Delivering wildlife habitat on productive agricultural lands cost‐effectively: The case of migratory shorebirds on California rice lands

Promoting wildlife habitat on working agricultural land is a growing conservation priority, and agri‐environment schemes are using payments for environmental services (PES) to reach these privately owned lands. PES can be ecologically effective, but also expensive, putting pressure on these schemes to be cost‐effective, maximizing the conservation value of their limited resources. This study assesses the cost‐effectiveness of four PES schemes in California that paid rice growers to provide temporary flood habitat on their working lands in support of shorebirds. It examines whether the schemes (1) paid for flood habitat only when needed (targeting); (2) paid for habitat that was actually delivered and would not otherwise be provided (additionality); (3) sacrificed as little habitat as possible when the habitat competed with production (ag/wildlife balance); and (4) fostered a commitment among growers to maintain the habitat after payments ended (permanence). Results show that the schemes fell short in each of the four goals, and they expose several factors that undermined their cost‐effectiveness. First, variable weather patterns altered the date in which the shorebird‐habitat gap emerged year‐to‐year, creating a dynamic and unpredictable target for the schemes to address. Second, growers' views on the compatibility of the flood habitat with their rice production varied widely and changed rapidly as the agricultural calendar progressed, making it a challenge to mitigate potential ag/wildlife conflicts. Third, water for flood habitat is expensive, and growers proved unwilling to shoulder that expense once payments ended. These results highlight the need for schemes to adopt design elements that can add flexibility, such as cancelation clauses, so that they can adjust to dynamic targets and adapt to changing ag/wildlife relationships. Results also suggest that PES schemes operating on working lands may require long‐term external support. That finding amplifies the imperative that schemes use their conservation resources cost‐effectively.

K E Y W O R D S ag/wildlife tradeoffs, agri-environment schemes, California, conflict mitigation, costeffectiveness, dynamic conservation, payments for environmental services, remote sensing, shorebirds, wildlife

| INTRODUCTION
Promoting wildlife habitat on privately owned, working agricultural lands is a growing conservation priority, as nature preserves prove insufficient for stemming biodiversity loss in these landscapes (Burger Jr et al., 2019;Donald & Evans, 2006;Poiani et al., 2000;Stralberg et al., 2011). To reach these lands, many agri-environment schemes use incentive-based tools, such as payments for environmental services (PES; Baylis et al., 2008;Cattaneo et al., 2005). With PES, schemes offer growers financial incentives to adopt practices that support socially desirable environmental services with no market value, such as wildlife habitat (Wunder, 2005).
In high-yield landscapes (e.g., conventionally cropped lands), PES schemes have taken two different approaches (Bat ary et al., 2015). Some pay growers to restore a matrix of diverse natural habitat on the unproductive lands that exist in these areas, such as field edges and waterway buffers. That tactic is especially effective for supporting a wide range of species (Ekroos et al., 2014). Others identify a particular species of concern and pay growers to deliver its specific habitat needs (Bat ary et al., 2015). When that species utilizes on-field (cultivated) lands, access to those lands for conservation purposes is temporary. So, schemes must enact dynamic conservation-inducing short-lived habitat to meet a short-term gap in the targeted species' needs (Bull et al., 2013;Reynolds et al., 2017). Both approaches (off-field permanent habitat and on-field, targeted, dynamic habitat) have proven ecologically effective, as both improve biodiversity outcomes (Bat ary et al., 2015). But PES schemes must also be cost-effective, maximizing the conservation value of every dollar they spend (Ansell et al., 2016;Engel, 2016;Ferraro & Pattanayak, 2006). Through that lens, schemes inducing on-field, targeted, dynamic habitat face especially difficult challenges.
To demonstrate cost-effectiveness, the PES literature has developed several specific goals, a subset of which are applicable to the wildlife-habitat case (Engel, 2016). Schemes should: (1) only pay for habitat that is delivered and would not otherwise be provided (additionality; Engel et al., 2008); (2) support habitat when and where it provides the greatest conservation value (temporal and spatial targeting; Johst et al., 2015;Wünscher et al., 2008); (3) offer growers the minimum fiscal incentive necessary to induce participation (costefficiency; Ansell et al., 2016); (4) sacrifice as little habitat as necessary to mitigate conflict with agricultural production and ensure participation (ag/wildlife balance; Ferraro & Pattanayak, 2006); and (5) foster a commitment among growers to maintain the habitat after payments end (permanence; Dayer et al., 2018).
Three of these goals are particularly challenging for schemes that deliver on-field, targeted, dynamic habitat. For one, accurate targeting can be especially complex because schemes must deliver habitat within a specific timeframe. As well, the ag/wildlife balance can be very sensitive and difficult to navigate because on-field habitat interfaces more directly with production. Finally, establishing permanence can face exceptionally high barriers because the cost of inducing habitat afresh each year can be higher than maintaining permanent habitat.
In the United States, some of the first PES schemes to implement an on-field, targeted, dynamic approach were designed to support habitat for migratory shorebirds on the rice fields of California's Sacramento Valley (MBCP, 2014). In this region, many growers use postharvest floods to accelerate the decomposition of their rice stubble, and those floods serendipitously provide habitat for many waterbirds, such as ducks and geese (Eadie et al., 2008). They do not, however, meet the needs of many migratory shorebird species. In the fall, the floods start after shorebird migrations have already begun; throughout the winter, they are often too deep; and in the late winter/early spring, they end before many species have finished passing through the region as growers dry out their fields to prepare for the next season's crop (CVJV, 2006;Dybala et al., 2017;Eadie et al., 2008;Shuford et al., 1998;Strum et al., 2013). Starting in 2005, the Natural Resource Conservation Service (NRCS), and later The Nature Conservancy (TNC), began offering growers conservation payments to maintain their floods longer and more shallowly than believed typical, and thereby help close the shorebird flood-habitat gap (MBCP, 2014;Reynolds et al., 2017).
This study assesses whether these PES schemes closed that gap cost-effectively (maximizing the conservation value of their payments). It examines whether the schemes delivered extra flooding when the floodhabitat gap developed (targeting); induced more shallow floods than historically available and closed the gap (additionality), supported as much flood-habitat as possible in the face of potential incompatibility with rice production (ag/wildlife balance), and engendered a commitment among growers to continue that extra flooding after the schemes ended (permanence). The study does not address whether the schemes offered the minimum payment necessary to ensure participation (cost-efficiency), which was out of the scope of this analysis. Results reveal that the schemes fell short in each of these goals. But findings also clarify why those problems occurred and point toward PES-design elements that can help conservation managers support wildlife habitat on cultivated lands more cost-effectively moving forward.

