The use of range size to assess risks to biodiversity from stochastic threats
Abstract
Aim
Stochastic threats such as disease outbreak, pollution events, fire, tsunami and drought can cause rapid species extinction and ecosystem collapse. The ability of a species or ecosystem to persist after a stochastic threat is strongly related to the extent and spatial pattern of its geographical distribution. Consequently, protocols for assessing risks to biodiversity typically include geographic range size criteria for assessing risks from stochastic threats. However, owing in part to the rarity of such events in nature, the metrics for assessing risk categories have never been tested. In this study, we investigate the performance of alternative range size metrics, including the two most widely used, extent of occurrence (EOO) and area of occupancy (AOO), as predictors of ecosystem collapse in landscapes subject to stochastic threats.
Methods
We developed a spatially explicit stochastic simulation model to investigate the impacts of four threat types on a dataset of 1350 simulated geographic distributions of varying pattern and size. We empirically estimated collapse probability in response to each threat type and evaluated the ability of a set of spatial predictors to predict risk.
Results
The probability of ecosystem collapse increased rapidly as range size declined. While AOO and EOO were the most important predictors of collapse risk for the three spatially explicit threats included in our model (circle, swipe and cluster), core area, patch density and mean patch size were better predictors for edge effect threats.
Main conclusions
Our study is the first to quantitatively assess the range size metrics employed in biodiversity risk assessment protocols. We show that the current methods for measuring range size are the best spatial metrics for estimating risks from stochastic threats. Our simulation framework delivers an objective assessment of the performance of hitherto untested but widely used measures of geographic range size for risk assessment.
Introduction
Spatially explicit stochastic threats such as extreme drought (Allen et al., 2010; Sheffield et al., 2012), disease outbreak (Fisher et al., 2012), coral bleaching (Hoegh‐Guldberg et al., 2007), fires (Stephens et al., 2013; Nowacki & Abrams, 2015), invasion of alien species (Blackburn et al., 2014), rapid land transformation (Vitousek et al., 1997) and tsunami (Cochard et al., 2008) can cause rapid species extinction or ecosystem collapse. Such threats typically produce large environmental perturbations over short time frames and trigger sudden reductions in species’ populations, suitable habitat or both (Lande, 1993). Similarly, stochastic threats can influence the ability of ecosystems to support component biota and sustain the processes that define them, primarily by reducing habitat area, increasing isolation, introducing novel ecological boundaries, facilitating invasions and reducing niche diversity (Ewers & Didham, 2006; Keith et al., 2013).
The capacity for species and ecosystems to persist after the occurrence of a spatially explicit threat is strongly related to properties of their geographical distribution (Gaston & Fuller, 2009; Keith et al., 2013). In general, widely distributed organisms or ecosystems that occur in several patches or across large geographic ranges are at lower risk from stochastic threats than those with highly concentrated distributions (IUCN, 2014; Bland et al., 2016). As a consequence, risk assessment protocols for both species and ecosystems incorporate range size criteria to evaluate risks from spatially explicit threats (Keith et al., 2004; Gaston & Fuller, 2009; Waples et al., 2013; Nicholson et al., 2015). For example, the IUCN Red List of Threatened Species and the IUCN Red List of Ecosystems assess species and ecosystems, respectively, against fixed thresholds of geographic range size (Mace et al., 2008; Keith et al., 2013; Rodríguez et al., 2015). The primary role of range size criteria in both Red List protocols is to identify species or ecosystems whose distribution is so restricted that they are at risk of extinction or collapse (the analogue of species extinction) from one or several threatening events (Rodríguez et al., 2015).
Extent of occurrence (EOO) and area of occupancy (AOO) are two standardized measures of geographic range size used in both IUCN Red List criteria (Gaston, 1991). EOO is measured by a minimum convex polygon encompassing all known occurrences of a species or ecosystem (IUCN, 2014; Bland et al., 2016). AOO is an index related to total occupied area, estimated by summing the areas of grid cells of a standard size (~2 × 2 km for species and 10 × 10 km for ecosystems; IUCN, 2014; Bland et al., 2016). Neither AOO nor EOO is intended to be a precise estimate of total area in which a species or ecosystem occurs (Gaston & Fuller, 2009). Rather, in risk assessment protocols such as the IUCN Red Lists, they function as standardized, complementary and widely applicable measures of risk spreading against spatially explicit threats (Gaston, 1991; Hartley & Kunin, 2003; Gaston & Fuller, 2009; Keith et al., 2013).
