Abstract
Decision analysis frameworks can be employed to deconstruct a complex problem or process into smaller components that can be readily understood and critically analyzed. These components can then be recombined to create a model of the process or problem which integrates qualitative and quantitative information from stakeholders and clarifies and communicates the decision framework to all parties. We apply simple decision frameworks to assessing the relative benefits and costs associated with conservation efforts of two species of endangered marine mammals. The first cases examines the utility of using satellite linked time depth recorders (SDRs) to study Maui’s dolphins (Cephalorhynchus hectori maui) in New Zealand, the second examines the utility of using acoustic deterrent devices to mitigate fisheries mortality of vaquita (Phocoena sinus) in Mexico. In both cases, risk-averse decision trees informed with data on the status of the species and techniques to be employed suggest that using some invasive methods and novel technologies with small populations of endangered marine mammals may not contribute significantly to conservation efforts. When applied to situations with more numerous populations, these methods are more likely to support conservation efforts. We conclude that decision framework analyses are useful tools for resolving contentious research and management issues associated with conserving small populations of marine mammal species at risk.
Introduction
Efforts to conserve species at risk often require field research to inform the process (Tracy & Brussard 1996) and the application of new technology and novel techniques. In many cases, this includes direct interactions with the animals in question. Some direct interactions are categorized as relatively benign, or “non-invasive” (Cuthill 1991), where animals are simply observed from a distance through techniques such as theodolite and video tracking (Denardo et al. 2001) and photo-identification (Markowitz et al. 2003). More “invasive” techniques (Cuthill 1991) are occasionally employed, which require capture and handling of animals and in some cases surgical procedures (Read & Gaskin 1985; Read & Westgate 1997; Westgate & Read 1998). While often providing significant insights into conservation problems, invasive techniques often carry with them a greater chance of influencing the system being studied (Cuthill 1991), and in some situations these effects may be at odds with the overall goals of the conservation project (de la Chenelière 1998). Similarly, conservation measures employing new technologies and novel techniques may not be universally applicable to all species and may result in counter-productive effects on the target system (e.g. Dawson et al. 1998). Balancing the benefits and costs of invasive work and the application of new technologies in conservation research is a complex task which integrates scientific uncertainties and human values. This may be especially true in situations where there is disagreement over which methods or techniques are appropriate, and where populations are small, threats significant and strong measures are required to conserve species.
Using a decision analysis framework to understand and direct conservation efforts can provide significant benefits in these situations (Maguire 1987). Through this technique, a complex problem or process can be deconstructed into components that are small enough to be readily understood and critically analyzed. These discrete components are methodically recombined to create a conceptual model of the process or problem which illustrates the possible events (including decisions, uncertainties and endpoints) and the relationships among them. The identification of the possible sequences and linkages of events clarifies the decision making process and provides a flexible and objective framework for identifying and evaluating explicitly the important components of such a process. This approach also provides a mechanism for establishing a common decision rule, such as maximizing expected utility or minimizing possible risks, which can be clearly understood by all stakeholders (Maguire 1987).
This technique was originally used in business management (Raiffa 1968) but is increasingly being applied in a number of scientific endeavours, including human medicine (e.g. Forrow et al. 1988), fisheries management (Harwood & Stokes 2003) and conservation biology (Maguire 1986; Peters & Marmorek 2001). In some cases the decision analysis framework to be employed is primarily qualitative, providing a visual roadmap of the various components of the process and treating model components and the relationships amongst them as broad categories and basic yes or no decisions (e.g. de la Chenelière 1998). In other situations, decision analyses can be highly quantitative, providing detailed statistical advice on how to proceed with management efforts (e.g. Peters & Marmorek 2001; Peters et al. 2001).
