Development of a Species Status Assessment Process for Decisions under the U.S. Endangered Species Act

Decisions that support the purpose of the Endangered Species Act (ESA 1973, as amended) require a scientific assessment of the risk of extinction of a species or, conversely, its probability of persistence (Carroll et al. 1996; Doremus and Tarlock 2005; Waples et al. 2013; definition of species from the ESA [1973] is presented in Table 1 along with definitions for other ESA-related terms, which are shown in italics at first occurrence). The purpose of the ESA (ESA 1973, 16 USC. §1531-1544) is to conserve threatened and endangered species and the ecosystems upon which they depend (section 2[b] of the ESA; Table 1). The Secretaries of the Interior and Commerce acting through the U.S. Fish and Wildlife Service (USFWS) and National Oceanic and Atmospheric Administration make decisions to fulfill that purpose. Analysts present results from a species assessment to decision-makers who then make a policy judgment based on the ESA and in light of that scientific information (Shaffer 1987; Rohlf 1991; Carroll et al. 1996; Doremus 1997; Vucetich et al. 2006; Gregory et al. 2013).

The past practice for species assessment within the USFWS often focused on threats without an explicit analysis of the response of the species to the threats (Andelman et al. 2004). Andelman et al. (2004) reviewed nine protocols for assessing species risk and concluded that the threats-focused assessment used at the time by the USFWS had low repeatability and transparency and was based on threat occurrence without explicit prediction of a species' future response to those threats. Also, a series of critiques on the role of science in ESA decisions have called for improved consistency and transparency in the procedures for assessing species status (Carroll et al. 1996; Andelman et al. 2004; Waples et al. 2013; Lowell and Kelly 2016; Murphy and Weiland 2016). Lowell and Kelly (2016) and Murphy and Weiland (2016) questioned the degree to which the mandate for best available science (USFWS and NMFS 1994) has been met under past practice. A team from the USFWS and the U.S. Geological Survey (including the authors and those listed in the Acknowledgments) with experience in species assessments, ESA applications, and decision analysis developed an assessment process to address issues raised by the critiques and to update the process to reflect the advances in scientific practices. The new process, termed the species status assessment, or SSA, is framed around a systematic analysis of the current and future responses of the species in question to stressors and conservation efforts. The USFWS directed the use of the SSA for ESA decisions through a USFWS Director's memorandum (USFWS 2016a). In this paper we document the SSA process, which is now an emerging practice within the USFWS (Earl et al. 2017; Evansen et al. 2017; McGowan et al. 2017). The SSA is relevant to species risk assessment broadly, but our primary focus is on ESA decisions conducted by USFWS.

At its most basic level, the SSA is species focused rather than threat focused, and we have designed it to analyze the available scientific information to estimate a particular species' current condition and forecast the species' future condition for ESA decisions (Figure 1). Threats are potential explanations for current and future condition rather than the endpoint of the assessment. Sequentially, the SSA results in a description of the fundamental ecological and evolutionary relationships between the species and its environment to hypothesize explanations for the current condition and provide a foundation for forecasting future condition. In the development of the assessment process, we incorporated the recommendations from current literature on species assessment regarding population viability analysis (Shaffer and Stein 2000; Redford et al. 2011; Waples et al. 2013; Wolf et al. 2015; Murphy and Weiland 2016), a broad role for modeling (Starfield 1997; Ruhl 2004; Addison et al. 2013), and scenario planning relevant to the decision context (Peterson et al. 2003; Duinker and Greig 2007; Goodwin and Wright 2014; IPBES 2016).

As pointed out by Doremus (1997), Rohlf (2004), Vucetich et al. (2006), Woods and Morey (2008), Waples et al. (2013), and others, ESA decisions involve both scientific and normative (policy) dimensions (Table 2). Conflating the roles of science and policy can create unnecessary confusion both within the agencies charged to make ESA decisions and with the public and partners who are affected by those decisions (Robbins 2009; Wilhere 2011; Waples et al. 2013; Boyd et al. 2016). The SSA results in a scientific report distinct from the application of policy, which is a departure from past USFWS practice in many instances (Waples et al. 2013).

In this paper, we focus on the conceptual framework for the SSA and present two case studies to illustrate how the process can be implemented. We avoid being overly prescriptive so that practices can adapt to specific cases and adopt innovative techniques and methodologies; SSA guidance is kept up-to-date and available online (USFWS 2016b). The structure of this paper first presents a process followed by example applications of that process (case studies). The case studies on eastern massasauga rattlesnake Sistrurus catenatus and Sonoran Desert tortoise Gopherus morafkai are among early applications, and we have used the insights gained to improve the SSA process. We point out the links between the SSA and the five factors of the ESA (Table 1) and conservation efforts. Finally, we discuss the roles of science and policy in ESA decisions.

The USFWS should initiate an SSA following receipt of a substantial petition for a species to be considered for endangered or threatened species designation; the SSA then follows the species (USFWS 2016a) so that the information and analyses are available for subsequent ESA determinations (Figure 1). The SSA results in a scientific report that describes the risk of extinction, which a decision-maker can then use along with policy judgment to determine legal status under the ESA (Doremus 1997; DeMaster et al. 2004; Doremus and Tarlock 2005; Vucetich et al. 2006; Robbins 2009). The initial SSA report can be adapted to the needs of a particular decision (listing, recovery planning, consultation, and permitting; Table 1) and updated with new data and information (Figure 1). The level of detail in an SSA will depend on the amount and quality of available data (Doremus and Tarlock 2005). In some cases, the available information is sufficient for assessing risk on a continuous scale. In other cases, because of data limitations the SSA can indicate only categorical levels of risk.

The SSA process involves three successive stages: 1) the ecology of the species, 2) the current condition of the species, and 3) the future condition of the species.

