Evidence, models, conservation programs and limits to management
James D. Nichols, U.S. Geological Survey, Patuxent Wildlife Research Center, Laurel, MD 20708, USA
Walsh et al. (2012) emphasized the importance of obtaining evidence to assess the effects of management actions on state variables relevant to objectives of conservation programs. They focused on malleefowl Leipoa ocellata, ground-dwelling Australian megapodes listed as vulnerable. They noted that although fox Vulpes vulpes baiting is the main management action used in malleefowl conservation throughout southern Australia, evidence of the effectiveness of this action is limited and currently debated.
Walsh et al. (2012) then used data from 64 sites monitored for malleefowl and foxes over 23 years to assess key functional relationships relevant to fox control as a conservation action for malleefowl. In one set of analyses, Walsh et al. (2012) focused on two relationships: fox baiting investment versus fox presence, and fox presence versus malleefowl population size and rate of population change. Results led to the counterintuitive conclusion that increases in investments in fox control produced slight decreases in malleefowl population size and growth. In a second set of analyses, Walsh et al. (2012) directly assessed the relationship between investment in fox baiting and malleefowl population size and rate of population change. This set of analyses showed no significant relationship between investment in fox population control and malleefowl population growth. Both sets of analyses benefited from the incorporation of key environmental covariates hypothesized to influence these management relationships. Walsh et al. (2012) concluded that ‘in most situations, malleefowl conservation did not effectively benefit from fox baiting at current levels of investment.’
In this commentary, I discuss the work of Walsh et al. (2012) using the conceptual framework of structured decision making (SDM). In doing so, I accept their analytic results and associated conclusions as accurate and discuss basic ideas about evidence, conservation and limits to management.
SDM is an approach to decision making that decomposes a decision problem into several components (Clemen & Reilly, 2001; Martin et al., 2009a): (1) objectives; (2) potential actions; (3) models of system response to actions; (4) estimates of system state variables (from monitoring); (5) an algorithm for selecting the appropriate action. SDM usually entails initial consideration of each component separately, followed by combination of the components to decide on the appropriate action. Although Walsh et al. (2012) did not describe their work in the specific terminology of SDM, their work included all of these ingredients needed to make good conservation decisions.
For objectives, Walsh et al. (2012) began by professing interest in the cost-effectiveness of conservation actions, providing a basis for selecting those actions that would ‘gain the greatest conservation outcome for a fixed budget or achieve a specified objective for the lowest cost’ (Walsh et al., 2012:1). These latter two types of objective function include both management cost and conservation outcome, where outcome was operationally defined as either the population size or population growth rate of breeding malleefowl. Thus, Walsh et al. (2012) clearly specified the key elements of their conservation objectives. They were similarly clear about specifying potential management actions. Although they noted multiple threats to malleefowl, including habitat loss and fragmentation, frequent fires and grazing competition, they focused on predation by introduced foxes and the corresponding action of fox baiting and control.
Models are key components of conservation programs, as they provide a means of predicting outcomes of alternative management actions. Most approaches to selecting the appropriate management action entail a comparison of outcomes predicted to result from the potential actions in order to identify the action that is best with respect to specified objectives. Walsh et al. (2012) clearly specify their interest in models that relate their management action, fox baiting, to their conservation outcomes, malleefowl population size and population growth rate. They indicate substantial uncertainty in these key relationships, motivating data analyses directed at estimating associated parameters, that is at parameterizing the management model.
Monitoring programs provide estimates of system state variables, such as population size of malleefowl and fox presence–absence, that are used for three primary purposes in conservation programs (Nichols & Williams, 2006). The first purpose is for making state-dependent decisions. Typically, application of a treatment such as fox control at a site for a particular year would be dependent on the number of malleefowl and foxes thought to be present. Although there was site-to-site variation in the number of years of fox baiting in the dataset of Walsh et al. (2012), it was not clear whether the decision to bait was based on current system state or not. The second purpose for monitoring is to assess the degree to which conservation objectives are being met; and the third purpose is to learn how well management models predict outcomes. Walsh et al. (2012) used their monitoring data for both of these purposes.
The final step of the SDM process usually entails use of the components mentioned earlier with some sort of algorithm for deciding which action is most appropriate. Optimization approaches are frequently used for this step. Because Walsh et al. (2012) focused on fox baiting, the final decision was simply to continue baiting or not. The results of Walsh et al. (2012) did not support continuation of baiting as a cost-effective action.
Dealing with uncertainty: monitoring, models, evidence and adaptive management
The work and analytic results of Walsh et al. (2012) were focused on models that related investment in fox control to conservation objectives of healthy malleefowl populations. These analyses were motivated by structural or epistemic uncertainty about the key relationships on which management programs are based. Walsh et al. (2012:2) noted the following: ‘When the effectiveness of a management action is estimated or assumed, there is no guarantee that investment in that action will achieve the conservation objective cost-effectively’. Of course, we hardly ever know with certainty the consequence of applying any management action (we can never guarantee results). However, there is a substantive difference between assuming and estimating management effects, and the distinction involves the concept of evidence. Assumptions may be based on faith, management lore and traditions, or evidence, and Walsh et al. (2012) argue the importance of basing predictions on evidence.
In the worlds of science and scientific management, the concept of evidence involves the correspondence, or lack thereof, of monitoring data to model-based predictions. So we may have competing models that make different predictions about the effects of a management action that we decide to apply, and we collect data to find out which model predicted results most closely. Or we may begin with a very general single model with a key parameter (e.g. a slope parameter relating fox baiting investment to malleefowl population growth) that we estimate using monitoring data. In both cases, we begin with models representing hypotheses about effects of management actions and compare their predictions with our best estimates from monitoring of what really happened. It is worth emphasizing that both models and monitoring are required in order to produce evidence and reduce uncertainty. For example, monitoring is sometimes wrongly equated with management. Monitoring, by itself, does not constitute evidence or reduce uncertainty in management. ‘Even though the malleefowl is one of the best-monitored species of conservation concern in Australia, we are still uncertain how to cost-effectively manage this species’ (Walsh et al., 2012:1). It is the comparison of predictions of models for management effects against monitoring data that allow us to reduce uncertainty.
Adaptive resource management (ARM) is a subset of SDM designed for iterative decisions characterized by substantial uncertainty (e.g. Walters, 1986; McCarthy & Possingham, 2007; Williams, Szaro & Shapiro, 2007). Walsh et al. (2012) concluded that malleefowl is an ideal species for an ARM program directed at resolving the uncertainty associated with the effectiveness of various management actions (including, but not limited to, fox control). I strongly support this conclusion, noting only that the approach to ARM could be either sequential (learn via experimentation and then apply what is learned in future management) or simultaneous (balance learning and management with each decision), with respect to the timing of learning and managing.
Limits to management
The failure of Walsh et al. (2012) to find strong evidence of the effectiveness of a favored management action is a disturbing result for at least two reasons. First, it demonstrates the dangers of managing in the absence of scientific inquiry and evidence, a common conservation situation throughout the world. Second, it places an urgent burden on managers to consider what other management actions might be appropriate for this species. Unfortunately this Walsh et al. (2012) result is not uncommon. The steps of developing explicit models of system response to management and then conducting analyses to either discriminate among competing models or estimate key parameters associated with general models have recently led various managers to the recognition that many previously favored management actions are indeed limited in their ability to produce desired outcomes (e.g. Martin et al., 2009b, 2011; Alisauskas et al., 2011; Johnson et al., 2011; McGowan et al., 2011). These kinds of results lend greater urgency to the task of implementing conservation programs that use defensible approaches to decision-making.