Detecting ecological traps in human‐altered landscapes: A case study of the thick‐billed longspur nesting in croplands

Abstract Conversion of the North American prairies to cropland remains a prominent threat to grassland bird populations. Yet, a few species nest in these vastly modified systems. Thick‐billed longspurs historically nested in recently disturbed or sparsely vegetated patches within native mixed‐grass prairie, but observations of longspurs in spring cereal and pulse crop fields during the breeding season in northeastern Montana, USA, suggest such fields also provide cues for habitat selection. Maladaptive selection for poor‐quality habitat may contribute to ongoing declines in longspur populations, but information on thick‐billed longspur breeding ecology in crop fields is lacking. We hypothesized that crop fields may function as ecological traps; specifically, we expected that crop fields may provide cues for territory selection, but frequent human disturbance would result in reduced reproduction. To address this hypothesis, we compared measures of habitat selection (settlement patterns and trends in abundance) and productivity (nest density, nest survival, and number of young fledged) between crop fields and native grassland sites during 2020–2021. Across both years, settlement patterns were similar between site types and occupancy ranged from 0.52 ± 0.17 SE to 0.99 ± 0.01 on April 7 and 30, respectively. Early season abundance differed by year, and changes in abundance during the breeding season appeared to be associated with precipitation‐driven vegetation conditions rather than habitat type. While an index of nest density was lower in crop than native sites, the number of young fledged per successful nest (2.9 ± 0.18 SE) and nest survival (0.24 ± 0.03 SE; n = 222 nests) were similar for crop and native sites. Collectively, the data did not support our ecological trap hypothesis: longspurs did not exhibit a clear preference for crop sites and reproductive output was not significantly reduced. Our results indicate that croplands may provide alternative breeding habitat within a human‐dominated landscape.


