Assessing prioritization measures for a private land conservation program in the U.S. Prairie Pothole Region

Private land conservation has become an important tool for protecting biodiversity and habitat, but methods for prioritizing and scheduling conservation on private land are still being developed. While return on investment methods have been suggested as a potential path forward, the different processes linking private landscapes to the socioeconomic systems in which they are embedded create unique challenges for scheduling conservation with this approach. We investigated a range of scheduling approaches within a return on investment framework for breeding waterfowl and broods in the Prairie Pothole Region of North Dakota, South Dakota, and Montana. Current conservation targeting for waterfowl in the region focuses mostly on the distribution and abundance of breeding waterfowl. We tested whether MaxGain approaches for waterfowl conservation differed from MinLoss approaches in terms of return on investment and which approach performed best in avoiding loss of waterfowl and broods separately. We also examined variation in results based upon the temporal scale of the abundance layers used for input and compared the region's current scheduling approach with results from our simulations. Our results suggested that MinLoss was the most efficient scheduling approach for both breeding waterfowl and broods and that using just breeding waterfowl to target areas for conservation programs might cause organizations to overlook important areas for broods, particularly over shorter timespans. The higher efficiency of MinLoss approaches in our simulations also indicated that incorporating probability of wetland drainage into decision‐making improved the overall return on investment. We recommend that future conservation scheduling for easements in the region and for private land conservation in general include some form of return on investment or cost‐effective analysis to make conservation more transparent.