| Study area
Historically, the Sacramento Valley was a massive seasonal wetland that provided habitat for millions of migratory birds and other wildlife on the Pacific Flyway. Early California settlers converted over 95% of those lands to agriculture and other uses, and wildlife populations plummeted (Frayer et al., 1989). Over time, rice became the dominant crop, and under that management system most growers burned their rice straw after harvest. In the 1990s, the state began to regulate that burning for air-quality purposes, which forced growers to adopt new residuemanagement strategies. Most growers now decompose the straw within their fields, and many add water to accelerate that decay. That increase in post-harvest flooding has increased the amount of habitat available for many waterbirds (Eadie et al., 2008;Elphick & Oring, 1998).
Much of that flooding does not, however, meet shorebird needs (CVJV, 2006). A large portion of it is deep, whereas shorebirds prefer shallowly flooded fields (<10 cm) to forage for food and roost (Strum et al., 2013). As well, the floods start to decline in January, whereas some shorebird species occupy the area into May (Dybala et al., 2017). In consequence, a gap emerges between flood-habitat availability and food-energy needs some time in late spring (Dybala et al., 2017;MBCP, 2014). Environmental managers would like to reduce that gap by supporting more shallow flooding on rice fields through the end of March (CVJV, 2006). Ideally, shorebirds will forage within those fields through the end of that month, and that behavior will reduce pressure on the nearby publicly managed wetlands. In early April, when those flooded fields dry out, the birds will move into the nature preserves and find sufficient resources to meet their needs through the rest of their stopover (CVJV, 2006). See Supporting Information for a regional map.
Between 2005 and 2015, NRCS managed three PES schemes that supported that plan: The Conservation Security Program (CSP); the Migratory Bird Habitat Initiative (MBHI); and the Waterbird Habitat Enhancement Program (WHEP). All three compensated growers for adopting flood practices designed to support waterbirds, including shorebirds. In 2014 and 2015, TNC managed an additional scheme with the same intent, called BirdReturns (Table 1).