Surprisingly, the thresholds and assumptions implicit in the geographic range size criteria have not been tested for either of the IUCN Red Lists (Mace et al., 2008; Keith et al., 2013), or any other risk assessment protocols that we are aware of (e.g. Berg et al., 2014; Kontula & Raunio, 2009). Indeed, the range size thresholds that delineate categories of risk in the IUCN Red List of Threatened Species were largely developed iteratively through trial applications of arbitrary values on only a few species (Mace et al., 2008). More than 58% of the 23,250 species listed as threatened on the IUCN Red List of Threatened Species were assessed against geographic range size criteria (criteria B1 or B2; IUCN, 2015), and an improved understanding of the relationship between range size metrics and risks of species extinction or ecosystem collapse is essential. Flawed range size measures could lead to systematic underestimates or overestimates of risks to biodiversity that could ultimately result in inappropriate allocation of conservation resources.
In this study, our aim was to examine the performance of the widely used geographic range size metrics, EOO and AOO, as predictors of ecosystem collapse in landscapes or seascapes subject to stochastic threats. We developed a spatially explicit simulation model and applied it to geographic distributions exposed to three stochastic and one non‐stochastic threat type. The distributions were randomly generated from factorial combinations of extent and dispersion to represent a wide range of extant distribution patterns. The threat types were parameterized from observed events which have transformed a range of ecosystems over the past century, allowing empirical estimation of collapse probabilities under the different threat treatments. We then investigated how the risk of ecosystem collapse varies with different types of threat and different distribution patterns, and evaluated the ability of the spatial metrics to predict ecosystem types at different levels of risk. Our simulation approach and findings are equally valid for species, but for simplicity, we refer to all geographic distributions as ecosystem distributions throughout this paper.
Methods
We formulated a grid‐based stochastic simulation model to investigate the probability that an ecosystem will collapse when confronted with stochastic threats. Our model was implemented in three stages (Fig. 1):
- Simulate a test set of ecosystem distributions of varying pattern and size to represent a wide range of historical and extant marine, freshwater and terrestrial ecosystem distributions;
- Apply simulated stochastic threats, parameterized by empirical observations of threats to ecosystems over the past century, to each ecosystem distribution; and
- Estimate the probability of ecosystem collapse as the proportion of simulation replicates that reduced the area of an ecosystem to zero.
Ecosystem distributions
We developed a test dataset of 1350 simulated ecosystems to represent a wide range of distribution extents (size) and patterns (e.g. linear, clustered and dispersed). Ecosystem distributions were simulated with the mid‐point displacement algorithm using the NLMpy Python package (Etherington et al., 2015). In each implementation, we randomly specified the level of autocorrelation (between 0.4 and 1) and number of classes (between 2 and 100), before extracting a single class to form a binary matrix representing the presence and absence of an ecosystem type (Etherington et al., 2015). The resulting dataset of distributions ranged from widely dispersed to fully contiguous (Fig. S1), representing ecosystems as diverse as single desert springs or oceanic islands to large, intact forest, reef or desert ecosystems (Figs S1 & S2).
All distributions were simulated within a landscape extent of 500 × 500 km, and we fixed the grain size (pixel size) of the analysis to 100 × 100 m, comparable to many existing ecosystem maps derived from field data, remote sensing or habitat suitability models (Fig. S3). To generate a range of distribution shapes and sizes, we varied landscape X and Y dimensions independently and randomly between 1 and 500 km. Our final dataset comprised a variety of clustered, dispersed and linear distributions with a mapped area ranging from 1.2 to 42,150 km2 (Table 1). The EOO and AOO values of these distributions, calculated in accordance with IUCN guidelines (Bland et al., 2016), spanned the thresholds for all threatened and non‐threatened categories of the IUCN Red List of Ecosystems (Fig. S2).
