Structured decision making to navigate trade‐offs between multiple conservation values in threatened grasslands

Managing biodiversity often requires making difficult trade‐offs, especially when threatened species and ecosystems overlap in their distributions, and management actions to promote their persistence varies between them. Trade‐offs reflect preferences and how much decision makers are willing to give up in benefits for one objective in exchange for gains in other objectives. Despite the increase of tools for exploring trade‐offs, decisions are often made without clear specification of preferences among objectives, reducing transparency and limiting clear communication of the rationale behind decisions. We used structured decision making to navigate trade‐offs between ecological objectives and management costs across three conservation reserves of protected grasslands in Victoria, Australia. The objectives included four nationally listed threatened species, two nationally listed threatened ecosystems, a group of locally threatened and rare non‐listed species and management costs. Alternative management strategies included various combinations of fire management and weed control. The consequences of alternatives were estimated using stochastic models, empirical data and expert judgment. Context dependent preferences were elicited using swing weighting from nine decision makers and stakeholders. While all species and ecosystems are valued, how they are weighted, and the resulting preferred management strategy is context dependent. Weights were variable across participants for all objectives. There was stronger alignment of weights for one of the ecosystems, Natural Temperate Grasslands, than other objectives. One of the threatened species, striped legless lizard, tended to be weighted higher than other ecological objectives, while management costs had the lowest weights. Stakeholder's weightings for objectives varied, however the influence on the rank order of management strategies was minimal. The structured approach to navigate trade‐offs identified management strategies that best address stakeholder preferences across multiple objectives. This approach offers improvements in evidence‐based decision making and provides a transparent and defensible rationale for selecting management interventions that considers all relevant objectives.

and provides a transparent and defensible rationale for selecting management interventions that considers all relevant objectives.