The first stage of an SSA is an exploration of the species' life history and ecology, which lays the foundation for the next stages of the process. Stage 1 results in a description of the life history, including trophic niches, reproductive strategies, biological interactions, and habitat requirements to determine how individuals at each life stage survive and reproduce. The SSA identifies areas representing significant ecological, genetic, or life history variation (i.e., the ecological settings) informed by historical as well as present distribution. The entire range of historical conditions under which the species was presumably self-sustaining serves as a starting point to understand how the species functions (or functioned) to maintain populations across its range (Seminoff et al. 2015)

The next stage of the SSA is to describe the current condition of the species' habitat, demographics, and distribution. Stage 2 results in an empirical description of the current 1) population structure, distribution, abundance, demographic rates, diversity (ecological, genetic, life-history diversity), and habitat, 2) changes from historical to current condition (i.e., trends), and 3) explanations or hypotheses of the causes and effects of stressors and conservation efforts that resulted in the current condition.

In the final stage, an SSA results in the prediction of the species' response to a range of plausible future scenarios of environmental conditions and conservation efforts. This step entails an analysis of future plausible scenarios of stressors and conservation efforts to project consequences on the ability of the species to sustain populations in the wild over time. The predictions start at the current condition estimated in stage 2 and project forward based on the information developed in stage 1 on how the species interacts with its environment. The metrics used for future condition align with metrics used in the prior stages and include demographics (abundance and population growth or productivity), distribution, and diversity (ecological, genetic, life-history), which are core autecological parameters that measure the relationships between a species and its environment. The numerical resolution and spatial and temporal scale of the metrics will depend on data availability and the information needed for the decision context.

The future condition is unavoidably uncertain because 1) future events are inherently probabilistic (aleatory uncertainty) and 2) the knowledge of the species' response to future scenarios is imperfect due to sampling and measurement error, competing hypotheses about ecological relationships, and imprecisely defined terms and categories (epistemic and linguistic uncertainty; Taylor et al. 2002; Carey and Burgman 2008; Lukey et al. 2010; McGowan et al. 2011; Murphy and Weiland 2016; Phillips-Mao et al. 2016). The scenarios developed in stage 3 represent an important tool for incorporating uncertainty in species risk (Peterson et al. 2003; Duinker and Greig 2007; Goodwin and Wright 2014; Rowland et al. 2014). Scenarios are designed to explore the response of the species to environmental stressors (including climate change) and interventions by conservation efforts that could ameliorate the stressors (Duinker and Greig 2007; Fordham et al. 2013; Gregory et al. 2013; IPBES 2016; Phillips-Mao et al. 2016). Uncertainty in forecasts within scenarios comes from variability in model predictions or expert judgments (Taylor et al. 2002; Refsgaard et al. 2007; McGowan et al. 2011; Drescher et al. 2013). The combination of variation among scenarios and uncertainty in forecasts within scenarios is used to explore the risk profile of the species, in other words, the plausible range in the species' response to future stressors and conservation efforts.

The SSA process applies the conservation biology principles of representation, resiliency, and redundancy (we refer to them here as the 3Rs) to evaluate the current and future condition of the species (Shaffer and Stein 2000; Redford et al. 2011; Waples et al. 2013; Wolf et al. 2015; Earl et al. 2017). In general, species risk will decrease, or at least does not increase, with increases in representation, resiliency, and redundancy. Shaffer and Stein (2000) composed a hierarchical order to the 3Rs as they relate to viability: first, conserve some of everything (i.e., representation) and then save enough to last (i.e., resiliency and redundancy). From a decision analysis perspective, we can view the 3Rs as means objectives for the overarching fundamental objective of sustaining populations in the wild. The fundamental objective is what we want to achieve, and the means objectives are essential ways to achieve what we want (Gregory et al. 2012). Shaffer and Stein (2000) related representation to the conservation of a species within the array of different environments in which it occurs or areas of significant ecological, genetic, or life-history variation, termed here as ecological settings (Carroll et al. 2010; Wolf et al. 2015). We suggest that for ESA decisions, representation uses diversity as a proxy for adaptive capacity. Resiliency refers to the ability of a population to withstand stochastic disturbance events; thus, resiliency is related to the demographic ability to absorb and bounce back from disturbance and persist at the population or meta-population scale. Redundancy spreads risk among multiple populations or areas to minimize the risk due to large-scale, high-impact (i.e., catastrophic) events. Thus, the 3R concept helps to construct a risk assessment that takes into account demographic factors, distribution or spatial structure, and diversity. Demographic factors (abundance, survival, productivity, and ultimately intrinsic population growth rate) contribute to the ability to absorb disturbance and persist (resiliency). Spatial structure contributes to redundancy through increased distributional extent by spreading risk across the broader landscape and adds to resiliency by increasing connectivity among meta-populations. Diversity, as represented in genetic, geographic, or life-history variation, contributes to adaptive capacity and can inform decisions related to the ESA concepts of distinct population segment (USFWS and NMFS 1996) and significant portion of the species' range (USFWS and NMFS 2014; Earl et al. 2017).

The National Oceanic and Atmospheric Administration uses abundance, productivity, spatial structure, and diversity as the criteria to determine viable populations in ESA-relevant assessments (Waples et al. 2013). The criteria relate to the 3Rs in this way: abundance and productivity correspond to resiliency, spatial structure contributes to resiliency and redundancy, and diversity relates to representation. Wolf et al. (2015) recommended the following metrics within the recovery planning context: abundance, productivity, and connectivity for resiliency; number of populations for redundancy; and occupancy across the gradient of genetic, ecological, or life-history diversity for representation.

The ESA identifies five statutory factors to consider as causes for endangerment (see section 3 of the ESA for definition of endangerment and section 4[a] for the five factors; Patrick and Damon-Randall 2008). The five factors are listed in Table 1. The ESA requires that a listing determination identify the factors causing endangerment, but the law does not specify how the factors should be considered. The SSA approach within each stage of the assessment is to hypothesize and evaluate the causal relationships between the factors and the species' response. Stage 1 describes the influence of habitat (factor A) and other natural factors (e.g., factor C: disease and predation) on the species' ecology. Stage 2 identifies and evaluates any of the factors that researchers have hypothesized to have led to the current condition of the species. Stage 3 incorporates the factors, which researchers have hypothesized to have population-level effects, into scenarios used to forecast the species' future condition. The SSA considers not just the presence of the factors, but assesses to what degree they influence risk. Because the SSA uses metrics for demographics, distribution, and diversity, the effect of multiple stressors is inherent in the assessment and helps to assess how populations and ultimately the species responds cumulatively to the interactive effects of stressors and conservation efforts included in the future scenarios.

