Variable precipitation leads to dynamic range limits of forest songbirds at a forest‐grassland ecotone

Abstract Boundaries between vegetation types, known as ecotones, can be dynamic in response to climatic changes. The North American Great Plains includes a forest‐grassland ecotone in the southcentral United States that has expanded and contracted in recent decades in response to historical periods of drought and pluvial conditions. This dynamic region also marks a western distributional limit for many passerine birds that typically breed in forests of the eastern United States. To better understand the influence that variability can exert on broad‐scale biodiversity, we explored historical longitudinal shifts in the western extent of breeding ranges of eastern forest songbirds in response to the variable climate of the southern Great Plains. We used climatic niche modeling to estimate current distributional limits of nine species of forest‐breeding passerines from 30‐year average climate conditions from 1980 to 2010. During this time, the southern Great Plains experienced an unprecedented wet period without periodic multi‐year droughts that characterized the region's long‐term climate from the early 1900s. Species’ climatic niche models were then projected onto two historical drought periods: 1952–1958 and 1966–1972. Threshold models for each of the three time periods revealed dramatic breeding range contraction and expansion along the forest‐grassland ecotone. Precipitation was the most important climate variable defining breeding ranges of these nine eastern forest songbirds. Range limits extended farther west into southern Great Plains during the more recent pluvial conditions of 1980–2010 and contracted during historical drought periods. An independent dataset from BBS was used to validate 1966–1972 range limit projections. Periods of lower precipitation in the forest‐grassland ecotone are likely responsible for limiting the western extent of eastern forest songbird breeding distributions. Projected increases in temperature and drought conditions in the southern Great Plains associated with climate change may reverse range expansions observed in the past 30 years.


| INTRODUC TI ON
Zones of changing temperature and precipitation can mark the boundaries of species' distributions (e.g., Zuckerberg et al., 2009).
Studying dynamics at species' distributional limits is important to understand threshold responses in areas of increased environmental stress (Holt & Keitt, 2005). Individuals at their distributional limits may be especially vulnerable to climate change because they occupy sites with prevailing conditions generally outside their optimal climatic niche (Glennon, 2014;Thuiller et al., 2005). Bird distributions are spatially dynamic in response to changes in temperature and precipitation over time, and this has been seen as evidence of dispersal to and colonization of new areas that fall within a species' climatic niche Huang et al., 2017;Tingley et al., 2009).
The influence of temperature on bird distributions has been well studied. In North America and Europe, some bird species and assemblages have shifted their latitudinal range limits and mean distributions northward in response to warming temperatures (Devictor et al., 2008;Hitch & Leberg, 2006;Thomas & Lennon, 1999;Zuckerberg et al., 2009). However, not all species exhibit significant distributional shifts in response to temperature, and some that do exhibit multidirectional shifts in response to precipitation or other environmental influencers (Currie & Venne, 2017;Huang et al., 2017). For example, breeding distribution of Henslow's Sparrow (Ammodramus henslowii) in the temperate United States has not shifted northward in response to rising temperature, suggesting other factors, including changes in precipitation, might be better predictors of range shifts for grassland birds (McCauley et al., 2017).
Drought-sensitive grassland birds in the southern Great Plains vary in their responses according to the temporal scale of droughts (Cady et al., 2019). Long-term data suggest some bird species are responsive to local climate changes and may shift their distributions in response to temperature, precipitation, or both depending on which climatic variable limits Net Primary Productivity (Tingley et al., 2009). Thus, there is potential for multiple climatic variables, including temperature, precipitation, and potential evapotranspiration (PET), to exert a strong influence over species' distributions (Barbet-Massin & Jetz, 2014).
In the southcentral United States, a marked precipitation gradient defines the broad ecotone between temperate forests of eastern North America and grasslands of the central Great Plains. A transitional landscape characterizes this region between the oak-hickory (Quercus spp. and Carya spp.) forests and the tallgrass prairies of the eastern Great Plains (Figure 1). Subject to periodic droughts and traditionally managed with fire and grazing, this region blends oak woodland, oak savanna, and tallgrass prairie vegetation (Rice & Penfound, 1950;Fuhlendorf et al., 2009). Long-term precipitation trends in the Great Plains are highly variable, and severe extended droughts have alternated with pluvial periods (Figure 2; Basara et al., 2013). Fire suppression has also changed forest composition and structure since the 1950s; stand density has increased, and mesophytic species such as elms (Ulmus spp.), red mulberry (Morus Many eastern forest songbirds reach the western extent of their breeding range in these transitional forests (Heinen & O'Connell, 2009;Reinking, 2004). These species may be sensitive to climate fluctuations at the forest-grassland ecotone (Clement et al., 2019). Based on Breeding Bird Survey population trends since 1966, the abundance of at least 15 eastern forest birds has increased and expanded westward into the southern Great Plains (Sauer et al., 2014). Expansion of these forest songbirds may have been facilitated by the 1980-2010 pluvial climate and the expansion of mesophytic tree species.
In the southern Great Plains, extended severe droughts occurred in the 1930s and 1950s. In contrast, between 1980 and 2010 F I G U R E 1 Map of the study area illustrating the transitional ecoregion between eastern oak-hickory forests and the grasslands of the Great Plains an unusually wet climate prevailed in Oklahoma (Figure 2; Basara et al., 2013). The southern Great Plains are projected to have more frequent and extended drought events in the near future (Shafer et al., 2014). To predict species' responses to projected climate change, it is important to understand historical responses to fluctuating and extreme climatic conditions (Araujo & Pearson, 2005). Our objectives were to (1) examine how western distributional limits of eastern forest songbirds shifted from historical drought periods to recent pluvial conditions and to (2) verify the reliability of historical projections of potential species distributions. We trained a climatic niche model for nine eastern forest songbirds using presence-only data from eBird and BBS from 1980 to 2010 and concurrent average climate data, and then projected the model onto 1952-1958 and 1966-1972 climate rasters, the latter evaluated with North American Breeding Bird Survey (BBS) data from 1966 to 1972. We hypothesized that historical droughts were responsible for longitudinal retractions of breeding distributions eastward and that pluvial periods created opportunities for expansion westward into the Great Plains.

