Henslow’s Sparrow shows positive response to prescribed fire rotation

We examined Henslow’s Sparrow ( Centronyx henslowii ) response to prescribed fire at 32 grasslands at Big Oaks National Wildlife Refuge in southeastern Indiana from 1999 to 2009. We burned grasslands in the spring between 1999 and 2007 and monitored Henslow’s Sparrows for up to 4 yr after treatment. We used linear mixed models to analyze our data. Henslow’s Sparrow counts were correlated with time since prescribed fire and grassland size. The estimated changes in mean Henslow’s Sparrow density relative to pre-burn densities were −0.19, 1.15, 0.74, and −0.68 birds ha −1 for 1–4 breeding seasons after a spring burn, respectively. We found that Henslow’s Sparrows preferred larger grasslands both during the first breeding season after prescribed fire, when vegetation was presumed to possess less litter and structural density, and during the fourth breeding season after fire, when vegetation was presumed to be a more ideal composition, but preferred smaller grasslands in the interim. Thus, grassland size shapes the magnitude of Henslow’s Sparrow population response to fire, with populations in smaller grasslands experiencing greater amplitude changes. Larger grasslands might provide more habitat diversity following prescribed fire, attenuating population change. On average, Henslow’s Sparrows responded positively to prescribed fire in a network of grasslands and cumulative net change in densities were highest three breeding seasons after a burn suggesting that maintaining this burn frequency would be beneficial to Henslow’s Sparrow abundance.


INTRODUCTION
The Henslow's Sparrow (Centronyx henslowii) global population has declined in the last half-century (Herkert and Glass 1999, Sauer et al. 2017, Herkert 2019, Herkert et al. 2020. Population declines are associated with loss or degradation of their primary habitat, which consists of grasslands with tall, dense grass, a thick litter layer, and little woody vegetation (Sauer et al. 2005, Herkert 2019, Herkert et al. 2020). The species is currently designated by the U.S. Fish and Wildlife Service as "the highest priority for grassland bird conservation in eastern and Midwestern North America" (U.S. Fish and Wildlife Service 2002, Rich et al. 2004, Sauer et al. 2005. They breed in scattered, local populations in northeastern and east central United States and southern Canada (Hands et al. 1989, Pruitt 1996, Herkert and Glass 1999. Woody plant encroachment is a major threat to the persistence of Henslow's Sparrows given their preference for tall grass habitat (Herkert et al. 2020). Although there are many methods for minimizing woody plant encroachment into grasslands, one of the more common methods is to use prescribed fire to prevent or delay it (Briggs et al. 2005).
Henslow's Sparrow habitat appears to follow a cycle of impermanent suitability after prescribed fire treatment in which there is an immediate negative effect on Henslow's Sparrow habitat in the months following a prescribed fire due to removal of grass and litter layer and decreased vegetation height (Herkert andGlass 1999, Hill andDiefenbach 2013). In subsequent years, habitat appears to improve due to regrowth of perennial vegetation and increased litter layer (Hill and Diefenbach 2013). However, fire-resistant woody species such as sweetgum (Liquidambar styraciflua L.), sour gum (Nyssa sylvatica Marshall), and sassafras (Sassafras albidum (Nutt.) Nees) are cleared aboveground during a burn but the subterranean root stalks are not destroyed (Basey and Badger 2004). Woody encroachment from root sprouts then ensues, degrading habitat quality. The cyclic nature of Henslow's Sparrow habitat quality in response to prescribed fire requires wildlife managers to identify a fire rotation schedule that optimizes the amount of suitable habitat available for grassland species while other habitat is undergoing fire treatment (Fuhlendorf et al. 2009). By burning different units in different years, the burn schedule will promote landscape heterogeneity, which has been shown to favor grassland bird species (Hovick et al. 2015). Sub-optimal fire-rotation schedules may be detrimental to metapopulation dynamics and the local assemblages of the breeding population (McCollough 1996, Hanski and Simberloff 1997, Hovick et al. 2015. The metapopulation of Henslow's Sparrows might be associated with a larger landscape than other grassland birds. Breeding Henslow's Sparrows appear somewhat nomadic from year to year and return rates to natal areas are not large especially for year old birds (Dornak et al. 2013, Ingold et al. 2009). Population stability might be best maintained in a network of grassland patches by using a burn rotation such that an optimal subset of grasslands is burned each year. This burn rotation should permit high-quality habitat to remain unburned while prioritizing degraded habitat for treatment (Fuhlendorf et al. 2009, Hovick et al. 2015. Several factors, such as size and isolation, might contribute to breeding Henslow's Sparrows use of patches following prescribed fire. Bakker et al. (2002) and Davis (2004) have shown that breeding density may decrease as grassland size decreases. Johnson and Igl (2001) found that large grasslands close to other grasslands are more likely to have high sparrow density. However, the density may be lower than predicted depending on when the patch was last burned (Bakker et al. 2002). Larger grasslands may also attract more birds than smaller grasslands following fire due to a decreased proportion of edge habitat, which has been shown to be unfavorable for Henslow's Sparrow occupancy (Helzer and Jelinski 1999, Davis 2004, Hill and Diefenbach 2013, Hill and Diefenbach 2014, Lituma et al. 2022. Therefore, we would expect that territory selection and breeding following a burn would increase with grassland patch size and decrease with distance to other grasslands. The interaction between grassland characteristics (e.g., size and degree of isolation) and temporal variation in habitat quality (varying with years since burn) will likely be strong factors to consider when determining the best burn regime to promote Henslow's Sparrow density.
To better understand how prescribed fire affects Henslow's Sparrows, we studied their response to a large-scale prescribed-fire regime, in which there are multiple units burned each year within a large complex. Our first objective was to estimate how Henslow's Sparrow densities varied in 4 breeding seasons following a prescribed fire. Our second objective was to examine which fire-and grassland-metrics were associated with changes in Henslow's Sparrow density. We predicted that larger and less isolated grasslands would correspond to larger densities of Henslow's Sparrows (Helzer and Jelinski 1999, Johnson and Igl 2001, Bakker et al. 2002, Davis 2004, Hill and Diefenbach 2013, 2014. We also hypothesized that Henslow's Sparrow densities would be greatest in the second and third breeding season after a burn, after which habitat quality would decline due to woody intrusion (Basey andBadger 2004, Hill andDiefenbach 2013). Our findings provide guidance for optimal burn strategies for managing Henslow's Sparrow habitat given varying habitat characteristics.

