Prioritizing populations based on recovery potential

For wide‐ranging species, it is often too expensive or politically challenging to effectively implement conservation action across their range. In these cases, conservation actions may be vigorously applied where the situation appears most dire, but inadvertently at the expense of where success is more probable. Consequently, it is prudent to use a prioritization approach that highlights areas of probable success. Using Southern Mountain Caribou as a target species, we develop a simple algorithm that integrates scaled habitat quality measures and population characteristics known to affect the demographics of caribou and weights them according to their relative importance as defined by expert opinion. The algorithm ranks subpopulations by their relative conservation status and, as a result, how likely they are to respond to additional conservation efforts and contribute to long‐term species persistence. Sensitivity analyses are then used to measure the implications of variance among key criteria and the potential variance in expert weighting. The transparent method quickly allows for real, or potential changes in criteria values, scaling, or their relative weighting, thus providing a baseline metric for conservation discussion, subpopulation comparisons, and adaptive management action. A web‐based application of the algorithm can be used directly or adapted for other species. This transparent framework can be used by conservation scientists and managers for prioritizing populations for receiving recovery actions to maximize long‐term conservation impact.


| INTRODUCTION
The ultimate goal of conservation is to maintain the diversity of life on earth. For species, this means preventing extinctions, maintaining viable populations, and where possible, recovering depleted populations. However, applied conservation remains primarily a defense game, focused on preventing the progression of extinction. As the culmination of several sequential extirpations, extinction repeats a familiar pattern where the rate of population decline increases as populations fragment into isolates. These fragmented populations are then subject to the initial causes of decline; exacerbated by Allee effects, their rates of decline accelerate as the populations shrink (Berec et al., 2007;Courchamp et al., 1999;McLellan et al., 2010). For conservationists and wildlife managers, this progression incites a sense of increasing urgency for action. When coupled with intense loss-aversion-bias (Kahneman et al., 1991) felt by the interested public (Mao et al., 2020), and likely the managers themselves, the default conservation strategy is to focus efforts and often limited resources, toward avoiding local extirpations starting with the most impending.
Paradoxically, by consistently focusing efforts on populations with the direst need of protection, we are also likely to invest heavily in populations with the highest probability of extinction, sometimes at the expense of those with a better opportunity for recovery. Conservation is expensive, politically challenging, and usually requires significant economic, social, and scientific costs (Adams et al., 2010;Polasky, 2008;Schneider et al., 2010). Even when an ecological system is well understood and recovery is predicted to be successful, implementation is subject to several social and economic caveats (e.g., Environment Canada, 2014). As a result, considerable thought has been directed at accounting for the probability of success in the cost of conservation action (Bottrill et al., 2009;Joseph et al., 2009;Naidoo et al., 2006;Schneider et al., 2010). When the costs of conservation are too high to be effectively distributed across the range of a species, prioritizing recovery efforts is necessary. For a species, an effective prioritization strategy will maximize the probability of conservation success given the reality of the population's status (e.g., population size and trend), degree of threats to recovery, and opportunity costs (social and economic).
Integrating subjective criteria, their relative importance, and the underlying data is the first step toward engaging multiple interest groups in conservation decisions (Failing et al., 2013). While such optimization and quantitative targets provide defensible outcomes (e.g., IUCN Species Survival Commission, 2012), decision-making is necessarily subjective because of complex socio-political contexts and biological uncertainty (Failing et al., 2013). To make the decision-making process more objective, multiple criterion approaches have been used to identify priority areas for conservation actions, and to explore how subjectivity and uncertainty impact decisions (Sutti et al., 2017). Multiple criterion approaches are highly valuable as they enable decision makers to integrate often complex criteria and data into an organized decision support framework where rules, data, and subjective values (weights) used to inform decisions are made explicit thus making the decision process more systematic versus purely subjective (Huang et al., 2011). Furthermore, a system-specific decision support framework may then be easily adapted for a variety of potential conservation strategies in the same system. For example, the same framework could be applied for spatially prioritizing habitat securement and alternatively for habitat restoration efforts while the specific criteria selected, and their relative weighting may shift to suit the specific conservation actions to be implemented.