The present paper introduces two case studies which apply a simple decision framework to assessing the relative benefits and costs associated with conservation efforts for endangered marine mammals. We developed these trees as a compromise between simple conceptual decision models and highly quantitative decision analyses, generating models similar to those used in other endangered species situations (Maguire 1986; Maguire 1987). This approach produces quantitative decision trees which are useful for resolving complex problems while still remaining accessible to scientists, managers and other stakeholders alike. In the first case we assess the use of an invasive and novel technique, in this case satellite tagging, on the Maui’s dolphin (Cephalorhynchus hectori maui) – a species listed as endangered by the World Conservation Union (IUCN) (Slooten & Taylor 2000). The second case assesses the application of a new technology – acoustic deterrent devices (ADDs) or more commonly referred to as pingers – to reduce the by-catch of vaquita (Phocoena sinus), also listed as endangered by IUCN (IUCN 1996). In both cases, scientific uncertainty associated with the efficacy of the techniques involved can manifest in various interpretations of the appropriate course of action. As well, non-expert stakeholders exhibiting different value systems may disagree over the best course of action when considering which research techniques and conservation measures are appropriate. By applying a decision framework analysis to these situations we can assess the decisions and outcomes explicitly and choose an integrative course of action which will maximize conservation benefits and minimize negative effects on the species (Maguire 1986).
These decision trees represent one of many possible models for such processes and they are not necessarily prescriptive (Maguire 1987). Rather, these examples are designed to illustrate the power and flexibility of using such a framework to help clarify thought processes and direct efforts to conserve endangered marine mammals. As a counterpoint to the vaquita and Maui’s dolphin scenarios, we use the same decision tree frameworks to assess the applicability of each process to harbour porpoises (Phocoena phocoena) in the Bay of Fundy/Gulf of Maine, a relatively numerous population that is not currently listed as endangered or threatened.
Species Profiles and Conservation Controversies
Maui’s dolphin(C. hectori maui)
Maui’s dolphin is a subspecies of Hector’s dolphin (C. hectori hectori) which only occurs along the west coast of North Island, New Zealand (Figure 1A). Maui’s dolphins are among the smallest of dolphin species with females reaching a maximum length of 1.7 m – males are slightly shorter. Female Maui’s dolphins reach sexual maturity at 7- 9 years of age and subsequently produce a single calf every two to four years. Maui’s dolphins are often found in shallow inshore regions and they feed primarily on benthic an epi-benthic fishes. The latest abundance survey indicates that the Maui’s dolphin population numbers fewer than 100 individuals and that their range and abundance continues to decline (Ferreira & Roberts 2003). Hector’s dolphins are currently listed as Endangered in the IUCN Red List of Endangered Species (Slooten & Taylor 2000) and the Maui’s dolphin population itself classified as Critically Endangered (Ferreira & Roberts 2003). The greatest threat to Maui’s dolphins is by-catch in commercial set-net gillnet fisheries and because of their low abundance and reproductive rate all sources of anthropogenic mortality pose a significant threat. The New Zealand Department of Conservation’s Auckland Conservancy recently formulated a research program designed to “establish a self-sustaining population of Maui’s dolphin throughout their natural range along the west coast of the North Island of New Zealand” (Ferreira et al. 2003). The overall research plan included a number of potential studies, including assessing habitat use by individual dolphins through the use of satellite linked time depth recorders (SDRs). The use of these satellite tags requires that individual animals be captured at sea and SDRs surgically attached to their dorsal fins. While SDRs and other electronic tags have proven useful in assessing the behaviour and distribution of other small cetaceans (Westgate & Read 1998) , there are risks associated with capturing animals (stress and mortality) and risks associated with the surgical procedures. In situations where populations are small and each individual an integral part of the surviving population, it may become necessary to balance the potential benefits of tagging animals against the cost of harming or killing individuals through the conservation process itself.
Vaquita
The vaquita is a small phocoenid cetacean whose range is restricted to the extreme northern portion of the Sea of Cortez (Figure1B) (Vidal et al. 1999). Very little is known about the natural history of vaquita, and what is known is based on data collected from less than 50 individual animals (Vidal et al. 1999). Vaquita appear to be the shortest species of porpoise, growing to a maximum length of approximately 1.49 m (Vidal et al. 1999). Female vaquita appear to be larger than males, and reach sexual maturity when they achieve a length of approximately 1.35 m (Vidal et al. 1999). Females appear to produce one calf annually, usually in late March or early April (Vidal et al. 1999). Vaquita feed primarily on squid and teleost fishes that are commonly found in the demersal and benthic zones of the shallow waters of the upper Gulf of California (Vidal et al. 1999). Vaquita are not numerous; the latest abundance survey produced an estimate of 567 animals, with a 95% confidence interval of between 177 and 1073 individuals (Jaramillo-Legorreta et al. 1999). Vaquita are also currently listed in the IUCN Red List of Threatened Species as Critically Endangered. Vaquita are taken incidentally in small-mesh gillnets (D’agrosa et al. 2000) and the greatest threat to the remaining vaquita is incidental mortality in fishing gear (Rojas-Bracho & Taylor 1999). Because vaquita numbers are so low, any significant increases in annual mortality pose a serious threat to this species (Rojas-Bracho & Taylor 1999). Acoustic deterrent devices have been employed in some fisheries to reduce by-catches of marine mammals (Barlow & Cameron 2003). While useful for reducing by-catch, ADDs do not prevent all incidental mortality (Bordino et al. 2002). In species at risk situations where populations are small – like the vaquita case – every individual is important to the population and by-catch reduction may not be the best approach to conserving the species. Indeed, there is some debate on the appropriateness of using ADDs in small population situations (Dawson et al. 1998).