| Bird species records
We generated potential distribution maps of nine passerine species that nest broadly across forested landscapes of eastern North America, reaching a western distributional limit in the Great Plains.
We used presence-only data from the community science network, eBird, and from the North American Breeding Bird Survey (BBS; Cooper et al., 2014;Sullivan et al., 2009). eBird is a community science project established in 2002 that engages millions of avocational bird watchers in the collection of semistructured data on the distribution and abundance of birds (Sullivan et al., 2014). The BBS is a standardized and systematic survey using specific methods and highly trained observers to estimate the abundance and trends of breeding birds in the United States, southern Canada, and northern Mexico. The BBS is a roadside survey established in 1966. BBS routes include 50 stops approximately 0.8 km apart where trained observers record counts of species detected during a 3-min listening period (Sauer et al., 2014). Each BBS route includes observation aggregations at five ten-stop segments.
Presence data from eBird and the BBS were downloaded for Eastern Wood-Pewee (Contopus virens), Acadian Flycatcher  1930-1939and 1952-1958, and an extended dry period occurred in 1962-1972(LSU 2012 data from routes in which species were detected within the first 10 stops (8 km) of each route. Locations associated with BBS presence data were assigned to the starting point of each survey route. To match the effort and resolution of eBird presence data with BBS 10 stop data, we selected eBird presence data between June and July from checklists that covered less than 8 km in less than 60 min with fewer than 10 observers (Strimas-Mackey et al., 2020). To reduce sampling bias in species climatic niche models, we spatially rarefied occurrence data by 40 km using SDMtoolbox (Boria et al., 2014; Brown, 2014).

| Climate variables
We retrieved climatic data from the PRISM Climate Group models (PRISM Climate Group, 2014). Environmental variables included annual cumulative precipitation, annual mean temperature, and spring (April-June) and summer (July-August) PET at 4km resolution (PRISM Climate Group, 2014; Thornthwaite, 1948). These environmental variables were selected as consistently relevant across bird species when developing species distribution models (Barbet-Massin & Jetz, 2014). Temperature, precipitation, and PET values were standardized based on their mean and standard deviation. We used average climate data across years for each of three time periods: 1980-2010, 1966-1972, and 1952-1958

| Maxent modeling parameters and evaluation
We modeled species' climatic niches in the program Maxent 3.4.4 using presence-only occurrences and climate data (Phillips et al., 2006;Phillips & Dudick, 2008). The Maxent algorithm estimates environmental parameters of a species' niche by sampling occurrences from its known distribution and contrasting them with randomly selected background samples (Elith et al., 2006;Phillips et al., 2006). For our models, we used the default Maxent ver. 3.4.4 settings for background sampling (10,000), as well as for features (environmental constraints of probability distribution), regularization multiplier, and maximum iterations and convergence threshold.
We randomly selected 70% of species' occurrences to train models and generate potential distribution maps from the 1980-2010 average climate data and 30% of occurrence points to test model performance on the 1980-2010 distribution map. A first set of models was run using all variables; variables that contributed less than 1.0% to overall accuracy gain in the initial model were removed from the final model. We evaluated the accuracy of 1980-2010 models by calculating the continuous Boyce index on the random 30% of presence data withheld as test points (Hirzel et al., 2006). Boyce index was calculated in program R version 4.1.0 using the package ecospat (Broennimann et al., 2021;R Core Team, 2021 (Hirzel et al., 2006). Omission error (percentage of test presences predicted absent) was also calculated to evaluate model performance. To calculate omission error, we converted Maxent outputs of probability suitability values between 0 and 1 to binary (suitable-unsuitable) maps by applying the 10-percentile training presence threshold rule.
This rule selects the Maxent value at which 10% of the occurrence locations used to train the model are predicted climatically unsuitable and then reclassifies all locations with values below this threshold as unsuitable and those above the threshold as suitable. The models were projected onto the 1952-1958 and 1966-1972 climates, and we applied the same 10% threshold rule to generate binary suitableunsuitable potential distribution maps for each time period. We used these maps to identify shifts in species range boundaries.