| INTRODUCTION
Numerous studies have demonstrated that the current network of public lands cannot adequately protect the persistence of biological patterns and processes (Joppa & Pfaff, 2009;Rodrigues et al., 2004;Theobald et al., 2016). As a result, private land conservation is being used more frequently as a land conservation strategy in global efforts to protect biodiversity; complementing existing protected area networks and facilitating the landscape-scale conservation of critical ecosystems (Bingham et al., 2017;Capano et al., 2019;Mitchell, Stolton, et al., 2018). Private land conservation can include areas that have a primary conservation objective (i.e., privately protected areas) in addition to areas that provide conservation value regardless of their original conservation objective (other effective area-based measures: Kamal et al., 2015;. Participation is often voluntary and encouraged through financial incentives (Kamal et al., 2015). This strategy has become particularly important in landscapes like the United States, where public land acquisitions have slowed considerably (USGS GAP, 2022), and land is predominantly private (e.g., over 60% of land in USA is private; Lubowski et al., 2006).
Easements are a popular private land conservation strategy (Parker & Thurman, 2019). In the United States, their use has elevated rapidly, with a 50% increase in area under easement since 2010 (Land Trust Alliance, 2020). Despite the increase in easements, the extent to which they are delivering the desired impact remains understudied (see Braza, 2017;Claassen et al., 2017;Nolte et al., 2019). Here, by impact we refer to the difference between the outcomes that were observed in the presence of easements and the outcomes expected in the absence of easements (Ferraro, 2009). Understanding the difference that easements make in achieving conservation objectives is an essential component of prioritizing future locations for investment. Impact thus captures both the benefits of acting, such as achieving desired biological outcomes, as well as the risk of loss in the absence of acting.
Ideally, prioritization approaches would consider program impact, in terms of both the benefits and risk of loss, alongside other factors like monetary costs, social processes, threats to biodiversity, and biological processes (Gaston et al., 2002;Naidoo et al., 2006;Pressey et al., 2007). However, many organizations continue to target areas based on simple biological metrics such as species richness or population abundance and ignore the other dimensions of program impact such as threats or costs of action (Pressey et al., 2007;Ryan et al., 2014). Approaches that ignore threats or probability of loss can result in the overallocation of limited budgets to areas under little to no threat of conversion (Visconti et al., 2010). This has been demonstrated for public protected areas, where the failure to consider threats has contributed significantly to a global conservation portfolio biased toward landscapes that do little to prevent land cover conversion (Joppa & Pfaff, 2009, 2011. The potential for such biases to occur in private land conservation are likely to be even more pronounced due to private landowners opting in land they were unlikely to convert in the absence of protection, ultimately diverting limited program funds to properties not under threat and minimizing opportunities to prevent land conversion (Börner et al., 2017;Ferraro, 2008;Moon & Cocklin, 2011;Selinske et al., 2015). Thus, having a tool that aids managers in prioritizing higher impact properties and scheduling conservation actions over time might improve overall conservation outcomes for private land programs (Anyango-van Zwieten, 2021; Gooden & Sas-Rolfes, 2020;Parker & Thurman, 2019).
Return on investment analysis is one such tool that has been consistently recommended to improve the allocation of limited resources while also addressing concerns about conservation impact (Boyd et al., 2015;Cook et al., 2017;Game, 2013). Scheduling conservation actions within a return on investment framework regularly focuses on two main tactics (Knight, 2008;Sacre et al., 2019;Visconti et al., 2010): minimizing the loss of benefits over a given budget trajectory (MinLoss) or maximizing the benefits (conservation impact) for a given budget, regardless of risk (MaxGain). Here, by scheduling, we refer to conservation priority setting over space and time (Pressey & Taffs, 2001). Benefit within a return on investment framework is most frequently defined as some form of biodiversity and past approaches have included genetic, taxonomic, species or ecosystem diversity, or abundance measures (Ando et al., 1998;Arthur et al., 2004;Carwardine et al., 2008;Grantham et al., 2008;Murdoch et al., 2007;Polasky et al., 2001;Siikamaki & Layton, 2007;Underwood et al., 2008;Wilson et al., 2006Wilson et al., , 2010. More recently, though, there have been calls to consider return on investment analyses and program evaluations that address conservation impact (Boyd et al., 2015;Pressey et al., 2021). However, integrating impact measures into return on investment have not been applied in practice (Boyd et al., 2015).
To address this gap in the literature and provide a framework for regional conservation managers, we assessed both MinLoss and MaxGain scheduling approaches for private land conservation using conservation impact within a return on investment framework. We applied these approaches to the United States Fish and Wildlife (USFWS) Small Wetlands Acquisition Program (SWAP). The USFWS SWAP easement program is a well-known private lands conservation strategy and one of the primary tools for addressing breeding waterfowl habitat loss in the Prairie Pothole Region, where over 80% of the landscape is privately owned . The program places wetlands and grasslands under perpetual protection via legally binding easements wherein the landowner(s) concedes certain development and land use rights to the USFWS in exchange for a one-time payment. While wetlands represent at least half of this program's deliveries and provide the basis of the carrying capacity for waterfowl (Baldassare & Bolen, 2006)-the target group of the program-only the grassland portion of the program has been assessed in terms of return-on-investment and conservation impact (e.g., impact: Braza, 2017;Claassen et al., 2017;return on investment: Walker, Rotella, Loesch, et al., 2013).
On paper, the current prioritization strategy for SWAP wetland easements (hereafter easements) is closest to a MaxGain strategy, although it includes concepts from both MinLoss and MaxGain approaches. It follows a heuristic scoring system that first selects wetlands with a high abundance of breeding waterfowl pairs (Figure 1). Pair abundance is implicitly expected to serve as a surrogate for hen brooding habitat (U.S. Fish and Wildlife Service, 2016;PPJV, 2017). Within the identified high abundance wetlands, the approach then prioritizes wetlands with the highest probability of drainage based on size and context of wetland (Cortus et al., 2011). In practice, this prioritization approach underscores several of the challenges commonly faced by return on investment analyses, including the incorporation of cost, threats, and appropriate definition of biodiversity targets (Boyd et al., 2015). First, the current approach does not explicitly include cost. While conservation managers might implicitly consider costs when evaluating easement opportunities, providing a transparent reporting of how different easements' priorities were affected by costs would be challenging at best (Auerbach et al., 2014;Murdoch et al., 2007).
Second, as the probability of drainage is not considered jointly with the biological value of the wetlands (pair abundance: Figure 1), wetlands with a high probability of drainage that are used infrequently by breeding pairs as refugia would not be considered a high priority for conservation. In other words, not giving the probability of drainage an equal weight with biodiversity in the prioritization schema means that while areas of high biodiversity are targeted, areas with low probabilities of drainage might still be prioritized for conservation. As mentioned previously, this is of particular concern for private land conservation programs due to landowners frequently F I G U R E 1 Copy of hierarchical decision tree used by U.S. Fish and Wildlife Service and conservation partners to determine priority of wetland easement requests from 2008 to 2017 where ducks refer to breeding dabbling ducks, small, at-risk references temporary, seasonal, or <1 acre semi-permanent wetlands, <25 acres references all other wetlands <25 acres in size, ES stands for endangered species priority, WDP stands for wetland dependent migratory bird priority, Y stands for yes and N for No (modified from U.S. Fish and Wildlife Service, 2016) self-selecting to conserve wetlands that they never intended to convert (Börner et al., 2017;Ferraro, 2008;Moon & Cocklin, 2011;Selinske et al., 2015).
Finally, while the outcome of interest for SWAP is ultimately waterfowl recruits (PPJV, 2017), the metric used for assessing success is a long-term average of breeding pair abundance and distribution (Reynolds et al., 2006: table 1;Reynolds et al., 2007;Niemuth et al., 2010). Neither annual nor intra-annual spatiotemporal dynamics are considered, which ignores previous studies that have demonstrated the importance of both to breeding waterfowl habitat use in this region (Janke et al., 2017;Johnson & Grier 1988;Johnson et al., 2010;Kemink et al., 2021). Breeding pairs depend on the temporary basins that become available early in the spring. Broods lean heavily on the deeper semipermanent and seasonal basins that tend to stay ponded throughout the entire summer .
These characteristics make SWAP ideally suited as a case study for examining prioritization strategies for a private land conservation program in a return on investment framework. Specifically, our objectives were to: (1) determine if MaxGain approaches for waterfowl conservation differed from MinLoss approaches in terms of return on investment and, if so, which approach performed best in avoiding loss of waterfowl pairs and broods separately; (2) determine if integrating annual variation in abundance improved return on investment for waterfowl pairs and broods separately; (3) compare the estimated benefits of the current easement targeting approach for waterfowl in the region to MaxGain or Min-Loss approaches; and (4) develop recommendations for efficient scheduling of management actions, given limited resources in any one year, that addresses the habitat needs of both pairs and broods simultaneously.