K E Y W O R D S
grasslands, structured decision making, threatened ecosystems, threatened species, tradeoffs, value preferences 1 | INTRODUCTION Conservation managers are faced with difficult trade-offs, especially when there are overlapping occurrences of multiple species and ecosystems with different management needs, where management cannot be optimized to maximize benefits across all the objectives of concern. In some instances, decision makers must accept losses in one species or ecosystem to secure gains in another (Keller & Simon, 2019). Most ecological management occurs without explicit recognition or documentation of such trade-offs (but for diverse examples see Butler et al., 2013;Lautenbach et al., 2017;van Wilgen & Richardson, 2014), even in cases where a strong evidence base is used to judge the likely consequences for individual objectives. Nonetheless, these trade-offs are unavoidable, and the way they are negotiated can change decisions and outcomes (Bradford & D'Amato, 2012;Dale et al., 2000).
Trade-offs reflect preferences among objectives. For decisions on managing threatened species and ecosystems, trade-offs will be influenced by a range of factors. These include the context for the decision including implicit or explicit recognition of socio-economic objectives (Runge, 2021) and spatial and temporal constraints; the conservation objectives of interest and their current threat status (i.e., species and ecosystems); the range of estimated benefits to each objective across the suite of alternative actions; whether the species or ecosystems have occurrences elsewhere or if they are the last remaining populations; the evidence base that underpins estimates of consequences; and influences from interested stakeholders. In practice, additional factors complicate the decision-making process further relating to who the stakeholders and decision-makers are and how they should contribute, whether there are other objectives such as management costs that should be considered, the feasibility of actions and the degree of tolerance among stakeholders and decision makers to uncertainty (Gregory et al., 2012;Riley & Decker, 2000). When trade-offs are made implicitly or intuitively, particularly for complex decisions regarding multiple objectives, it is unclear if appropriate consideration has been given to all objectives of interest and what factors have influenced the decision. Such approaches lack transparency and limit clear communication with stakeholders and the broader public (Keeney, 1982).
Since the 1980s, decision science has emerged as an important field within conservation practice for navigating complex decisions and promoting the use of science to support evidence-based decision making for biodiversity (Failing et al., 2007;Harwood 2000;Maguire, 1986;Possingham, 1997;Ralls & Starfield, 1995). Frameworks for formulating decision problems such as multi-criteria decision analysis (Keeney & Raiffa, 1993), structured decision making (SDM) (Hammond et al., 1999), cost-benefit analysis (Arrow et al., 1996) and systematic conservation planning (Margules & Pressey, 2000) have been applied to numerous conservation problems around the globe (e.g., Converse et al., 2013;Joseph et al., 2009;Maguire et al., 1987;Moore et al., 2011;Ringma et al., 2017). Broadly, these frameworks use a set of steps that disaggregate the decision into discrete elements: (i) defining the decision context, (ii) defining objectives, (iii) identifying management alternatives, (iv) estimating consequences of alternatives, (v) exploring trade-offs between objectives, (vi) implementing decision and monitoring outcomes, and (vii) updating the decision within an adaptive management framework (Garrard et al., 2017;Gregory et al., 2012;Hammond et al., 1999;Keeney & Raiffa, 1993). These decision frameworks provide a critical link between the objectives underpinning the decision and the data and models that predict the consequences of management alternatives. Importantly these frameworks enable data and modeling components underpinning a decision to be clearly distinguished from stakeholder preferences for different objectives.
The availability of ecological models and quantitative decision support tools within the conservation sciences has provided a strong evidence base for evaluating the consequences of alternative actions (Ferrier et al., 2016;Nicholson et al., 2019). Population viability analysis models are common tools for evaluating management alternatives for single threatened species (Regan et al., 2011;Taylor et al., 2017). For threatened ecosystems, predictive models that link interventions and ecosystem consequences are more rare, but may include approaches such as state and transition models (Rumpff et al., 2011) and Bayesian belief networks (Trifonova et al., 2017). When addressing multiple species, species distribution models (Guisan et al., 2013;Liu et al., 2013) underpin spatial optimization methods for identifying spatial priorities for biodiversity management given limited budgets and other constraints (Margules & Pressey, 2000). Conservation planning tools like Zonation (Moilanen, 2007) and Marxan (Ball et al., 2009) identify spatial priorities for protection, while adaptations of these tools address a broader suite of management options beyond reserve design and are increasingly being used to inform conservation priorities (Carwardine et al., 2012;Thomson et al., 2020).
While conservation decision tools have advanced significantly over recent decades, the contribution of value preferences remains opaque. Despite the existence of methods for capturing stakeholder preferences and subsequent trade-offs among objectives (Keeney & Raiffa, 1993), there are limited examples in the conservation literature (but see King et al., 2015;Sinclair et al., 2021). To promote more defensible evidencebased decision making, more real-world applications demonstrating the effective use of decision analysis tools that include judgments and trade-offs are urgently needed (Fuller et al., 2020;Runge, 2011).
In this paper we use SDM to navigate complex tradeoffs for management of seven ecological objectives (five species and two ecosystems) that are unevenly distributed across three conservation reserves in grasslands on the outskirts of Melbourne in south-eastern Australia. All three reserves support natural grasslands (i.e., steppe) in a landscape that is otherwise largely agricultural and periurban. Grasslands are one of the most threatened ecosystems globally, with losses stemming from conversion to agriculture, urbanization and altered fire regimes (Blair et al., 2014). In Australia, the natural temperate grasslands are at risk of continued loss and deterioration as a result of weed invasion and the build-up of excessive biomass (Humphries et al., 2021;Stuwe & Parsons, 1977), and they are listed as critically endangered under Australian environmental legislation (DPCD, 2009). They are generally dispersed in many small reserves which support a range of threatened species and are often intensively managed with high levels of community engagement (Reid, 2015).
We use SDM to frame and navigate the management problem, and ecological models and expert judgment to estimate consequences of management options, which include various options focused on burning and weed control. We capture preferences from key stakeholders and decision makers to explore trade-offs among objectives, to ultimately agree on a management strategy where all relevant objectives are considered.

| METHODS
We used the steps and principles of SDM outlined in Hammond et al. (1999) to develop a shared understanding of the decision context, the objectives to be addressed, a suite of potential management alternatives, evaluation of the impact of those alternatives on the objectives of interest and the potential trade-offs between them. These steps are outlined below and serve as a guide for applying SDM in practice for similar types of decision problems.
All stages that involved decision makers and stakeholders were undertaken during two workshops on separate days. The first workshop, attended by 12 stakeholders, focused on scoping the decision context and developing a shared understanding. The second workshop focused on quantitative trade-offs process with a subset of nine stakeholders. Stakeholders included seven representatives from two public land management agencies responsible for delivery of conservation outcomes for threatened grassland species and ecosystems as well as five government agency scientists with expertise in grassland ecology, modeling, and SDM.

