from the many hundreds (or perhaps thousands) that existed in the Holocene (Steadman, 1995; Curnutt and Pimm, 2001), there is a high conservation imperative to protect these iconic birds.

While flightlessness was not a predictor of overall vulnerability on islands, we did find that flightless rails were more vulnerable to overhunting specifically. Flightless bird species tend to be more threatened than volant species, and this is generally associated with predator naivety (Duncan et al., 2002; Steadman, 2006; Boyer, 2008). This is consistent with our findings where we found that rails, whether flying or flightless, were more acutely impacted by introduced predators when naïve and lacking anti-predator behaviours. Although this is not a surprising result and has been suggested in the literature (Balmford, 1996), little empirical work has been done on this. The role of naivety for island birds to withstand systematic threat from introduced predators and to other threats (e.g., overhunting) is an area of research that deserves more attention.

Island size

Contrary to our expectations, large islands had more rails threatened by overhunting and habitat loss than smaller islands. One explanation might be that large islands have the potential to be more similar to continents in terms of threats, and some threat types would be absent from smaller islands (Manne et al., 1999). For example, rates of habitat loss and fragmentation could be far greater on large islands (Didham et al., 2005), and it might be that only larger islands have sufficient resources to support introduced predators and competitors. However, this could also be simply a result of a (pre)historical selection bias, wherein extinctions have already wiped-out rails from the majority of small islands, due for instance to smaller maximum population sizes (see Green, 1996 for similar results with Anatidae).

Conclusion

Extinction events in rails were not random, with some life-history traits being selected through human-driven extinction filters. However, the pattern of contemporary vulnerability differs from past extinctions. Threatened species today are represented mostly by slow-reproducing and island endemics, while the pattern of vulnerability is essentially random on islands. Returning to our original question ‘can understanding past avian extinctions help to forecast extinction risk in birds?’, it seems to be a quixotic goal to draw conclusions from past extinctions to predict future extinctions, at least for rails. Contemporary species have survived intense extinction filters that were driven by a few threats only, but the increase in the diversity and intensity of modern anthropogenic threats is likely to be responsible for the change in pattern. Beyond endemism, there is little commonality in which persisting rails are now vulnerable, although the roles of genetic stochasticity (Evans and Sheldon, 2008; Sarre and Georges, 2009) and extinction debt due to past habitat loss and modification, warrant further investigation. Overall, islands appear to be the most sensitive unit of conservation for rails, and preserving the remaining island endemic and flightless species is clearly a high conservation priority, which would also greatly benefit most island wildlife (Graham et al., 2017), including the 69 subspecies of rails (47%) that are island endemic. Protecting islands also opens the possibility for rails to naturally, or via human assistance, recolonise islands on which they became extinct (Curnutt and Pimm, 2001; Morris et al., 2021).