M. P. Keating et al.
Henslow's Sparrow response to prescribed fire 3

Study Area
We conducted our study at Big Oaks National Wildlife Refuge (BONWR; 220 km 2 ), located in Jefferson, Jennings, and Ripley counties in southeastern Indiana, USA ( Figure  1). BONWR was designated as a Globally Important Bird Area by the Audubon Society because it provides breeding habitat for the fifth largest population of Henslow's Sparrows in the world (National Audubon Society 2010).
BONWR was formerly known as the Jefferson Proving Ground (JPG) and used for military ordnance testing from 1941 through 1995. The site was managed by the U.S. Army until 1996 (Havlick 2011) and Army staff at JPG maintained GMUs using prescribed fire, disking, mowing, and herbicides until 1995 (Gibbes et al. 2017). Subsequently, the site was managed by the U.S. Fish and Wildlife Service and was designated as a National Wildlife Refuge in 2000. After JPG closed, mechanized management tools (i.e., disking and mowing) were largely discontinued due to the possibility of detonating remnant unexploded ordnance, limiting BONWR staff to prescribed fire as the primary management tool (Basey and Badger 2004). Since 1998, GMUs used in this study were managed using a burn rotation, burning an average of ~250 ha yr -1 . Burn frequency at individual GMUs varied, but each was burned up to two times between 1998 and 2005. Spring prescribed fire at BONWR was conducted between late-February to mid-April.