| Study area
The Prairie Pothole Region is a grassland-wetland ecoregion encompassing over 700,000 km 2 that spans the United States-Canada (US-CA) border and includes parts of five US states and three CA provinces. This region is also one of the most anthropogenically altered landscapes in the world because of the predominance of private land ownership and highly arable soils (Doherty et al., 2018;Hoekstra et al., 2004). Estimates suggest that agricultural drainage has caused the loss of up to 89% of the wetlands in some parts of the region (Dahl, 1990(Dahl, , 2014, and the rates of native grassland losses have been compared to historic deforestation rates in the tropics (Wright & Wimberly, 2013). Despite these losses, the remaining habitat in the Prairie Pothole Region serves as the breeding grounds for a large proportion of the continent's shorebirds and grassland nesting obligates (Niemuth et al., 2008;Peterjohn & Sauer, 1999). The region is particularly important for breeding waterfowl and, in some years, over 50% of the breeding duck population counted in an annual survey are in this region (Johnson & Grier, 1988;Zimpfer et al., 2009).

| Study species
While 15 species of waterfowl breed in the Prairie Pothole Region, conservation planning by partner organizations tends to focus on five of these: the mallard (Anas platyrhynchos), gadwall (Mareca strepera), northern pintail (A. acuta), northern shoveler (Spatula clypeata), and bluewinged teal (S. discors) because of data availability and the importance of the region to their nesting success. To ensure our results paralleled the needs of regional organizations, we focused our efforts on these five major waterfowl species. Pair and breeding population estimates for these species track closely with May pond numbers (U.S. Fish and Wildlife Service, 2017, 2018Janke et al., 2017;Kemink et al., 2021) while brood estimates are often more closely related to both pond density and the size of occupied basins (Carrlson et al., 2018;Kemink et al., 2019;Walker, Rotella, Schmidt, et al., 2013).

| Overview of scheduling scenarios and simulations
To address our research objectives, we considered the return on investment across all possible combinations of targeting strategy (MaxGain, MinLoss), benefit function (broods, pairs), temporal resolution (averaged or annual estimates of broods and pairs), and rate of wetland loss. This resulted in a total of 24 possible scheduling scenarios. A number of methods of varying mathematical complexity for prioritizing and scheduling conservation exist and range from integer linear programing (e.g., Schuster et al., 2020) to the use of expert opinion (McKay et al., 2020). We used deterministic simulations of wetland drainage to estimate and compare the return on investment for the 24 easement scheduling scenarios in terms of the impact for both breeding waterfowl and broods (Table 1). We also used simulations to estimate the potential return on investment of the current scheduling approach as represented by the placement of wetland easements from 2008 to 2017 for comparison (Table 1).
To compare outcomes for each simulation we calculated the impact of each strategy, defined as the observed actual conservation outcomes (factual) and outcomes that could have occurred in the absence of intervention (counterfactual; Ferraro, 2009;Pressey et al., 2015). Simulations were necessary for our analysis because habitat layers representing wetland conditions before and after the application of the easement program were not available-so traditional quasi-experimental methods were not viable. Rather, we identified what were likely to be the highest and lowest bounds of the true drainage rates in the region (0.57% and 0.09%/year, respectively; Dahl, 2014;Oslund et al., 2010) and applied simulations to explore the potential range of impacts. We also included a third, very inflated drainage rate (1.00%/year) that was well beyond those recorded in the literature to assess sensitivity of the analysis to dramatic changes in wetland habitat loss.
In the following sections, we provide further detail about the simulation approach applied and assumptions of rate of loss (drainage rates). First, we describe the equations used to prioritize investment within each targeting strategy. Then, we give background information about the geospatial layers used within these equations to represent the wetlands, the cost of conservation, and the abundance of breeding waterfowl and broods. Finally, we define the metrics that we use to evaluate the impact of targeting strategies in each scheduling scenario.