The SSA incorporates conservation efforts in the assessment, and results in a description of how conservation efforts influence the current condition of the species. The SSA can incorporate conservation efforts in the scenarios used to forecast the species' future condition. In the context of a listing decision, for example, decision-makers can use the SSA to evaluate sufficient regulatory mechanisms that satisfy factor D of the five factors (Table 1). An evaluation of the species' response to conservation efforts, including those with uncertain implementation and effectiveness, may help to develop and evaluate conservation strategies. Prior to listing, the assessment of what could be done to improve the condition of the species provides opportunities to carry out candidate conservation (Table 1) actions in advance of future ESA decisions. In addition, if the species is subsequently listed for protection under the ESA, then the conservation strategies evaluated in the SSA can be used in recovery planning.

Projecting the future condition of the species, which is integral to all ESA decisions, and thus to an SSA, relies broadly upon models and modeling (Starfield 1997). The resolution of the available information (including covariates, response variables, and uncertainties) and purpose of the modeling determine the type and complexity of the models. The utility of a model to an SSA and an ESA decision depends on how the available information, derived from data or expert judgment, is analyzed and how the outputs are interpreted. Uncertainty and low-quality data do not prohibit the utility of a model as long as the sensitivity of model outputs to violations of the underlying assumptions or sources of uncertainty are assessed and effectively communicated to the decision-makers. Even simple models can be quite useful. For example, conceptual models (expressed as influence diagrams) are useful for illustrating life history or graphically relating environmental factors to a species' condition. Also, McCarthy et al. (2004) found that predictions of extinction risk were less biased when based on an explicit model compared to subjective judgment, even if the model is necessarily simplistic because of sparse data. Habitat models can translate ongoing stressors to future habitat upon which the species depends (Copeland et al. 2009). Population models can project future condition as a function of future stressors and conservation (Akçakaya and Sjögren-Gulve 2000; Runge et al. 2007; Murphy and Weiland 2016). Models provide an explicit, transparent, and repeatable method of analysis, which facilitates a thorough peer review of both the SSA methodology and results (Rohlf 1991; Starfield 1997; Ruhl 2004; Addison et al. 2013). Models also provide a structure to integrate new information in subsequent assessments. Within an SSA the specific models should be parsimonious and built to meet the specific needs of the SSA within the decision context (Starfield 1997; Burgman and Yemshanov 2013).

As Burgman (2016) advised, only after all other sources of data are exhausted should an assessment turn to expert judgment to fill in gaps. But data gaps are common in endangered species assessments. Formal elicitation of expert judgment is a complicated endeavor that involves the careful determination of specific information needs that may call for expert judgment, identification and preparation of experts, and elicitation and characterization of uncertainty in judgments. Fortunately, recommended practices for eliciting expert judgment have been recently published (U.S. Environmental Protection Agency 2011; Drescher et al. 2013; Burgman 2016). The first considerations when eliciting expert judgment is to determine the information needs and the format with which to acquire that information—a workshop, interviews, or questionnaires. Identification of experts can start with a review of relevant literature. Professional credentials, position, the areas of expertise, and relevant experience can be used for selection criteria to help ensure that scientific experts are familiar with the topic and that the choices were transparent, unbiased, and captured a broad diversity of expertise and professional judgments related to the topic. Analysts develop detailed technical questions on topics germane to the assessment. After establishing a common understanding of the information context, analysts ask experts through in-person, phone, or online discussions, about facts and information based on their individual, professional knowledge on specific topics. Analysts do not seek group consensus. Instead, the variation in judgment among experts is an important source of uncertainty. Techniques to capture uncertainty within and among experts include four-point elicitation or likelihood point method coupled with the Delphi method (Burgman 2016).

The SSA results inform decision-makers, who apply ESA policies to the decision at hand (Table 2). The decision-makers compare the risk inferred from the SSA to the relevant ESA regulatory standards and definitions (Waples et al. 2013; Murphy and Weiland 2016). To take full advantage of the SSA, the decision-makers need to accept the overall analytical process and understand the SSA results, as well as the strength of data and any assumptions used to develop any models used for estimation or prediction. Therefore, it is important that SSA biologists and decision-makers discuss and agree to the metrics, future scenarios, and time frames as influenced by the Office of the Solicitor's guidance on foreseeable future (USDOI 2009; Table 1). It is incumbent upon the analyst to present the levels of uncertainty for the future condition to the decision-makers in the agreed-upon metrics. A practical interpretation is that the uncertainty represents the plausible range in the future condition of the species across the scenarios, including prediction variance within each scenario. The decision-maker also needs to understand the underlying assumptions and data limitations to avoid inappropriate conclusions and inference (Murphy and Weiland 2016). The characterization of uncertainty, such as the probability associated with confidence level or the quantitative interpretation of what is plausible, relies on scientific judgments that can be based on professional norms (Anderson et al. 2001; Doremus and Tarlock 2005; Refsgaard et al. 2007). For instance, Anderson et al. (2001) advised that a confidence level (1-α) percentage should be explicitly reported but acknowledged that there are acceptable options for the particular level (e.g., 90, 95, 99%). Heuristically, the level of uncertainty is related inversely to the quality of available data and information (Runge 2011; Williams and Johnson 2015). More and better data would, in principle, reduce uncertainty. However, policy judgments have to be made in the face of that uncertainty due to data limitations or time constraints or because some uncertainty (e.g., environmental variability) is irreducible even with more research (Murphy and Weiland 2016). Furthermore, policy judgment may not be sensitive to the level of uncertainty. For example, the appropriate policy choice can be evident in spite of the uncertainty, in which case, allocating time or funds to reducing the uncertainty would not provide value to the decision-making process (Williams and Johnson 2015).

The eastern massasauga rattlesnake (EMR) became a candidate for listing due to multiple factors associated with habitat modification and loss of populations across its range. Given the broad distribution of the species and the inconsistency in amount and quality of demographic information across the various populations, we considered the SSA to be fairly complex. A peer-reviewed report documented the assessment (USFWS 2016c).

