Three grassland bird species’ responses to fire and habitat structure in southern Illinois, USA suggest broad benefits of grassland size and plant diversity

. Grassland birds are the most rapidly declining bird guild in North America, largely due to extensive habitat loss and fragmentation. Because many grassland bird species have different habitat preferences, managing grasslands to provide habitat for a range of species can be a challenge. We used four years of data from southern Illinois, USA grasslands to estimate the influence of prescribed fire and habitat structure on nest survival, nest density, and abundance of three grassland bird species with different habitat preferences: Dickcissel ( Spiza americana ), Field Sparrow ( Spizella pusilla ), and Common Yellowthroat ( Geothlypis trichas ). We found that Dickcissels exhibited the strongest response to prescribed fire, as nest density and nest survival both increased after previously undisturbed grasslands were burned. Fire may have also benefitted Common Yellowthroats and Field Sparrows by reducing woody cover and increasing bare ground, both of which were characteristics associated with nest survival for these birds. Dickcissel abundance was positively related to plant diversity within a grassland and agriculture in the surrounding landscape (within 400 m of a grassland patch), and negatively related to edge-interior ratio. Field Sparrows demonstrated a positive association with woody cover and proximity to forests. Common Yellowthroats were associated with tall vegetation and agriculture in the surrounding landscape. Both Field Sparrows and Common Yellowthroats associated positively with habitat characteristics that reduced nest survival, suggesting potential adaptive mismatches. Our results suggest that periodic prescribed fire, increased plant diversity, and larger patch size may simultaneously benefit a broad variety of grassland bird species with different habitat preferences.


INTRODUCTION
Grassland species are the most rapidly declining bird guild in North America (Rosenberg et al. 2019), largely due to habitat loss from the spread of agriculture and urbanization (Samson et al. 2004). With the amount of habitat now greatly reduced, it is critical that remaining grassland patches are effectively managed to create high-quality habitat for birds, which requires a detailed understanding of grassland bird-habitat relationships. However, the ways in which grassland characteristics affect birds can vary dramatically among bird species with different habitat preferences and life histories. For instance, species such as Upland Sandpipers (Bartramia longicauda), Lark Sparrows (Chondestes grammacus), and Horned Larks (Eremophila alpestris) require early successional habitat structure maintained by regular disturbance, including short vegetation and minimal woody encroachment, whereas species such as Clay-colored Sparrows (Spizella pallida) and Henslow's Sparrows (Ammodramus henslowii) prefer grasslands at a more advanced successional stage or with a shrub component (Grant et al. 2004, Winter et al. 2005, Fuhlendorf et al. 2006).
Here, we sought to improve the understanding of grassland birdhabitat relationships by examining the influence of grassland characteristics on abundance, nest density, and nest survival of birds with differing habitat preferences in our study area. We were also interested in identifying any habitat characteristics that benefitted all species to inform management actions for grasslands where multiple species of interest are present. Birds select habitats in a hierarchical fashion and are affected by processes at multiple spatial scales (Johnson 1980, Jones 2001. Therefore, we considered habitat features across several scales: landscape, patch structure, within-patch characteristics, and nest site. We also considered responses to prescribed fire, a common management tool. We examined multiple bird population metrics to gain a more complete understanding of how habitat characteristics are affecting the focal bird species because bird abundance, nest density, and nest survival may all respond to different variables and are decoupled in many systems. For instance, birds may select nest sites with characteristics that do not optimally promote nest survival for a variety of reasons (reviewed by Chalfoun and Schmidt 2012). Similarly, bird abundance may not accurately reflect the number of active nests because not all adult birds make nesting attempts during the breeding season (e.g., Zimmerman 1966, Best 1977, Harrison et al. 2009, and males of polygynous species such as Dickcissels may be associated with multiple nests (Zimmerman 1966). We focused our analysis on three common grassland bird species with differing habitat preferences in our study area: Dickcissels (Spiza americana), Field Sparrows (Spizella pusilla), and Common Yellowthroats (Geothlypis trichas). Dickcissels are obligate grassland-breeding birds that generally prefer grasslands in open landscapes with minimal woody encroachment. In contrast, Field Sparrows and Common Yellowthroats are facultative grassland species that often associate with shrubbier grasslands, dense vegetation, and grassland-forest edges.
We predicted that all bird species would respond positively to litter depth, vegetation height and density, and plant diversity at local spatial scales. Litter depth and vegetation biomass aid in nest concealment (Fogarty et al. 2017, Monroe et al. 2019) and can provide a beneficial thermoregulatory climate for altricial nestlings (Lomas et al. 2014). A diverse floral composition in grasslands can benefit birds by increasing arthropod biomass (Haddad et al. 2001, Benson et al. 2007) and by creating complex vegetation structure that may provide preferential nesting habitat for multiple species (Johnson and Sandercock 2010). We also predicted that all species would be positively associated with patch size because larger grasslands should provide more resources and space for both edge-and interior-associated species, and smaller patches may host higher nest predator densities (Johnson and Temple 1990).
We predicted that Dickcissels would negatively associate with woody cover (Grant et al. 2004, Thompson et al. 2014) and forest edges (Hughes et al. 1999, Winter et al. 2000, Jensen and Finck 2004, and respond positively to prescribed fire (Fuhlendorf et al. 2006, Londe et al. 2019. Previous studies have suggested that obligate grassland bird abundance increases in open landscapes; thus, we expected Dickcissel abundance and nest density to increase in patches surrounded by agriculture and grassland (Zimmerman 1971, Davis et al. 2013 and to decrease in patches surrounded by forest and shrubland (Hughes et al. 1999, Osborne andSparling 2013).
We predicted that Field Sparrows and Common Yellowthroats, unlike Dickcissels, would associate with woody edges and shrubs because both species frequently inhabit grasslands with woody components (Sousa 1983, Dinsmore et al. 1984, Grant et al. 2004 and forest edges (Vickery et al. 1994). We expected that Field Sparrow and Common Yellowthroat abundance and nest density would decrease in recently burned grasslands because fire can reduce the shrub cover and dense vegetation cover that these birds prefer (Dechant et al. 2002a, Grant et al. 2004, Murray and Best 2014. At the landscape scale, we predicted that Field Sparrows and Common Yellowthroats would respond positively to trees in the surrounding landscape (Dechant et al. 2002a, Grant et al. 2004).