| Simulation of scheduling scenarios
Half of the scenarios we simulated used MaxGain approaches, and the other half used MinLoss approaches for scheduling easement conservation (Table 1). The Min-Loss approaches included a variable to represent the probability of each wetland being drained that ranged from 0 (no probability of drainage) to 1 (most likely to be drained) that was calculated using the size of individual wetlands and their surrounding landscape composition ( Figure 2): such that area is the geospatial footprint of the wetland j in question and PC i is the percent of the landscape surrounding the basin that was defined as cropland by the annual U.S. Department of Agriculture's Cropland Data Layer in year i (U.S. Department of Agriculture National Agricultural Statistics Service 2008Service , 2009Service , 2010Service , 2011Service , 2012Service , 2013Service , 2014Service , 2015Service , 2016Service , 2017. We defined a landscape as an area 10.36 km 2 in size as this is the home range of a mallard and a metric commonly used in waterfowl studies (Baldassare & Bolen, 2006). Thus, smaller wetlands surrounded by higher percentages of cropland were more likely to be drained and large, difficult to drain basins had an almost zero probability of drainage. Wetland drainage in our simulations completely removed wetland footprints; no partial drainage was permitted.
Within the MaxGain and MinLoss prioritization strategies, we compared the return on investment of prioritizing based upon breeding waterfowl versus brood distributions as recent research has suggested that breeding waterfowl distributions in the region might not adequately represent areas important for brood conservation  Table 1). We also tested whether the conservation impact improved if we prioritized using annual predictions of breeding waterfowl and brood abundance versus a single layer for each that represented the average abundance across all 10 years of interest. Thus, in MaxGain strategies that used averaged abundance predictions, the return on investment equation used to prioritize wetlands for conservation was: where μ represents the average abundance on wetland j from 2008 to 2017 and cost represents the easement cost specific to year i on wetland j. For MaxGain strategies that used annual abundance predictions, the return on investment equation used to prioritize wetlands for conservation was: where abundance represents the annual abundance on wetland j in year i and cost represents the easement cost specific to year i on wetland j. In contrast, for MinLoss simulations that used averaged abundance predictions, the return on investment equation used to prioritize wetlands for conservation was: where μ represents the average abundance on wetland j from 2008 to 2017, P(d) represents the probability of wetland drainage on wetland j in year i, and cost represents the easement cost specific to year i on wetland j. And finally, for MinLoss simulations that used annual abundance predictions: Note: Selection strategies include MaxGain, MinLoss, and the current selection strategy represented by the 2008-2017 easement placements. Equations are also referenced in-text and describe how each wetland is prioritized within a simulation where P(d) refers to probability of wetland drainage, i represents year, j refers to wetland, abundance stands for annual abundance in year i on wetland j, and μ stands for average abundance from 2008 to 2017. Temporal resolution of abundance layers refers to whether wetland abundance estimates were averaged across the 10-years or whether annual estimates were used. Prioritized life history stage indicates whether the selection strategy equation was calculated using the abundance layer represented by the number of breeding ducks (male and female) in the spring (April and May) or number of broods later in the summer (July and August).
where abundance represents the annual abundance on wetland j in year i, P(d) represents the probability of wetland drainage on wetland j in year i, and cost represents the easement cost specific to year i on wetland j.
Simulations that looked at the impact of the current (2008-2017) prioritization strategy used the following equations for the average:  Table 1); (2) sorted all wetlands by decreasing total return on investment; (3) selected wetlands for protection with the highest return on investment estimates until budget was expended (set based on the annual observed U.S. Fish and Wildlife Service budget for the same period); and (4) drained wetlands indicated by the wetland drainage simulation. We coupled each simulation with a counterfactual simulation representing the wetland drainage that would have occurred without protection (factual being the opposite here and meaning the actual conservation outcomes : Ferraro, 2009, Pressey et al., 2015. Finally, we used data about the location of easements purchased from 2008 to 2017 in the United States Prairie Pothole Region, to evaluate the return on investment of the current prioritization strategy in terms of breeding waterfowl and broods for comparison.