Study Design and Data Collection
A subset of GMUs was selected for spring prescribed fire each year, such that only a proportion of GMUs were burned each year. Spring burn windows were more reliable than fall burn windows due to more favorable conditions, and there are prohibitions on burning between May and mid-September to avoid disturbance of the Indiana bat (Myotis sodalis), an endangered species (U.S. Fish and Wildlife Service 2019). For these reasons, GMUs burned in the fall burns were withheld from our analysis. GMU selection was based on burn history, weather conditions during the burn season, and accessibility by fire personnel. Only GMUs that were accessible by road and had not been burned in the previous 3 yr were selected for prescribed fire, which resulted in 26 GMUs being monitored for 3-4 breeding seasons following spring prescribed fire ( Figure 1, Supplementary Material Table 1). There were other GMUs that did not meet these criteria and were used to assess the predictive performance of the models we developed. Prescribed fire was implemented via drip torches along roads. Several additional ignition techniques, including fuses and hand launchers, were used to maximize the burn coverage.
We performed count surveys for Henslow's Sparrows on previously burned GMUs for up to 4 yr post-burn. Transect surveys were performed adjacent to GMU areas and were located on 52 single-lane unpaved roads or trails of varying lengths (Figure 1). Some larger GMU areas had multiple transects. Transects were spread throughout the refuge and the mean transect length was 795 m (range: 169-2,062 m). Our survey data consisted of counts of observed and calling Henslow's Sparrows. We acknowledge that conducting surveys along roads may induce sampling bias, particularly when making comparisons across many years of observations (Bart et al. 1995) and edge-and interior species (Keller and Fuller 1995). However, work in similar grassland habitat did not find significant differences between counts from road and non-road surveys (Rotenberry and Knick 1995). Additionally, we were not making comparisons across species, and our treatment (prescribed fire) also followed the road network.
Single observers walked transects counting the number of detected Henslow's Sparrows heard or observed within 150 m of the transect. If both sides of the transect contained a GMU, then each transect would contain a total width of 300 m. We used 150 m as the maximum distance from transects because detection probability of singing sparrows beyond this distance would likely decrease (Herkert and Glass 1999). Each transect was visited once in June between 0600 hours and 1000 hours (Eastern Daylight Time). Breeding seasons 1, 2, 3, and 4 yr following a burn correspond to 3, 15, 27, and 39 mo after a spring burn has occurred.
We recognize that observers might not have detected all Henslow's Sparrows that were present due to imperfect detection probability (i.e., the data may have contained false negative errors). However, we took extensive measures to train observers, used the same observers for multiple years, and routinely cross-checked observer counts within surveys for consistency. Observers were BONWR biologists that completed extensive training prior to sampling. Training included new observers simultaneously walking transects with experi-enced observers and counting singing Henslow's Sparrows independently, and comparing results when sampling concluded. When differences in detection were negligible, new observers began conducting surveys on their own and additional cross-checks occurred throughout the study. We did model detection as a function of distance to determine if our counts were an accurate representation of the number of Henslow's Sparrows in a patch and found that the detection probability was 1.0, which supports our use of the raw count data (Supplementary Material).
We collected Henslow's Sparrow counts at 26 GMUs for 3-4 breeding seasons following spring prescribed fire. After an initial burn and count survey, 6 of the 26 GMUs were burned and monitored a second time during our study. Thus, we had 32 opportunities to examine the effect of prescribed fire on Henslow's Sparrow density. The six transects that were surveyed twice were first surveyed from 1999 to 2002 and resurveyed in 2004-2007 (n = 2) and 2005-2008 (n = 4). To ensure that burning a grassland a second time was comparable to the first burn, we performed a paired t-test to compare densities within a grassland for each year post burn across the two monitoring periods, regardless of year of survey, and did not detect significant differences using an α value of 0.05 (p = 0.17, 0.19, and 0.34, for 1, 2, and 3 yr post burn, respectively).

Statistical Analysis
We developed a suite of a priori hypotheses describing postburn differences in Henslow's Sparrow density by conducting a literature review and using expert opinion, and each hypothesis was represented by a linear mixed model with varying numbers of covariates. Our response variable was the change in density from sampling period t relative to pre-burn density for t = 1, …, 4 breeding seasons following burns (i.e., response t = density t − pre-burn density, such that the response could be positive or negative). We used a random effect for each transect to account for non-independence of withintransect samples and allowed transect densities to have different baseline densities regardless of predictors. We used calendar year of burn treatment to account for the effects of inter-annual variation in vegetation characteristics and precipitation on Henslow's Sparrow density (Yurkonis et al. 2019) and other unobserved contributors to population-level changes. We also examined whether Henslow's Sparrow's density was increasing or decreasing linearly through time using year as a continuous covariate, which would account for systematic unobserved changes occurring at the population or study area level. We used the log of the GMU size to account for the effects of GMU size burned on Henslow's Sparrow density and we used the distances between GMU centroids to its nearest neighbor to represent distance between grasslands. We used the lme4 package (Bates et al. 2015) in R Statistical Software (R Core Team 2020) to fit the models.

Model Selection and Assessment
We used Akaike Information Criterion (AIC) to select the most parsimonious model among our a priori set. We considered the top model to be the one with the lowest AIC value, but we also considered all models within 2 AIC units of the top model to be competing models. We assessed model assumptions for our data by examining the normalized residuals for the top model. We also assessed the predictive performance M. P. Keating et al.
Henslow's Sparrow response to prescribed fire 5 of the top model by comparing density estimates of Henslow's Sparrows from withheld data (the GMUs that did not meet the study inclusion criteria) to predicted density estimates using a likelihood score function (Lee et al. 2018). The withheld data consisted of GMUs that had been burned < 3 yr prior to the start of our study and GMUs that were burned in the fall as opposed to the spring.