Southern Mountain Caribou (Rangifer tarandus caribou) are a threatened population of woodland caribou that live in the interior mountainous areas of British Columbia and the west-central Rocky Mountains of Alberta, Canada. Once a contiguous population, they are now delineated into three groups reflecting ecological and evolutionary differences: Northern, Central, and Southern (COSEWIC, 2011). Caribou in the Northern and Central Groups live in low snowfall areas, primarily foraging on terrestrial lichens in winter. Caribou in the Southern Group live in high snowfall, rugged mountainous areas at the southern extreme of the species range (Stevenson & Hatler, 1985). They are dependent on oldgrowth coniferous forests that vary in age from hundreds to over a thousand years old. Following a continuous decline over the last century, and a precipitous decline in recent decades, caribou in the Southern Group are Endangered and $1250 remain (Province of British Columbia, 2021b). Despite federal and provincial recovery objectives for achieving self-sustaining caribou populations, five of the 17 sub-populations have become extirpated since 2003, three have fewer than 10 caribou, and most have historic declines (Province of British Columbia, 2021b).
Caribou population decline has been attributed to landscape modification increasing the prevalence of early seral forests and resulting in amplified predation rates via altered inter-trophic relationships (Wittmer et al., 2007). Predator density is mostly dependent on the density of their primary prey such as moose (Alces alces) and whitetailed deer (Odocoileus virginianus), which thrive in early seral forests. Increased populations of primary prey resulting from landscape alteration, mostly forest harvest, have led to increased apparent competition with caribou Serrouya, Kellner, et al., 2017;Wittmer et al., 2005). As a result, habitat alteration in core habitats, where caribou feed and live, and in interspersed matrix habitats, directly affects caribou survival because it influences the predator-prey system that operates at a much broader scale. A warming climate has also driven the expansion of white-tailed deer distribution northward (Dawe & Boutin, 2016), increasing predator density in its wake (Latham et al., 2011). Furthermore, increased linear features such as forestry roads and transmission lines have accelerated the ingress of predators and predation rates (Dickie et al., 2017(Dickie et al., , 2022Whittington et al., 2011). Considering only ecological parameters, the recovery of subpopulations in the Southern Group is contingent on (1) the degree and extent of habitat degradation from landscape alteration in core and surrounding matrix habitats which affect the populations of primary prey species and predators (Lochhead et al., 2022;Wittmer et al., 2007); (2) the degree and extent to which habitat alteration is permanent, for example, agricultural areas (Apps & McLellan, 2006); (3) the age and extent of non-permanent or temporary habitat alteration such as forest harvest and wildfire; (4) degree that populations are connected to larger, more stable populations, thereby providing meta-population function and intermittent demographic or genetic rescue (Charlesworth & Willis, 2009); (5) the current population size and trend (Middleton & Nisbet, 1997;Wittmer et al., 2010); and (6) the latitude of the population reflecting its relative susceptibility to changing climate and associated increase in white-tailed deer density (Dawe & Boutin, 2016).
Recent attention to the accelerating habitat loss across Southern Mountain Caribou range (Nagy-Reis et al., 2021;Palm et al., 2020), and renewed public pressure to conserve old-growth forests (Hunter, 2021) often motivated by indigenous reconciliation, has spurred governments to consider and implement additional habitat protections (Government of Canada, 2020). Beginning in the 1990s, land-use planning resulted in increased protection of older forests within the Southern Group; however, these measures were focused on core habitat and were insufficient to produce self-sustaining caribou populations across the entire range . Recent pressures for old-growth forest protections have led governments to propose the deferral of old-growth forest harvest across the province (Province of British Columbia, 2021c), temporarily increasing habitat protection for mountain caribou. The possibility of additional old-growth protection incentivized us to develop an evidence-based system for identifying locations where old-growth protection will have the maximum benefit to Southern Mountain Caribou.