| Step 1-Decision context
The decision problem focusses on conservation management of threatened grassland ecosystems and individual threatened species that reside within them on the periphery of Melbourne, which are impacted by urban development. Under the Melbourne Strategic Assessment program (MSA), the Victorian State Government has agreed to a range of conservation measures for these areas, including the establishment and management of new reserves, to achieve a suite of objectives related to the improvement of ecosystems and the persistence of threatened species (DPCD, 2009;Dripps, 2010).
We developed a decision statement in collaboration with stakeholders to describe the scope of the decision, including what factors triggered the decision, relevant stakeholders and ultimate decision maker/s, spatial and temporal scope, potential management actions, any constraints, objectives of interest and recognition of uncertainty. The purpose of the statement is to ensure all stakeholders have a shared understanding of the scope of the decision problem and any key assumptions. The key content of the decision statement is provided in the following text.
The geographic scope of this study encompasses three discrete reserves: Mount Cottrell Nature Conservation Reserve (NCR) (44 ha; À37.76, 144.66), One Tree East Reserve (436 ha; À37.83, 144.53) and Little Raven Reserve (121 ha; À37.93, 144.55). Each reserve supports multiple conservation objectives including nationally listed and non-listed threatened species, and ecosystems ( Table 1). All three reserves contain the nationally listed critically endangered natural temperate grassland (NTG) ecosystem, in varying condition. Two reserves (Little Raven and One Tree East Reserve) also contain occurrences of the nationally listed critically endangered seasonal herbaceous wetlands (SHW). All reserves contain different combinations of populations of nationally listed threatened species including the striped legless lizard (Delma impar), golden sun moth (Synemon plana), largefruit groundsel (Senecio macrocarpus), and spiny riceflower (Pimelea spinescens subsp. spinescens). In addition, two reserves (Mt. Cottrell NCR and One Tree East reserve) contain a range of other state listed threatened species or locally rare species of concern. These were treated as a single objective, since the workshop participants agreed that they share common ecological responses and requirements and low population sizes (Table 1). These species are referred to as non-listed flora. Including sites with differing composition of species and ecosystems helps to highlight the potential contextdependent nature of trade-offs. If a species is present at all three sites and another species is only present at one site, managers may give a higher weight to the species at T A B L E 1 Objectives with associated performance measures and their occurrence across three sites.

Objectives
Performance measures one site because its management at that one site is the only opportunity for benefit. Treating each site as a separate decision problem does not easily allow aggregated benefits for each species and ecosystem to be evaluated. This is especially relevant where species occur at multiple sites. This point has real-world implications. In practice, managers must often allocate resources across a portfolio of sites. Capturing three sites within this case study provides greater clarity in how context-dependent trade-offs can be navigated. The trigger for this decision arose from an annual requirement for public land management agencies to decide on a set of actions to achieve conservation and other management goals. The ultimate decision maker is the MSA team in the Victorian State Government, who commits funding for management actions. This decision is based on the knowledge from Parks Victoria (the assigned on-ground managers), consultant and government ecologists, Local Government representatives, community groups, and aims to increasingly draw on the expertise of Traditional Owners. Workshop outcomes from this study were provided as informal advice for the MSA team.

| Step 2-Objectives and performance measures
Capturing what stakeholders care about is integral to any decision-making process (Keeney & Raiffa, 1993). To aid evaluation of alternatives, we developed specific, measurable objectives that describe what is to be achieved. We presented an initial list of potential objectives to decision stakeholders, based mostly on previously published objectives for the MSA program (DELWP, 2015).
Eight fundamental objectives were included in our evaluation of management actions (Table 1): management costs, conservation outcomes for four nationally listed species and two threatened ecosystems and a suite of locally threatened and rare non-listed flora within the grassland ecosystem (considered as a single objective as they have common ecological requirements, grazing intolerance and low population sizes). The objectives were a subset of the overarching objectives for the MSA program relevant to the three reserves. Workshop participants indicated other objectives were of interest (such as Traditional Owner engagement with country, cultural, recreational and other ecological objectives) but we confined our study to the agreed core ecological objectives included here, which are the remit of the MSA (DPCD, 2009). For each objective we identified performance measures (Table 1)