| Planning units
Data layers that identify wetland drainage and separate it from the natural wetland hydrodynamics in the Prairie Pothole Region are not currently available. The large number of wetlands in the region (over 2 million) coupled with their small size have created processing and mechanical roadblocks to using remote sensing for wetland identification until recently (Sahour et al., 2021). Consequently, we used an adjusted version of the National Wetland Inventory spatial layer (U.S. Fish and Wildlife Service, 2014;PPJV, 2017) to represent potential wetlands available for conservation easements in our analysis. The NWI is a static geospatial layer created through the digitization of high-altitude imagery, the use of supplemental sources, and field data. Prior to all analyses, we removed wetlands that were protected perpetually before 2008 either by easements or fee title purchase, and only kept wetlands designated as semipermanent, seasonal, or temporary. Finally, we removed all wetlands >10.12 ha large, because although the USFWS does place easements on large wetlands (>10.12 ha), these are typically not considered to be at-risk of drainage or loss compared to smaller wetlands (PPJV, 2017). The resulting NWI vector layer was used at the start of all simulations (year = 2008).

| Breeding waterfowl and brood abundance
We used breeding waterfowl population and brood count data (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017) to develop spatiotemporal abundance models described in detail elsewhere . Here, breeding waterfowl refers to the male and female ducks arriving to the breeding grounds in the spring (April and May) and broods refers to the ducklings hatched later in the breeding season (July and August). The abundance models were used to calculate model-based predictions of the cumulative loss of carrying capacity for waterfowl breeding populations and broods from 2008 to 2017, for wetlands identified as drained from the static adjusted NWI layer used in the habitat loss scenario (PPJV, 2017). Because we were making predictions to a static wetland layer, methods for obtaining model-based predictions differed slightly from those in the original publication . For each year's predictions, we assumed that all wetland footprints were 100% full and we used these footprints to calculate the input variables to the models which included May pond count (breeding waterfowl and brood models), July wet area (brood), basin area (brood), and basin regime (brood: Kemink et al., 2021). We used the USDA Cropland Data Layers to estimate the perennial cover input variable in the brood models and left other variables at their mean values .
To create smoothed layers of mean abundance per km 2 we applied a moving window analysis to the modelbased predictions of abundance. Although the breeding waterfowl and brood models created posterior predictive distributions with credible intervals, we used the median of these distributions as input for the moving-window analyses. We then used the smoothed abundance layer to calculate breeding waterfowl and brood abundance at each wetland that was drained or protected in the analysis (e.g., wetland carrying capacity). We used a smoothed surface rather than individual wetland values because we were limited by the spatial grain of the breeding population data which were available only at the survey segment level . These estimates of abundance were used as the measure of benefit in the return on investment for each selection strategy in our analysis.

| Cost data
Financial offers made to landowners are based upon a complicated formula designed by the USFWS to maintain equity with current market prices while also ensuring high (50%-80%) acceptance rates of easements (Supporting Information). We used a layer developed recently by Nolte (2020) to represent the estimated 2010 market value of land. Because there were small areas in this layer that contained no data, we filled these missing values using the focal statistics tool in ArcMap 10.8 by taking the maximum of the values within a 3 x 3 cell window surrounding the location in question. Cells were 480 x 480 m large (Figure 2). We then extracted the value of the resulting raster to the centroid of all wetlands within the wetland layer we developed. Then, to calculate the USFWS wetland easement cost per parcel, we applied USFWS index values to the estimated market values per wetland. Index values are developed on an annual basis by a group of USFWS realty employees for the purpose of maintaining a landowner easement offer acceptance rate between 50% and 80% (Table 2; Supporting Information). We also adjusted for annual inflation using the consumer price index (CPI: US Department of Labor Bureau of Labor Statistic; Table 2). We identified annual budget limitations by totaling all digitized wetland easement data costs such that the amount of money available in each simulated year was the same amount spent in each actual year.