| Variable | Description | Mean and range (n = 1350) |
|---|---|---|
| Derived variables | ||
| Probability of collapse | The proportion of simulations that resulted in ecosystem collapse | 0.090, 0–0.585 |
| IUCN Red List of Ecosystems criterion B | ||
| Extent of occurrence | The area of a minimum convex polygon encompassing all occurrences of the ecosystem (km2) | 16210, 3.87–152,100 |
| Area of occupancy | The number of 10 × 10 km grid cells area occupied by > 1 km2 of the ecosystem | 56.7, 0–576 |
| Distribution metrics | ||
| Largest patch index | The area of the largest patch divided by total landscape area | 0.613, 0.0147–1 |
| Mean patch area | Average area of all patches of the ecosystem (km2) | 9.789, 0.029–707 |
| Mean perimeter area ratio | The mean ratio of patch area to patch perimeter | 0.032, 0.001–0.035 |
| Mean shape index | Global mean of patch perimeter divided by the square root of patch area, adjusted by a constant | 1.322, 1.118–3.057 |
| Minimum patch area | Area of smallest patch | 0.283, 0.01–280.5 |
| Number of patches | Total number of patches | 899.7, 1–18,530 |
| Patch density | Number of patches divided by the total ecosystem area | 1.07 × 10−6, 1.41 × 10−9–3.45 × 10−6 |
| Proportion landscape core | Total ecosystem area divided by core area | 0.641, 0–0.991 |
| Splitting index | A measure of habitat fragmentation; total ecosystem area divided by the sum of patch area squared, summed across all patches | 5.538, 1–391.6 |
| Total area | Total area of ecosystem distribution (km2) | 1997, 1.17–42,150 |
Simulating stochastic threats
We formulated four threat types to represent a wide range of ‘threat syndromes’ (Burgman et al., 2007). To simplify the model, each application of a threat resulted in the complete transformation of that part of the ecosystem distribution covered by the threat footprint (Fig. 2; Appendix S1 in Supporting Information). This assumption is realistic for the most severe types of threat observable in real landscapes and seascapes. To incorporate stochasticity into the model, spatial footprints of threats were applied at random locations within the mapped ecosystem distribution and were allowed to vary in size within the specified ranges detailed below.
For the first threat type (circle), we applied a single circular disturbance representing contiguous threats such as airbursts, volcanic eruptions, coral bleaching events, disease outbreaks, deforestation events, contiguous fires and mining operations (Fig. 2b). In each simulation replicate, an integer (1–18 km) was randomly sampled from a uniform distribution to set the disturbance radius, with the maximum corresponding to the estimated size of the 1908 meteorite strike in the Tunguska River region of Siberia, which felled trees over ~1000 km2 (Chapman & Morrison, 1994). The location of the centre of the circular disturbance was specified by randomly sampling coordinates within the mapped ecosystem distribution.
For the second threat type (cluster), up to 20 circular disturbances, each with maximum possible radius of 20 km, were applied in each simulation replicate to represent multiple instances of threats such as forest clearing, logging coupes, mining operations, gas wells, patchy fires or dispersed outbreaks of pests or disease (Fig. 2c). The total size of this threat was limited to 1250 km2, corresponding to the estimated area of unregulated gold mining recently discovered within the Madre de Dios region of the Peruvian Amazon (Asner et al., 2013). The location of each circle in the cluster simulation was specified according to the methods described for the circle threat.
The third threat type (swipe) represented threats such as tsunami, urban expansion, desertification, an invasion process, lava flows and similar catastrophes (Fig. 2d). We parameterized the swipe threat from empirical observations of the impact of the Indian Ocean tsunami of December 2004, the largest tsunami in recorded human history, which resulted in the destruction of many coastal ecosystems across Southeast Asia (Cochard et al., 2008). The tsunami had a maximum flow depth of 32.5 m, which would travel about 24.8 km in ecosystems that present no resistance (Cochard et al., 2008). However, all ecosystems present resistance so, in agreement with observations of the impact of the 1883 Krakatau tsunami on coastal ecosystems, we randomly sampled an integer up to a maximum of 12 km from the edge of the distribution (1–12 km; Cochard et al., 2008).