| Step 3-Management strategies
Eleven management actions were derived via discussion among the stakeholders. These alternatives focussed on different combinations of managed burns and weed control. Burning regimes vary with respect to frequency (fire return interval) and season. Three fire frequencies were explored with different burning windows: short-burn window of 1-3 years, medium-burn window 2-4 years and long-burn window 3-7 years (Table 2). Burning can occur in Spring or Autumn or both or not at all. Weed control can occur in the presence or absence of burning (A2 and A3, Table 2). Weed control is done via spot spraying of herbicide, with four possible states in each year: none, spraying grassy weeds only (Poaceae, which are usually the most abundant exotic species), or spraying both broadleaf (dicot) and grassy weeds (Table 2). We also include a do-nothing action (A1) which is used to compare benefits for all other actions (A2-A11).
Different combinations of actions were applied to the three reserves to create a portfolio of management strategies (Table 3). Nine management strategies were created that focused on the existing management (i.e., status quo [S1]), aspirational burning (S2) or were designed to benefit particular species or ecosystems more than others (S3-S9). This approach enabled us to select combinations of actions likely to be most effective for one or more objectives, across the three reserves of interest.

| Step 4-Estimating consequences of management strategies
We used species-and ecosystem-specific predictive models, in combination with expert judgments, to generate estimates of outcomes for each objective and each of the management strategies. While management decisions are made annually, we used a 50-year time scale for estimating the consequences of different management actions, which assumes that the actions are applied annually across the time period. This time frame is long enough to represent species and ecosystem persistence, but still within the limits of human experience.
For the nationally listed species, we used population viability analysis (PVA) models specifically developed for each species to evaluate management outcomes within the MSA program (Regan et al., 2021). PVA models were developed in RAMAS GIS (Akçakaya & Root, 2013) and were parameterized using currently available data on each of the species either within the MSA management area or related sites in other parts of the species' range. Any data gaps were filled via a structured expert elicitation using the 4-point process, where experts are asked to estimate parameter values by providing plausible upper and lower bounds and a best guess along with their level of confidence that the true value falls within their estimated bounds (Speirs-Bridge et al., 2010). Judgments were averaged across three to six experts depending on the species (Regan et al., 2021). Initial conditions of the models, such as initial abundances and carrying capacities, were updated to reflect the conditions of the three reserves based on inventory and monitoring data (unpublished data). The performance measures used from the models were the change in average abundance after 50 years given a management action compared to no management over 50 years.
For the threatened ecosystems, we estimated consequences using a condition metric, an ecosystem model and expert judgment. For the NTG ecosystem, we quantified consequences using a published condition metric that incorporates cover and richness of specific groups of plant species (Sinclair et al., 2015). We used initial data from recent surveys of the three reserves and predicted the future magnitude of these variables (and hence condition) using an unpublished model that uses Random Forests to relate grassland species composition to multi-year grazing and fire regimes (Sinclair et al. unpublished). No model exists that describes the response of SHW to different management options. Thus, a single expert (Author Sinclair) estimated the consequences of each action for this ecosystem. Similarly, the non-listed flora species were treated as a group and benefits from each management strategy were estimated by a single expert (Sinclair).
Costs for each management action (and strategy) were estimated using unpublished cost models in consultation with Parks Victoria (PV), who undertake the T A B L E 3 Alternative management strategies considered in trade-off analysis made up of combinations of actions from Table 2  management. These models were informed by a combination of expert opinion and previous expenditure on similar actions by PV. Costs for burning consider labor, administration (e.g., public notification of burns, planning time) site preparation (e.g., establishment of temporary fire breaks) and equipment (slashers, tankers, etc.) for both pre-burn preparation, actual burn implementation, and safe mop-up operations. The costs were scaled to each of the three reserves, considering their size and location with regard to roads and permanent firebreaks. Weed control costs for a given season consider labor, vehicle, and chemical costs. Labor and chemical use is scaled to the cover of target weed species and the size of each reserve. For strategies which included ongoing weed control over many years, we calculated total costs by applying the cost model sequentially to the year-on-year cover predictions of the grassland ecosystem model, without considering inflation or a discount rate.
Benefits of strategies were calculated as the difference in estimated outcomes with and without management (Table 2: A1) for each alternative management action (Table 2: Actions 2-11). We then summarized the benefits in a consequence table to enable clear comparison of the effectiveness of contending management options. The predicted outcomes for the two ecosystems were considered separately for each reserve (Mt. Cottrell NCR, Little Raven Reserve, and One Tree East Reserve), since the ecosystem condition performance measure does not explicitly aggregate different condition scores across areas of different size.