| INTRODUC TI ON
In North America, grassland birds have experienced steeper longterm declines than any other avian guild during the past 50 years (Rosenberg et al., 2019;Sauer et al., 2020), and more than 79% of grasslands have been lost since the early 1800s (Samson & Knopf, 1994;White et al., 2000). While factors such as fire suppression, overgrazing, desertification, and the introduction of non-native plant species have contributed to degradation of native prairies, conversion to cultivated cropland remains one of the greatest threats to grassland ecosystems (Blann, 2006;Ellis et al., 2010;Knapp et al., 1999;Wright & Wimberly, 2013). Indeed, intensification of agricultural practices is considered a leading driver of grassland bird population declines worldwide Quinn et al., 2017;Wilson et al., 2005). However, some species do use crop fields for nesting and foraging (Best et al., 1997;Davis et al., 2020). Birds nesting in crop fields face a myriad of hazards: farming operations (e.g., tilling, discing, and harvest) may result in nest destruction (Devries et al., 2008;Santangeli et al., 2018), and soil instability in crop fields may lead to nest failures during floods and heavy rains (Van Pelt et al., 2017). Herbicide and pesticide applications may directly harm adults and nestlings or indirectly affect bird populations by reducing invertebrate food resources (Loss et al., 2015;Pimentel et al., 1995). In addition, most crops grow rapidly into dense monocultures, with vegetation conditions changing from short stature (5-15 cm tall) with large amounts of bare ground to tall (60-70 cm), closed-canopy conditions within 2-3 months of seeding (Wilson et al., 2005).
Ecological traps occur when there is a mismatch between habitat selection cues and habitat quality (Battin, 2004), and are most commonly identified where human activities produce novel environmental cues or alter habitat quality associated with a particular cue (Hale & Swearer, 2016;Robertson et al., 2013;Simon & Fortin, 2019). An ecological trap differs from a demographic sink in that animals often exhibit strong preference for trap habitat (Gilroy et al., 2011;Pulliam, 1988). Such maladaptive selection leads to negative fitness consequences and reduced population viability (Battin, 2004;Gilroy et al., 2011;Schlaepfer et al., 2002). The ideal free distribution theory that underpins source-sink population models assumes animals exhibit optimal habitat selection when distributing themselves among habitat patches and that the fittest individuals obtain the highest-quality territories. In reality, individuals likely select habitat according to evolutionarily predisposed cues, and ecological traps are attractive because they provide such cues (Abrams et al., 2012;Delibes et al., 2001;Fletcher Jr et al., 2012;Hale et al., 2015;Hale & Swearer, 2016).
The thick-billed longspur (Rhynchophanes mccownii; hereafter "longspur"; Figure 1) is a grassland songbird endemic to the shortand mixed-grass prairies of North America (Knopf, 1996). Like most obligate grassland birds, populations of longspurs have declined precipitously (4% per year on average) since the advent of the North American Breeding Bird Survey (Rosenberg et al., 2019;Sauer et al., 2020), but mechanisms driving the decline are poorly understood. Longspur habitat is patchy within native mixed-grass prairie, resulting in clustered distributions at the landscape level (i.e., high local densities in some areas yet absent from other superficially similar areas; Greer & Anderson, 1989;Lipsey, 2015). As such, longspur habitat preferences limit distributions at regional scales and make this a focal species for federal conservation efforts (Somershoe, 2018). Unlike many other grassland birds, longspurs have a unique preference for recently disturbed or sparsely vegetated habitats and historically relied on large-scale disturbance regimes to maintain suitable habitat patches through spatial-temporal interactions of soil, precipitation, fire, and intensive periodic defoliation by native herbivores (e.g., bison [Bison bison] and locusts [chiefly Melanoplus spretus]) (Felske, 1971;Samson et al., 2004;Shaffer et al., 2019;. However, these dynamic processes that once shaped prairie ecosystems are largely absent in today's Northern Great Plains (Fuhlendorf & Engle, 2004;Hovick et al., 2015;Samson & Knopf, 1996).
Cropland, or cultivated land annually seeded to crops, is now the dominant disturbance factor producing extensive bare ground at large scales in the Northern Great Plains. Sporadic reports of longspurs occurring in crop fields during the breeding season (Felske, 1971;Shaffer et al., 2019) suggest that some croplands may resemble suitable nesting habitats as longspurs arrive on the breeding grounds in April. However, the selection of crop fields during territory establishment could contribute to population declines if reproductive success is depressed through (1) destruction of nests by discing, seeding, and herbicide/pesticide application; and (2) abandonment of nests or territories in response to rapid changes in vegetation conditions or disturbance by farming activity. If crop fields provide attractive nesting habitats where longspurs experience low reproductive success, modern cultivated landscapes may be operating as ecological traps for this shortgrass prairie specialist.
Identifying an ecological trap involves demonstrating habitat preference (i.e., disproportionate selection) that results in reduced vital rates (e.g., nest survival). Importantly, high population densities do not necessarily equate to preference for that habitat type (Van Horne, 1983;Vickery et al., 1992), and comparisons of relative selection metrics between habitat types are needed to discern preferences. A demonstrated preference for crop fields coupled with reduced reproductive success relative to native prairie would indicate an ecological trap with significant implications for longspur conservation (Lloyd & Martin, 2005;Robertson & Hutto, 2006). Delayed settlement in crop fields after native sites are occupied, coupled with lower reproductive success, would suggest crop fields simply provide spill-over habitat, which is characteristic of source-sink populations (Gilroy & Sutherland, 2007).
We conducted a 2-year field study to evaluate whether cultivated fields, primarily cereal and pulse crops, operate as ecological traps for nesting longspurs within the core of their breeding distribution in northeastern Montana, USA. We compared settlement patterns, temporal trends in abundance, and nest density between crop fields (hereafter "crop sites") and nearby native prairie (hereafter "native sites"), assuming that earlier settlement and higher use of crop sites indicated that either longspurs preferred crop sites for nesting or that native sites were limited (Robertson & Hutto, 2006). We measured reproductive success in both habitat types to assess the relative quality of crop sites as breeding habitats for longspurs. Our objectives were to compare the following in crop and native sites: (1) settlement patterns of territorial male longspurs; (2) longspur abundance and nest densities, as well as changes in abundance over the breeding season; (3) nest survival and fledging success; and (4) structural changes in vegetation during the breeding season. For our ecological trap hypothesis to be supported, we first predicted that crop sites would become fully occupied by longspurs earlier than native sites, indicating earlier settlement and preference for crop habitat. Second, we predicted that longspurs would occur in higher abundance in crop sites early in the season, but abundance would decline over the growing season as crop height increased because longspurs are known to avoid tall, dense vegetation. We predicted that nest densities would reflect bird abundance and longspurs may abandon nests in crop fields when vegetation height increased. Lastly, we predicted that rates of daily nest survival and fledging success would be lower in crop sites due to hazards from farm machinery and increased exposure to weather and predators.