| Uncertainty around the shorebird habitat gap
These schemes were challenging to design because, to date, it has not been clear when or why the late winter/early spring shorebird flood-habitat gap actually develops. There are three main reasons for that uncertainty. First, it is well understood that growers dry out their fields in the early spring to prepare for the next planting season, but the exact timeframe in which that process occurs has not been clear. Several factors influence the timing. Flooding relatively late can be competitive with rice production because wet ground delays spring planting, and planting late can lower yields (Linquist & Espe, 2015). But flooding late can also be complimentary with production because flooding longer ensures more straw decomposition. It also offers more opportunities for hunting and birding (Strum et al., 2013). Soil type impacts the timing decision as well because some soils take longer to dry out than others, and soils vary within the region (Jones et al., 1950). How many growers end their flooding relatively early or late is not known.
Second, water supplies for flooding are not always available, affordable, or reliable. Growers get their postharvest flood water from three sources: irrigation districts, groundwater and rainfall. Many separate federal, state and private irrigation districts serve the region, and they all operate under different water-rights regimes, but most stop delivering surface water by February 1st. Those growers who want to flood longer have three choices. They can make special arrangements to purchase surface water from a district (if available). They can pump groundwater. Or, they can rely on rainfall. If rains fail and growers cannot afford the two alternatives, their fields dry out despite their preference for floods. It has not been clear how many growers rely solely on rainfall, but many likely fit that profile as flood levels in February vary year-to-year by weather condition (Schaffer-Smith et al., 2017).
Third, conservation science has yet to determine how much shallowly flooded rice habitat shorebirds actually need. An increasing number of studies are examining this issue, but the answer remains unknown (e.g., CVJV, 2006; Dybala et al., 2017;Golet et al., 2018;Strum et al., 2013). Within the context of these uncertainties, some research suggests that the flood-habitat gap emerges by mid-February (MBCP, 2014). More recent studies indicate that the gap may not develop until as late as mid-March (Dybala et al., 2017;Golet et al., 2018).

| The PES schemes
Each of the four NRCS and TNC schemes supported a unique set of practices and protocols, but their rules concerning extended-flood practices overlapped significantly ( Table 1). The NRCS schemes required that growers fill their fields with water by their start date, and thereafter supported two possible types of flood management: active or passive. For active flooding, growers had to maintain a specified water depth (5-10 cm) on their contracted fields throughout the entire length of the contract. When the weather was dry and flood levels dropped below the minimum, growers had to pump or purchase water to replenish them. PES payments were relatively high to cover the cost of that water. For passive flooding, growers kept their fields plugged (i.e., kept boards in their tailwater structures to stop drainage) and then relied on rainwater to replenish them. When rains failed, the fields were allowed to dry out. Payments for this practice were relatively low. See Supporting Information for NCRS practice payment schedules showing differentiated payments. By supporting both passive and active flooding, the NRCS schemes addressed two possible reasons for the flood habitat gap: (1) growers choosing to stop flooding (unplugging their fields) too early because they wanted to reduce the risk of delays in their spring planting; and (2) growers not replenishing their floods during droughts because the cost of pumping or purchasing water was too high. TNC supported active flooding only.
Three of the schemes included parameters designed to mitigate potentials conflict between late flooding and early spring planting. MBHI and WHEP used variabledrawdown rules as an ag/wildlife tradeoff (Table 1). Under those rules, growers could unplug 25% of their fields (or lower the depth of their floods by 25%) each successive week in February. TNC handled the issue through a conservation auction. It solicited bids from growers for flooding their fields in several temporal increments (4-8 weeks in duration) within the February-April timeframe (Table 1). Growers bid on the increments in which they were willing to flood, and submitted relatively low-or high-cost bids, depending on the cost of water and on how complementary or competitive they considered flooding within those increments. TNC was then able to choose the lowest-cost bids and use the savings to help cover the higher-cost bids later in the season, when flooding becomes more competitive with production (Reynolds et al., 2017).

| Data
To identify the flood-habitat gap and assess how well the PES schemes reduced it, I used four sources of data: Landsat satellite images (downloaded from USGS), historical weather data (purchased from Weather Source), anonymous scheme-contract databases (provided by NRCS and TNC) and a recent biological study of shorebird habitat needs (Dybala et al., 2017). To understand growers' views concerning the impact of extended flooding on their rice management regimes, and to gauge growers' intensions after the schemes ended, I conducted two surveys, in 2011 and 2015. The surveys yielded 47 structured questionnaires from growers who adopted extended-flood practices through one of the NRCS schemes. Twenty-four growers also responded to openended questions, providing qualitative data. See Supporting Information for details on the survey protocols and instruments.