Considering the complexity of the systems, the variation in specific threats among subpopulations, and the broad extent of Southern Mountain Caribou range, understanding the relative peril of each subpopulation is necessary for recovery planning and prioritizing conservation action (Bottrill et al., 2009). Here we develop a simple additive algorithm in a multiple-criterion approach that integrates scaled measures of habitat quality and population characteristics known to affect the dynamics of mountain caribou so that populations can be systematically compared and ranked according to ecological status. The algorithm allows for criteria to be weighted according to expert opinion and the literature to account for the unequal importance of each criterion on subpopulation stability and recovery potential. In this application, criteria weights were assigned with the specific objective to identify where additional habitat protections would most likely benefit caribou. Because total habitat gains from protection would be consistent regardless of where they were applied, the implied assumption is that costs associated with fixed additional habitat protection are also consistent among subpopulation areas. By quantifying each criterion and its weighted contribution we determine the sensitivity of the outcome to each criterion or weighting decision. Finally, the flexibility of the tool quickly allows for simple updates to the weighting of criteria as well as changes in landscape composition and population characteristics thereby simultaneously providing a measurable yardstick for conservation improvement and adaptive management. In circumstances where political, social, or economic resources are limited this tool provides a systematic, repeatable, and adaptive means of ranking populations so that resources can be allocated to where conservation efforts are most likely to succeed.

| METHODS
We considered the Southern Group of Southern Mountain Caribou populations for developing our population status assessment and prioritization strategy. We used provincially defined subpopulation boundaries (Province of British Columbia, 2021a) resulting in the delineation of 17 subpopulations ( Figure 1; Table 1). Subpopulations that were extirpated since 2005 were included because there remains interest in their eventual recovery (Province of British Columbia, 2021b).

| Criteria development and weighting
For each subpopulation, several population and landscape characteristics were considered as potentially important for determining its status and recovery potential including: (1) subpopulation size, where larger sizes are associated with increased probability of persistence (Middleton & Nisbet, 1997). Population sizes were obtained from the most recent subpopulation surveys (2020-2022) using consistent methods (Table 1; Wittmer et al., 2005). (2) Proximity to large, protected areas where both core and matrix habitats are conserved because adjacent caribou populations benefit from intact habitat (in this case Wells Gray Provincial Park N 52.10 , W120.00 ; Figure 1); measured as the distance between subpopulation and park centroids. (3) Latitude at the center of the population range, favoring more northerly populations as they are less likely than southerly populations to be affected by expanding white-tailed deer populations (Dawe & Boutin, 2016;Laurent et al., 2021); measured as meters north using Universal Transverse Mercator (UTM) projection system for assigning coordinates (Table 1, Figure 1). (4) Population areas with a high level of protection in the Timber Harvesting Land Base (THLB; Province of BC), hereafter referred to as habitat protection, which includes areas with merchantable timber regardless of the current seral stage but excludes lowproductivity sites that are uneconomical to harvest (Wittmer et al., 2007). As a result, subpopulations with higher levels of extant habitat protection are assumed to have lower future habitat alteration than those with less protection; these subpopulations are more likely to recover (Table SI-1.1). (5) Habitat alteration was assumed to reduce recovery potential in proportion to the relative area altered within each subpopulation; data were obtained from federal habitat disturbance mapping that spanned western Canada caribou ranges (Canadian Wildlife Service, 2012). We assumed that social and economic opportunity costs are consistent among subpopulation areas, partly because our system lacks consistent, qualitative data describing these costs. However, where quantitative socio-economic data are available, they could be incorporated as additional criteria. Accounting for social and opportunity costs is likely closely associated with the specific conservation action in conservation efforts.