| Study area and site selection
Our research was conducted in northern Valley County, Montana, USA, which lies within the core of the remaining breeding distribution of longspurs. The climate in this region is semi-arid with long, cold winters and short, hot summers producing frequent thunderstorms, hail, and flash floods (Cooper et al., 2001). Average daily temperatures range seasonally from below 0 to 25°C. Annual precipitation averages 25-35 cm and typically comes as rain in late May and early June (Lenard et al., 2006;PRISM, 2022). The region is at about 915 m in elevation. Clay shale is the most abundant substrate, and the landscape is dominated by glacial till (Cooper et al., 2001).
Northern Valley County is characterized by large expanses of poor soils unsuitable for cultivation and more productive areas used primarily for cereal and pulse crop production.
Our study area incorporated native grassland in the western portion and cultivated fields in the eastern portion ( Figure 2).
Land ownership on native sites was a mix of federal lands and private ranches and included the Bitter Creek Wilderness Study Area.
Crop types included primarily cereals (spring wheat) and pulse crops (pea and lentils, lentils often planted with flax). An additional crop type, summer fallow, was included in which spring wheat farming was conducted on a 50:50 rotation with 50% of the acreage fallowed each year to conserve soil moisture and nutrients (M. Sather, USFWS, pers. comm.). Less abundant crop types included oilseed (canola) and cover crops, and a mix of random seed types (i.e., spring wheat, flax, lentil, beets, and barley) planted during years a field was rested. Pulse crops are often planted between wheat rotations to add nitrogen, conserve soil moisture, and disrupt weeds, pests, and diseases (Long et al., 2014;Miller et al., 2002). Pulse crop rotations have replaced summer fallow over most of the region (M. Sather, USFWS, pers. comm.).
We used a geographic information system (ArcMap 10.7.1; ESRI, 2019) to select study plots in crop and native sites. We selected plots at two different levels: 64-ha plots for initial surveys (Appendix S1), which were designed only to locate populations of breeding longspurs each year, and 16-ha plots, which were generated within areas where longspurs occurred and were used to monitor longspur abundance and nest success ( Figure 2).
Within initial plots (64 ha) was found to be occupied by longspurs (Appendix S1); we used ArcMap to generate smaller, 16-ha survey plots within which we conducted abundance surveys and nest searching for the entire nesting season (Figure 2). Due to the patchy distribution of longspurs on native sites, our plot selection methods differed slightly for crop and native sites. We randomly selected between 20 and 30 occupied crop fields and generated a single 16-ha plot within the center of each field. This ensured crop plots were ≥200 m from field edges, roads, and other plots. In occupied native sites, we delineated large patches of habitat by mapping the extent of occupied areas on foot with a GPS unit and later transferred this information to ArcMap. Patch edges were defined by the presence/ absence of singing longspurs and were typically coupled with subtle changes in vegetation structure. We then overlaid a grid of 16-ha cells over occupied patches and used ArcMap to randomly select between 20 and 30 cells from these patches. Only non-adjacent cells were used to ensure plots were ≥200 m apart.
Our plot selection process in crop sites ensured that the distribution of crop types surveyed was representative of the study area, although it was limited to the land parcels where we were able to obtain permission to conduct sampling. All crop types except summer fallow were operationally similar throughout the breeding season. Crop sites, regardless of crop type, were structurally similar in April prior to seeding with all fields containing residual stubble (10-15 cm) from the previous year's crop. Timing of seeding was variable depending on land ownership but began in April and was completed by early June. Some fields were seeded directly into the previous year's stubble (no till), but most fields were disked prior to seeding (minimum till).
Most producers conducted a weed control spray immediately prior to seeding. Post-emergence herbicide applications were administered in most fields in June. In all summer fallow fields, tilling was typically F I G U R E 2 Map of study area illustrating song meter deployment sites (e.g., SM_Crop_20) and 16-ha plots used for abundance surveys and nest searching on crop and native habitat sites in Valley County, Montana, 2020-2021. Clustering of native plots is due to patchy distribution of longspurs in native habitats as these plots were randomly selected from areas known to be occupied by thick-billed longspurs.
conducted in unplanted portions two to three times between May and July for weed control. Some producers also used weed control spray in summer fallow fields. Harvest of all crops occurred in August or September, after the nesting season was complete. Because crop types were operationally similar, our goal was to obtain a representative sample across the study region instead of focusing on a particular crop type, although spring wheat was the dominant crop on the landscape and was disproportionally represented in our study plots (Table 1; Table S1). All native sites were grazed by cow/calf pairs after May 15; while cattle use of specific plots could not be quantified, grazing pressure was similar across all native sites.

| Longspur settlement
We deployed 24 autonomous acoustic recorders (Wildlife Acoustics model SM4, Maynard, MA; hereafter "song meters") to assess settlement patterns of territorial male longspurs on the breeding grounds.
Song meter locations were not part of the site selection process described above, as they were deployed before longspurs had arrived and surveys could begin. We consulted local biologists and used observations from the USFWS Breeding Bird Survey (BBS) and eBird (Sauer et al., 2020;Sullivan et al., 2020) to identify specific points annually used by longspurs. We deployed 8 and 16 song meters in 2020 and 2021, respectively, with half (4 in 2020, 8 in 2021) in crop sites and half in native sites ( Figure 2). We selected sites that were no more than 25 km apart to minimize regional variation in weather patterns between site types. No crop fields were planted during deployment; fields were structurally similar and contained 10-to 15-cm-tall stubble from the previous year's crop. In 2020, song meters were placed in fields that contained the following crops the previous year: spring wheat (3) and flax (1); in 2021: spring wheat (5), lentil (1), and summer fallow (1). We deployed song meters from April 6 to 29, 2020, and April 7 to 30, 2021, to capture territory establishment. In Montana, male longspurs arrive about 2 weeks before females (early April), but do not typically settle onto territories until late April .
We affixed song meters to 1.8-m t-posts at a height of 1.2 m and covered each microphone with an extra layer of foam to reduce recorded wind noise. We programmed song meters to collect a 3-min recording every half hour starting 15 min before sunrise and ending by 9:00 a.m. to coincide with morning breeding choruses of longspurs , resulting in six 3-min recordings collected each morning. Upon removal from the field, a trained technician manually reviewed each 3-min recording and documented longspur presence.
We excluded any recordings in which ≥25% of the recording length was obscured by wind or other noise.