| Flood patterns (with-and without PES support) under wet and dry conditions
With the Landsat imagery, I produced 16 years of floodcover maps (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015). These maps depicted just rice lands and identified three cover classes: shallow (2.5-10 cm), deep (10+ cm), and no flood (<2.5 cm). See Supporting Information for details on how I created those maps, including pre-processing, classification, and accuracy assessments. I then split those maps into two groups, based on whether they were taken during wet-or dryweather conditions. In wet conditions, both active-and passive-flood practices produced and maintained flooding, so both forms of management were visible (as floods) in the maps. In dry conditions, the passively flooded fields eventually dried out, leaving only the actively flooded fields visible in the maps. I considered conditions "dry" if less than 2.5 cumulative centimeters of rain fell (and "wet" if more) in the 14 days prior to the satellite flyover. I then calculated the hectares of flooding observed in each map and ordered that data by day of the month, not by year (Table 2). Next, using the scheme-contract databases, I calculated the number of hectares under PES-contract for flooding. I aggregated hectarages across all four schemes, differentiating passive-and active-flood practices, and delineating hectarages by week in order to account for variable drawdown rules (Table 3). The databases did not specify how growers chose to manage their drawdowns. I assumed that all growers unplugged their fields (instead of reducing flood depths) because that approach dries out fields fastest.
I then produced two flood-pattern graphs, for wet and dry conditions, using bars to display flooded hectarages, by date, throughout the season (Figure 1). These graphs differentiate PES-supported (red) and unsupported (blue) flooding at each date (the bars). In wet conditions, I made that distinction by assuming that all PES-contracted flooding (active or passive) was shallow and visible in the maps, and that any additional visible flooding was unsupported ( Figure 1b). In dry conditions, I assumed that all PES-contracted flooding (active or passive) was shallow and visible in the maps during the first two weeks of February because all contracts mandated full flood-up on February 1st. After the 14th, if dry conditions persisted, I assumed that only actively flooded fields would remain flooded and visible in the maps, whether contracted (red), or unsupported (blue) (Figure 1a).
Finally, I estimated a shallow-flood pattern for wet and dry condition (Figure 1: gray shading). These gray-shaded areas roughly encompass the lowest and highest hectarages of unsupported shallow flooding (blue bars) observed over the 16 years. They represent typical flood patterns for the region, without PES intervention.

| Timetable in which growers unplug their fields
To help determine when growers unplug their fields, I analyzed a subset of maps that could reveal plugged fields. I selected five maps in which the Landsat satellite flyover dates occurred two weeks after a large rain event (>2.5 cm). These rain events were large enough to replenish plugged fields, and the dry conditions extended long enough to allow unplugged fields to drain. For each map, I reported the hectarages of flooded fields as a proportion of that same season's peak flood hectarages (in December or January) in order to account for varying amounts of rice production year-to-year. I also subtracted out any PES-supported hectarages, as that flooding would not necessarily have occurred without external support (Table 4).

| A benchmark for the floodhabitat gap
To identify shortfalls in shallowly flood rice land for shorebirds (the flood-habitat gap), I derived a minimumhectares benchmark from Dybala et al., 2017. That study estimated habitat availability for the entire Central Valley. It then used a model to identify when that habitat becomes insufficient to meet shorebird energy needs at baseline and 2Â baseline populations. It concluded that those deficits emerged in April and mid-March respectively (Dybala et al., 2017). In the 2Â baseline case, it estimated that the mid-March deficit emerged while 35,500 ha of shallowly flooded rice fields were still available in the Central Valley. Of that total, 94% of those floods (33,350 ha) were in this study's area of analysis.
I used that estimate as a benchmark, based on the premise that, if flood hectarages drop below 33,350 ha, even before mid-March, a habitat gap is likely emerging. If rains continue to fail and additional irrigation is not applied, food energy will not last long, although how long is not clear from available evidence. Dybala et al. (2017) anticipated a 33% increase in energy demand on March 1 as spring-migrating populations begin passing through. Hence, I applied a correction factor, decreasing the benchmark by 33% (to 22,240 ha) prior to March 1. This T A B L E 2 Flood-cover estimates (ha) from the satellite imagery.