Methods
We constructed decision frameworks which model the process of: 1. using SDRs to study Maui’s dolphins and, 2. deploying ADDs in attempts to reduce bycatch of vaquita in commercial fisheries. This basic architecture incorporates information on the status of the species, the status of the techniques being applied, and includes probabilities associated with the likelihood of various effects that may manifest during either process. We recognize that this list is not necessarily complete, but represents a suite of ideas associated with the current conservation problem.. We calculated the net value of payoffs and path probabilities using DATA 4.0 by TreeAge software.
Decision tree design and calculations
In the design of our decision tree, traditional guidelines were followed:
-
Trees flow from left to right. Decision trees are horizontal structures, ordered from left to right, in which each successive set of branches represents the outcomes of an event or decision.Several types of “nodes” are used. Each branch line in the tree has a node located at the right-hand end of the branch. Each node represents a decision, an uncertain event, or a final outcome.
- Adecisionnode (square) is used to indicate a choice facing the decision maker, which will be made based on a strict interpretation of the value of each alternative. Achancenode (circle) is used to represent an uncertain event with multiple possible outcomes. Aterminalnode (red triangle) is used to denote a final outcome – the end of a path in the model, herein referred to as ascenario. Branches emanating from a decision node represent the available options.All choices being considered should be represented and the choices must be designated such that none overlap.Branches emanating from a chance node represent the possible outcomes of the event. Again,the possible outcomes of a single event are represented in such a way that all possibilities are covered and none overlap – and that their probabilities sum to 1.0.Terminal nodes are assigned a value, referred to as a payoff. The payoff represents the net value (e.g., cost or benefit) of a particular scenario – i.e., the series of events leading up to that endpoint in the model. Only two payoffs were incorporated into these trees: benefits to conserving the population and costs to conserving the population. This simple set of payoffs creates a basic cost/benefit analysis with the common decision rule of maximizing conservation benefits. As a common decision rule, we can eliminate all scenarios where the expected value to conservation drops to below zero.
To calculate the decision tree, one works backward, from right to left. The value of each node is determined as follows:
- The value of a terminal node is its payoff value. For the cost-benefit analysis, payoff values are calculated by subtracting the cost of each scenario from it’s benefit. For this type of analysis, the cost and benefit units must be the same. The value of a chance node is equal to itsexpected value, which is calculated by weighting the values of each of its branches by their respective probabilities and summing the results. For the present tree, the expected value (EV) at the node in Figure 3 labeled A is calculated as such:
- The value of a decision node is equal to the value of its best option.
Decisions and Outcomes
For each decision tree, we included decisions, outcomes and payoffs pertinent to the specific research question, incorporating available information on the status of the population and the techniques to be applied.
SDRs Tree
Risk Status and Population size
For this tree, these decisions relate to whether the species is currently “at risk”, whether the animals represent a “small population”, and whether or not the technique being applied is novel or proven. Several metrics can be used to assign the “at risk” status of Maui’s dolphins for the purposes of this tree. As mentioned previously, they are currently included under the endangered listing of Hector’s dolphins by the IUCN (Slooten & Taylor 2000). At a national level Maui’s dolphins were listed as threatened in 2001 under the New Zealand Marine Mammals Protection Act (1978) (Ferreira & Roberts 2003). Finally, Dawson et al. (2001) concluded that this subspecies of Hector’s dolphin is vulnerable to extinction. For the purposes of this framework, Maui’s dolphins could be considered “at risk.” The entire population of Maui’s dolphins numbers approximately 100 animals (Ferreira & Roberts 2003), which are separated out into a number of management units (Burkhart & Slooten 2003). A recent population viability analysis indicates that three of the four management units of Maui’s dolphin are not abundant enough to persist without strong conservation measures (Burkhart & Slooten 2003). For the purposes of the present tree, Maui’s dolphins could be considered a “small population.”