Study area
Our study took place at Burning Star State Fish and Wildlife Area (37°52' N, 89°12' W, hereafter Burning Star), a former surface coal mine comprising 1824 ha of both reclaimed and undisturbed land in northeast Jackson County, Illinois, USA (Fig. 1). Landcover types present at Burning Star include mid-and latesuccessional forest (936 ha), agriculture (456 ha), wetland (66 ha), and restored tallgrass prairie (110 ha), as well as several freshwater lakes (223 ha; Illinois Department of Natural Resources 2018). Fieldwork occurred on 10 restored prairie patches (hereafter sites) within Burning Star, which ranged in size from 1.8 to 35.9 ha (mean ± standard deviation [SD] = 10.99 ± 10.57 ha). Sites were contiguous grassland patches bounded by roads or a change in landcover composition (e.g., forest, agriculture, lake). Common grasses on the restored prairie sites included warm-season natives such as big bluestem (Andropogon gerardii), Indian grass (Sorghastrum nutans), and switchgrass (Panicum virgatum), as well as cool-season non-native grasses such as smooth brome (Bromus inermis), Kentucky bluegrass (Poa pratensis), and foxtail (Setaria sp.). Common forbs included Canada goldenrod (Solidago canadensis), annual ragweed (Ambrosia artemisiifolia), and nonnatives such as sericea lespedeza (Lespedeza cuneata) and sweet clover (Melilotus sp.). Management actions such as prescribed fire and ungulate grazing have mostly been absent at these sites since restoration was conducted shortly after 1992, although two sites were burned in the final two years of our study (Site 10 in February 2020 and Site 8 in March 2021). The general lack of management has resulted in the establishment and encroachment of woody shrubs such as eastern red cedar (Juniperus virginiana), blackberry (Rubus sp.), honey locust (Gleditsia triacanthos), and non-native autumn olive (Elaeagnus umbellata). A concurrent study at Burning Star using continuously recording camera systems (Cox et al. 2012) on a subset of grassland bird nests has identified North American racers (Coluber constrictor), prairie kingsnakes (Lampropeltis calligaster), black kingsnakes (Lampropeltis nigra), and raccoons (Procyon lotor) as the primary nest predators in the area (Appendix 1).