Due to the varying effects of habitat alteration age, habitat type, and whether habitat was expected to recover, we further stratified our criteria so that weighting could be differentially applied among these subcriteria. First, habitat alteration was defined as permanent or temporary, where permanent habitat alteration was defined as alteration that is not expected to recover, including agriculture, urban areas and residences, roadways, mining activities, and transmission lines. Temporary habitat alteration includes non-permanent changes: wildfire, pest-affected areas, and forest harvest, which are expected to recover. To allow for variable weighting by the age of temporary habitat alterations, these were further delineated into (1) habitat alteration <40 years which is marginal caribou habitat (Mumma et al., 2021;Serrouya et al., 2011); (2) habitat alteration <10 years indicating prolonged early-seral vegetation supporting moose populations and, as a result, associated with increased predator densities . Finally, we used core and matrix definitions developed for the Canadian Federal Recovery Strategy (Environment Canada, 2014) and delineated Provincially (Province of BC data) to further categorize both alteration and protected areas into core and matrix habitats (Franklin & Lindenmayer, 2009). Core habitats are defined as the area that encompasses the annual range of a subpopulation and provides sufficient suitable range to support life history requirements and matrix habitats are areas adjacent to core habitat that, when altered, support primary prey populations and associated predators which affect the caribou subpopulation (Environment Canada, 2014). This resulted in 34 potential population and landscape criteria including 7 potential criteria where protected forest or habitat alteration criteria were pooled by habitat types (core and matrix) and/or alteration types (Table SI- Area-based criteria were simply the proportion of area within each subpopulation boundary. For example, the total disturbed area was the sum of the total disturbed area in the matrix and core habitats as a fraction of the total area. Non-area-based criteria, including population size, latitude, and distance to Wells Gray Park, were normalized across subpopulations so that weighting can be applied without consideration of absolute criteria values; features with small values will have similar importance to features with very large values when the algorithm is run. To normalize a criterion, the minimum value of a criterion was deducted from the raw value, and then T A B L E 1 Subpopulations of Southern Mountain Caribou considered for prioritization, most recent population size estimates from most recent provincial population surveys (BC Government, 2020, the latitude of the subpopulation defined as the UTM meters north at the center of the subpopulation, and the area of the subpopulation. divided by the difference between maximum and minimum criterion values, thus ensuring that all criteria in the data set range from 0 to 1. Area-based criteria were not normalized to maintain the ability to directly compare criteria among subpopulations. However, in some circumstances, users may prefer to normalize all criteria to homogenize variation among all criteria. To satisfy this nuance, we repeat the same process with all criteria normalized, including area proportions, and report these methods with associated results in Supporting Information SI-2, and as a toggle option in the web application (Prioritization Shiny). Criterion weighting was determined using the authors' professional opinions based on the literature but could be changed by users of the web application (Table SI-1.1; Prioritization Shiny). For Southern Mountain Caribou, permanent alteration in both core and matrix habitat, especially agricultural land, is associated with caribou population decline, as is forest harvest (Apps et al., , 2013Apps & McLellan, 2006;Lochhead et al., 2022;Serrouya et al., 2021;Wittmer et al., 2007). Recent wildfire also has negative effects on caribou, however, due to the lack of direct association with linear features the impact is less severe (Johnson et al., 2020;Stewart et al., 2020;Wittmer et al., 2007), therefore was weighted half that of permanent or harvest alteration. Larger population sizes and relatively short distances to large, protected areas were both weighted positively (Elmhagen & Angerbjörn, 2001;van Oort et al., 2011). Criterion weights can take any value from 0 to 100 but are constrained to sum to 100 to retain comparability among algorithm iterations. Some factors considered as potential criteria but were not included in this iteration of the algorithm by the authors were given a weighting of 0. Applying correlation matrices to compare selected criteria allows users to identify the degree to which criteria may have similar effects on ranking, where redundancies exist, or where the nuances between highly correlated criteria are desirably reflected when applying subjective weighting (Table SI-1.1).

| Prioritization algorithm
The relative likelihood that a population would recover because of additional habitat protections was estimated by the equation: where H i is the rank score in population i, ω j is the weight of criterion j, and criterion C ji is the value of criterion j in population i. For criteria where high values are considered detrimental to caribou recovery, such as core habitat alteration, the inverse (1 À C) is used before summation.

| Algorithm sensitivity
Recognizing the potential subjectivity of the weightings to variation in expert knowledge, and potential changes in population status or landscape composition, we assessed the sensitivity of the rank outcome to variation in criterion value and weighting. Sensitivity to weighting variation was evaluated by allowing weights to vary between 0 and 5 for each criterion but constrained so that each value is resampled with equal probability across all criteria and so that each iteration sum to 100, as specified in the algorithm design to retain comparability among alternative iterations. We then bootstrapped the algorithm 10,000 times to obtain the frequency distribution of rank outcomes for each population.