Proven or novel technique
The final decision node we present in the SDRs tree incorporates information on the status of the technique to be used and is more difficult to evaluate. This node governs the following outcomes (likelihood of capture mortality and likelihood of procedural effects) and is meant to describe the status of tag attachment processes, the tags themselves and the likelihood of their failure. For most cases, there are no data to support either choice, and deciding whether a technique is proven or novel may rely on subjective measures that must be agreed upon by stakeholders. One potentially useful metric may be the maturity of the method. Multiple successful deployments and optimized experimental durations reflect the utility of refining satellite tagging techniques to minimize procedural effects. Maui’s dolphins have not been tagged previously, nor have Hector’s dolphins.
Robustness to capture
In many cases the likelihood of capture mortality is an unknown, but there are examples from the literature that may be informative for guiding our decisions. Capturing marine mammals in the wild poses several challenges to the researcher and even the most mature capture programs still suffer mortalities (Ref). In novel situations, higher mortality rates have been recorded (e.g. seven finless porpoises Neophocaena phocaenoides died during capture attempts for a radio telemetry study in the Shishou Semi-Natural Reserve, China in 1993 – see Wang Ding et al. 2000). One in 25 animals died in a scientific tagging program conducted with heavyside’s dolphin Cephalorhynchus heavisidii in southern Africa. In the absence of species specific data, we use relative high and low probabilities to inform the tree and resolve the optimal path. In novel situations, the probability of capture mortality is set arbitrarily to 0.7, in proven situations the probability of capture mortality is set at 0.3. These probabilities relate to the utility of refining techniques, but are also designed to be risk averse by helping to minimize the likelihood of committing type II errors when assessing uncertainty associated with the effects of capture on study animals (see discussion).
Procedural effects and likelihood of normal behaviour
Similarly, there are no species specific data on the likelihood of procedural effects but there are again examples from the literature that can help guide the development of this model. We know that in proven technique situations, procedural effects can result in abnormal behaviour that can be both short term and long term. As well, the use of proven techniques still occasionally results in obvious damage from tags to the dorsal fins of small cetaceans (seamamms ref.). The short and long term effects of such damage are unknown, but considering the importance of dorsal fins as hydrodynamic stabilizers and as thermal windows for temperature regulation (Pabst et al. 1995; Rommel et al. 1994), they are likely to produce significant behavioral effects until the tags are shed and the wounds healed. Considering these data, we set the likelihood of procedural effects as high (0.7) for novel situations and low (0.3) for proven techniques. Similarly, in novel situations we set the likelihood of normal behaviour lower (0.4) than in proven situations (0.6), the likelihood of abnormal behaviour higher (0.4) than in proven situations (0.3), and the likelihood of procedure-related mortality higher (0.2) than in proven situations (0.1). Again, these relative probabilities help to minimize the chance of committing type II errors when assessing uncertainty associated with procedural effects of equipping study animals with satellite tags and allow the tree to resolve an optimal path when no species-specific data are available.
ADDs Tree
Risk Status and Population size
In the ADDs tree, these decisions also relate to whether the species is currently “at risk” and whether the animals represent a “small population.” Vaquita are endemic to the upper Sea of Cortez (Vidal et al. 1999) and they are listed as critically endangered by the IUCN (IUCN 1996). For the purposes of the present tree the vaquita could be considered as a species at risk. Similarly, vaquita could be considered a small population. The latest population estimate is approximately 570 animals (Jaramillo-Legorreta et al. 1999) with no variability in examined mitochondrial DNA (Rosel & Rojas-Bracho 1999; Taylor & Rojas-Bracho 1999).