RESULTS
The empirical mean change in Henslow's Sparrow density for all transects during the first, second, third, and fourth breeding seasons were −0.09, 0.28, 0.11, and −0.01 Henslow's Sparrows ha −1 , respectively ( Figure 2). AIC indicated that our top model for changes in Henslow's Sparrow density was most influenced by the time since burn, the size of the GMU, and their interaction ( Table 1). The variance estimate for the random intercept parameter was 0.11 in the top model, suggesting high variation among transects. Time since burn was included in the top model and was the only predictor in our second-best model (ΔAIC = 0.7), which collectively received an Akaike weight (w i ) of 0.54, indicating a correlation between Henslow's Sparrow density and time since burned (Table 1). The estimates for the effect of time since burn followed a pattern similar to our empirical mean changes in density ( Figure  2). Our estimates for the effect of time since burn indicated a slight decline in the first breeding season but with a large amount of uncertainty (−0.   Figure 3). These results indicate a positive relationship between grassland size and time since burn, but it is important to note that the confidence intervals overlap zero. We found little support for the hypotheses that the calendar year of a burn and distance to other GMUs influenced changes in Henslow's Sparrow density. Models that contained these covariates had large AIC values compared to the top model and the confidence intervals for the effect sizes overlapped  zero. Our assessment of predictive ability indicated that there was a strong positive correlation between predicted values from our model and actual density estimates from our withheld GMUs (β 1 = 0.24; 95% CI: 0.14, 0.32; R 2 = 0.13).