Vegetation surveys
We performed two vegetation surveys per site, in mid-May and mid-July, in each field season from 2018 to 2021. Vegetation surveys occurred along permanent 90-m transects established within each site, with each site containing a minimum of two transects. For sites larger than 20 ha, another transect was added for each additional 10 ha. For example, sites between 20 and 30 ha had three transects, and sites between 30 and 40 ha had four transects. To ensure that both the edge and interior of each site were represented in the surveys, the first transect was established along an edge and oriented toward the site's interior. Subsequent transects began at randomly selected points within the site interior and followed the same orientation as the first transect. Each transect consisted of 10 sampling points spaced 10 m apart. At each sampling point, we measured vegetation density using a Robel pole (Robel et al. 1970), following the methodological recommendations of Fisher and Davis (2010). We also estimated the percent cover of grasses, forbs, plant litter, woody vegetation, and bare ground within a 20 x 50 cm quadrat, and measured litter depth using a standard ruler. We identified to species all plants within the quadrat at each sampling point and calculated plant diversity using the Shannon index.

Point count surveys
We conducted 5-min point count surveys at designated locations within each grassland site once per week from early May until late July. During point count surveys, we recorded all birds that were seen or heard, along with the estimated distance and direction of each individual. Each site contained three point count locations that were separated by approximately 100 m. Only one location per site was used for a point count survey each week. Surveys rotated among within-site locations every week so that each location within a site was visited once every three weeks. We ensured a minimum of 500 m between point count locations at different sites to minimize the chance of double-counting individuals. We conducted all surveys between sunrise and 10:00 am local time, with a minimum of five days between surveys at the same site. Surveys were not conducted in inclement weather such as heavy rain, wind (> 3 Beaufort scale), or fog.

Nest searching and monitoring
We searched for grassland bird nests at each site 10 times per field season, with sites being searched approximately once per week. A designated nest search area of approximately 3 ha was delineated within each site. For sites that were smaller than 3 ha, the entire site was searched. We searched for nests using rope dragging (Winter et al. 2003), and nests that were discovered were monitored every three to four days until they either fledged or failed. We determined nest age by floating eggs in a cup of water to estimate developmental stage (Westerskov 1950). If nests were found in the nestling stage, we aged nestlings based on appearance (e.g., Jonsomjit et al. 2007:9-14) and assumed a 12-days incubation length to determine when the nest was initiated. Twelve days is a typical incubation length for Field Sparrows (Best 1978), Dickcissels (Long et al. 1965), and Common Yellowthroats (Stewart 1953). To estimate nest-site vegetation structure, we conducted vegetation surveys at nest sites within one week of a nest becoming inactive. Apart from our standardized nest searching efforts, nests were occasionally found incidentally in the sites while conducting other field surveys. We monitored those nests as well, and they were included in our nest survival analysis. However, only nests found while nest searching were used to calculate nest density.

Habitat variables
We quantified habitat variables representing four spatial scales: nest site, within-patch, patch, and landscape ( Table 1). The nestsite scale and associated predictor variables were only considered for the nest survival analysis. Nest site and within-patch variables were derived from vegetation surveys. We also included as nest site variables distance to edge and distance to nearest shrub, calculated in ArcMap 10.6 (ESRI 2018) using drone imagery of the study area obtained in October 2019 and 2021. The imagery from 2021 was only used for Site 8 in 2021 and Site 10 in 2020 and 2021 to account for any reduction in shrub cover that may have occurred post-fire. To estimate the effect of prescribed fire on the grassland bird community, we included fire as a withinpatch variable. The fire variable was categorical, with the value for unburned sites set to 0 and for sites that were burned that year or the previous year (N = 3) set to 1.
Patch-scale variables included patch size (ha), edge-interior ratio (perimeter [m] / area [ha]), and the proportion of patch edge comprising forest, agriculture, water, or roads. Landscape-scale variables consisted of the proportion of five different habitat types within a 400-m buffer around each site: forest, agriculture, water, grassland (including buffers around agricultural fields and roads), and development (roads, buildings, parking lots, and mowed lawns). Buffers around patches were limited to 400 m to minimize redundancy in landscape variable values because patches are located in close proximity to each other ( Fig. 1). Although this buffer size is smaller than those used in other landscape-scale studies, it is an appropriate size for measuring proximate landscape features that may affect grassland birds (Cunningham and Johnson 2006). Additionally, it allows us to account for the territory size of the focal bird species, which are generally ~0.4 ha for Dickcissels (Harmeson 1974, Verheijen et al. 2019), ~0.76 ha for Field Sparrows (Best 1977), and ~0.15 ha for Common Yellowthroats (Kirsch et al. 2007). We calculated the percent cover of different habitat types by digitizing the study site and surrounding area in ArcMap 10.6 at a 1:1500 scale, using Maxar basemap imagery from 2017 provided by ArcGIS. Prior to statistical analysis, all variables were standardized to have a mean of 0 and a standard deviation of 1.