Abundance surveys
We conducted six rounds of line transect surveys within each 16-ha survey plot from May 10 to July 15, 2020-2021. Survey rounds were separated by ≥5 days. Observers walked a U-shaped line transect within each plot, starting 100 m inward from a randomly selected plot corner ( Figure S1; modified from Golding & Dreitz, 2017;Igl & Johnson, 1997). We recorded perpendicular distance and direction from the transect line for each longspur seen or heard. Estimated distances were recorded in bins: 0-25, 26-50, 51-75, and 76-100 m.
We walked at a pace of 2-3 km h −1 and completed each transect within 30 min. Observers were trained to avoid double-counting birds. Surveys began 15 min before sunrise and were completed by 9:00 a.m. We did not conduct surveys if wind speed was >25 km h −1 or if it was raining. We recorded survey covariates including observer, cloud cover, temperature, wind speed, date, GPS starting point, and transect start/end times.

| Nest phenology, survival, and reproductive output 2.2.3.1 | Nest searching
We searched for nests within the same 16-ha plots to assess reproductive effort throughout the entire nesting season  on days without precipitation, and observers were randomly assigned a group of plots to search each morning. Observers alternated between crop and native sites during subsequent days and used behavioral observations to find nests (Martin & Geupel, 1993;Winter et al., 2003). We observed longspurs from a distance of ≥30 m and moved to a new plot after 60 min if no female longspurs were detected. In addition, we supplemented behavioral nest searching with standard rope-dragging methods (Klett et al., 1986;Koford, 1999).

| Nest monitoring
Upon finding a nest, we recorded the geographical coordinates and marked the nest location with 15-cm bamboo stakes placed 2 m north and east of the nest to aid in relocation. Nests were checked every 2-4 days until fledging or failure (Martin & Geupel, 1993;Ralph, 1993

| Habitat conditions
We collected vegetation measurements at two spatial scales, the nest site and 16-ha survey plot. Measures were collected at each nest within 3 days of fledge or expected fledge for failed nests. In addition, we randomly selected 3 and 10 habitat sampling points within survey plots in crop sites and native prairie sites, respectively.
Vegetation conditions in crop sites were fairly homogenous and required fewer sampling points (Swicegood, 2022). We measured vegetation conditions three times throughout the longspur breeding season, once each in May, June, and July. At each sampling point, we recorded visual obstruction reading (VOR) in each cardinal direction from a distance of 4 m and a height of 1 m (Robel et al., 1970).

| Longspur settlement
We used multi-season occupancy models with our song meter data to evaluate whether longspur settlement patterns differed between crop and native sites (Mackenzie, 2006;MacKenzie et al., 2003).
We combined data from both years due to small sample sizes, and each day represented a primary period (24) while each 3-min recording represented a secondary period (6). We fitted models using the "colext" function in R package "unmarked" (Fiske & Chandler, 2011;Kéry & Chandler, 2016) and used information theory to evaluate support for competing models representing hypotheses about detection probability, initial occupancy, and settlement patterns (MacKenzie et al., 2003). We evaluated support for our a priori models in a phased approach. First, we evaluated how well a fully parameterized model fit the data and estimated a variance inflation factor (ĉ) using the mb.gof.test in the R package "AICcmodavg" from 500 bootstrapped simulations (Mazerolle, 2020). We found evidence of moderate overdispersion (ĉ = 1.9) and inflated estimated standard errors by √ ĉ. We based subsequent model evaluation and inference on the quasi-Akaike's information criterion adjusted for finite samples (QAIC c ; Burnham & Anderson, 2002).
We then developed models that evaluated the effects of survey conditions on detection probability. Variables hypothesized to influence detection included daily precipitation, minimum daily temperature, minutes past sunrise, and Julian day. We evaluated a quadratic effect of minutes past sunrise because bird detections were previously found to be highest mid-morning . We did not include an effect of observer because recordings were annotated after removal from the field by a single primary observer.
Because all detection covariates we measured are known to affect the detectability of songbirds, we used a backward selection approach based on QAIC c to eliminate uninformative parameters (Arnold, 2010;Montgomery et al., 2021;Pagano & Arnold, 2009).
Models with large relative weights (w i ) and QAIC c values ≤2 from the best-fit model were considered equally parsimonious (Arnold, 2010;Burnham et al., 2011;Devries et al., 2008). After we identified a parsimonious sub-model for detection, it was retained in subsequent evaluations of occupancy and settlement.
Because some birds had already arrived at the study area prior to song meter deployment, we evaluated whether initial occupancy differed by habitat type (crop vs. native) before evaluating the effects of habitat type and Julian day on settlement probability. Our candidate set included two models for initial occupancy and five models for settlement probability, which included an interaction term for habitat type and day. We hypothesized that settlement rates would change over the season differentially by habitat type.
We also hypothesized that probability of site abandonment would be low because longspurs are unlikely to abandon a breeding site . Therefore, we did not include any models with covariates on settlement probability. Model selection was again based on QAIC c ; when supported models differed by one parameter, we considered this parameter uninformative (Arnold, 2010;Burnham & Anderson, 2002). We used empirical Bayes methods to derive estimates of latent occupancy, or the true proportion of sites occupied, from the most parsimonious model for each primary period from predicted posterior distributions using the "ranef" function in R package "unmarked" (Fiske & Chandler, 2011).
All analyses were performed using R Statistical Software (v 4.1.2; R Core Team, 2021).