Dry conditions
Wet conditions Lettered events are used in Figure 1, signified with the same letters. Not all observations fit. b Shallow = 2.5-<10 cm; Deep = >10 cm.
benchmark (depicted as a solid green line in the floodpattern graphs) is not definitive, but represents current understanding.

| Assessing cost-effectiveness
I evaluated whether the schemes met four goals of costeffectiveness: temporal targeting, additionality, ag/wildlife balance and permanence. To assess temporal targeting, I first characterized the scheme's target. Using the floodpattern graphs, I determined both when the habitat gap emerged (when unsupported flooding fell below the Dybala et al. benchmark) and why it occurred (droughts or unplugged fields) in order to clarify which types of practices could reverse that decline. The schemes were considered on target if they induced extra flooding after the flood-habitat gap emerged. They induced additionality if they raised flood levels above typical flood patterns and exceeded (or got substantially closer to) the Dybala et al. benchmark.
To assess the ag/wildlife balance, I first determined when flooding shifted from being complimentary to competitive with production. That timing varied by grower, but I focused on the aggregate, regional level. I interpreted plugged fields and unsupported flooding as signs of compatibility, and unplugged and rapidly draining floods as signs of incompatibility. I defined the timeframe of the regional shift as the point at which growers (not droughts) drove the decline in flooding and caused flood levels to drop below the Dybala et al. benchmark. Next, I assessed whether the timing of the NRCS schemes' variable drawdown rules aligned with that compatibility shift and sacrificed as little flooding as necessary to ensure participation. For the TNC scheme, I examined whether its conservation auction (and higher payments) were able to overcome rising incompatibility and secure significant participation in the late Spring. To address permanence, I analyzed the survey data to ascertain whether growers' perceptions about late flooding changed during the schemes, and whether they intended to maintain any of the extended-flood practices after the schemes ended.

| The target
The date by which the flood-habitat gap emerged varied by weather. In dry conditions, typical flood hectarages (gray shading) fell below the Dybala et al. benchmark (solid green line) as early as February 17 (Figure 1a). In wet conditions, that crossover occurred around March 10 ( Figure 1b). Hence, the start date for the gap was dynamic and unpredictable, depending on the weather year to year.

April
Week 1 Active 1063 Note: Hectarages are aggregated across all four PES schemes (CSP, MBHI, WHEP, and TNC). Growers under VD contracts are assumed to have unplugged 25% their fields each week.
One factor driving the decline in flooding was growers unplugging their fields. Table 4 provides a window into when that happened. It shows 87% of fields flooded (and thus plugged) on February 1, 53% on February 14, 48% on February 22, and then 8% on March 4 (Table 4). This evidence indicates that growers unplugged about 40% of the region's rice fields some time within the first and third week of February, and then unplugged 40% more within the last week of February and the first week of March.
A second factor causing the decline in flooding was weather, and that weather interacted with the presence of plugged and unplugged fields. In dry conditions, flood levels dropped below the Dybala et al. benchmark by F I G U R E 1 Flood patterns.
February 17, even though over half of the region's fields were likely still plugged through February 14 (Figure 1a; Table 4). This evidence clarifies that few fields were managed under active-flood practices, maintaining floods with groundwater or purchased surface water. Instead, most were managed under passive-flood practices, drying out when rains failed. In wet-conditions, overall flood levels remained well above the Dybala et al. benchmark into the second week of March, even though over 90% of rice fields were unplugged by March 8 (Figure 1b; Table 4). In wet conditions, the fields drained slowly and extended the availability of flood-habitat. After March 8, flood-levels dropped rapidly, even in wet weather and slow drainage conditions (Figure 1b). That steep decline suggests that growers took additional proactive measures to accelerate the drainage.
Most growers practice passive, not active, flooding in February because once growers lose access to irrigation district water (by February 1), pumping or purchasing extra water is expensive, and they have other options. Two growers explained: I don't know many [growers] that would feel good about firing up ground water sources, just because the costs would get out of control really quick.
We used to flood this when water was a little cheaper. It worked pretty good […] But the price of water started going up.
[It] became cheaper to use a different chopper, make the chop finer, and just chisel it in, and it wouldn't take so long to dry out in the spring.
Some growers leverage revenue from hunting to cover the cost of flooding, explaining: We rent some […] fields to duck hunters for enough money to cover the PG&E.
But ducks prefer deep water, so hunter-supported flooding does not necessarily support shorebird flood habitat.
This evidence clarifies that the causes of the floodhabitat gap changed. For the first 3 weeks of February, droughts alone produced the gap. By the last week of February, growers became an important driver, taking measures to dry out their fields and prepare for spring planting. These results reveal a dynamic target. The schemes needed to start producing extra flooding after mid-February, but only during droughts and only with active-flood practices. By early March, they needed to support extended flooding in both wet and dry conditions.