| Step 5-Trade-offs
Objectives were weighted using a multi-attribute method known as swing weighting (von Winterfeldt & Edwards, 1986). This process involves creating hypothetical scenarios based on the best and worst benefits of the alternative management strategies in question. The hypothetical baseline scenario has all the objectives performing at their worst. Participants are given the opportunity to "swing" one objective from its worst amount to its best amount. Each workshop participant selected their choice, and their selected objective is given a rank of one. Each participant then swings the second most important objective from its worst amount to its best amount and assigns a rank of two; and so on until all objectives are ranked. The highest ranked objective is then assigned 100 points and the remaining ranked objectives are assigned an amount between 100 and 0 by each expert, in accordance with how important they are judged to be, relative to the top ranked objective.
The swing weightings for the NTG and SHW were treated separately across the three reserves because of our inability to aggregate condition scores across reserves. The purpose of the weightings was to rank objectives given the outcomes of all the management scenarios. When normalizing the swing weights, we applied a proportional factor to NTG (i.e., 0.33) and SHW (i.e., 0.5) scores to account for these objectives and associated weights that were captured at separate reserves for the two ecosystems. If we did not do this, an objective such as NTG if ranked highly, is likely to be ranked highly across all reserves it is present in, over inflating the assigned weights compared to the other objectives that were weighted based on the consequences aggregated across all the reserves they were present in. To normalize scores across all objectives, we divided each score by the sum of all the scores, so they summed to 1.0. We were then able to aggregate weights for NTG and SHW across reserves to allow comparison of weightings across all objectives.
The swing weighting was undertaken in a single workshop by nine workshop participants. The group discussed the management strategies and predicted consequences across all the objectives. The degree to which each species and ecosystem also occurred outside the three reserves (globally, and within the MSA conservation reserves) was discussed, as it was deemed that understanding the broader context may influence the weightings.
The consequences of each objective were also normalized to a relative score between 0 and 1 across all management strategies, where the worst consequences for an objective assigned zero and best consequence assigned 1.0 and all other consequences relative to those values. The scores were assigned based on whether the objective was to maximize or minimize. For example, the objective to minimize costs, assigned a score of one to the strategy that had the lowest cost and zero to the highest cost strategy. For all other objectives that aimed to maximize benefits, the strategies with the highest benefit were scored as one and the lowest performing strategies were scored as zero. The normalized consequences were then multiplied by the weights and summed to give an overall performance of each management strategy. For this step, the site-specific weights for NTG and SHW were used. We explored strategy performance using no weights (i.e., same as assuming equal weights), average weights across all nine participants and individual weights.

| Weights on objectives
Normalized weights varied between 0.02 and 0.24 (unitless) across all objectives and participants (Figure 1).
Although this range may seem small, it is actually quite large relative to the normalizing process (i.e., where the weights sum to 1.0). Relative weights varied across stakeholders (Figure 1). Individual weightings indicate some alignment in stakeholder preferences for different objectives, with NTG having the narrowest range across all participants, suggesting more agreement among the participants than for other objectives. The striped legless lizard ranked highest with the large-fruit groundsel a close second, but weights for both these objectives were variable. Management costs ranked lowest in relative importance. Golden sun moth and spiny rice-flower also had highly variable weights with the non-listed flora and SHW having moderately variable weights.