Statistical analyses Bird abundance
All statistical analyses were performed in R 4.2.1 (R Core Team 2022). We conducted species-specific analyses for Dickcissels, Field Sparrows, and Common Yellowthroats. To model bird abundance, we considered point count data from all four years and treated each site-year combination as a separate sampling location (Ahlering and Merkord 2016). We analyzed counts at the site level, combining counts across all three locations within a site. We estimated the effects of habitat variables on bird abundance with N-mixture models, which simultaneously model abundance and detection probability, using the 'pcount' function in the R package unmarked (Fiske and Chandler 2011). We first selected a probability distribution by creating three null models set to different distributions: Poisson, zero-inflated Poisson, and negative binomial. The three models were ranked by AIC c , and the distribution of the top-ranked model was used in all subsequent models for each focal species. Once a probability distribution was selected, we compared detection submodels while maintaining a null abundance submodel. The detection candidate set consisted of four submodels: date, observer, date + observer, and null. After being ranked by AIC c , the top-ranked detection submodel was used in subsequent models.
For the abundance submodels, we employed a two-phase model selection process. The first phase involved identifying the most influential habitat predictor variables at each spatial scale (Table  1). We followed recommendations from Arnold (2010)

Nest survival
We used daily survival rate (Dinsmore et al. 2002) to estimate the effects of habitat variables on nest survival. Daily survival rate models were developed using the nest survival module in the RMark package (Laake 2013), which interfaces with Program Mark (White and Burnham 1999). Before adding habitat covariates to the models, we accounted for any effects of nest age on daily survival rate by comparing linear and quadratic terms of nest age against a null model for each focal species. The most informative variable for nest age based on AIC c was considered the ecological null model and used in subsequent habitat models. We created our habitat models using the two-phase model selection process described earlier. For the nest survival analysis, daily survival rate is reported as the model-averaged estimate ± standard error (SE).

Nest density
We calculated nest density using the following equation derived from Arnold et al. (2007): where N is the number of nests found in the nest search area, HA is the size of the nest search area (ha), DSR is the average daily survival rate for the site, and d is the average nest age at first discovery. The DSR d term in the equation allowed us to account for nests that failed before they could be discovered by our field crews (McPherson et al. 2003, Arnold et al. 2007). Adjusting nest density estimates to account for imperfect detection is an important step for reducing bias and improving the accuracy of estimates (McPherson et al. 2003, Tyre et al. 2003. Although the use of eq.1 accounts for nests that were missed due to prior failure or variation in predation rates among study sites, it does not completely remove sources of bias because the rate at which nests are found can vary among species and observers. Thus, nest density estimates presented here should be interpreted with a degree of caution.
We estimated associations between habitat variables and nest density by creating a candidate model set using the two-phase model selection process described above.
Year and site were included in all models as random effects to account for unmodelled temporal and spatial variability in nest density. Models were created in the R package glmmTMB (Brooks et al. 2017), specifying a normal distribution, and ranked by AIC c in the R package MuMIn (Bartoń 2019). Nest density estimates were log-transformed prior to modeling to improve normality and homogeneity of variance. We examined the potential for an edge effect on nest density by calculating nest density for three different groups: nests within 20 m of an edge, nests between 20 and 40 m from an edge, and nests > 40 m from an edge. When calculating nest density, HA was set to the area (ha) of each group. Because we did not have the data to calculate group-specific daily survival rate within a site, we assumed a constant daily survival rate among groups for each site. We conducted a one-way ANOVA and Tukey's HSD test to determine whether differences in nest density were statistically significant between groups at the α = 0.05 level.