The EMR occupies wet meadows, fens, and bogs in the Midwest and northeastern United States and southern Ontario, Canada (Seigel 1986; Kingsbury et al. 2003). The particular ecological needs of EMRs vary with season. During the hibernation period, the EMR requires a moist subterranean space below the frost line to avoid desiccation and freezing (Sage 2005), while during the active season, the EMR needs a mosaic of shaded and sunny areas for thermoregulation, abundant prey, and areas to escape predators (Johnson 2000).

To assess EMR viability, we analyzed its historical, current, and projected future abundance and distribution. We began with describing the breadth of adaptive capacity across the EMR range. We investigated variation in habitat use, prey, venom, climate, and genetics as potential indicators of variation in adaptive capacity. We ultimately determined that breadth of adaptive capacity can be captured by a wide distribution of populations within three genetically diverse regions identified by Ray et al. (2013): 1) the western analysis unit (WAU) consisting of populations in Minnesota, Missouri, Iowa, Wisconsin, and Illinois; 2) the central analysis unit (CAU) consisting of populations in Indiana, southern and central Michigan, Ohio, and far southwestern Ontario; and 3) the eastern analysis unit (EAU) consisting of populations in New York, Pennsylvania, northern Michigan, and the remaining portions of Ontario.

Next, we assessed the change in the number and distribution of populations from before 2014 to current (2014–2016) and future (10-, 25-, and 50-y) time periods in each of the three analysis units (AUs). We assessed the status of historical populations (extant, extirpated, or unknown) and the health of the extant populations. We relied on the results of Faust et al. (2011) and supplemental information garnered since 2011 to assess the health of the populations. Faust et al. (2011) built an age-based, stochastic population model for a hypothetical healthy EMR population and determined how various influences affect EMR vital rates. We derived the demographic parameters, the prominent influences, and the effect of such influences on vital rates from empirical data and expert judgment. The prominent influences identified and analyzed were habitat loss, vegetative succession, fragmentation, road mortality, hydrologic alteration, human harassment, collection, ineffective management regimes, and habitat restoration. Using elicited site-specific information, Faust et al. (2011) generated estimates of population growth rate, ending population size, and probability of quasi-extirpation (adult female population size ≤ 25) for all populations with sufficient data (57 populations). We used these results to identify populations considered healthy (i.e., self-sustaining). We defined a self-sustaining population as having 1) an adult female population size > 50, 2) a positive population growth rate, and 3) a probability of persistence greater than 0.90 over 25 y despite the stressors acting upon it. We extrapolated the results to infer the health of the nonmodeled populations by multiplying the proportion of modeled populations meeting our self-sustaining criteria by the number of extant populations in each AU.

We evaluated the change in adaptive capacity over time by calculating the spatial extent of occurrence range-wide and within the three AUs. We used ArcGIS to draw polygons around clusters of counties with EMR populations and summed the area of all polygons within and across AUs. We evaluated EMR redundancy by assessing the vulnerability to catastrophic events. We consulted the literature and species experts to identify the natural and anthropogenic catastrophic events that would likely lead to population extirpation. Experts identified drought, flooding, and disease (rapid and widespread epidemic) as potential catastrophic events. We had insufficient information on flood (specifically, the magnitude of flood that would lead to extirpation) and disease risk (notably, the likelihood of disease outbreaks, the factors that affect disease spread, and the magnitude of impact on EMR populations) to include either in our analysis. Thus, drought was the only catastrophic event analyzed. To calculate the risk of extirpation due to drought, we used an extinction risk model developed by Ruckelshaus et al. (2002).

Historically, there were 558 EMR populations scattered across parts of Ontario, New York, Pennsylvania, Ohio, Michigan, Indiana, Illinois, Missouri, Iowa, Wisconsin, and Minnesota. Currently, there are 347 extant populations in 10 states (Figure 2). Within the WAU, 20 of 72 historical populations are extant, and of these, six are self-sustaining. In the CAU, 256 of 350 historical populations are extant, of which 65 are self-sustaining. In EAU, 71 of the 136 historical populations are extant, 30 of which are self-sustaining. Range-wide, EMR spatial extent declined from the historical period by 41%, with 70, 33, and 26% decreases in the WAU, CAU, and EAU, respectively.

Due to time constraints, we ran only one future scenario; we assumed the magnitude of impact and frequency of the prominent influences would continue into the future. To identify the number of populations likely to persist under the continuation scenario, we assumed that populations that met the criteria for being self-sustaining at years 10, 25, and 50 would persist for those three time periods. Using these results for the 57 modeled populations, we then extrapolated to the remaining extant populations by multiplying the proportion of modeled populations that were self-sustaining by the total number of currently extant populations in each AU to estimate the number of EMR populations that we projected to be self-sustaining at years 10, 25, and 50.

Range-wide, we forecasted population losses to continue into the future (Figure 3), with 263 populations forecasted to be extirpated by year 50. Of the populations projected to persist, we forecasted one population in WAU, 47 in CAU, and six in EAU to be self-sustaining by year 50. We projected the spatial extent of EMR to decline. Range-wide, we forecasted the special extent of occurrence to decline by 80% by year 50; within the AUs, we projected extent of occurrence to decline by 91, 64, and 89% in the WAU, CAU, and EAU, respectively. The risks of AU-wide extirpation due to catastrophic drought remained near zero for CAU and EAU due to a combination of low drought risks and the number of populations, but the probability of WAU-wide extirpation within 25 y ranged from 0.02 to 0.82 depending upon the drought severity.

The abundance and distribution of the EMR have declined from its historical condition and is forecasted to continue to decline into the future. Relative to historical conditions, currently there is a 38% reduction in the number of extant populations with predicted reductions to reach 85% by year 50. These losses have not been uniformly distributed. The WAU, which historically represented 28% of the EMR range, today represents 14% of the species' range, and by year 50, is predicted to represent 12% of the range with one self-sustaining population persisting. Catastrophic drought greatly increases the risk of extirpation, resulting in a 0.96 probability of extirpation of the WAU. Similarly, the EAU historically comprised 36% of the range, but by year 50 we project it to comprise 19% with six self-sustaining populations persisting, representing a 96% loss of the historical populations. In the CAU, we predicted 78% of the historical populations to be extirpated, with 47 populations projected to be self-sustaining. Although populations are projected to persist, EMR range is projected to contract, with high likelihood for the extirpation of populations from the western portion of the range and substantial losses in eastern portion of the range. These losses are likely to lead to considerable decreases in adaptive capacity, which may impair the ability of EMR to adapt to near-term and long-term changes in its environment (e.g., novel diseases and predators, habitat alteration due to invasion of exotic species), thereby increasing its vulnerability to extinction.