| Temporal targeting and additionality
Results expose three observed cases in which the schemes were on target and induced additionality. All three occurred in dry conditions. The first case, event "n," happened on February 22, 2014 (Figure 1a [event n]). On that date, 38,303 ha of shallow floods were visible in the Landsat imagery (Table 2: dry conditions, event n). The PES schemes supported 9434 ha of active flooding on that date, which represented 25% of the observed shallow flooding (Table 3: Week 3, 2014). That extra flooding boosted levels well above typical patterns (Figure 1a: gray shading). Assuming the same rate of decline after that event (the slope of the gray shading), that additional flooding precluded an imminent drop below the Dybala et al. benchmark in the following days. Event "m," illustrates the same type of impact (Figure 1a-event m). In the third case, event "o," the schemes produced 60% of observed flooding, and the region reached 64% of the benchmark, instead of just 28% (Figure 1a-event o). These cases demonstrate the power of PES to deliver habitat on working lands.
T A B L E 4 Proportion of all flood-managed rice fields that are still flooded (plugged) throughout the late winter/early spring. Note: Peak flood levels were identified from Dec/Jan. images. Satellite flyover days occurred $2 weeks after a large rain event (>2.5 cm). Flood areas exclude any PES-supported flooding.
In all other observed cases, the schemes failed to induce additionality. Most commonly, they were off target in two different ways. First, they were off-target when they delivered extra floods at a time when sufficient floods were already provided. This problem occurred whenever the schemes supported flooding before mid-February (in wet or dry conditions) because typical flood patterns already exceed the Dybala et al. benchmark (Figure 1a: events g, h, and i and Figure 1b-event e). It also occurred whenever the schemes supported flooding (active or passive) between February 17th and March 10th in wet conditions because, once again, typical flood patterns already exceeded the benchmark (Figure 1bevents g and i). Notably, between 2006 and 2015, conditions were wet for 60% of days within that timeframe, so over half the time no PES flooding was needed. Second, the schemes were off-target when they ended too early. The three NRCS schemes encountered this problem, ending by March 1 which was only 2 weeks into the 6-week habitat gap.
The schemes also failed to induce additionality when they paid for flooding that was not delivered. This problem occurred any time they paid for passive flooding during dry conditions in late February. Event "p" in the dry pattern graph (February 25, 2015) illustrates the problem (Figure 1a). On that date, the schemes supported 3391 ha of active flooding and 2130 ha of passive flooding (Table 3, Week 4, 2015). With only active-flood practices able to induce flooding in dry conditions, the schemes only boosted flooding by 3391 ha. That additional hectarage barely lifted flood levels above typical patterns, and it was too little to raise levels much closer to the Dybala et al. benchmark and close the habitat gap.

| Ag/wildlife balance
Results reveal a dynamic ag/wildlife relationship, as floods shifted from being complimentary to competitive with rice production within the late winter/early spring. In early February, 87% of fields were flooded, which implies regional-level compatibility with production. In the last week of February, growers became the principal driver of the flood-habitat gap, which marks the onset of incompatibility for the region as a whole. By March 8, growers drained their fields very rapidly, even in conditions in which that was difficult to pull off. That signals strong, regional-level incompatibility (Figure 1b).
To mitigate any incompatibility and encourage participation, MBHI and WHEP offered variable drawdown rules, allowing growers to start drawing down their floods in the second week of February (Table 1). Evidence suggests that, while some growers may not have participated without those rules, many may have. First, enough growers practiced passive flooding into the last week of February (without PES support) to hold flood levels far above shorebird needs into early March when it rained (Figure 1b). Second, between 2006 and 2011, the CSP enrolled up to 12,711 hectares for active flooding through the end of February with no variable drawdown rules (Table 3). That hectarage was significant, representing 57% of the Dybala et al. benchmark. Notably, two of the three events demonstrating additionality in this study (events "m" and "o" in dry conditions; Figure 1a) were supported by the CSP alone. Third, over 85% of surveyed MBHI/WHEP growers (n = 22) reported that they would like to participate in a future scheme that required active flooding (without variable drawdown) through the end of February.
Furthermore, results show that variable drawdown sacrificed so much flooding that the rules undermined additionality. Event "p" in the dry pattern graph (February 25, 2015) illustrates this problem (Figure 1a). Over 9000 ha were under contract that year for active flooding (Table 3, Week 1, 2015). But, 80% were under WHEP contracts with variable drawdown rules. By the last week of February only 3391 of those hectares had to be flooded and that was insufficient to induce additionality.
TNC's approach to addressing the ag/wildlife relationship, with its conservation auction, avoided the challenge of predetermining when floods became incompatible with production. Growers revealed that information through the price of their bids. Nevertheless, results show that the BirdReturns scheme did not induce significant flooding in March or April. Despite the scheme's ability to accept high-cost bids, it enrolled less than 2000 hectares within that timeframe (Table 3). That represented less than 10% of the Dybala et al. benchmark and failed to raise flood levels significantly closer to the Dybala et al. benchmark (Figure 1a,b-event s and event o, respectively). As well, those contracts were for fields cycling out of rice and into fallows or other crops (Golet et al., 2018;Reynolds et al., 2017). That experience reinforces the assertion that flooding becomes highly incompatible with rice production by early March, and that barrier may be impervious to economic incentives.