Bird abundance
We detected 361 Field Sparrows, 340 Dickcissels, and 237 Common Yellowthroats during our point count surveys. Detection probability was estimated at 0.27 ± 0.04 SE for Dickcissels, 0.09 ± 0.02 SE for Field Sparrows, and 0.08 ± 0.04 SE for Common Yellowthroats. The null model was the top detection submodel for all three species (Appendix 2), so no detection variables were used in final abundance models. Dickcissel models used the negative binomial distribution, whereas Field Sparrow and Common Yellowthroat models used the Poisson distribution. Our goodness-of-fit tests indicated adequate fit of competitive models (ΔAIC c < 4) for Dickcissels (P ≥ 0.074), Field Sparrows (P ≥ 0.075), and Common Yellowthroats (P ≥ 0.526) that appear in final model sets.
The top model in the Dickcissel candidate set included edgeinterior ratio and plant diversity (w i = 0.54,  (Table 2). Common Yellowthroat abundance was positively related to agriculture (β = 0.11, −0.06 to +0.28) and forb cover (β = 0.1, −0.05 to +0.24), and negatively related to litter depth (β = −0.13, −0.33 to +0.07) and edge-interior ratio (β = −0.09, −0.27 to +0.08). The wide confidence intervals for these variables indicate a high level of uncertainty regarding the parameter estimates and their effect on Common Yellowthroat abundance. Cumulative model weights of each variable for all species are shown in Appendix 3.

Nest survival
Through four field seasons, we found and monitored a total of 207 grassland bird nests. We found 175 nests during standardized nest searches, including 107 Field Sparrow nests, 39 Dickcissel nests, and 29 Common Yellowthroat nests. Another 32 nests were found incidentally: 29 Field Sparrow nests, 1 Dickcissel nest, and 2 Common Yellowthroat nests.
The effect of nest age on birds can vary among species due to species-specific behavioral and ecological traits. The ecological null model for Field Sparrows and Common Yellowthroats was an intercept-only model, suggesting that nest age did not have a measurable effect on nest survival for these birds at Burning Star. The ecological null for Dickcissels included a linear term for nest age. Daily survival rate decreased with nest age (β = −0.1, 95% confidence interval: −0.20 to +0.01), resulting in model-averaged daily survival rate estimates for Dickcissels changing throughout the season (range of mean ± SE = 0.17 ± 0.28 to 0.99 ± 0.01).
Mean daily survival rate for Dickcissels was 0.79 ± 0.10. Daily survival rate estimates were 0.90 ± 0.01 for Field Sparrows and 0.95 ± 0.02 for Common Yellowthroats. Predation was the most common cause of nest failure (N = 138 nests), accounting for 93% of failed nests. Nine nests failed due to abandonment, one nest was destroyed by inclement weather, and one nest failed with nestlings dying of an unknown cause.
Two models in the Dickcissel candidate set had more support than the null. The top model contained the additive effects of fire and within-patch vegetation height (w i = 0.10, Table 3), and the second-ranked model contained percent water at the landscape scale (ΔAIC c = 0.24). Dickcissel daily survival rate was positively associated with fire (β = 1.09, 0.06-2.13), vegetation height (β = 0.41, −0.02 to +0.85), and water (β = 0.25, −0.05 to +0.56). Other influential variables in the model set included water edge (β = 0.25, −0.09 to +0.58), forest edge (β = −0.32, −0.88 to +0.25), and litter depth at the nest site (β = −0.37, −0.97 to +0.23). The Field Sparrow model set was dominated by distance to edge, as this variable was present in all competitive models (ΔAIC c < 4). Daily survival rate was positively associated with distance to edge (β = 0.38, 0.10-0.66), indicating that nests farther from an edge had higher survival. Additionally, model results suggested a positive relationship between daily survival rate and nest site grass cover (β = 0.22, −0.06 to +0.49), within-patch grass cover (β = 0.23, −0.01 to +0.47), and within-patch ground cover (β = 0.31, −0.08 to +0.70), and a negative relationship with road edge (β = −0.11, −0.36 to +0.14) and water in the landscape (β = −0.12, −0.41 to +0.17). In the Common Yellowthroat model set, only three models ranked above the null. Distance to shrub (β = 0.66, −0.18 to +1.49) was the most influential variable and was present in two of the top three models. Other variables in top models included litter depth at the nest site (β = 0.73, −0.21 to +1.66), vegetation height (β = −0.39, −0.92 to +0.13), and distance to edge (β = 0.52, −0.17 to +1.20). Vegetation density (β = −0.56, −1.31 to +0.20) and road edge (β = −0.38, −0.92 to +0.16) were also present in models that had similar support as the null. Similar to the abundance models, these variable estimates contain wide confidence intervals, which suggests a high level of uncertainty regarding the relationship between habitat parameters and Common Yellowthroat nest survival. Cumulative model weights of each variable for all species are shown in Appendix 4.