| Evaluation metrics
To estimate the return on investment of each scheduling scenario in terms of avoided loss we estimated the average treatment effect on the treated (ATT) at the end of each 10-year simulation using a difference-in-difference estimator (Manski, 2003). Here, ATT refers to the difference between the cumulative lost carrying capacity of breeding ducks or broods in the absence of easements (counterfactual) and the cumulative lost carrying capacity of breeding ducks or broods in the presence of easements (factual): factual abundance lost i where bird stands for either breeding waterfowl or broods, factual abundance lost i is the number of breeding waterfowl or brood abundance on drained wetlands in the presence of easements in year i, and counterfactual abundance lost i is the number of breeding waterfowl or broods on drained wetlands in the absence of easements in year i. We also present results in terms of percent avoided loss of breeding waterfowl and broods using the ATT bird estimate: We calculated the ATT and % avoided loss for both breeding waterfowl and broods regardless of the life history stage used to prioritize selection. For example, in simulation 1 (Table 1), the average number of breeding waterfowl per km 2 was used to prioritize wetlands for conservation. However, we calculated the ATT for both the average number of breeding waterfowl and broods per km 2 . This allowed us to determine if there were trade-offs for using just breeding waterfowl distributions rather than both breeding waterfowl and brood distributions to inform conservation prioritization strategies.
Finally, for each scheduling strategy we included a calculation for the remaining U.S. dollars (USD), which describes the total amount of money in USD remaining from the budget provided (2008-2017) after applying each 10-year simulation.
remaining USD 2017 ¼ X 2017 i¼2008 Budget i À total spent si Here, the budget is the total amount of money provided at the beginning of each year i, the total spent si is the total spent in a given year (i) specific to a prioritization strategy (s: MaxGain, MinLoss, or current [2008). We calculated this metric to ensure the strategies were comparable in terms of dollars spent as well as to determine whether one strategy would be more efficient than the other at using up the allotted funds over the 10-year simulations.

| RESULTS
At i = 0 (beginning of 2008), there was 834,415.50 ha in the at-risk wetland layer (NWI) and an associated 5,537,706 breeding waterfowl and 2,257,591 broods. Without any additional wetland easement conservation from 2008 to 2017 the simulated annual drainage rates resulted in a total of 7684.93 ha (0.09% annual loss rate), 47,996.83 ha (0.58%), or 83,438.1 ha (1.00%) being T A B L E 2 Index values developed on an annual basis by a group of United States Fish and Wildlife Service realty employees for the purpose of maintaining a landowner easement offer acceptance rate between 50% and 80%  -price-index-and-annual-percent-changes-from-1913to-2008. drained. In the results that follow, we present results from the highest drainage rate (1.00%) in text (Tables 3  and 4) but also provide results for the medium and low drainage rates in Tables S1-S4.
Across all simulations, we saw the most support for a scheduling approach to prioritization using a MinLoss strategy for both breeding waterfowl and brood abundance (Tables 3 and 4; Figures 3 and 4) Note: Temporal resolution indicates whether averaged or annual abundance layers were used for prioritization. The column "Prioritized life history stage" identifies whether breeding waterfowl or brood layers were used to prioritize areas for conservation in the scheduling strategy. The columns "Difference" and "% avoided loss broods" demonstrate the results of the difference-in-difference estimator and the return on investment of each strategy in terms of avoided loss of broods. Results are only shown for the inflated rate of wetland drainage (1.00%/year: total of 83,438.16 ha). The life history stage used to prioritize wetlands for conservation in each scenario is identified in the third column except for with regards to the current approach wherein this does not apply. Results from simulations with medium and low drainage rates can be found in the Supporting Information. Note: Temporal resolution indicates whether averaged or annual waterfowl abundance layers were used for prioritization. The column "Prioritized life history stage" identifies whether breeding waterfowl or brood layers were used to prioritize areas for conservation in the scheduling strategy. The columns "Difference" and "% avoided loss breeding ducks" demonstrate the results of the difference-in-difference estimator and the return on investment of each strategy in terms of avoided loss of breeding waterfowl. Results are only shown for the inflated rate of wetland drainage (1.00%/year: total of 85,447.9 ha). The life history stage used to prioritize wetlands for conservation in each scenario is identified in the third column except for with regards to the current approach wherein this does not apply. Results from simulations with medium and low drainage rates can be found in the Supporting Information.
strategies had less than $2000 remaining of their total provided budgets compared to the MaxGain strategies which had $10,983.49-$15,929.85 unused funds remaining after 10 years (Figure 4). While the MinLoss strategy always demonstrated a higher percent avoided loss than the MaxGain strategy, we note that the relative difference in avoided loss between the two strategies decreased with decreasing rates of drainage (see Tables S1-S4). Within both the MinLoss and MaxGain strategies, the estimates for ATT and percent avoided loss regarding use of annual versus average spatiotemporal abundance prediction layers demonstrated support for use of the annual layers. Regardless of whether brood or breeding waterfowl layers were used for prioritizing conservation strategies, using the annual abundance layers always provided a higher ATT and avoided loss in both groups (breeding waterfowl and broods: Tables 3 and 4; Figure 3).
Of the wetlands selected for protection in the MinLoss annual prioritization strategies (nbrood = 635,401 wetlands; nbreeding = 606,645 wetlands), on average 10% overlapped between the two life history strategies each year and, by the end of the 10-year simulations 57% of the basins selected for protection were the same across both strategies. Within the MinLoss strategy simulations, using one life history stage as a prioritization surrogate for the other always resulted in a decrease in percent avoided loss (Table 3, Figures 3 and 5). Our results also suggested that the 2008-2017 wetland easements selected using the current prioritization approach provided values of percent avoided loss that were lower than both the MinLoss and MaxGain prioritization strategies (Tables 3 and 4; Figure 3). Similarly, wetlands selected for protection using the current prioritization strategy had a lower probability of drainage on average (43.35%) than those selected for protection using the MinLoss approach (73.5%) and using the MaxGain approach (51.28%).