DISCUSSION
We found that Henslow's Sparrow densities were highest in breeding seasons 2 and 3 after a prescribed burn. We found little evidence that densities of Henslow's Sparrows were lower than pre-burn densities during the first growing season after a prescribed fire on average. This suggests that our data were insufficiently precise to detect a strong correlation between decreased Henslow's Sparrow densities 3 months post-fire, if one exists. One possible explanation of the high uncertainty in our estimate is the short duration between our spring burns and early Henslow's Sparrow arrivals. Henslow's Sparrows begin arriving at BONWR from their wintering area as early as a few weeks after the time GMUs are burned and we collected our data at these burned GMUs 2 and 3 months after the prescribed fire. Therefore, we can conclude that either Henslow's Sparrows had reached BONWR 3 months after the fire, but the recently burned habitat was mostly unsuitable, or the habitat became suitable, but Henslow's Sparrows did not occupy recently burned habitat until after the survey period, when breeding occurs (Crimmins et al. 2016). Alternatively, as time since burn progresses four breeding seasons and longer, some grasslands are invaded by woody vegetation and their suitability for breeding Henslow's Sparrows decreases dramatically (Herkert 2019). Smaller grasslands are especially prone to these edge effects and Henslow's Sparrow breeding densities in some of these GMU would be similar to recently burned areas. The size of our burn complex and the variation of GMU sizes may have also limited our estimation power during the first breeding season. Despite the high uncertainty in our estimate, it is possible that a slight initial decrease occurred, as is supported by other studies of grassland-passerine density response to fire (Herkert and Glass 1999, Powell 2006, Thatcher et al. 2006, Palasz et al. 2010, Herakovich et al. 2021, Lituma et al. 2022. Change in Henslow's Sparrow density was greatest during the second year after a prescribed fire, which was earlier than reported elsewhere (e.g., Herkert and Glass 1999). There are several possible explanations of why this may have occurred. First, grasslands in our study area were comprised primarily of broomsedge species (Andropogon virginicus L, A. ternarius Michx, A. gyrans Ashe), forbs, and invading woody vegetation on acidic soils. The vegetation makeup may lead to densities of grasses and forbs that were structurally ideal with a commensurate reduction of woody vegetation the first few years following a fire than vegetation at other, less acidic study areas. Therefore, ideal vegetation characteristics for Henslow's Sparrows may have been reached earlier at our study area due to the characteristics of our soils and dominant vegetation (Basey andBadger 2004, Palasz et al. 2010). Second, our study area consisted of a complex of many GMUs of varying sizes. A subset of these GMUs was burned each year and thus our burn complex was larger than other studies (e.g., Hill and Diefenbach 2013, Hovick et al. 2015, Herakovich et al. 2021. It is possible that the difference in response to prescribed fires at our study area was due to the differences in burn complex sizes (Wiens 1989) and their effect on heterogeneity and metapopulation assemblages (Fuhlendorf et al. 2009, Hovick et al. 2015. Further, it is possible that uneven disturbance from the fire treatment altered within-patch heterogeneity, and Henslow's Sparrows were responding to that patchiness. Third, other regional differences in and around study areas (e.g., climate, invertebrate populations, farming practices) may have contributed to the observed differences, and Henslow's Sparrow response to fire may be regionally variable depending on landscape characteristics.
Response of Henslow's Sparrow density to prescribed fire was amplified in smaller GMUs (Figure 3). This difference may have been due to larger GMUs containing more heterogeneity in grassland structure compared to smaller GMUs (Lindgren and Cousins 2017). For example, if large GMUs are more likely to contain areas of vegetation that do not burn (e.g., wet areas because of the poor drainage qualities of the soil), then these unburned areas might continue to provide suitable habitat for Henslow's Sparrows immediately after a burn. Smaller GMUs that do not contain the same extent of structural heterogeneity could be less likely to hold pockets of suitable habitat that Henslow's Sparrows can continue to use after a burn. Likewise, structural heterogeneity in grasslands could provide patches of suitable habitat within grasslands near the end of their burn cycle (i.e., 39 mo; Hovick et al. 2015). These patches might remain suitable for Henslow's Sparrows after other areas have become unsuitable (i.e., too woody). Although there were similar directional changes in density, GMU size appeared to attenuate the impacts of prescribed fire, resulting in more stable sparrow densities through time in larger GMUs. Larger GMUs potentially have more habitat heterogeneity which may buffer against the temporal cycle of negative, then positive effects of prescribed fire on Henslow's Sparrow populations. It is important to note that our results do not suggest that distance between grasslands was a significant factor in Henlow's sparrow density response. This lack of isolation effect may be due to high proportions of suitable grassland habitat in surrounding areas (Andrén 1994), supporting our burn rotation strategy. Based on our results Henslow's Sparrow density increased by a net 1.7 birds ha −1 3 growing seasons after a fire, and 1.0 birds ha −1 4 growing seasons after a fire. Although the amplitude of the effect varied with GMU size, these data suggest that a 3-yr burn interval provides the maximum cumulative number of sparrows for this management rotation, and the increased heterogeneity produced by the prescribed fire regime should benefit avian biodiversity across the refuge (Powell 2006, Hovick et al. 2015, Herakovich et al. 2021, Lituma et al. 2022). However, Crimmins et al. (2016) found only nominal support to suggest time since burn influenced Henslow's Sparrow nest survival, which combined with our antithetical results for density, suggests prescribed fire operates at the scale of Henslow's Sparrow site selection versus nest success. Further investigation is required to examine how Henslow's Sparrow density relates to survival rates and reproductive success through density-dependent mechanisms, because our study was limited to inference on density which may not be an indicator of habitat quality ( Van Horne 1983, Stephens et al. 2015, Buderman et al. 2020. We cannot address the sustainability of repeatedly burning grasslands using the same time interval. In many systems, consistent burning schedules are difficult to achieve due to funding and regulatory restraints. Other research supports that burning grasslands more frequently may lead to overall lower Henslow's Sparrow densities but could also prevent woody intrusion and allow grasslands to persist with larger landscape dominance. Nevertheless, our study has highlighted the importance of understanding scale and heterogeneity when assessing grassland bird response to management regimes. BONWR currently hosts over 300 breeding pairs of Henslow's Sparrows, and fire is still the primary tool to combat woody encroachment which remains an issue (Joseph Robb, Project Leader, personal communication). Continued long-term monitoring of repeated burning at BONWR will likely reveal additional population impacts of repeated use of prescribed fire for Henslow's Sparrow management.

Management Implications
Our results suggest that a 3-yr burn cycle (i.e., burning in years 1 and 4) would be beneficial to Henslow's Sparrow densities, and that densities respond positively to prescribed fire in small grasslands as well as large grasslands. They also suggest an amplified response (both positively and negatively) of Henslow's Sparrow density in small grasslands, meaning, after a fire, smaller grasslands will initially have a smaller increase in density of Henslow's Sparrows, but then have a higher net gain in density, compared to larger grasslands. If large grasslands are important for population stability, and small grasslands provide the highest post-burn densities, but are much more variable, this suggests that a mosaic of grassland sizes might be important for long-term maintenance of Henslow's Sparrow metapopulations in our system and similar systems (Fuhlendorf et al. 2009). Similarly, it supports that heterogeneity is important for grassland bird species metapopulations (Andrén 1994, Powell 2006, Hovick et al. 2015, Herakovich et al. 2021, Lituma et al. 2022). However, more information on the reproductive success of individuals in small grasslands is needed to ensure that they do not act as population sinks before deciding on a habitat management plan that maintains a mosaic of small and large grassland habitat with optimal vegetation conditions for Henslow's Sparrows.

Supplementary material
Supplementary material is available at Ornithological Applications online.