DISCUSSION
Despite its limited extent in the study area, prescribed fire had a clear impact on grassland birds at Burning Star. Dickcissel nest density increased dramatically post-fire ( Fig. 3): almost half (47%) of Dickcissel nests were found in sites that were burned earlier that year (Site 10 in 2020 and Site 8 in 2021). Nest survival Fig. 2. Relationship between nest density and distance to edge for grassland birds nesting at Burning Star State Fish and Wildlife Area. Means ± 95% confidence limits. Nest density for Field Sparrows was significantly higher (P < 0.05) within 20 m of an edge compared to ≥ 40 m from an edge.
for Dickcissels was also positively related to fire. The effects of prescribed fire on grassland birds change with fire frequency, and annually burned grasslands may provide poorer habitat for Dickcissels compared to grasslands burned at less frequent intervals (Reinking 2005, Churchwell et al. 2008. However, the two prescribed fires in this study both occurred on previously undisturbed grasslands, which likely contributed to the immediate positive response in Dickcissel nest density and survival. As an obligate grassland bird, Dickcissels have evolved to rely on regular disturbances, including fire, in the grasslands they inhabit (Askins et al. 2007). Fire encourages new vegetative growth and can have a positive effect on net primary productivity (Blair 1997), which may provide advantageous vegetation structure for Dickcissels compared to undisturbed grasslands (Fuhlendorf et al. 2006). However, the vegetative growth encouraged by fire may be temporarily suppressed when grazers are present, as grazers will focus their foraging on newly burned areas Shaw 1998, Fuhlendorf andEngle 2001). Therefore, the immediate positive effect of prescribed fire on Dickcissels may have been subdued if grazers were present in the study area (e.g., Powell 2008, Monroe et al. 2016. The two facultative grassland species exhibited differing responses to prescribed fire. Field Sparrow abundance was negatively associated with fire, indicating a preference for grasslands in later successional stages that have not been disturbed for several years. Fire can reduce woody cover and herbaceous vegetation biomass, both of which are positively associated with the occurrence or abundance of this species (Dechant et al. 2002a, and citations therein). Field Sparrow nest density exhibited a strong positive relationship with woody cover in our study (Table  4). However, abundance alone is not necessarily a reliable indicator of habitat quality (Van Horne 1983, Chalfoun and  (Table 4). Means ± 95% confidence limits; dots represent raw data points.
Schmidt 2012). For instance, despite the strong association between Field Sparrow nest density and woody cover, we detected no evidence for a positive effect of woody cover on nest survival for Field Sparrows or any of our other focal species. Common Yellowthroat daily survival rate was positively associated with distance to shrub and negatively associated with vegetation height and density, suggesting that the reduction of woody vegetation and plant biomass by prescribed fire may benefit nest survival for this species. Additionally, previous studies suggest that nest survival for grassland-nesting birds is negatively affected by woody cover because of its association with nest predators (Bakker 2003, Graves et al. 2010, Klug et al. 2010. Previous research from our study area (Glass and Eichholz 2022) has linked snake abundance to woody cover in these grassland sites, likely due to thermoregulatory benefits and protection from overhead predators that shrubs provide to snakes. The reduction of plant litter post-fire may have also benefitted Field Sparrow nest survival, as daily survival rate for Field Sparrows was positively associated with bare ground.
Edge effects were also detected for multiple focal species. Distance to edge was positively related to daily survival rate for both Field Sparrows and Common Yellowthroats (Fig. 4), indicating that nests farther from an edge had higher survival rates. This result is consistent with the popular concept that nests near forest-field Table 4. Summary of the final model sets estimating the relationship between nest density and habitat variables for grassland birds at Burning Star State Fish and Wildlife Area. All models include year as a random effect. K = number of variables in the model, ΔAIC c = difference in AIC c between a given model and the top model in the candidate set, w i = Akaike model weight. NULL refers to the intercept-only nest density submodel. See Table 1  edges may be subjected to higher rates of predation (Johnson andTemple 1990, Renfrew andRibic 2003) due to higher activity of generalist nest predators near these edges. However, despite common assumptions concerning edge effects on nest survival, studies investigating grassland birds often fail to find such a relationship (reviewed by Benson et al. 2013). The edge effects on nest survival we detected at Burning Star may instead relate more to road edges than forest edges. Previous studies in Midwestern U.S. grasslands have indicated that snake and raccoon abundance may be positively associated with road edge (Barding andNelson 2008, Glass andEichholz 2022). The roads at Burning Star are lightly used by humans; thus, they may serve as travel lanes for mammalian mesopredators (Barding andNelson 2008, Wysong et al. 2020) and may attract snakes due to their beneficial thermoregulatory environment (Gibbons andSemlitsch 1987, Rosen andLowe 1994). Our model results yielded limited evidence that road edge was negatively related to Field Sparrow and Common Yellowthroat nest survival, suggesting that the use of these roads by nest predators may lead to higher rates of incidental nest predation. However, this effect is likely limited to areas where roads experience minimal use. Roads with more consistent traffic are unlikely to attract nest predators because the increase in disturbance levels and road mortality may reduce populations of snakes (Row et al. 2007) and mesomammals (Fahrig and Rytwinksi 2009) and lead to behavioral avoidance of roads (Shine et al. 2004, Fahrig andRytwinski 2009).