| Permanence
Results indicate that some passive flooding, but no active flooding, has likely persisted since the payments ended. All of the surveyed active-flood adopters (n = 44) stated that they did not intend to maintain active flooding through February once the payments ended. However, some of that active flooding may have converted to passive flooding. All surveyed MBHI/ WHEP participants (n = 26) stated that they intended to revert to passive flooding, and 80% (n = 20) clarified that they intend to unplug their fields later than they had before participating in the schemes (Figure 2). Sixty-seven percent of the CSP participants (n = 14) reported that they would consider adopting passive flooding on at least some of their land through March 1 for no payment.
Evidence suggests that this partial permanence may have occurred because growers reassessed the economic impact of flooding on their management systems. Half of surveyed MBHI and WHEP participants (n = 13) said that flooding through the end of February did not delay their spring planting at all, and nearly half (n = 12) described the impact as moderate. Only one grower considered the delay significant. Half (n = 13) thought that the extended floods improved rice-residue decomposition, and almost half (n = 12) reported that their labor costs did not rise significantly. Other evidence suggests partial permanence may have occurred because growers care about shorebird conservation. Nearly 95% of MBHI/ WHEP participants (n = 25) expressed a sense of responsibility or interest in supporting waterbird conservation, and over 70% (n = 18) reported seeing higher shorebird populations on their fields during the schemes.

| DISCUSSION
The four PES schemes examined in this study implemented an on-field, targeted, dynamic approach to wildlife conservation on working agricultural lands-inducing short-term flood habitat to reduce a short-term gap in shorebird habitat needs. Results show that that approach can be extremely challenging to implement cost-effectively, as the schemes fell short in each of the four goals examined: targeting, additionality, ag/wildlife balance and permanence. In appraising the reasons why they fell short, this study affirms several well-understood challenges, but also highlights new issues that have been less recognized. The lessons learned from this case point toward PES-design elements that can help environmental mangers develop more cost-effective schemes moving forward.

| Designing dynamic conservation under uncertainty and variability
The schemes' first challenge was to induce extra flooding after the flood-habitat gap emerged, but results show that the schemes produced flooding when it was not needed (mistargeting) and they supported practices that failed to produce flooding when it was needed (no additionality). The reasons were twofold. First, the schemes were designed in the context of uncertainty, when the reasons for, and the timing of, the floodhabitat gap were not fully understood (Section 2.2). Uncertainties such as these are especially likely in working, high-yield landscapes because access to private lands is limited (Engel, 2016). In this case study, designers for the NRCS schemes addressed that challenge by casting a broad net, commencing payments at the flood-habitat gap's earliest possible emergence, and supporting practices that would cover all possible reasons for its development (i.e., growers unplugging fields too early and droughts). That approach made it inevitable that some PES resources would miss their mark. One PES design element that can help mitigate that problem is offering contracts of especially short length for the first few years of a scheme's operation in order to build in an opportunity for managers to refine scheme parameters as the nature of the target becomes clearer (Engel, 2016).
The second reason for the miss-targeting was the schemes' lack of responsiveness to weather, which altered the timing and cause of the flood-habitat gap from yearto-year and created an unpredictable moving target for scheme intervention. A PES design element that can add more flexibility in that context is a cancelation clause. In this case study, for example, a cancelation clause would allow schemes to cancel support for flooding when February rains are ample. TNC adopted this approach for its BirdReturns scheme after the timeframe of this study (G. Golet, personal communication). Another possible design element is an option (Hansen et al., 2014). With that tool, schemes can compensate growers for the right to require them to provide wildlife habitat (e.g., floods), and then exercise that option only if and when a habitat shortfall actually occurs (e.g., a drought). These types of tools would allow schemes to implement dynamic conservation with more temporal precision in conditions of unpredictable variability. F I G U R E 2 The dates by which growers unplugged their fields before (blue) and after (green) participating in the schemes. These data represent surveyed MBHI/WHEP participants (n = 26).