The effect of a criterion is directly related to the variance of criterion values among populations as well as the variation in the weighting scale. We assessed the sensitivity of the rank outcomes to each criterion by holding the weighting of all but one criterion constant and iteratively changing the weight of that criterion to 0, 5, 10, and 15 and calculating the rank outcomes of the populations for each iteration. This was repeated across all criteria and the absolute difference between each rank output and the mean rank for that population was calculated. We then calculated the mean variation in rank outcome for each criterion across all populations. All analyses were conducted using program R (V.4.1.2; R Core Team, 2021).

| RESULTS
Our algorithm to prioritize populations for habitat protection identified the Hart Ranges, Wells Gray North, and Redrock-Prairie Creek as the populations most likely to benefit from additional habitat protection considering their current ecological status (Figure 2). The most southern and recently extirpated populations, South Selkirks, Purcells South, and Purcells Central were the lowest-ranked populations (Figure 2). A large difference in output values between sequentially ranked subpopulations indicates more confidence (i.e., less variation) in potential rankings. For example, Monashee and Purcells Central populations had large differences between them as did Hart Ranges and Wells Gray North Populations, suggesting large differences between subsequent rank outcomes ( Figure 3) and F I G U R E 2 Southern Mountain Caribou population prioritization rank according to expert-based weighted prioritization algorithm for determining where further old-growth conservation would be most beneficial. Input criteria include (a) population size, (b) distance to Wells Gray Park protected area (Photo: wellsgraypark.info), (c) latitude, (d) forest harvest alteration, (e) fire alteration, (f) agricultural land (Photo: maplogs.com), (g) permanent roadways (Photo: britishcolumbia.com). Permanent and transient alteration are further delineated into alteration in core and matrix habitat (h). Transient alteration is also delineated by age of regeneration as can be seen in panel d. Criteria are weighted according to expert opinion to produce a prioritization output. See Figure 1 caption for unabbreviated subpopulation names.
F I G U R E 3 Difference in algorithm value between subsequent ranks (purple) and cumulative algorithm value (teal). For example, the Hart Ranges is the top-ranked Southern Mountain Caribou population thus the basis for comparison for all other populations (teal) or the next-ranked population Wells Gray North (purple, right). Small differences in algorithm value between a population and the next highest rank indicate little difference in algorithm output (e.g., Redrock-Prairie Creek and Wells Gray South). See Figure 1 caption for unabbreviated subpopulation names. increased confidence rank outputs. Smaller differences between rank outcomes, for example, Narrow Lake and Columbia South, or Purcells South and South Selkirks, indicate more similar algorithm outputs and thus more equivocal ranking (Figure 3).
The sensitivity of rank outcomes to weighting variation was highest in the Wells Gray North and the Hart Ranges where the mean variation in ranking from the average rank across all weighting iterations was 1.59 and 1.58, respectively. Both populations showed skewed frequency distributions in their rank outcomes across variable weighting iterations indicating that in some weighting combinations these populations would have much lower ranking outcomes however, it is unlikely that professionally based weighting would be simulated by random weighing inputs (Figure 4). Nevertheless, this outcome suggests that variation in weighting would most likely affect these populations. Frisby-Boulder and Redrock-Prairie Creek varied in mean rank by just 0.62 and 0.63, respectively, despite varying weights (Figure 4), indicating that the rank of these populations was less sensitive to differences in expert weighting.
The largest variation in scaled criteria values were latitude, distance from Wells Gray Park, and population size, in general, these resulted in a higher influence on the rank outcome ( Figures SI-1.1 and SI-1.2). Iteratively varying weights across all criteria showed that the distance to Wells Gray Park had a large effect on rank outcome particularly if this criterion was not considered at all or ranked very highly. The proportion of habitat protected also had potentially large effects on the outcome, as did the temporary habitat alteration, especially in matrix habitats. Forest harvest alteration did affect rank outcome variability, but its effect was less than one rank point on average ( Figure SI-1.2).