| Longspur abundance
We used open-population distance sampling models to estimate longspur abundance in crop and native sites and assess whether abundance changed differentially over the breeding season (Royle et al., 2004;Sollmann et al., 2015). We fitted models using the "distsampOpen" function in R package "unmarked" (Fiske & Chandler, 2011) and used information theory to evaluate support for competing models representing hypotheses about detection, initial abundance, and trends in abundance over the breeding season (Sollmann et al., 2015). Data were analyzed separately for the 2 years because differences in weather and drought conditions were likely to produce different population responses.
We evaluated support for a priori models in a phased approach.
First, we used null models with the "trend" dynamics parameterization to estimate the best-fitting detection function and mixture type based on our data. We then evaluated how well a fully parameterized model fit the data and estimated a variance inflation factor (ĉ) using the Nmix.gof.test in the R package "AICcmodavg" from 500 bootstrapped simulations. We used the hazard rate function and Poisson mixture type for all subsequent models, inflating estimated standard errors by √ ĉ and basing model evaluation and inference on the quasi-Akaike's information criterion adjusted for finite samples (QAIC c ; Burnham & Anderson, 2002). We found evidence of moderate overdispersion (ĉ = 1.9 for 2020 data and ĉ = 1.7 for 2021 data).
We then developed models to evaluate the effects of survey conditions on detection probability. Variables hypothesized to influence detection probability included observer, wind speed, temperature, and start time (minutes past sunrise). We evaluated a quadratic effect of start time because bird detections are usually highest 1-2 h after sunrise . Initial screening indicated that detection was variable across observers, so we separated observers into two groups for each year ("high" and "low" detection rates) based on relative coefficient estimates from a full model to reduce the number of parameters in candidate models while retaining observer effects on detection. We used the backward selection approach described previously to eliminate uninformative parameters and identify a parsimonious sub-model for detection probability, which was retained in subsequent evaluations of abundance and seasonal trends.
Finally, we developed a set of four a priori models to test the effect of habitat type on both initial abundance and seasonal trend. Model selection was again based on QAIC c (Burnham & Anderson, 2002). We used Bayesian methods to derive true abundance estimates from the most parsimonious model for each survey round from predicted posterior distributions using the "ranef" function in R package "unmarked" (Fiske & Chandler, 2011 (Badyaev et al., 2003;Hébert, 2002). Initiation date was estimated based on clutch size, hatch date, or chick age and assuming an incubation period of 12 days . For nests found after clutch completion but destroyed before hatch, we assumed initiation to be 6 days prior to the midpoint of the active period. We plotted nest initiation dates to visualize patterns of nest initiation between crop and native sites and to assess differences between years.

| Nest survival
We used the nest survival model in program MARK to model daily nest survival rate (DSR), and we fitted models in the R package "Rmark" (Laake, 2013;Rotella et al., 2004;White & Burnham, 1999).
We built and evaluated a set of 15 competing models representing a priori hypothesized relationships between DSR and habitat type (crop or native), nest initiation date, and year (2020, 2021). Our models included all combinations of habitat type, initiation date, and year.
We also included a model with a quadratic effect of initiation date because other studies have shown DSR to be higher or lower midseason (Skagen et al., 2018;Weintraub et al., 2016). We predicted that DSR may exhibit a pseudo-threshold response in crop sites only, being low for nests initiated early and leveling off after fields were planted. Therefore, we included a model with a pseudo-threshold effect of initiation date and one including an interaction term with habitat type. We evaluated the relative support of models using Akaike's information criterion corrected for finite sample size (AIC c ; Burnham & Anderson, 2002). To estimate nest survival probability, we used a 26-day nesting cycle beginning with the start of the laying period and multiplied DSR for each daily interval over a 25-day period from nest initiation to fledging (e.g., DSR 25 for constant model).
We calculated standard error for nest survival estimates using the Delta method (Powell, 2007).