| Designing for a dynamic ag/wildlife balance
The schemes' second challenge was to mitigate potential ag/wildlife conflicts while sacrificing as little habitat as possible. That goal was especially difficult to meet because, as results revealed, flooding for shorebirds shifted from being (1) complementary with production and cheap to (2) complementary but expensive to (3) competitive and expensive in a matter of weeks. To operate cost-effectively in that context, the schemes needed to adjust their compensation and conflict-mitigation strategies as those shifts occurred. Adding to that complexity, individual growers (1) have different views about those ag/wildlife compatibilities; (2) they face different costs in executing the practices; and (3) they have sole knowledge of their views and costs (Ansell et al., 2016;Engel, 2016).
The NRCS schemes did not accommodate that diversity or adjust to that shifting ag/wildlife relationship. Their fixed timeframe for variable drawdown sacrificed more shorebird habitat than necessary to mitigate a conflict that did not emerge until weeks later. TNC's conservation auctions were more flexible and struck a more cost-effective ag/wildlife balance. They let growers choose the timeframes in which they would participate, and they mitigated conflict by paying growers higher rates for floods that growers considered more competitive with their production.
It is already well understood that, as a PES design element, auctions can increase cost-efficiency by reducing the gap between scheme payments and growers' minimum price for executing an environmental service (Ansell et al., 2016;Engel, 2016). This study amplifies the potential for auctions to help schemes adapt to dynamic ag/wildlife relationships. In addition, auction protocols can be modified to enhance flexibility. For example, in the context of this study, schemes might put shorter contracts out for bid (e.g., 2-3 week increments). That way, growers could submit separate bids for each increment and adjust their rates as their views on compatibility change. The contracts could also offer variable drawdown rules as an option within each increment. Schemes could then prioritize bids that meet two criteria instead of one: the lowest cost and the least variable drawdown. These design elements would forfeit less habitat while still promoting participation.
All of the PES-design elements identified above (e.g., cancelation clauses, options, and auctions with short contract increments) would add complexity, and complexity can discourage participation (Whitten et al., 2013). These elements might also raise administrative costs. Those tradeoffs would have to be examined before reaching any conclusions about their overall cost-effectiveness (Armsworth et al., 2012;Wünscher et al., 2008).

| Addressing persistence
The schemes' third challenge was to induce permanence, fostering a commitment among growers to continue flooding after the payments ended. Expecting growers to take on those costs is somewhat contradictory to the PES approach in the first place, as PES is based on the premise that growers will not provide the service unless compensated . Nevertheless, U.S.based schemes typically limit contract durations and restrict re-enrollments, and those rules expose the underlying expectation that PES will act as a transition payment, encouraging growers to provide the habitat themselves permanently (Dayer et al., 2018).
The results of this study indicate that growers intend to continue passive flooding later than they had in the past, but they will not continue any active flooding. This finding is not definitive, but if growers follow through with that intension, that extended passive flooding would represent partial permanence. The literature identifies two pathways to permanence: economic and non-economic. When wildlife practices are able to offer growers economic benefits, PES is expected to serve as a bridge, either providing enough time for those benefits to build up and become apparent; or affording growers the opportunity to experience the benefits directly and thereby change their perception of the practices (Jacobson, 2014;Kuhfuss et al., 2016). The results of this case reveal participants recalibrating and lengthening the time in which they consider flooding compatible with, and beneficial to, production.
When financial benefits are not likely, hopes hinge on the prospect that growers will take on the cost of the practices anyway, either because they develop a strong intrinsic interest in the wildlife (Guillem & Barnes, 2013), or because extrinsic social norms nudge them in that direction (Kuhfuss et al., 2016). The results of this study show that most participants harbor a strong sense of responsibility to support shorebirds. Those values likely supported their decision to participate in the schemes in the first place (Laney & Moses, 2020). During the schemes, participants witnessed higher shorebird populations, and that experience may have verified for them that extended floods are an effective way to fulfill those values, reinforcing their commitment to maintaining their passive floods longer.
Unfortunately, this study clarifies that the conservation value of this partial persistence is minimal. The only type of permanence that can have any added conservation value in February or early March is active flooding in dry conditions, and no scheme participants expressed a willingness to take on that expense. Society will need to compensate them for those costs on a long-term basis. These insights support Dayer et al.'s (2018) recommendation that policymakers accept a more permanent role for PES (loosening restrictions on contract durations and reenrollments) in situations where permanence is not tenable. This reality also magnifies the imperative that schemes operate as cost-effectively as possible.

AUTHOR CONTRIBUTIONS
This is a single-author manuscript.