| Estimated outcomes of management strategies
Estimated outcomes for the species, ecosystems and management costs varied across strategies. No single management strategy was best for all objectives (Table 4). There F I G U R E 1 Box and whisker plots of the variation in normalized swing weights for each objective across nine participants for each objective. See Table 1 for description of objectives. Site-specific normalized weights for NTG and SWH have been adjusted by a factor proportional to the number of objectives and then aggregated across sites.
T A B L E 4 Estimated benefits for each objective for the nine management strategies (S1-S9) compared to no management. were also no redundant objectives (i.e., an objective not influenced by any of the strategies). Some strategies produced higher benefits for some species and ecosystems than others. For example, there was a fourfold increase in the change in condition score for NTG between the worse and best performing strategy, ranging between 8.97 and 35.01. Similarly, for the golden sun moth, benefits, measured as change in abundance, ranged from 313,780 to 895,898. For some objectives, the benefits of management strategies were less. For example, four of the nine management strategies (S3, S5, S6, and S7) were predicted to result in zero benefit to striped legless lizard (Table 4). This was primarily due to low abundances of this species in the MSA reserves at the time of this study. Similarly, benefits for spiny-rice flower were low, due to low population sizes and the long lifespan of this species, with predicted changes in abundance ranging between 1 and 4 across the strategies. Changes in condition scores for SHW were also modest ranging from 10 to 19 depending on the management strategy.

| Trade-offs
Management strategy performance with no weightings revealed strategy S4 performed best (as judged by the sum of standardized benefits across all objectives), closely followed by strategies S5 and S2 (Figure 2). These best performing strategies covered a range of actions, but all included both burning and weed control. Strategy S4 included long burn intervals in either autumn or spring followed by weed control at Mt. Cottrell NCR and medium burn intervals in autumn and weed control for Little Raven and One Tree East reserves. Strategy S5 F I G U R E 2 Management strategy performance: Relative benefit for each objective for alternative management strategies (described in Table 3). (A) No weights, (B) Average weights across nine participants. NTG, natural temperate grassland; SHW, seasonal herbaceous wetland. See Table 1 for description of objectives and performance measures used to evaluate the relative benefit of alternative management strategies.
comprised short burn intervals in either season followed by weed control for all three reserves, while strategy S2 focused on medium burn intervals in Autumn with weed control across all three reserves (Figure 2a). The worst performing strategy was S6 that focused on medium burn intervals in spring followed by weed control in Mt. Cottrell NCR, long burn intervals and weed control in either season for Little Raven and medium burns and weed control in autumn at One Tree East reserve. The distribution of benefits also varied across the objectives (Figure 2a). For the top-performing strategy (S4), benefits were largely attributed to striped legless lizard, the non-listed flora and management costs. For S5, benefits were distributed more evenly across the plant species and golden sun moth with minor benefits attributed to other objectives. For strategy S2, the distribution of benefits was more evenly spread across all objectives ( Figure 2).
When weights were applied to objectives, the top performing strategy switched from S4 to S5, which was previously the second-best performing strategy ( Figure 2b). The worst performing strategy also switched from S6 to S7, which was previously the second-worst strategy. Strategy S4 focuses on a shorter burn interval compared to S5 and benefits the plants more than other species. The low weighting on costs seemed to have a large influence on the switch between S5 and S4. Strategy S2 rounded out the top three performing strategies where weighted benefits were spread more evenly across objectives but tended to be attributed more to striped legless lizard, and the non-listed flora (Figure 2b).
While individual stakeholder's weightings for objectives varied (Figure 1), their influence on the rank order of management strategies was minimal ( Figure S1). The three top performing strategies given individual stakeholder weightings were consistent (i.e., S5, S4, and S2), with slight variations in rankings ( Figure S1). Pairwise correlations of the ranks of management strategies between individuals ranged between 0.88 and 1.00 averaging 0.95 (Table S1).