| Reproductive output
We calculated an index of nest density for each plot by dividing the number of nests located in each plot by the total search effort (hours) for that plot. We report the mean and standard deviation of relative nest density for each habitat type. Incidental nests located outside of survey plots and nests found via rope-dragging methods were excluded from this calculation. We were unable to account for detectability of nests with behavioral search methods, and it is possible that detectability differed in crop and native sites.
We tabulated maximum clutch size for all nests with known fates and the number of young fledged per successful nest. The number of young fledged was recorded as the number of chicks present 8-10 days after hatching, the typical fledging time for longspurs .
We developed a set of generalized linear models to analyze the effects of habitat type and initiation date on the number of young fledged per successful nest using a Poisson distribution with a log link. Our five candidate models included an interaction term to test if the number of young fledged differed by both habitat type and initiation date. Nests were removed from analysis if the number of young fledged was unknown. We evaluated relative model support using AIC c (Burnham & Anderson, 2002) and used the best-fitting model to estimate the number of young fledged per successful nest.

| Habitat conditions
We used generalized linear models to test hypotheses that specific vegetation attributes differed significantly between crop and native sites, longspur habitat changed structurally over the summer as plants grew, and such changes were more extreme in crop sites than in native sites. Variables included VOR, bare ground cover, grass and forb cover, litter cover, and litter depth.
For proportional response data (e.g., percent coverages), we used the binomial distribution and logit link function to fit GLMs (Chen et al., 2017). For all other vegetation measures, including VOR and litter depth, we used the identity link and log-transformed the response variables to meet the assumptions of linear regression (Dunn & Smyth, 2018). For each habitat variable, we built and evaluated the same set of five competing models representing a priori hypothesized relationships between habitat type and survey round. We evaluated relative model support using AIC c . We based inferences on effect sizes from a single top model and calculated model-averaged estimates when models shared support (ΔAIC c ≤ 2; Burnham & Anderson, 2002).

| Longspur settlement
We deployed song meters at 8 and 16 sites in 2020 and 2021, respectively, half in crop fields and half in native sites. Due to equipment malfunction and failure of longspurs to establish territories at some sites, we were able to obtain data from two song meters in native sites and four song meters in crop sites in 2020, and seven song meters in native sites and seven song meters in crop sites in 2021. Overall, we collected >37 h of useable recordings in 2020 and >100 h in 2021.

| Detection probability
The top model for detection probability contained an effect of Julian day, minimum temperature, and a quadratic effect of minutes past sunrise (QAIC c w i = 0.97; Table 2). Detection probability increased with Julian day (β = 0.99 ± 0.13 SE) and increased in response to minimum temperature (β = 0.08 ± 0.02). Detection probability was highest at ~90-100 min past sunrise, or 1.5 h after sunrise ( Figure S2).

| Initial occupancy and settlement probability
We found no evidence for an effect of habitat type on initial oc-  (Figure 3).

| Longspur abundance
In 2020, we conducted 287 longspur abundance surveys at 24 crop sites and 22 native sites from May 14 to July 19. We observed 5.4 ± 4.4 (mean ± SD) male longspurs in crop sites and 4.2 ± 3.3 in native sites. In 2021, we conducted 325 surveys at 25 crop sites and 25 native sites from May 10 to July 14. We observed an average of 3.8 ± 3.2 and 3.2 ± 2.3 males per plot in crop and native sites, respectively.
Confidence intervals for the effect sizes for other covariates on detection overlapped 0; therefore, only observer group was retained in subsequent abundance modeling (Arnold, 2010).

| Initial abundance and seasonal trends
We found support for an effect of habitat type on both initial abundance and seasonal trend for data collected in 2020 (QAIC c w i = 0.91; during the first survey round and 9.4 (8.4-10.7) during the sixth round ( Figure 4).
In 2021, we found no support for an effect of habitat type on either initial abundance or seasonal trend, with the null model carrying the most support (QAIC c w i = 0.54; Table 3 Figure 4).

| Nest phenology, survival, and reproductive output
In 2020, we conducted nest searching in 47 (25 crop and 22 native) plots. We found nests in 88% of crop plots and 82% of native plots.
In 2021, we conducted nest searching in 55 (27 crop and 28 native) plots. We found nests in 63% of crop plots and 68% of native plots (Table S1). We located 240 total longspur nests, 111 in crop sites and 129 in native sites. Of these, 174 were located using behavioral cues of adults, 14 using rope-dragging methods, and 52 were incidental finds while observers were conducting other fieldwork. We

| Nest phenology
Patterns of nest initiation were similar within crop and native sites each year, but median initiation dates in native sites were 6-11 days later than median dates in crop sites ( Figure 5). In addition, the first and third quartiles were 6-10 days later in native sites. In 2020, median initiation date was May 29 (IQR = 25 days, n = 68) and June 9 (IQR = 26 days, n = 71) in crop and native sites, respectively. Longspurs nested through mid-July and there were two prominent peaks in nest initiation. In 2021, median date of initiation was May 28 (IQR = 17 days, n = 28) in crop sites and June 3 (IQR = 13 days, n = 55) in native sites. Nesting slowed significantly in early July, and there was only one main peak in nest initiation.
Notably, the interquartile distance for initiation dates was 32% shorter in crop sites and 50% shorter in native sites during the 2021 drought.