Despite the potential risk to nest survival presented by roads at Burning Star, neither Field Sparrows nor Common Yellowthroats displayed any behavior associated with road avoidance. Instead, Common Yellowthroat nest density was higher in grasslands with a greater proportion of roads around their perimeter, whereas Field Sparrows tended to nest near edges regardless of edge type (Fig. 2). Although edge avoidance is common in grassland birds (Bollinger and Gavin 2004, Davis 2004, Renfrew et al. 2005, Dickcissel was the only species at Burning Star for which there was some evidence of edge avoidance, as its abundance was negatively related to edge-interior ratio. Consequently, Dickcissel nest survival was less affected by edge proximity than Field Sparrows or Common Yellowthroats. All of the focal species responded to within-patch plant composition, and our results demonstrate the diverse effects that plant composition may have on grassland birds. Field Sparrow daily survival rate was positively associated with grass cover both at the nest site and within a grassland patch. The tillering growth structure of grasses may provide better ground-level concealment than forbs (Crabtree et al. 1989), particularly in early summer, when the cool-season grasses are fully grown, but common forbs such as goldenrod, sweet clover, and sericea lespedeza are still growing. This structure may result in greater protection from ground-dwelling nest predators such as mammals and snakes for nests with more grass cover than forb or woody cover. Grass seeds are also an important food source for Field Sparrows and make up a significant portion of their breeding season diet (Evans 1964, Allaire andFisher 1975). In contrast, Dickcissel nest density and Common Yellowthroat abundance were positively associated with forb cover. Forb cover has been previously linked to increased nest density and survival for Common Yellowthroats (Patterson andBest 1996, Fletcher andKoford 2002) and Dickcissels (Patterson andBest 1996, Dechant et al. 2002b), and may create preferred nest structure for grassland birds, in addition to encouraging arthropods (Benson et al. 2007, Rzanny et al. 2013. Plant diversity was also influential in our study and appeared to increase Dickcissel abundance. Previous studies (Barnes et al. 1995, Johnson and Sandercock 2010, Port and Schottler 2017 have documented a positive effect of plant diversity on grassland birds, including Dickcissels (Osborne andSparling 2013, Conkling et al. 2017). Grasslands with high floristic diversity may support more arthropods than grasslands with lower diversity (Fulbright et al. 2013, Nelson et al. 2017, which represent an important food source for Dickcissels (Mitchell et al. 2012). Additionally, plant diversity increases the structural complexity of grasslands (Barnes et al. 1995, Johnson andSandercock 2010), potentially providing better structure for nest sites, foraging, and perching.
The 400-m landscape buffer we used is smaller than those of other landscape-scale studies, which may have limited the inferential abilities of our models at this scale. However, we still detected diverse effects of landscape composition on the grassland birds in the study area. Dickcissel nest density and abundance, and Common Yellowthroat abundance, were positively associated with agriculture, although we detected no relationship between agriculture and nest survival. These associations are consistent with previous research, as several grassland bird species, particularly Dickcissels (Zimmerman 1971, Walk et al. 2010, Reiley and Benson 2019, have been documented associating with agriculture during the breeding season. Although pesticide drift from agricultural fields can reduce arthropod populations in neighboring grasslands, potentially negatively affecting habitat quality for grassland birds (Mineau and Whiteside 2013), the lack of trees or shrubs in crop fields likely makes agriculture a more preferred landscape than shrubland or forest. Additionally, agricultural fields and other open areas around a grassland patch may be interpreted by birds as contiguous grassland habitat. The appeal of such landscapes can be strong enough that some birds will preferentially select grassland patches with low horizons over larger patches (Keyel et al. 2012, Marshall et al. 2020. Field Sparrow abundance, in contrast, was negatively associated with agriculture and positively associated with forest in the landscape. This result is not surprising, given that Field Sparrows are understood to prefer proximity to forests (Vickery et al. 1994, Dechant et al. 2002a, and a larger proportion of agriculture in the landscape corresponded with a smaller proportion of forest at Burning Star (r = −0.85). Water in the surrounding landscape had a positive association with Dickcissel nest survival, though this result may be influenced by the facts that burned sites tended to be located near lakes (Fig. 1) and the variables representing fire and percent water in the landscape were highly correlated (r = 0.63).