Sensitivity to ADDs and comparisons with similar or related species
There are two different decision nodes in the ADDs tree: whether the species is known to be sensitive to the ADDs to be deployed and in the absence of these data, whether there may be comparative data from similar or related species to help guide the process. For vaquita, ADDs have not been previously deployed to reduce by-catch, and it is currently unknown whether these animals are sensitive to any acoustic devices used in marine mammal – fishery interactions. Acoustic deterrent devices have been used in a number of commercial fisheries to reduce the by-catch of a related species, the harbour porpoise (Phocoena phocoena) (Kraus et al. 1997), and the data generated through these trials may be useful for informing any attempts to use them for vaquita conservation. For the purposes of this tree, it is not known if vaquita are sensitive to ADDs and there maybe useful information in comparisons with ADD-use in mitigation measures for harbour porpoises.
Aversive responses and achieving by-catch reduction targets.
For outcomes in the ADDs tree, probabilities are included for the likelihood of an aversive or agonistic/attractive response to ADDs. Aversive responses can be defined as a significant avoidance reaction to nets with ADDs, where animals are excluded from an area around the nets and resulting in significantly fewer physical interactions with the nets (Anderson et al. 2001; Carlstrom et al. 2002). Agonistic/attractive responses are described as those when animals either investigate the ADDs, or charge them rapidly in an aggressive manner (Anderson et al. 1999). In cases where previous research on the species has shown success in reducing by-catch, the probability of aversive reactions is high (0.9) and agonistic/attractive responses low (0.1). In situations where the effects are unknown and no appropriate comparative data are available, there are equal probabilities (0.5) of aversive and agonistic/attractive responses. Finally, because ADDs do not eliminate by-catch and animals may habituate to the sounds, it must be recognized that aversive responses will not necessarily achieve the level of by-catch reduction required to conserve the population (Dawson et al. 1998). To account for this, the probability of the sustainable by-catch outcome is lower (0.3) in small population situations as compared to large populations (0.7). As in the SDRs tree, these probabilities are also risk averse, helping to minimize the chance of committing type II errors when assessing uncertainty associated with the level of by-catch reduction afford by ADDs.
Payoffs
For payoffs in the SDRs tree and the ADDs tree, we include costs to conservation (hereafter referred to simply as costs, with values of 0 being low, 5 being medium and 10 being high) and benefits for conservation (hereafter referred to simply as benefits, using the same numerical spread as above). For the SDRs tree, the cost of mortality (through capture or procedural effects) is high and the benefit low. In cases where the animal recovers from the procedure and behaves normally, the costs are low and the benefits high. Where tagged animals recover from the procedure and behave abnormally, the costs are medium and the benefits low. For the ADDs tree, the cost of aversive responses is low and the benefit high as animals will avoid the devices. For agnostic/attractive responses the benefits are low and the costs are high as animals will ten to interact with the gear and become entangled. These payoffs are also modified for previous decision nodes. For example, there are greater costs associated with losing an animal from a “small population” (+3 to costs) and from a species at risk (+2 to costs).
Results
Using SDRs to study Maui’s dolphins
The decision tree framework for using SDRs with Maui’s dolphins is presented in Figure 2. As described above, the red triangle nodes are terminal nodes, the green circles are chance nodes and the blue squares are decision nodes. As this tree was calculated, it presents an optimal path (pink lines and solid nodes) and the probabilities associated with it. Sub-optimal paths are separated by hatch marks on lines connecting nodes.
The present tree suggests that according to the available data and decisions made, invasive techniques should be used on species not at risk, with a proven technique (black arrows). In this scenario (B), the highest probability (0.49) is associated with no capture mortality and no procedural effects and results in a payoff of 10.0 (terminal node labeled C). This scenario closely resembles a tagging program on harbour porpoises conducted in the Bay of Fundy. In comparison with Maui’s dolphins, harbour porpoises in the Bay of Fundy/Gulf of Maine are not at risk (Read 2002), they represent a large population (Waring et al. 2002) and the capture and surgical techniques used would be considered proven, after refinement over several years of the program (indeed, basic guidelines for handling animals have been published – see Wong et al. 2002). If this tree were applied “as is” to the Maui’s dolphin situation, the model would requiring forcing down one or more of the less optimal paths. According to the available data, the model should be forced down species at risk path (D), and the small population path (E). At this point, the optimal path follows the proven technique path (F). If the tree is calculated with these cumulative forcings along the novel technique node (scenario G), the highest probability (0.70) is associated with a capture mortality terminal node which provides a net payoff of -5.0 (H). If we were to apply our common decision rule in the species at risk scenario, using SDRs with small populations or with novel techniques in large populations produces negative values to conservation. These results suggest that using SDRs to study Maui’s dolphins is not likely to contribute significantly to conservation efforts.