TA B L E 3 Model selection results from
open-population distance models for detection probability (p), initial abundance (N 1 ), and seasonal trends (λ) of thick-billed longspur populations.

| Nest survival
The null model of constant daily nest survival was the best supported in the candidate set (AIC c w i = 0.18; in crop and native sites, respectively. The mean number of young fledged per successful nest was 3.0 ± 1.1 SD in crop and 2.8 ± 0.9 in native sites (Table S2). The null model was the best-supported model in our candidate set of generalized linear models for number of young fledged per successful nest (AIC c w i = 0.49;

| Habitat conditions
We observed significant differences in vegetation conditions between crop and native sites that varied across survey rounds (Appendix S2). Visual obstruction reading (VOR) estimates (cm ± SE) in 2020 changed from 0.81 ± 1.42 in May to 17.81 ± 1.43 in July in crop sites and from 1.95 ± 1.51 to 2.61 ± 1.46 in native sites. In 2021, VOR estimates (cm ± SE) changed from 0.72 ± 1.35 to 1.48 ± 1.35 in crop sites and from 0.68 ± 1.34 to 0.28 ± 1.34 in native sites. Bare ground coverage was significantly lower on native sites than on crop sites during both years. Estimated bare ground (% ± SE) in 2020 was 45 ± 6 in crop and 10 ± 4 in native sites. In 2021, estimated bare ground was 42 ± 6 in crop and 14 ± 4 in native sites.
Estimated litter coverage in 2020 (% ± SE) was 25 ± 5 in crop sites and 8 ± 3 in native sites. In 2021, estimated litter coverage was 26 ± 5 in crop and 11 ± 3 in native sites. Estimated litter depth in 2020 (mm ± SE) changed from 4.66 ± 1.30 in May to 0.58 ± 1.31 in July in crop sites and from 1.57 ± 1.34 to 1.05 ± 1.32 in native sites.
In contrast to expectations, some farming activities, including rolling (field leveling) and spraying, did not result in nest damage or abandonment. Longspur nest bowls were constructed below the soil surface, and we found that farming activities that did not disturb the soil did not negatively affect nests regardless of nest stage (n = 9 nests in leveled fields). Although nests were active during herbicide application, spraying did not directly result in losses of eggs or nestlings. However, we did not assess potential indirect effects of herbicide or fertilizer spraying on nestling growth rates or subsequent fledgling survival. Flooding and hail destroyed nests in both crop (n = 8) and native sites (n = 4). Nest abandonment was often due to partial predation, weather, brown-headed cowbird parasitism, and possibly frequent disturbance by predators or perceived predators. On a few occasions in native sites (n = 3 nests), we found all nestlings apparently uninjured but laying outside the nest. These nestlings never survived and were never returned to the nest by adult longspurs. We suspect this to be the activity of brown-headed cowbirds or other passerine nest predators (Pietz et al., 2012;Pietz & Granfors, 2000).

| Longspur abundance and use of crop fields
Our prediction that suitability of crop fields as nesting habitats would decline over the growing season was supported by observed trends in longspur abundance. We are not aware of other studies that specifically investigated how attractiveness of cereal or pulse crop fields changed over the breeding season for thick-billed longspur, but similar trends have been observed for Eurasian skylarks (Alauda arvensis) nesting in cereal crops (Donald & Vickery, 2000) and for horned larks (Eremophila alpestris) nesting in winter wheat fields  we were unable to distinguish true nest density from our ability to detect nests.
We found relatively fewer nests in summer fallow fields compared to spring wheat and lentil fields (n = 11 total nests in summer fallow fields). The fallow portions of summer fallow fields pose additional hazards to ground-nesting birds because they are tilled multiple times during the breeding season. Therefore, nests in unplanted portions of summer fallow fields have a higher risk of being destroyed later in the season, whereas in other crop types, longspurs were only vulnerable to this particular hazard during the early nesting season. It is also possible that frequent tilling of fallow fields results in different soil conditions, generating different invertebrate resource availability than that found in minimum-tilled fields (Kladivko, 2001;Stinner & House, 1990).