CONCLUSIONS
Although Dickcissels, Field Sparrows, and Common Yellowthroats were associated with different habitat characteristics, all birds responded positively to various metrics of patch interior habitat availability. Nest survival for Field Sparrows and Common Yellowthroats increased with greater distance from an edge, and bird abundance for all focal species was related positively to patch size (Field Sparrows) or negatively to edgeinterior ratio (Dickcissels, Common Yellowthroats). Plant diversity also likely benefitted the bird community at Burning Star. Field Sparrow nest survival increased with grass cover, whereas Dickcissel nest density and Common Yellowthroat abundance increased with forb cover, illustrating the varying preferences in vegetation composition displayed by grasslandnesting birds. While increasing grass and forb cover simultaneously may not always be possible, encouraging plant diversity could help to provide the preferred nesting substrate and vegetation structure for a wide variety of grassland birds (Johnson and Sandercock 2010, Osborne and Sparling 2013, Wilson et al. 2022. Additionally, plant diversity was positively related to Dickcissel abundance.
Prescribed fire increased Dickcissel nest density and survival, although it also discouraged Field Sparrow abundance, at least in the short time frame of the study. However, prescribed fire may have benefitted Common Yellowthroats and Field Sparrows by reducing vegetation biomass and woody cover and increasing bare ground. Vegetation height and density, and proximity to shrubs, were negatively associated with Common Yellowthroat daily survival rate. Field Sparrow daily survival rate responded negatively to woody cover and positively to bare ground.
We suggest that managers and decision makers interested in increasing grassland bird density should be aware of trade-offs inherent in management decisions and consider which species or group of species is the highest priority. Actions such as prescribed fire and shrub removal are important tools for maintaining grasslands and benefit many grassland bird species (Fuhlendorf et al. 2006, Askins et al. 2007, Coppedge et al. 2008). Implementing these actions may result in an initial reduction in abundance or nesting density of grassland birds that associate with shrubs or dense vegetation, though our results suggest that such actions may still benefit the nest survival of these species, especially if used sparingly. Other actions such as encouraging plant diversity and increasing patch size may benefit a wide variety of grassland bird species, regardless of functional group or specific habitat preference. 1 estimating the relationship between habitat predictor variables and abundance of bird species at Burning Star State Fish and Wildlife Area. In this first phase of model selection, we created a set of models comprising all single variables and all pairwise combinations of variables within a spatial scale. Variables with cumulative model weights ≥ 0.2 were used in the second model selection phase. The sign in parentheses next to the bolded model weights indicates the direction of the predictor variable's effect on the response variable.

Variable
Cumulative