| DISCUSSION
We quantified relevant population and habitat quality attributes for 17 Southern Mountain Caribou populations F I G U R E 4 Priority ranking frequency from bootstrapped algorithm iterations with varying criteria weights for each Southern Mountain Caribou population considered in the analysis. Panels are ordered from highest to lowest priority outcomes from expert weighted algorithm. Heart Ranges (HaRa), Red-Rock Prairie Creek (RRPC), North Caribou (NoCa), Narrow Lake (NaLa), Barkerville (Bark), Wells Gray North (WeGN), Wells Gray South (WeGS), Groundhog (Grou), Columbia North (CoNo), Central Rockies (CeRo), Columbia South (CoSo), Frisby Bouder (FrBo), Monashees (Mona), Central Selkirks (CeSe), Purcells Central (PuCe), Purcells South (PuSo) and South Selkirks (SoSe). and, considering the relative importance of each attribute, we developed an algorithm to rank the populations according to their likelihood of recovery. Our goal was to help decision-makers answer questions like "where would we allocate any new old-growth habitat protection for mountain caribou?" We identified the Hart Ranges as the population with the highest potential for recovery. It is the most northern and largest population with $32% of the remaining mountain caribou and has a stable trend (Table 1; Province of British Columbia, 2021a). The Wells Gray North population was ranked second. The high ranking of this population was influenced by its proximity to Wells Gray Park and the relatively high level of habitat protection overall, however, this population also had among the highest habitat alteration in both core and matrix habitats. The lowest-ranking populations, South Selkirks, Purcells South, and Purcells Central (Figure 2), were penalized by their southerly latitudes, distance from Wells Gray Park, and recent extirpation status ( Table 1). The potential variation in rank outcome highlights the importance of considering both the independent and relative influences of population and landscape criteria on recovery potential.
Our method provides a transparent basis for comparing populations using consistent metrics as well as a platform from which to base management discussions and decisions. As knowledge expands, shifts in the importance of criteria can be incorporated and measured against the original algorithm. Future deviations from this baseline allows for comparing outcomes of recovery initiatives among populations and management treatments. However, the subjective nature of assigning weights to contrasting criteria is a double-edged sword. On the one hand, it lacks repeatability, and although the sensitivity analysis helps to address this shortfall by quantifying any magnitude of subtle individual tendencies in weightings, it remains subject to individual variation. On the other hand, the adaptability of weightings is an opportunity to quantify the effects of differing opinions and values on the prioritization of certain populations. A logical extension of this algorithm is to incorporate a mechanistic tool such as an integrated population model that explicitly links habitat attributes with population change (e.g. McNay et al., 2022). The algorithm would then less subjectively answer such questions as "given a fixed amount of incremental habitat protection, where would it best be allocated to maximize absolute population growth?" However, the disadvantage of such an approach is reduced flexibility and less opportunity for incorporation into land-use planning where diverse interests and stakeholders have influence. In instances where finer-scale prioritization initiatives are desired, the algorithm could be applied using pixel-based data thus allowing for variation in priority levels within the subpopulations.
Prioritizing populations for recovery provides landscape-level direction to scale up management activities across the range of Southern Mountain Caribou. For caribou subpopulations that were declining rapidly and then subject to recovery actions, the maximum increase in population growth was attained by the simultaneous implementation of multiple recovery actions and high levels of management intensity (Heard & Zimmerman, 2021;Lamb et al., 2022;McNay et al., 2022;Serrouya et al., 2019). In these cases, the efficacy of management actions was measured at the population level or smaller. Scaling up management beyond the population level may decrease conservation effectiveness by introducing setbacks commonly afflicting large organizations (Mills et al., 2019) or unforeseen constraints implicit in larger areas. Even if financial resources are sufficient, epistemological and administrative constraints in large organizations can cause communication breakdowns, lack of coordination, and loss of direction (Canback et al., 2006;McAfee & McMillan, 1995). Bureaucratic constraints may limit the ability to build and maintain broad-scale and long-term political will required to implement effective management until landscapes and populations have been restored sufficiently to relax intensive management efforts. Positive outcomes garner more public support for long-term conservation action than mixed outcomes (Decker et al., 2010;Orians et al., 1997). Like mediocre outcomes from broad-scale efforts, applying costly management efforts and opportunity costs to the populations less likely to succeed may also erode public confidence in conservation efforts and thwart future efforts in populations with a higher probability of success (Gerber, 2016). This is potentially even more important for old-growth dependent species like mountain caribou that require long-term recovery action to restore landscapes sufficiently and reduce more controversial management actions such as predator reductions. We argue that by concentrating management action on highly ranked subpopulations, we increase the likelihood of success in those subpopulations and ultimately leading to an overall improvement in the vigor of the Southern Mountain Caribou metapopulation.