The final threat type (edge effect) contrasts with three stochastic threat types described above. This threat reduced the ecosystem distribution uniformly from exposed edges, representing the growing impact of agriculture, urbanization, wetland decline, degrading effects of adjacent land uses, species invasion or uniform clearing across all edges of an ecosystem distribution (Fig. 2a). At each time step of the simulation, 100 m of an ecosystem was removed around its full perimeter. The number of time steps in each simulation replicate was a randomly sampled integer (1–20), allowing a maximum incursion of 2 km in any single replicate. According to a review of forest edge impacts, 99% of documented edge impacts in tropical and temperate forests were expressed within the first 2 km from the forest edge (Broadbent et al., 2008).
The simulation model was implemented in R (version 3.2.2; R Core Team, 2013) with the use of packages ‘raster’ (Hijmans, 2015), ‘sp’ (Pebesma & Bivand, 2005) and ‘rgeos’ (Bivand & Rundel, 2015).
Data analysis
We used boosted regression trees (BRT) to assess EOO, AOO and other spatial metrics of ecosystem distributions as predictors of the ecosystem collapse. BRT is a machine learning method widely used in ecological analyses, primarily owing to its flexibility in handling large, diverse datasets with many explanatory variables and complex relationships between predictors and response variables (Elith et al., 2008; James et al., 2013). Many studies have used BRTs as a modelling approach for investigating the underlying drivers of extinction risk in species, principally as a classification problem (with threatened status as the discrete response variable; Sullivan et al., 2006; Murray et al., 2011; Davidson et al., 2012; Bland et al., 2015), and to investigate the drivers of continuous responses such as the relative increase in extinction risk (Pearson et al., 2014).
We used probability of ecosystem collapse as the BRT response variable, calculated as the proportion of simulations in which the mapped area of an ecosystem was reduced to zero, based on 500 simulations of each threat type for each of the 1350 ecosystem distributions (2000 simulations per ecosystem). We first established the initial state (i.e. pre‐threat) EOO and AOO of each ecosystem (Fig. 1). To estimate EOO, we determined the area of a minimum convex polygon enclosing the entire ecosystem distribution (Bland et al., 2016). For AOO, we counted the number of 10 × 10 km grid squares that were occupied by > 1 km2 of the ecosystem (Bland et al., 2016). Secondly, we tested a set of supplementary spatial metrics, including the number of patches, core area and mean patch size, drawn from the ‘ClassStat’ function in R package ‘SDMTools’ (VanDerWal et al., 2014). From an initial set of 38 spatial metrics, we removed those that were highly correlated (Pearson's |r| > 0.7; Dormann et al., 2012), choosing to retain those which we judged most interpretable. We also removed metrics with near zero variance (Kuhn, 2008). This yielded a final set of 10 covariates for the analysis (Table 1; Fig. S4). However, in our dataset, AOO was collinear with EOO (|r| = 0.80) and the proportion of the ecosystem buffered by ≥ 100 m (proportion landscape core; |r| = 0.82; Fig. S4). BRTs tend to arbitrarily select just one variable when offered two or more highly correlated variables, which can lead to lower perceived importance of the other correlated variables (Cutler et al., 2007; Kuhn, 2008). Therefore, to allow assessments of both AOO and EOO in our analysis, we ran two separate models: one including EOO and proportion landscape core, and the other including AOO. Covariates were log‐transformed where required and standardized by subtracting the mean and dividing by the standard deviation.
As the threat type was an intrinsic process that was expected to influence collapse probabilities, the dataset was subset by threat type for four independent analyses. We partitioned each dataset into training sets (50%) and testing sets (50%), and selected the number of trees using 10‐fold cross‐validation (Table S1; Ridgeway, 2015). After running each BRT analysis, we assessed the importance of each covariate in predicting collapse probability using standard variable importance methods employed in BRTs (Friedman, 2001; Elith et al., 2008). We used the ‘caret’ and ‘gbm’ R packages to run the BRTs (Kuhn, 2008; Ridgeway, 2015).