| DISCUSSION
When managing for multiple conservation objectives, tailoring actions that optimize outcomes for all objectives is challenging and, in many cases, impossible. The cognitive burden of dealing with complex decisions directs our focus to the objectives believed to be most important, at the expense of others (Árvai et al., 2014). This tendency risks perverse outcomes that come from not adequately considering all objectives of interest (Árvai et al., 2014) or "status quo traps" where decisions are biased towards the options of common experience, and fail to fully evaluate less common alternatives (Hammond et al., 2003). SDM facilitates a genuine participatory approach for navigating complexity and providing stakeholders with a voice. The process helps to provide a clearer rationale for decisions and better understanding of the trade-off process so that stakeholders are more comfortable with decisions, even if their personal preferences do not match the ultimate decision (Brown & Ferguson, 2019;Gregory & Keeney, 2002;Liu et al., 2012).
For many decisions, not all objectives are considered of equal importance (Gregory & Long, 2009). However, without formal elicitation of the relative weights of objectives, such variation is often left undetected, or weighting of objectives is not considered (Keeney & Raiffa 1993). This may result in considerable misunderstanding about the reason behind strategy selection. In this study, we found that weights for objectives varied across stakeholders. These differences among stakeholders suggest that decisions that emerge from structured decision making processes will be specific to, and representative of, the stakeholder group who participated, and that they form part of the problem context. Had a different set of stakeholders been involved (e.g., only reptile experts, or only wetlands experts, or economists), the results may have been very different. Thus, it is imperative that the stakeholders are carefully selected to cover all relevant interests and whose contributions are ultimately recognized by the decision maker.
The swing weighting approach used in this study derives weights that are dependent on the specific decision context (von Winterfeldt & Edwards, 1986). Participants ranked objectives based on their understanding of the decision context and the estimated best and worst outcomes across the management strategies under consideration rather than an abstract or context free measure of overall importance of an objective. Under the MSA program, there is a requirement to manage all listed threatened species and ecosystems. At the onset of this research participants expressed the difficulty of that task given they care about all threatened species and ecosystems, but they overlap and have different burning preferences. Quantification of the outcomes of the alternative management strategies allowed participants to weight objectives and the species and ecosystems that are represented by them, more easily because the weighting represents in part the benefits of management within the decision context rather than an overall preference for one species or ecosystem over another.
Preferential weights were also likely to be influenced by factors other than the benefits across management strategies, including the extent to which each species or ecosystem existed outside the three reserves, either within the broader MSA management area or further afield. For example, at the time of this study, the striped legless lizard had only been detected in one reserve within the MSA area (Mt. Cottrell NCR), despite extensive monitoring. While the species has a broader distribution outside of the MSA area (Department of the Environment, 2012), stakeholder discussions revealed that if it occurred at only one location within the MSA management area, then it was important to manage it appropriately. This information is likely to have contributed to the higher weights for that species even though the benefits of the management strategies considered were small. Golden sun moth and SHW had lower relative weights, which may be partly attributable to the fact that they are found at multiple locations within other MSA reserves not considered here, as well as further afield. While we did not test the relative contribution of these factors explicitly, they were discussed at length as part of the participatory approach and were likely contributing factors in the allocation of weights. This emphasizes the importance of clear specification of the decision context and stakeholder discussion so that preferences are assigned from the same information base. It also highlights that if the decision context changes, for example when new data and information become available, or new management strategies or objectives are considered, then this is likely going to influence the weights, so preferences will need to be elicited again.
While the weights for objectives varied among stakeholders, the influence of these weights on the rank of the management strategies was minimal. The top three strategies were consistent across all stakeholders although there were slight changes in the rank order of the top two strategies. This may not be the case for different types of decision problems, particularly those with a broader suite of actions or different objectives. Understanding how weights influence the rank of management strategies is extremely useful because it helps to better direct discussions towards those factors that influence the decision the most, rather than focusing on elements of the decision problem that have little influence. If weights vary but ranks of options do not, then this may make it easier to arrive at a decision, despite disagreement among stakeholders.
For this study, we were fortunate to have quantitative models for the species, one ecosystem and management costs to estimate consequences for each action. For the other objectives, consequences were estimated by a subject matter expert. Ideally, we would have preferred to conduct a structured expert elicitation process using either the 4-point elicitation process (Speirs-Bridge et al., 2010) and the IDEA protocol (Hemming et al., 2018) with multiple experts, however, we were constrained by resourcing. Our alternative was to omit the relevant species and ecosystems from the analysis which we did not want to do given they are an important part of the grassland community. The use of varying information sources for estimating consequences demonstrates the versatility of the approach used, such that different types of data sources can be integrated to inform decisions.