Using ADDs to conserve vaquita
The decision tree for using ADDs for vaquita conservation is presented in Figure 3. The present tree suggests that the optimal scenario is one where ADDs are used to mitigate fisheries interactions with species that not at risk, with large populations and are known to be sensitive to the devices being deployed (black arrows). In this scenario (I), the highest probability (0.81) is associated with an aversive response and sustainable by-catches with a payoff of +10 (J). This scenario is similar to the ongoing use of ADDs to reduce by-catch of porpoises in commercial fisheries in the Bay of Fundy/Gulf of Maine. Compared to vaquita, porpoises are not at risk and have a large population in the Bay of Fundy/Gulf of Maine. As well, harbour porpoises are known to be sensitive to ADDs (Kastelein et al. 2000) and will avoid nets which have active ADDs attached (Culik et al. 2001; Kraus et al. 1997) . It must be noted here that harbour porpoises by-catch is reduced, not eliminated, by the use of ADDs (Dawson et al. 1998; Kraus et al. 1997) and that harbour porpoises may actually habituate to ADDs over time (Cox et al. 2001). If the tree is applied to the vaquita situation, it also will require forcing down one or more sub-optimal paths. For example, the model would be forced down the species at risk path (K), down the small population path (L) and then down the ADDs untested path (M) where the expected value to conservation has dropped to -6.3. If we apply the common decision rule to this sample tree, we would eliminate the use of ADDs in small population scenarios where they have not been tested previously. This also suggests that, according to our analysis, using ADDs in small population situations may often be counter-productive.
Discussion
The results of our decision analysis suggest that the use of ADDs and satellite tagging techniques in research and management pertaining to small populations of endangered species may not always contribute significantly to their conservation. These results are not meant to be prescriptive, nor applicable in all cases. Rather, these examples are useful for illustrating the utility of decision analysis frameworks for conflict resolution in contentious conservations scenarios such as those surrounding the use of invasive techniques and novel technologies in conservation research and management. Clearly, each situation will require a unique analysis which includes information from all stakeholders. In some situations, competing trees may be created to illustrate differences in stakeholder views (Maguire & Boiney 1994) and sensitivity analyses completed to illustrate how changes to probabilistic nodes (and stakeholder opinions) effect the outcome of models (Maguire 1987). As well, decision trees like those presented here may address only a small part of a larger environmental conflict that requires the application of other dispute resolution techniques (Maguire & Boiney 1994).
The benefits of this technique to resolve complex decision-making processes are obvious. It is flexible, allowing scientists and managers to incorporate various forms of information into an explicit conceptual model. This technique has been successfully applied in terrestrial endangered species situations. For example, Maguire (1987) used decision trees to assess alternative management options for endangered tiger populations. While also not prescriptive, Maguire’s (1987) trees were useful to organize and communicate information relevant to various management options and provided a stable framework for assessing the utility of competing management options including changes to the size of ecological reserves, instituting active population control measures for some populations or simply maintaining the status quo. These trees also included probabilistic nodes associated with the uncertain effects of active management techniques that were informed with quantitative data and the careful use of subjective information derived from stakeholders (Maguire 1987). We applied a similar approach in our analyses, informing uncertain probability nodes with quantitative estimates and subjective estimates based on information from other species where available.
In the present examples, real-world values for the likelihood of capture mortality, the effects of tagging procedures on animals or the likelihood of reaching sustainable by-catch levels can be incorporated to tune the model appropriately. However, in uncertain situations where there are no data available to inform probability nodes, there are still ways to design trees to resolve optimal paths. In our case, we used relative probabilities that are based on previous nodes and knowledge derived from other case studies (as described above). These are referred to as subjective probability estimates (Maguire 1987). These subjective probability estimates were also subjected to an ethical analysis of uncertainty and potential outcomes to make them risk-averse.