| Timing of nesting
We found that median nest initiation occurred 6-11 days earlier in crop sites despite similar settlement patterns for the two habitat types. Longspurs appeared to shift timing of nesting in crop sites, and perhaps this phenological shift is beneficial in habitat that changes to become unsuitable late in the breeding season. In native prairie habitats, longspurs select territories on south-facing slopes during the early breeding season where snow melts and the ground warms faster (Felske, 1971;Greer, 1988). Bare ground cover was higher in crop sites than native sites throughout the breeding season, and exposed soils warm faster than vegetated soils (Shaffer et al., 2019;Song et al., 2013). Earlier warming of crop fields may allow earlier nest initiation and egg laying resulting from favorable microclimatic conditions or an earlier invertebrate food supply (Felske, 1971;Greer & Anderson, 1989). However, we did not assess thermal or other microclimatic conditions at nests. Furthermore, the range of nest initiation dates and therefore length of nesting period was significantly shorter during the drought of 2021. Longspurs are known to forego nesting or experience lower reproductive success during periods of extreme drought (Felske, 1971 (Perlut et al., 2008;Sedgwick, 2004), however, low site fidelity in breeding areas for longspurs may make estimation of these vital rates difficult (Sedgwick, 2004;

| Conservation implications and recommendations
Other studies have demonstrated the importance of croplands as nesting habitats for passerines (Best et al., 1995;Donald et al., 2006;Lokemoen & Beiser, 1997). For example, cereal fields are thought to be the single most important extant habitat for skylarks in Europe (Donald & Vickery, 2000). Given that croplands now occupy a vast proportion of arable regions worldwide, we should encourage management of these lands in ways that enhance nesting habitats for passerines. In particular, management strategies that modify the timing or degree of disturbance in cropland may provide direct benefits to ground-nesting birds. Odderskaer et al. (1997) found that artificially creating open areas (7 m 2 ) within cereal fields resulted in higher skylark densities and extended nesting seasons. Recent studies in North America have indicated that autumn-seeded wheat, or "winter wheat", may be better for nesting waterfowl and passerines than spring-seeded wheat due to the increased cover winter wheat provides early in the nesting season and elimination of the need for spring tillage Devries et al., 2008;Skone et al., 2016). Although not a viable option for wheat cropping in northcentral Montana, shifting to alternative plantings may be an option in other regions (M. Sather, pers. comm.). However, the appeal of winter wheat depends on species preference for vegetation structure. Species like longspurs that prefer short, sparse vegetation are more likely to prefer spring wheat over winter wheat (Martin & Forsyth, 2003), as winter wheat has taller stature during territory establishment. In Europe, skylarks and yellow wagtails (Motacilla flava), which also prefer short-stature vegetation, abandoned autumnseeded crops for spring-seeded crops late in the season as crop height increased, resulting in lost nesting opportunities in autumnseeded crops (Eggers et al., 2011;Gilroy et al., 2010).
Although overall nest survival rates were similar in crop and native sites, reproductive output might be improved with modified farming practices that minimize the destruction of early nests. Early seeding (<10 May) and reducing summer fallowing should result in greater use and higher nest survival by longspurs in crop fields. Pesticides, including fungicides, can harm adults and nestlings (Martin et al., 1998;McEwen & Ells, 1975;Mineau & Whiteside, 2013). Thus, we recommend reduction or elimination of widespread application of herbicides, pesticides, and fungicides when possible, until their potential impacts on the survival of nesting longspurs can be evaluated.

| CON CLUS IONS
The loss of native grasslands through conversion to cropland is a primary driver of grassland bird population declines, including thickbilled longspurs (Blann, 2006;Ellis et al., 2010;Samson et al., 2004;Wright & Wimberly, 2013). Nevertheless, we show that longspurs breeding in a region of relatively intact grasslands in northeastern Montana appear to nest successfully in nearby crop fields. While hazards for nests (e.g., farming activities and predation) differed between crop and native sites, overall seasonal reproductive output and nesting success were similar.
Due to the loss of large-scale, historic disturbance regimes on native sites, crop fields may effectively expand nesting opportunities in a region where native habitat has been confined to patches with arid soils. Minor modifications to cultivated land management practices in this region could benefit longspurs, particularly in relation to timing of mechanical operations. Appropriate management of cultivated fields for conservation depends entirely on the species and agricultural system, making ubiquitous recommendations across species and systems inappropriate. In addition, such management strategies should be restricted to cultivated fields to preclude deleterious effects on grassland obligate species . Given the great weight of evidence that conversion to cropland is detrimental to grassland bird populations, we strongly recommend against any conversion of native prairie to benefit longspurs. Additionally, future research should explore management practices that promote dynamic patterns of disturbance, bare ground, and short grass in native prairies, especially in early spring when longspurs select territories. Further investigation into population demographics, body condition, and resource availability may provide additional insight into the relationship between longspurs and cropland. We are also appreciative of the private landowners who allowed access to their land for data collection.

CO N FLI C T O F I NTER E S T S TATEM ENT
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
Sampling locations, data, and scripts used for analysis are all avail-