To investigate collapse probabilities in relation to risk categories, we also assessed the initial status of each ecosystem under criteria B1 (EOO) and B2 (AOO) of the IUCN Red List of Ecosystems Categories and Criteria. For criterion B1, the EOO thresholds are critically endangered (CR) ≤ 2000 km2; endangered (EN) ≤ 20,000 km2; vulnerable (VU) ≤ 50,000 km2; and near threatened or least concern (NT or LC) > 50,000 km2); and for B2, the AOO thresholds are CR ≤ 2 10 × 10 km grid cells; EN ≤ 20 10 × 10 km grid cells; VU ≤ 50 10 × 10 km grid cells; and NT or LC > 50 10 × 10 km grid cells (Bland et al., 2016). In practice, listing under criteria B1 or B2 of the Red List of Ecosystems requires that at least one of three subcriteria is met in addition to meeting a threshold for either EOO or AOO (Rodríguez et al., 2015; Bland et al., 2016). Therefore, to achieve an assessment outcome for each simulated ecosystem, we assumed that at least one of the subcriteria within B1 or B2 was always met. Following the simulation, we recorded the proportion of simulations that resulted in ecosystem collapse for each combination of initial threat category and threat type.
Results
The simulation of 2,700,000 stochastic threat events across the 1350 distributions resulted in 241,519 ecosystem collapse events. Probabilities of collapse were heavily skewed towards zero in three of the four threat types (Table 1; Fig. S5). The exception was edge effect, for which 58% of ecosystems had nonzero probabilities of collapse, compared to 4.8% for circle, 4.2% for swipe and 3.1% for cluster. As expected, the probability of ecosystem collapse increased with decreasing EOO and AOO across all threat types (Figs 3 & 4). Ecosystems subjected to edge effect also exhibited increasing collapse probabilities with decreasing EOO and AOO but with greater variance than the other three threat types (Figs 3 & 4).
Boosted regression trees identified several spatial properties of ecosystem distributions that influenced their probability of collapse. Three distribution metrics emerged as highly important for predicting the probability of collapse: EOO, AOO and proportion core area (Fig. 5). For the stochastic circle, cluster and swipe threats, EOO and AOO were consistently the most important predictors of ecosystem collapse (Fig. 5). However, both analyses indicated that EOO and AOO were relatively unimportant for predicting the probability of collapse due to the edge effect threat. In that case, proportion core area, patch density and mean patch area were the most important predictors of collapse, indicating that distributions with smaller patch sizes were at higher risk of collapse than ecosystems with larger, contiguous patches with a greater total extent as core area (Fig. 5). In all cases, spatial metrics describing the shape of the ecosystem distributions contributed little to the BRT models (Fig. 5).
Collapse probabilities also varied with the initial status of the ecosystem (Figs 3 & 4). Ecosystems in higher threat categories based on criterion B exhibited higher probabilities of collapse than those in lower categories (Figs 3 & 4). For example, ecosystems classified as critically endangered had mean probabilities of collapse ranging from 0.006 (cluster) to 0.487 (edge effect), whereas near threatened or least concern ranged from 0 (circle, cluster, swipe) to 0.27 (edge effect; Fig. S6). Edge effect had the highest propensity to cause ecosystem collapse, with collapse probabilities > 0.2 regardless of the initial status of the ecosystem. Ecosystems initially listed as endangered, vulnerable or least concern were highly resilient to cluster and circle threat types, with no ecosystem collapses recorded in any of the simulation replicates.
Discussion
Our results show that the widely applied spatial metrics, EOO and AOO, were consistently the most important predictors of ecosystem collapse for the three spatially random threat types implemented in the simulation model. This was not the case for the edge effect threat type, which generated more collapse events and led to greater probabilities of collapse. The edge effect threat type was specifically included because it represents widely occurring threats to both ecosystems and species with expected impacts that relate directly to the properties of geographic distributions targeted by our study (Millennium Ecosystem Assessment, 2005; CBD, 2014). Our analysis indicated that ecosystems did not collapse by edge effect if a major proportion of their total area was buffered core area, thus comprising large, contiguous patches that are more robust to edge effects than small patches (Haddad et al., 2015). In this case, the impacts of edge effects are not readily detected by assessing EOO and AOO alone, suggesting that trends in core area or patch size distribution should also be employed in risk assessment protocols. In the IUCN Red List of Ecosystems, decline from edge effects is unlikely to go undetected; these metrics can be included in assessments through measures of change in biotic processes (criterion D), or through overall change in geographic distribution (criterion A; Bland et al., 2016; Keith et al., 2013). In IUCN Red List of Threatened Species assessments, such metrics may be considered in evaluating the ‘severely fragmented’ subcriterion under criteria B and C.