Prior to a 2-d in-person meeting with the USFWS decision-makers, the analysis team provided a written SSA report (USFWS 2016c) and presented a summary via a webinar of the methods, results, and the implications of uncertainty. This provided an opportunity for the decision-makers to ask questions, request inclusion of alternative scenarios, and explore different assumptions prior to making a determination of whether EMR meets the definition of threatened or endangered under the ESA. The decision-makers asked about the rationale underlying the definition of a self-sustaining population and our extrapolation approach. The decision-makers were satisfied with the rationale for underlying assumptions and with the range of uncertainty we modeled. The decision-makers recommended that EMR warranted protection under the ESA, and it was designated a threatened species (USFWS 2016d). The determination attributed habitat loss as the greatest cause of current and future condition and noted that emergent disease, collection and persecution of individuals, and climate change contribute to the risk of extinction.

The Sonoran Desert tortoise (SDT) was determined to be a candidate for listing under the ESA in 2010 due to a preponderance of different potential threats to the species (USFWS 2010). We anticipated this to be a complex SSA as the geographic scope encompassed Arizona and Sonora, Mexico, which we incorporated into geospatial and demographic modeling (USFWS 2015a, McGowan et al. 2017). Interest in the decision from external parties was quite high due to the implications of a listing determination.

The SDT is a long-lived tortoise that ranges from Arizona to central Sonora, Mexico. Recent genetic analysis delineated the SDT as a species distinct from the Mojave Desert tortoise (Murphy et al. 2011). SDTs utilize rocky slopes at higher elevations and soil types that facilitate excavation of burrows for shelter and nesting, though they sometimes use natural cavities (Van Devender 2002). The occurrence of drought has potentially large effects on tortoise demographics, as seasonal monsoon rains and annual green-ups provide forage resources that improve survival and reproduction (Averill-Murray et al. 2002; Sullivan et al. 2014).

Spatial subdivisions cannot clearly delineate population boundaries for the SDT. Though potential barriers to movement exist, there was not sufficient genetic evidence of spatial structure to guide population delineation (Edwards et al. 2004; Edwards et al. 2015). Therefore, we divided the species' range into two populations for this analysis, separating the United States from Mexico because the species faces different threats and management strategies in these two regions.

The geographic range of the SDT is unchanged compared to historical conditions. We estimated the abundance of the two populations by first evaluating habitat quality across the range with a geospatial model that used land cover vegetation type, slope, and elevation to classify areas into primary, secondary, and tertiary potential habitat classes (Figure 4A). Estimated population densities vary considerably (from two to 20 adult SDTs/km²). We assumed that primary habitat would sustain densities at the high end of that range (17 adult SDTs/km²), secondary in the middle (9 SDTs/km²), and tertiary at the low end (2 SDTs/km²). Based on a range of habitat qualities and reported population density, we extrapolated Arizona abundance ranged from ∼ 310,000 to ∼ 640,000 and Mexico from ∼ 160,000 to ∼ 330,000 adult males and females (USFWS 2015a). Using a population projection model, we ran different scenarios with low and high starting population sizes to convey to decision-makers how we expected uncertainty in current population size to influence future condition.

We elicited conceptual models of SDT ecology from species experts to identify key demographic parameters, habitat requirements, and environmental factors that might affect tortoise populations (Figure 4B). Through a series of meetings with federal, state, and academic biologists we mapped the environmental factors, such as food availability, precipitation, invasive grasses, and cattle grazing, that influence tortoise demographics and, therefore, their population viability. We initially used the conceptual modeling to explore all the factors that might affect tortoises, but focused on the most important components that would affect viability. For example, though we initially explored the effects of increasing regional temperatures due to anticipated climate change on sex ratio in tortoise populations (nest temperature determines sex), experts agreed that temperature effects on adult survival via drought would manifest much sooner and more severely than sex ratio effects on the populations, so we eliminated temperature effects on sex ratio from our stochastic simulation model.

Our analysis identified potential primary threats: drought and expected decreases in precipitation due to climate change, increases in frequency and intensity of wildfire, and habitat loss and degradation due to urban expansion and human population growth and invasion of nonnative plants (Figure 4A). Literature indicates that tortoise survival is highly susceptible to drought (Zylstra et al. 2013), causing up to a 10% decline in annual survival of adult and larger juvenile tortoises. Urbanization and wildfire could limit habitat availability and reduce the quality of remaining habitat through associated disturbances (Figure 4). Nonnative plants degrade habitat quality and increase wildfire intensity and effects. Because the rates of change in climate or urbanization in the future are uncertain, we made these effects variable over time in the projection model and used the model to explore multiple scenarios of climate change and habitat loss.

We built a population viability model to simulate tortoise populations and measure population resiliency into the future under stochasticity and parametric uncertainty (McGowan et al. 2017). The matrix population model accounted for three life stages (small juveniles, large juveniles, and breeding adults; McCoy et al. 2014). The parameters in the model (e.g., survival, fecundity, etc.) varied annually to represent environmental variability and also applied parametric uncertainty functions to account for imperfect data and observation errors that affect parameter estimates (McGowan et al. 2011). To represent climate change, we incorporated a randomized drought function that determined what proportion of the population experienced drought and the magnitude of the drought effect on the survival rates.