| DISCUSSION
We examined wetland easements in the USFWS SWAP to evaluate how current conservation scheduling approaches in the Prairie Pothole Region compared to MinLoss and MaxGain scheduling approaches. We simulated 24 scheduling scenarios across a range of potential wetland drainage rates and calculated the return on investment in terms of the avoided loss of breeding F I G U R E 3 Avoided loss in terms of (a) breeding waterfowl and (b) broods when prioritizing by either breeding waterfowl (black or white bars) or broods (dark gray or light gray). Results from the simulation using the highest drainage rate are shown (1.00%/year). The red line presents the avoided loss in terms of breeding waterfowl (a) and broods (b) for the scheduling strategy that was used on the landscape in the region from 2008 to 2017. This value was calculated using annual abundance layers. Results for the current strategy are not shown for averaged layers as differences between the two are not detectable on the graph F I G U R E 4 Graph of eight simulations and the remaining USD after the last year (2017). Bars represent which life history stage and strategy was used to prioritize wetlands waterfowl and broods. Our results suggested that a Min-Loss approach that explicitly included both costs and threats to biodiversity outcomes could improve the efficiency of the current spatial prioritization and scheduling processes. They also revealed that within this MinLoss approach, there was not strong evidence to support the use of breeding waterfowl as a surrogate to prioritize brood conservation and vice versa.
Despite the similarities between the current 2008-2017 prioritization and MaxGain approaches, the avoided loss estimates from our MaxGain simulations were always higher. The main difference on paper between MaxGain and the 2008-2017 prioritizations using the actual easement locations was the inclusion of wetland cost per ha. However, one would still expect the overall avoided loss of these strategies to be similar because the first step in the 2008-2017 hierarchical prioritization approach is identifying areas of high importance to breeding waterfowl. Instead, the higher avoided loss exhibited by the MaxGain scheduling strategy in our results suggests that the process of self-selection by landowners into this program is influencing overall conservation impact. In fact, the mean probability of wetlands' drainage on easements selected for protection under the current scheduling strategy was slightly lower than the probability of drainage on easements selected for protection in the MaxGain strategy. This suggests landowners are leveraging their access to information about land management (information asymmetries) to sell easements on wetlands that they never intended to drain (Ferraro, 2008).
The study area within the Prairie Pothole Region of North Dakota, South Dakota and Montana and the wetlands selected for protection based upon a 10-year simulation using annual layers of breeding waterfowl population abundance and targeting MinLoss of breeding waterfowl numbers (blue) and using annual brood abundance and targeting MinLoss of broods (gold) The current conservation scheduling strategy addresses the threat of wetland drainage implicitly through the hierarchical decision-making process (U.S. Fish and Wildlife Service, 2016; Figure 1). Because this strategy is hierarchical, it automatically prioritizes biodiversity (breeding waterfowl abundance) rather than avoided loss of breeding waterfowl, which can lead to inefficient and ineffective interventions (e.g., program impact: Ferraro, 2009;Pressey et al., 2015). As evidence of this inefficiency, the MinLoss strategy, which considered the threat of wetland drainage and budget limitations jointly, outperformed both the MaxGain strategy and the current prioritization approach. Others have demonstrated that MinLoss tends to outperform Max-Gain prioritization approaches (Costello & Polasky, 2004;Drechsler, 2005;Pressey et al., 2004;Wilson et al., 2006); especially when habitat loss is ongoing and spatially variable (Adams & Setterfield, 2015;Murdoch et al., 2007;Visconti et al., 2010).
While the conditions in which MinLoss is a preferable strategy reflect those outlined in our simulations (spatially variable and continuous habitat loss rates), it is worth noting that MaxGain has been demonstrated to be more efficient in alternative conditions when the threats to habitat are not spatially variable and/or when there is substantial uncertainty in conservation funding or opportunity (Costello & Polasky, 2004;Wilson et al., 2006). In practice, an organization's prioritization strategy will likely fall somewhere in between a MinLoss and Max-Gain approach (Sacre et al., 2019) and the resource allocation to each will depend on biodiversity targets and the time horizon for ecological objectives (Armsworth, 2018;Sacre et al., 2019). If organizational goals favor high immediate gains, a targeting strategy weighted toward MinLoss to protect high risk areas might be preferable (Armsworth, 2018;Sacre et al., 2019). While there are often contextual factors for organizations choosing a mixed targeting approach, such as exhibited by the SWAP approach, our analysis demonstrates that these can result in lower conservation gains. Our results found that both MaxGain and MinLoss approaches resulted in higher avoided loss relative to the current prioritization approach. This emphasizes the need to consider benefits, threats, and cost jointly rather than hierarchically. Thus, improvements to the current targeting approach could be made by embracing our simple to calculate and implement return on investment approach.
Although our results suggested that threat, cost, and biological information should be considered jointly, they also indicated that scheduling conservation for breeding waterfowl and broods should be considered separately. Regardless of which scheduling strategy was used in the simulations, the use of surrogacy for setting conservation priorities for both breeding waterfowl and brood habitat was never strongly supported by measures of avoided loss. Even though our simulations represented the bestcase scenarios (wherein all ponds are 100% wet all season), our results demonstrate that a surrogacy approach could decrease conservation impact. For example, prioritizing wetlands for conservation based on brood abundance provided an avoided loss of return on investment of 0.58% for broods compared to 0.03% when we prioritized using average pair abundance as a surrogate (Table 4).
The practical application of prioritization schemes representing two life history stages, that also integrate strategies like MinLoss, still represents a logistical challenge. Actual easement interventions traditionally purchase wetland easements by parcel (see Figure 1 and Supporting Information) wherein landowners will cede a contiguous area of their property containing wetlands for sale as an easement. While this is an efficient way to acquire a large area for conservation, our analysis demonstrates that it does not necessarily provide the highest impact possible in terms of conservation outcomes. Rather, the sale of easements by individual wetlands, likely scattered across discontinuous portions of property would be more effective in terms of biological impact. We acknowledge, though, that this assessment does not account for the additional administrative and enforcement costs that such a dispersed approach would require.
Costs above and beyond the acquisition cost of easements such as staff time and enforcement costs, would be a crucial next step for analyses to include and were not possible to include at the appropriate level of detail in this analysis (Armsworth, 2014;Naidoo et al., 2006). Restrictions on where and how certain funds are spent provide challenges for assessing the full return on investment of this program as well. Migratory Bird Conservation Fund dollars, for example, cannot be used to purchase wetland easements beyond a certain extent in North Dakotan counties (U.S. Fish and Wildlife Service, 2016). Further, a spatiotemporally explicit layer of risk of wetland drainage would be invaluable to similar future analyses in this region. Finally, leakage or spillover effects could influence overall program impact and could be considered in future decision making (Oestreicher et al., 2009;Pfaff & Robalino, 2012).