Owing to constraints on the extent of our model domain, EOO and AOO were highly correlated with our test dataset of simulated ecosystems. While the range of distributions that we considered are common in real landscapes and seascapes, other types of distributions can also be observed. Sparsely distributed ecosystems such as springs, inselbergs and island archipelagos typically have a large EOO and small AOO, whereas EOO and AOO tend to be highly correlated with contiguous ecosystems such as forests or marine pelagic systems. These exceptions illustrate that both EOO and AOO are useful for identifying ecosystems that may be differentially vulnerable to stochastic threats. In particular, where threatening processes cannot be reliably predicted, the complimentary use of both EOO and AOO allows assessment of spatial risk spreading (EOO) and the potential impact of directional or contagious threats (AOO) to be assessed relatively independently (Gaston & Fuller, 2009).
Our analysis assumed that subcriteria for criterion B were met in all scenarios, but this will not always be the case in real ecosystems. The subcriteria require qualitative evidence of continuing decline or an estimate of the number of threat‐based locations of the ecosystem (Bland et al., 2016). They play an important role in distinguishing at‐risk ecosystems from those that persist naturally in small, stable geographic ranges (Keith et al., 2013). When thresholds for EOO or AOO are met, but none of the subcriteria are met, an ecosystem will be listed as near threatened (Bland et al., 2016). The range size subcriteria for the Red List of Threatened Species serve a similar role (Mace et al., 2008).
Category thresholds of both EOO and AOO broadly reflected risks of collapse, with highest collapse probabilities associated with ecosystems classified as critically endangered and lowest with vulnerable and near threatened/least concern. Unsurprisingly, probabilities of collapse under the three stochastic threat types were negligible for ecosystems classified as vulnerable and least concern, but may have been underestimated for two reasons. First, constraining the size of each threat type to within the range observed in nature restricted the ability of some threat types to cause collapse in the majority of larger ecosystem distributions. For instance, the maximum specification of circle (18 km radius) was less than the EOO threshold for critically endangered (2000 km2). Limiting the size range of threats was necessary to attain realistic rates of ecosystem collapse by stochastic threats, emulating the very low rates of collapse by stochastic threats recorded in natural ecosystems. Similarly, developing an adequate dataset of geographic distributions required the simulation of ecosystem distributions across all possible size classes of criterion B. This issue does not influence the ability of the analysis to identify the spatial predictors of collapse. Of course, it is also possible that some threats have larger footprints than those used in our simulations, and these may be expected to generate higher risks of collapse. Secondly, the combinations of ecosystem distribution patterns and threat regimes that were explored in our simulation model are likely to be a conservative subset of plausible combinations that may be observed in real landscapes and seascapes. We explored only one threat type in each simulation, but real ecosystems are typically exposed to multiple threats of different types that interact additively or synergistically (Brook et al., 2008). Many threats also have a spatio‐temporal pattern of recurrence (e.g. fire regimes) with lagged ecological recovery times or cumulative effects (Strayer et al., 2006), which we did not explicitly model in this study. Development of models involving a broader range of threatening processes and their interactions is therefore required to reasonably evaluate the EOO and AOO thresholds employed in the two IUCN Red Lists. This is an obvious pathway for future work that will fill a significant remaining knowledge gap regarding the adequacy of the Red List criterion B thresholds.
To simplify the formulation of the problem, our simulation model assumed that ecosystem distributions were clearly delineated and mapped with error‐free methods, allowed only a binary response to threat events and required the use of generic threat types parameterized from observed threats. Further model development will therefore focus on (1) allowing threatening processes to operate with increasingly complex interactions, (2) estimating future changes in extant ecosystem distributions under different threat regimes, (3) incorporating continuous data of ecosystem condition and (4) investigating sensitivities of range size criteria to data of a different grain size and quality. In addition, constraining the simulation to landscapes or seascapes of different sizes will inform the application of the spatial criteria to subglobal domains (such as countries or regions) and assist in the formulation of objective range size thresholds for national and regional Red List assessments (Gärdenfors et al., 2001; Kontula & Raunio, 2009; Keith et al., 2015).