where P_drought is the proportion of the population exposed to drought and is the survival rate of adults for the full population, given the proportion that was exposed to drought. The drought effect in a specific year is DE_t, which was modeled as a random variable between 0.8 and 0.99 and which simulates a 1 to 20% reduction in survival due to the drought in any given year, to represent differing drought severity (spatially and magnitude) from year to year. This drought function enabled simulations with increasing drought frequencies that could result from future climate change. We also incorporated into the model a ceiling-type density-dependence function that prevented the population from exceeding the abundance allowed by the available habitat (Morris and Doak 2002). The density ceiling would reduce productivity to 0 if the population exceeded the threshold, allowing us to limit population growth without speculating on the mechanisms of density dependence. The density threshold was set at the maximum population size possible if all the available habitats were occupied at the highest empirically observed densities (USFWS 2015a; McGowan et al. 2017). We used the model to simulate the effects of nine scenarios related to climate-driven drought effects, habitat loss from urban expansion, and potential benefits of proposed management actions. Habitat loss scenarios lower the density-dependent ceiling over time to mimic loss of habitat that could be occupied, and positive management scenarios were simulated by counteracting or stabilizing habitat loss (i.e., increasing or stabilizing changes in the density-dependent ceiling). Also, the model included random variation in annual parameter values to represent environmental stochasticity and added parametric uncertainty to survival and fecundity parameters (McGowan et al. 2011).

Model outputs included the median population trajectories and the 2.5 and 97.5 percentiles of the population trajectories (Figure 5), and we also reported the probability of quasi-extinction under each scenario over a 200-y time span. We measured the probability of quasi-extinction as the proportion of simulation replicates that fell below a predetermined minimum population size (e.g., with 1,000 replicates and a quasi-extinction probability of 0.05, 50 replicates fell below the threshold and 950 did not). We evaluated both 2 and 4% of the initial population estimates as quasi-extinction thresholds, offering two quasi-extinction thresholds to the decision-makers to more fully describe risk of extinction. In other words, we predicted the probability that the population would decline to 2 or 4% of its initial size estimate in the future. Quasi-extinction thresholds are theoretically supposed to represent the point at which a population is so small that extinction is unavoidable; however, we do not know what that threshold is for SDT populations, so we presented results for two different thresholds. We selected the 2 and 4% thresholds using input from species experts and managers. At 2 or 4% of current abundance, we presumed that population densities would be so low or the population would be so patchily distributed that population would be ecologically and functionally extinct.

Our habitat analysis predicted reductions in overall potential habitat available and degradation in habitat quality over time. Our population model predicted that SDT abundance in both populations would, on average, decline through time with some continual habitat degradation; however, there was wide variation in predicted outcomes (Figure 5). We ran the model under multiple future scenarios that varied maximum habitat availability based on potential habitat conditions and management actions, and we varied survival rates under different magnitudes of climate change effects. In all, we ran the model under nine different scenarios for each of the two populations. The model results also indicated that extinction probability was very low (< 0.02) under most scenarios over the next 100 y (Figure 5) and virtually no extinction probability within 50 y (USFWS 2015a). We predicted small population declines (measures of resiliency) due to some habitat loss and degradation and drought impacts from climate change. However, we predicted relatively small changes in the overall distribution of the species (measures of redundancy and representation).

In addition to an extensive written SSA report detailing population assessments, habitat modeling, and population viability modeling (USFWS 2015a), we held a 2-d interactive meeting with the analysis team and USFWS decision-makers to present the results of the SSA. In the meeting, analysts presented information on species biology, model structure, projection scenarios, and figures and tables of model output, and they explained their rationale for any scientific judgments. For example, we presented figures depicting the future median population size, the 95% confidence interval of population size, and the proportion of trajectories that declined to the quasi-extinction threshold for 18 different future scenarios (Figure 5). The analysis team responded to decision-makers' clarifying questions and explained the many areas of uncertainty, using figures and projection scenarios. The decision-makers settled on a foreseeable future of 50 to 75 y and considered predicted changes in abundance, distribution and diversity over that time frame. The decision-makers recommended that SDT was not warranted for listing under the ESA as a threatened or endangered species, and it did not receive ESA protection (USFWS 2015b).

We presented case studies to illustrate how the SSA process can be implemented. The spatial distributions for both EMRs and SDTs cover multiple states and cross international borders and some population data were available. For EMRs, representative areas were identified by three genetically informed regions (Ray et al. 2013). Resiliency was assessed using population-specific modeling; EMR condition was projected at 10, 25, and 50 y under the scenario of a continuation of existing stressors and conservation efforts. Redundancy was evaluated by assessing the likelihood of losing all populations within a representative area due to catastrophic drought. The analysts gave decision-makers the opportunity to run additional scenarios to describe the full risk profile for the species. Going through the SSA iteratively can reveal that the scenarios had not sufficiently captured the risk profile, suggesting additional scenarios for analysis.

For SDTs, evidence indicated an absence of strong genetic or population structure. However, exposure to threats and management strategies differed across the international border into Mexico. Thus, we structured the assessment within representative areas defined by the international border. We assumed the density of adult tortoises to be a function of habitat category as influenced by precipitation, wildfire, and urban expansion. We used a demographic model to project SDT abundance and quasi-extinction during a 200-y time span for combinations of drought, urban development, and conservation effort scenarios.

Although the case studies were in the context of listing decisions, the intent of the SSA is to develop a scientific analysis, which decision-makers can then use as a basis for informing the various ESA decisions (Figure 1). Much of the information of an SSA represents the state of knowledge of a species and its ecology, which would change with new data but not with a particular decision. However, from a decision analysis perspective, the SSA predicts the consequences that arise from the decision options, which depend on the particular ESA decision (Runge 2011; Gregory et al. 2013). For example, the decision to protect a species under the ESA relies on an assessment of the likelihood of extinction assuming levels of stressors and conservation efforts without ESA protections in place (Doremus 1997; Waples et al. 2013). In contrast, for an already protected species, decisions relate to recovery planning or consultation and permitting (Figure 1; Steiger 1994; McGowan 2013). Recovery planning (Table 1) requires a comparison of actions to identify the set of actions or strategy that offers the best chance to reduce species risk to a level that recovers the species (Boor 2013; McGowan et al. 2014). Interagency consultation on federal actions and permitting of nonfederal actions (cf Consultations and Permits in Table 1) requires an evaluation of species risk with and without a proposed project (Runge et al. 2008; McGowan and Ryan 2010). Thus, an SSA adapts to the decision context (Figure 1). Scenarios can be used to adapt an SSA to a particular decision context. For example, scenarios for interagency consultation compare the response of a species to alternative project designs and conservation measures in combination with future threats. Scenarios for recovery planning incorporate alternative recovery actions to identify those that will most likely achieve the goal of species recovery.