Caribou range is broad, spanning areas with varying economies, politics, and interest in conserving caribou. A caveat of this algorithm is that it is ecologically driven and without considering local interests and the opportunity cost of management actions in each area. As a result, the probability of successful recovery may be subject to unforeseen challenges arising from varied sociopolitical barriers (Ban et al., 2013). In areas facing substantial opposition to management actions, this method does not account for overcoming socially driven obstacles or how they would affect prioritization. In the opposite circumstance, where recovery efforts are welcome and there is strong local interest, but ecological circumstances are less favorable for recovery, a population's probability of success may be underestimated. Considerations should be made for circumstances where the sociopolitical will is dissociated from the biologically derived priority. In some regions, local political will may be sufficient to drive caribou recovery initiatives in that area.
Our research points to a deficit in quantitative socioeconomic data that could assist in prioritizing conservation actions, were they available, they could easily be integrated into the weighting of ecological criteria or as additional spatially derived criteria. Accounting for costs would likely be closely associated with the specific conservation action applied; for example, reducing timber harvest would have different costs to different people and communities than restricting recreational activities would.
Most critiques of conservation prioritization are aimed at implementation at the species level, where prioritization or triage favors some species, potentially at the expense of others (Hayward & Castley, 2018). Critics argue it has the potential to signal that species loss is acceptable (Buckley, 2016) or that the economic valuation of species devalues their intrinsic worth (Justus et al., 2009) and induces generational inequalities (Weiss, 1990). However, in the case of Southern Mountain Caribou, where prioritization is among populations of the same species and the goal of conservation is to preserve the existence of a species in perpetuity, then prioritization may be a means to the same goal (Schneider et al., 2010). Nevertheless, subpopulation loss and range contraction represent the erosion of biodiversity as well as a diminution in ecosystem services (Ceballos et al., 2017;Sanderson, 2019), therefore prioritization should be framed as a solution for preventing additional loss rather than for dismissing the recovery of populations and for contributing to shifting baselines for conservation action and success (Rodrigues et al., 2019).
The situation for Southern Mountain Caribou remains perilous; 29% of subpopulations are recently extirpated, 18% have fewer than 10 individuals (Province of British Columbia, 2021b). Yet, despite these patterns, caribou recovery remains controversial (e.g., Schneider et al., 2010) because of their enormous range that spans billions of dollars of natural resources (Hebblewhite, 2017), and due to politically controversial predator, and/or alternative prey reductions needed to stop short term population declines. Implementing the necessary actions at the spatial and temporal scales to recover caribou is an enormous undertaking, especially all at once. When the costs of conservation are this high and dollars are messily entwined with emotions, small successes in building and maintaining public confidence are imperative to reaching broad-scale conservation. Likewise, when expensive interventions are unsuccessful, they erode public and political confidence and increase the likelihood of extinction. In this case, prioritizing recovery action to apply effective, consistent, and high-quality management action is not only the way to save some subpopulations, but also likely the only way to save them all.
The plight of Southern Mountain Caribou is not unique; on earth, more than a quarter of terrestrial large mammal species are at risk (Ceballos et al., 2005). Globally, there have been significant increases in policy responses to biodiversity loss such as increased protected areas, responses to invasive species, and sustainable resource investment (Stuart et al., 2010). Efforts to address biodiversity loss occur across many spatial and temporal scales, from protection of global biodiversity hot spots (Myers et al., 2000) to protection of species and local ranges (Schneider et al., 2010). The applicability of repeatable systematic approaches for supporting conservation action decisions, such as this one, is wide ranging. At any scale, public support for conservation is increased by making conservation decisions transparent and defensible.
AUTHOR CONTRIBUTIONS Michelle L. McLellan contributed to analyses and wrote the initial manuscript and led editing process. Melanie Dickie, Kathryn L. Zimmerman, Darcy Peel, Bevan Ernst contributed to research design, data collection and manuscript preparation. Stan Boutin and Robert Serrouya contributed to research design and manuscript preparation. Marcus Becker contributed to shiny application development.