Although we focused our study on ecosystems, our approach to assessing the performance of the geographic range size metrics has potential for application to other biodiversity risk assessment protocols. In particular, the same spatial metrics that we have investigated are employed for assessing risks to species, and we expect that all of our findings are equally informative for species listing protocols such as the IUCN Red List of Threatened Species. Our model was specifically developed to operate with many types of spatial data, including distribution maps, remote sensing data and maps of disturbance histories, which offers a means to quantitatively evaluate stochastic drivers of extinction risk for species listed under geographic range size criteria. For example, our approach could elucidate patterns of extinction risk from stochastic threats, providing support to Red List status assessments while exploring the limitations in the application of this criterion (Gaston & Fuller, 2009; Joppa et al., 2015).
To our knowledge, this is the first quantitative evaluation of the spatial criteria that are used across the globe to evaluate risks to species and ecosystems. Our modelling framework delivers clarity to the derivation of geographic range size metrics and contributes to a fundamental goal of maintaining global lists of threatened biodiversity that reflect genuine risks of ecosystem collapse or species extinction. This goal demands ongoing efforts to improve existing biodiversity listing protocols, which in turn must be accompanied by parallel efforts to objectively evaluate their performance.
Acknowledgements
This project arose at the first meeting of the IUCN Red List of Ecosystems Committee for Scientific Standards, held in Finland in March 2015. The project was supported by an Australian Research Council Linkage Grant LP130100435 and co‐funded by the International Union for the Conservation of Nature, MAVA Foundation, NSW Office of Environment and Heritage, and the South Australian Department of Environment, Water and Natural Resources. All simulations were computed using the Linux computational cluster Katana, supported by the Faculty of Science, UNSW Australia.
References
Biosketches
The authors are members of the IUCN Red List of Ecosystems Committee for Scientific Standards. Their research focuses on developing and testing risk assessment methods for ecosystems and species, including the IUCN Red List of Ecosystems categories and criteria, adopted by the IUCN in 2014 (http://iucnrle.org).
Author contributions: This project arose from the first meeting of the IUCN Red List of Ecosystems Committee for Scientific Standards, held in Finland in early 2015. N.M., D.K and M.B developed the simulation model; N.M. and L.B. conducted the data analysis; N.M. led the manuscript with contributions by all authors.
Citing Literature
Number of times cited according to CrossRef: 15
- Michael Sievers, Ryan M. Pearson, Mischa P. Turschwell, Melanie J. Bishop, Lucie Bland, Christopher J. Brown, Vivitskaia J.D. Tulloch, Jodie A. Haig, Andrew D. Olds, Paul S. Maxwell, Rod M. Connolly, Integrating outcomes of IUCN red list of ecosystems assessments for connected coastal wetlands, Ecological Indicators, 10.1016/j.ecolind.2020.106489, 116, (106489), (2020).
- Michael Sievers, Mahua Roy Chowdhury, Maria Fernanda Adame, Punyasloke Bhadury, Radhika Bhargava, Christina Buelow, Daniel A. Friess, Anwesha Ghosh, Matthew A. Hayes, Eva C. McClure, Ryan M. Pearson, Mischa P. Turschwell, Thomas A. Worthington, Rod M. Connolly, Indian Sundarbans mangrove forest considered endangered under Red List of Ecosystems, but there is cause for optimism, Biological Conservation, 10.1016/j.biocon.2020.108751, 251, (108751), (2020).
- Md Anwar Hossain, Heini Kujala, Lucie M. Bland, Mark Burgman, José J. Lahoz‐Monfort, Assessing the impacts of uncertainty in climate‐change vulnerability assessments, Diversity and Distributions, 10.1111/ddi.12936, 25, 8, (1234-1245), (2019).
- Pablo M. Lucas, Manuela González‐Suárez, Eloy Revilla, Range area matters, and so does spatial configuration: predicting conservation status in vertebrates, Ecography, 10.1111/ecog.03865, 42, 6, (1103-1114), (2019).