The cost of completing an assessment is an important consideration regardless of regulatory context because of workload relative to available agency capacity (Rohlf 2004; Murphy and Weiland 2016). The SSA explicitly analyzes the response of a species to stressors and conservation, which was not always included in the past threat-focused process. We suggest that the additional analytical demands will be at least offset by the efficiencies produced by 1) relying on the analysis for multiple decisions as the SSA follows the species and 2) helping to defend decisions due to the improved consistency and transparency of the supporting science. In our experience, the initial SSA usually takes longer than the past threat-focused process, but the SSA report is available for future decisions on the same species (Figure 1). However, efficiency comparison of the past process and the SSA is not straightforward because science and policy were often conflated in the past, resulting in more extended Federal Register notices and allowing for few opportunities for input by outside experts. While the SSA completed before an endangered species designation may take longer than subsequent updates, it presumably leads to improved analyses, extensive expert input, clear decision processes, and short Federal Register notices. Nevertheless, insufficient institutional capacity can impede the use of best available science in decision-making (Burgman 2015; Lowell and Kelly 2016; Murphy and Weiland 2016). Solutions to insufficient capacity include building analytical capacity through hiring, training, and collaboration with science institutions (Burgman 2015) and by ensuring that analytical practices are applied efficiently. It is reasonable to expect that, all else equal, the scientific rigor of an SSA is a direct function of analytical capacity relative to the demand on time, effort, and expertise; this capacity per demand relationship underlies the conclusions reached by Lowell and Kelly (2016). In our experience, the most demanding SSAs, which are associated with high data availability, wide range, and spatial complexity and extent of stressors, constitute a minority, perhaps about 10%, of the workload. In the interest of parsimony, an SSA, especially for low- or moderate-complexity situations, can characterize risk on a categorical rather than on a continuous scale and can reduce the number and complexity of future scenarios. However, simplification of an SSA potentially compromises scientific and legal defensibility.

The SSA process is based on methods for risk assessment and scenario planning that have been developed over the past decades (Shaffer 1981, 1987; Carroll et al. 1996; Akçakaya and Sjögren-Gulve 2000; Shaffer and Stein 2000; Peterson et al. 2003; DeMaster et al. 2004; Keith et al. 2004; Duinker and Greig 2007; Runge et al. 2007, 2008; Patrick et al. 2008; McGowan and Ryan 2009; ; Fordham et al. 2013; McGowan 2013; Doak et al. 2015; Wolf et al. 2015; IPBES 2016; Murphy and Weiland 2016; Phillips-Mao et al. 2016). Fundamental to all ESA determinations is a basic understanding of the ecology, current condition, and future condition of a species, which is what the SSA is designed to provide. Importantly, the SSA does not stop at an assessment of threats but moves to answer the next natural question: What do the projected stressors and conservation efforts mean for the future condition of the species or its risk of extinction?

Recognizing the distinct roles of science and formal policy (e.g., application of regulatory standards) is a prerequisite to providing a transparent and consistent species assessment (Doremus 1997; Ruhl 2004; Doremus and Tarlock 2005; Gregory et al. 2013; Waples et al. 2013). The SSA results in a scientific report to the decision-maker about the condition of a species, and then the decision-maker applies the policy to make the decision. Conflating the roles of science and policy can create unnecessary confusion both within the agencies charged to make ESA decision and with the public and partners who are affected by those decisions (Waples et al. 2013; Boyd et al. 2016).

While it is important to distinguish the roles of science and policy, it is essential that the scientific information matches the decision context. When decision-makers understand the analytical processes and results then the assessment will be more useful in making ESA decisions. Therefore, regular communication between the decision-makers and the analysts to achieve a common understanding of metrics, future scenarios, time frames, and implications of uncertainty helps to ensure that the assessment is informative to the decision-makers. Any scientific analysis involves judgments, which determine underlying assumptions and parameters (Doremus and Tarlock 2005). For example, confidence levels, quasi-extinction thresholds, and stressor levels to define scenarios are judgments to be evaluated, peer-reviewed, and communicated to the decision-maker with the aim of providing a transparent assessment. In the case studies, future conditions under multiple quasi-extinction thresholds and the rationale underlying the choice of population metrics were communicated to the decision-makers.

An SSA empirically evaluates species risk (Table 2). However, whether decision-makers deem that level of risk to be unacceptably high, leading to ESA protection, is an inherently normative determination (Doremus 1997; Vucetich et al. 2006). Consistency and transparency in ESA decisions emerge from two sources: the scientific analysis and the policy application. In the context of ESA determinations, normative judgment is expressed through the ESA legislation, guidance for interpreting the ESA, agency practice, and clarifying case law (Rohlf 2004). In total, these policies provide an understanding of the ESA's regulatory standards, which are then applied to make the decision. The SSA affects only the consistency and transparency of the risk assessment and does not determine policy standards and definitions. Currently, the policies do not contain explicit standards for making management judgments, and policies can change over time.

In summary, the design of the SSA process is intended to improve the consistency and transparency of the scientific analysis of the available biological information to support policy-based ESA decisions. The degree to which the SSA represents progress can be gauged relative to the baseline, which is the earlier threat-focused analysis (Andelman et al. 2004). The SSA includes explicit analyses of the species' response to stressors through a description of the ecology, estimation of the current condition, and forecasts of the future condition under multiple scenarios. Decision-makers apply the policy-guided interpretation of the ESA to the SSA results to make ESA determinations. Both science input and policy application contribute to consistency in ESA decisions. The SSA results in a scientific report distinct from policy judgment, which contributes to streamlined, transparent, and consistent decision-making and allows for greater technical participation by experts outside of the USFWS. As a consequence, we believe the SSA provides better scientific analysis that will in turn improve ESA decisions.