For our trees, we weighed subjective probability estimates to minimize the chance of committing type II errors when addressing uncertainty associated with the research or management technique. This maximizes protection of the public good from potential harmful effects of a particular research or management technique. In the Maui’s dolphin case, the underlying uncertainty relates to whether or not SDRs will kill or harm individual animals. Here we present this uncertainty as a null hypothesis:
H0: Tags will not kill or significantly harm dolphins
In a type I error situation, we would have rejected the null hypothesis when it is actually true. In other words, we rejected the notion that SDRs are harmless, even though they actually are harmless. In type II situations, we would have accepted the null hypothesis when it is actually false. In this case we accepted the notion that SDRs are harmless, when in fact they are not. When weighing the use of SDRs with small populations versus the use of a less invasive technique – and in situations of uncertainty – researchers should be trying to minimize the chance that they actually harm animals, even if they probably won’t (Type II error). This explicitly acknowledges that any attempts to test for harmful effects of tags on dolphins are likely to have small samples sizes and low statistical power and the chance of committing a type II error is therefore high. In these situations, it may often be advisable to reject the null hypothesis to minimize type II error. Risking type I errors is preferable, where animals are not tagged – even though it won’t cause them harm – as a less invasive technique can be implemented. A similar process was followed for the use of ADDs in vaquita conservation.
There is precedence for this thinking in the conservation literature (Noss 1986) and there are good prima facie reasons for minimizing type II errors in applied conservation research (Shrader-Frechette 1994). For example, it is generally accepted that society should strive to prevent acts that cause harm before it promotes acts that enhance welfare (Shrader-Frechette 1994). If the public good is represented by a growing population of Maui’s dolphins, then it is more important to protect the public from harmful research than to save it from harmless impacts – there is no direct harm to the public by not tagging the dolphins, as other methods can be used to answer the question. Furthermore, these situations may exhibit ethical asymmetry (Shrader-Frechette 1994). Contractors who conduct tagging research receive greater benefits (financial compensation and primary literature publications) than the general public, yet the general public bears the risks/costs of the research if animals are harmed or killed. This is not fair (Shrader-Frechette 1994), and researchers should seek solutions that minimize ethical asymmetry. Finally, it is generally accepted that the public has rights to protection from research that can cause incompensable harms (research that harms the public/environment in ways that cannot be compensated for – see Shrader-Frechette 1994; Thomson 1986). In other words, since researchers cannot compensate the public for the loss of animals from the Maui’ dolphin population that result from their research, they should seek solutions to minimize the chance of those losses.
Decision nodes used in trees can be based on various types of information including scientific data, governmental and international status designations, ethical considerations or other stakeholder values. For example, the “at risk” status of a species could be based on statistical calculations of sustainable anthropogenic removal levels relative to a conservative population abundance estimate. Where removal levels exceed sustainable limits, a population could be designated “at risk”. An example of this would be the calculation of potential biological removal levels as used in the stock assessment process for marine mammals in the United States (Wade 1998). Where removal levels exceed PBR, the stock is afforded a new designation and management actions are triggered (Wade & Angliss 1997). At risk status could also be based on quantitative assessments of the magnitude of the threats faced by the species or population. This could include frequency or magnitude of by-catch, extent of contaminant loading in tissues (Baird 2001) or the amount of essential habitat lost or available. In the absence of these types of data, the “at risk” status could also be relative, based on comparisons to similar or related species in the same geographical region or under similar threats. Endemism might also be useful for informing decisions related to the status of a species.
There are also several metrics useful for informing the population size node. Small populations could be defined through an international or national standard (e.g. a percentage of an overall species population or a minimum population management target), through a quantitative population modeling method technique such as the population viability analysis conducted by Burkhart & Slooten (2003) or, as described above, through comparison with others species when the required data are not available for quantitative assessment.
Expandability is another advantage of applying this type of decision tree framework to endangered species conservation issues. While we employed these trees to assess when it would be advisable to employ a certain technique, decision trees are most often used to simultaneously assess a variety of possible or competing techniques, providing guidance on which should be used to maximize or minimize the chosen payoffs. In the case of the present trees, including a competing research or management technique, such as aerial surveys and photo-identification in the case of Maui’s dolphin, or immediate reductions in fishing effort in the vaquita situation, can further resolve which methods will provide the best solution to the problem at hand. Furthermore, both trees could be informed with information on the monetary costs associated with each technique, which may provide greater resolution on the utility of these research and management techniques when used with small populations of endangered species.
Acknowledgements
We would like to thank Liz Slooten, Steve Dawson, Randy Reeves, Simon Childerhouse, Kerry Irish, and others. This work was supported by the Duke University Marine Laboratory.