- Milan Chytrý, Michal Hájek, Martin Kočí, Pavel Pešout, Jan Roleček, Jiří Sádlo, Kateřina Šumberová, Jan Sychra, Karel Boublík, Jan Douda, Vít Grulich, Handrij Härtel, Radim Hédl, Pavel Lustyk, Jana Navrátilová, Pavel Novák, Tomáš Peterka, Alena Vydrová, Karel Chobot, Red List of Habitats of the Czech Republic, Ecological Indicators, 10.1016/j.ecolind.2019.105446, 106, (105446), (2019).
- William José Agudelo-Hz, Nicolás Urbina-Cardona, Dolors Armenteras-Pascual, Critical shifts on spatial traits and the risk of extinction of Andean anurans: an assessment of the combined effects of climate and land-use change in Colombia, Perspectives in Ecology and Conservation, 10.1016/j.pecon.2019.11.002, (2019).
- Cintia E Celsi, Liliana M Giussani, Geographical distribution and habitat characterization of Poa schizantha (Poaceae), a narrow endemic of the coastal sand dunes of the southern Pampas, Argentina, Botanical Journal of the Linnean Society, 10.1093/botlinnean/boz069, (2019).
- Wu-Bing Xu, Jens-Christian Svenning, Guo-Ke Chen, Ming-Gang Zhang, Ji-Hong Huang, Bin Chen, Alejandro Ordonez, Ke-Ping Ma, Human activities have opposing effects on distributions of narrow-ranged and widespread plant species in China, Proceedings of the National Academy of Sciences, 10.1073/pnas.1911851116, (201911851), (2019).
- Tsuyoshi Takeuchi, Takuma Sasaki, Masahiro Mitsuhata, Takuya Kiyoshi, Mana Nishimoto, Tetsuro Nomura, Jun-ichi Takahashi, Low mitochondrial DNA variation in the endangered bumble bee Bombus cryptarum florilegus , Journal of Apicultural Research, 10.1080/00218839.2019.1614735, (1-6), (2019).
- Nicholas J. Murray, David A. Keith, Lucie M. Bland, Renata Ferrari, Mitchell B. Lyons, Richard Lucas, Nathalie Pettorelli, Emily Nicholson, The role of satellite remote sensing in structured ecosystem risk assessments, Science of The Total Environment, 10.1016/j.scitotenv.2017.11.034, 619-620, (249-257), (2018).
- Lucie M. Bland, Kate E. Watermeyer, David A. Keith, Emily Nicholson, Tracey J. Regan, Lynne J. Shannon, Assessing risks to marine ecosystems with indicators, ecosystem models and experts, Biological Conservation, 10.1016/j.biocon.2018.08.019, 227, (19-28), (2018).
- Frank T. Breiner, Ariel Bergamini, Improving the estimation of area of occupancy for IUCN Red List assessments by using a circular buffer approach, Biodiversity and Conservation, 10.1007/s10531-018-1555-5, 27, 9, (2443-2448), (2018).
- Md Anwar Hossain, José J. Lahoz‐Monfort, Mark A. Burgman, Monika Böhm, Heini Kujala, Lucie M. Bland, Assessing the vulnerability of freshwater crayfish to climate change, Diversity and Distributions, 10.1111/ddi.12831, 24, 12, (1830-1843), (2018).
- Nicholas J. Murray, David A. Keith, Daniel Simpson, John H. Wilshire, Richard M. Lucas, Remap: An online remote sensing application for land cover classification and monitoring, Methods in Ecology and Evolution, 10.1111/2041-210X.13043, 9, 9, (2019-2027), (2018).
- Lucie M. Bland, Tracey J. Regan, Minh Ngoc Dinh, Renata Ferrari, David A. Keith, Rebecca Lester, David Mouillot, Nicholas J. Murray, Hoang Anh Nguyen, Emily Nicholson, Using multiple lines of evidence to assess the risk of ecosystem collapse, Proceedings of the Royal Society B: Biological Sciences, 10.1098/rspb.2017.0660, 284, 1863, (20170660), (2017).