Reference S1. Faust L, Szymanski J, Redmer M. 2011. Range wide extinction risk modeling for the eastern massasauga rattlesnake (Sistrurus catenatus catenatus). Alexander Center for Applied Population Biology, Lincoln Park Zoo, Chicago, Illinois.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S1 (2430 KB PDF); also available at https://www.researchgate.net/publication/299645160_Range-wide_Extinction_Risk_Modeling_for_the_Eastern_massasauga_Rattlesnake_Sistrurus_catenatus_catenatus_Technical_Report_to_the_US_Fish_and_Wildlife_Service?showFulltext=1&linkId=5703be2908ae646a9da9bf00 (2,431 KB PDF).

Reference S2. [IPBES] Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. 2016. Summary for policymakers of the methodological assessment of scenarios and models of biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. Ferrier S, Ninan KN, Leadley P, Alkemade R, Acosta LA, Akçakaya HR, Brotons L, Cheung W, Christensen V, Harhash KA, Kabubo-Mariara J, Lundquist C, Obersteiner M, Pereira H, Peterson G, Pichs-Madruga R, Ravindranath NH, Rondinini C, Wintle B, editors. Secretariat of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services, Bonn, Germany. 32 pp.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S2 (10989 KB PDF); also available at http://www.ipbes.net/sites/default/files/downloads/pdf/SPM_Deliverable_3c.pdf (10,990 KB PDF).

Reference S3. Rowland ER, Cross MS, Hartmann H. 2014. Considering multiple futures: scenario planning to address uncertainty in natural resource conservation. U.S. Fish and Wildlife Service, Washington, D.C.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S3 (12714 KB PDF); also available at http://www.cio.noaa.gov/services_programs/prplans/pdfs/ID232_Final_Product_Green_Turtle_Technical_Memorandum_SWFSC_No_539.pdf (12,715 KB PDF).

Reference S4. Runge MC, Sanders-Reed CA, Fonnesbeck CJ. 2007. A core stochastic population projection model for Florida manatees (Trichechus manatus latirostris). U.S. Geological Survey Open-File Report 2007-1082.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S4 (544 KB PDF); also available at https://pubs.er.usgs.gov/publication/ofr20071082 (545 KB PDF).

Reference S5. Seminoff, JA, Allen CD, Balazs GH, Dutton PH, Eguchi T, Haas HL, Hargrove SA, Jensen MP, Klemm DL, Lauritsen AM, MacPherson SL, Opay P, Possardt EE, Pultz SL, Seney EE, Van Houtan KS, Waples RS. 2015. Status review of the green turtle (Chelonia mydas) under the U.S. Endangered Species Act. NOAA Technical Memorandum, NOAANMFS-SWFSC-539, National Marine Fisheries Service, La Jolla, California.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S5 (6467 KB PDF); also available at http://www.cio.noaa.gov/services_programs/prplans/pdfs/ID232_Final_Product_Green_Turtle_Technical_Memorandum_SWFSC_No_539.pdf (6,468 KB PDF).

Reference S6. U.S. Department of the Interior. 2009. The meaning of “Foreseeable Future” in Section 3(20) of the Endangered Species Act. Solicitor memorandum M-3702, Office of the Solicitor, Washington, D.C.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S6 (886 KB PDF).

Reference S7. U.S. Environmental Protection Agency. 2011. Expert elicitation task force white paper. U.S. Environmental Protection Agency Science Policy Council, Washington, D.C.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S7 (1561 KB PDF); also available at http://www.fws.gov/endangered/esa-library/pdf/ESAall.pdf (1,562 KB PDF).

Reference S8. [USFWS] U.S. Fish and Wildlife Service. 2015a. Species status assessment for the Sonoran Desert tortoise. Version 1.0, September 2015. Southwest Region U.S. Fish and Wildlife Service, Albuquerque, New Mexico.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S8 (18235 KB PDF); also available at https://www.fws.gov/southwest/es/arizona/Documents/SpeciesDocs/SonoranTort/SDT_SSA_Report_September2015_Chp1-3.pdf (18,236 KB PDF).

Reference S9. [USFWS] U.S. Fish and Wildlife Service. 2016a. Long-term listing transformation and five guiding principles of the Unified Listing Team. Director's memorandum 16 March 2016, Washington, D.C.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S9 (97 KB PDF).

Reference S10. [USFWS] U.S. Fish and Wildlife Service. 2016b. USFWS species status assessment framework. U.S. Fish and Wildlife Service, Falls Church, Virginia.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S10 (1326 KB PDF); also available at https://www.fws.gov/endangered/improving_ESA/pdf/SSA%20Framework%20v3.4-8_10_2016.pdf (1,327 KB PDF).

Reference S11. [USFWS] U.S. Fish and Wildlife Service. 2016c. Species status assessment for the eastern massasauga rattlesnake (Sistrurus catenatus). U.S. Fish and Wildlife Service, Falls Church, Virginia.

Found at DOI: http://dx.doi.org/10.3996/052017-JFWM-041.S11 (2465 KB PDF); also available at https://www.fws.gov/midwest/endangered/reptiles/eama/pdf/SSAFinalV2July2016EMR.pdf (2,466 KB PDF).

We thank Greg Breese, Debby Crouse, Jonathan Cummings, Cat Darst, Beth Forbus, Carey Galst, Craig Hansen, Michael Horton, Donald Imm, Michelle Kissling, Mary Parkin, Mark Pavelka, Erin Rivenbark, Steve Morey, Marjorie Nelson, Tara Nicolaysen, Laura Ragan, Angela Romito, Justin Shoemaker, Jesse Wild, and all additional members of the SSA Framework Implementation Team for their contributions to the development of the SSA. We thank Jeff Servoss, Michael Martinez, Shaula Hedwall, Brian Woolridge, Katie Boyer, Richard King, Frank Durbian, and Trisha Crabill for their contributions to the assessment case studies. We also thank Resit Akçakaya, Jean Cochrane, Jonathan Cummings, Holly Doremus, Marty Miller, Steve Morey, Jeff Newman, Mike Runge, Mark Shaffer, Robin Waples, Rollie White, an anonymous reviewer, and the Associate Editor for helpful comments on earlier versions of this paper. U.S. Fish and Wildlife Service funded this work.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.