| CONCLUSION
We tested scheduling approaches within a return on investment framework for the USFWS SWAP, an important conservation program in a landscape dominated by private land ownership . Our results provided support for the use of a MinLoss scheduling approach over the hierarchical approach used currently and a MaxGain approach. We suggest that future scheduling for the SWAP consider a MinLoss approach to prioritize wetlands for conservation. Future research exploring solutions to existing information asymmetries in the current program model would be valuable. Reverse auctions, for example, have proven successful in other studies at addressing this challenge (Brown et al., 2011;Liu, 2021).

AUTHOR CONTRIBUTIONS
Kaylan M. Kemink, Robert. L. Pressey, and Vanessa M. Adams conceived the project and study design with assistance from Sarah K. Olimb, Todd Frerichs, and Randy Renner. Aidan M. Healey conducted data collection and organization with assistance from Boyan Liu. Kaylan M. Kemink conducted the analysis and wrote the manuscript with assistance and editing from Robert. L. Pressey and Vanessa M. Adams.

ACKNOWLEDGMENTS
This paper was made possible with support from Ducks Unlimited, Inc. and support for Robert. L. Pressey was provided by the Australian Research Council. We thank AJ for his review and comments on the manuscript. We also thank the AE and EIC for their comments as well as two anonymous reviewers whose suggestions greatly improved this manuscript. We would also like to thank the numerous staff of the North Dakota, South Dakota, and Montana realty offices who assisted with the data acquisition process. The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service.

CONFLICT OF INTEREST STATEMENT
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
All data used in this manuscript are publicly available through government websites or prior publications. The spatial data on wetland easements were collected via a Memorandum of Understanding with the USFWS and so are not available for dissemination. Aidan M. Healey https://orcid.org/0000-0001-7274-367X Boyan Liu https://orcid.org/0000-0002-7646-5299