This study does not explicitly incorporate uncertainty in the estimated consequences. It is likely the uncertainty is considerable and may have resulted in different preferred strategies depending on participant's tolerance to uncertainty. Participants that are risk averse may prefer strategies that have the least uncertainty even if it results in lower mean consequences. Participants that are more tolerant to uncertainty may choose strategies that are more uncertain but with the possibility of large gains. Future studies that elicit preferences given uncertain consequences are warranted.
In this study, we only considered ecological objectives and management costs. Grasslands are valued for a range of ecosystem services, biodiversity, social, and cultural objectives, including those of Traditional Owners (Bengtsson et al., 2019;Farrar et al., 2020;Gott et al., 2015;Reid, 2015). The method adopted in this study does not preclude inclusion of these other objectives. However, it does require clear specification of objectives and performance measures that can be quantified and that are sensitive to the management options under consideration. Additional management strategies that benefit those objectives may also need to be crafted to ensure the suite of alternatives captures all the objectives of interest as well as the identification or procurement of relevant data to predict outcomes of alternatives management strategies on those objectives.
Performance measures for all objectives were framed as benefits calculated as the difference in estimated outcomes with and without management. Benefits were quantified as increase in population size (for species) or improvement in condition (for communities). Benefits were used instead of direct consequences to provide a view of improvement of the alternative strategies compared to no management. This improvement can be masked if no management also results in some improvement. For many decision problems, no management may be a good option, particularly if costs of alternative actions are potentially prohibitive. In the selection of alternative actions, only actions that were deemed affordable were included. Thus, none of the strategies were cost prohibitive. When defining the scope of the study, there was a desire to actively manage the three sites, so a "no management" action was considered out of scope as part of the management scenarios. Instead, we developed a status quo scenario which captured the existing management regime for the three sites. We acknowledge that for other studies that might want to use our methodology, this might not be the case and including a no management action and using the direct descriptors of consequences could be preferred.
Structured decision making tools such as those used in this study help to identify areas of disagreement encourage discussion and help to resolve disagreements. Visuals such as Figure 2 capture not only the overall relative benefits from each strategy, but also the distribution of benefits across each objective and how they change once weights are applied. These are useful when discussing and developing the rationale for decisions.
Additional information such as practical considerations regarding on-ground implementation that may not be captured in the models or estimates of consequences can be discussed in conjunction with the results. In this study the top performing strategy when average weights were considered was Strategy 5. This strategy consisted of short burn intervals followed by weed control. However, in practice, short burn intervals can be difficult to implement if seasonal conditions make it unsafe to burn or if residential homes are close to the burn site. Fortunately, Strategies 4 and 2 also resulted in high benefits but had longer burn intervals that have less implementation issues, although the distribution of benefits for those two strategies differed from Strategy 5. Strategy 2 represents an inspirational burning regime. Even though it was not the optimal strategy, participants preferred it over the other strategies because it performed well, and the distribution of benefits were more evenly spread across species and ecosystems compared to the other strategies.
The overall preference for a strategy that included ease of implementation and equity of benefits across objectives suggests there were other unspecified objectives underpinning the decision. Ease of implementation could have been considered as a fundamental objective incorporated using a constructed scale. Equity of benefits across objectives could have been considered as a process objective that would help guide the development of alternative strategies that were more equitable rather than focusing on strategies that benefit one or two objectives. Without the structured process and the accompanying analysis, it is unclear if those unspecified objectives would have been revealed. This insight highlights the need to reiterate through the SDM steps to ensure all objectives are captured and update as more information about the decision context is revealed.
Government agencies are increasingly required to justify the use of public funding and be transparent about why decisions are made. SDM provides the scientific justification for decisions but recognizes the complexity of the problem and the role of judgment values and preferences which are often not communicated effectively. The context dependent nature of consequences of management and subsequent trade-offs are best communicated by quantifying the outcomes of alternative actions using the available evidence. For this case study the actions were nuanced focusing on different burning times and frequency, but there were still differences of opinion about burning at times that may not benefit particular species. The consequence table enables a more direct conversation about those differences and the trade-off analysis allows participants to specify which benefits are more desired in a quantitative and transparent manner.

| CONCLUSION
Evidence-based decision making is challenging but improves the likelihood of better outcomes and enfranchises stakeholders (Gregory & Long, 2009). Reframing routine management decisions as problems to be systematically addressed is immensely useful (Árvai et al., 2014). A structured participatory process, such as that demonstrated here, helps to avoid cognitive biases (Hammond et al., 2003), facilitates complete expression of objectives and complete exploration and interrogation of management options, and provides an explicit and transparent approach to performing trade-offs within an adaptive management framework (Brown & Ferguson, 2019). This process increases the likelihood of better decisions (Brown & Ferguson, 2019;Conroy & Peterson, 2013), it also builds collaborative links between managers and stakeholders at different levels, and fosters public confidence in management decisions that would otherwise be opaque.

DATA AVAILABILITY STATEMENT
Data are available via contacting the corresponding author.