| Independent validation of historical projections
We used BBS presence data from the first 10 stops of [1966][1967][1968][1969][1970][1971][1972] roadside surveys and locations of route starting points as independent evaluation points to indicate the predictive power of historical projections of species' potential climatic distributions (Pardieck et al., 2015). We evaluated the reliability of 1966-1972 species' estimated potential distributions by calculating the continuous Boyce index on concurrent presence data from BBS. This validation was not possible for model projections of 1952-1958 potential distributions due to the lack of independent datasets of species' presences for that time frame; thus, the model performance is the only indicator of reliability of these estimates. contribution to the Maxent model (  Figure 5).

| D ISCUSS I ON
We found that changes in climate conditions resulted in longitudinal shifts in estimated potential distributions of eastern forest songbirds along a broad forest-grassland ecotone. Our results describe westward expansion of species' potential distributions from historical drought conditions in 1952-1958 and 1966-1972, to recent pluvial conditions in 1980-2010 in the southcentral United States. These results are consistent with at least one study that found high rates of colonization and a positive occupancy trend for Eastern Wood-Pewee in the southwest portion of its range in the late 1990s (Clement et al., 2019 Warbler, and Northern Parula have increased locally in that period (Sauer et al., 2014).
We found the western limits of potential distributions of eastern forest songbirds along a forest-grassland ecotone were sensitive to precipitation and temperature changes under historical drought conditions. Precipitation most strongly influenced the potential distribu- Notes: Presence-only data from 1980-2010 were related to concurrent average annual precipitation (P) and temperature (T), as well as spring and summer potential evapotranspiration (PET) across the eastern temperate forest and central Great Plains ecoregions of the United States.
"." Indicates a variable contributed <1% to a climatic niche model in the full model and was not included in the final reduced model.

TA B L E 2
Percent contribution of climate variables to accuracy gain of species climatic niche models for nine eastern forest songbirds F I G U R E 4 Potential species distributions estimated with Maxent models, showing westward shifts in species estimated range margins from the severe drought of 1952-1958 (red) to the pluvial period of 1980-2010 (blue) for Eastern Wood-Pewee, Acadian Flycatcher, Whiteeyed Vireo, Yellow-throated Vireo, Red-eyed Vireo, Louisiana Waterthrush, Black-and-white Warbler, Kentucky Warbler, and Northern Parula composition. Ecotone shifts are driven by climate patterns (Allen & Breshears, 1998;DeSantis et al., 2010;Risser, 1995). Tree species richness, as well as the basal area and tree density of eastern red cedar, elms, red mulberry, black hickory, and sugarberry have increased (DeSantis et al., 2010). During this period of mesophication, many eastern forest songbirds expanded farther west into the southern Great Plains (Sauer et al., 2014).
Structural attributes of forests are important determinants of forest bird density (Bakermans et al., 2012). Forest cover, soil moisture, and canopy height are among the most important environmental predictors shaping songbird community composition in this transitional ecotone (Cavalieri et al., 2009). Increased precipitation during pluvial periods may have contributed to changes in forest composition, structure, and extent at the forest-grassland ecotone, thereby facilitating the recent western expansion of eastern forest songbirds' breeding range. Among the focal species included in this study, Kentucky Warbler is positively associated with increased ground litter, understory density, and canopy height (Bakermans et al., 2012). Eastern Wood-Pewee occupies forests with greater tree density and canopy cover and landscapes with greater forest cover overall (Holoubek & Jensen, 2015;Kendrick et al., 2013).
Acadian Flycatcher, Northern Parula, and Red-eyed Vireo are mature forest species associated with higher canopies (Reidy et al., 2014).
Forest patches, especially expanding from riparian gallery forests, have increased the amount of land area in the southern Great Plains with attributes that are potentially attractive to bird species typically associated with temperate forests of eastern North America (Knight et al., 1994).
Although pluvial periods can create conditions amenable to the expansion of forest birds in the Great Plains, droughts can have the opposite effect. Water-limited forests are especially vulnerable to temperature increases and drought which can increase the background rate of tree mortality, cause widespread forest die-off, and induce rapid shifts in woodland ecotones (Allen & Breshears, 1998;Allen et al., 2010;Breshears et al., 2005;van Mantgem et al., 2009).
Drought events have been responsible for oak mortality in savannas of Minnesota and oak decline in the Ozarks of Missouri (Faber-Langendoen & Tester, 1993;Voelker et al., 2008). In the 1950s, drought stress caused significant tree mortality in the oak savannas of central Oklahoma, where 11.8% of trees died within a stand (Rice & Penfound, 1959). The forest-grassland boundary of the southcentral United States is also sensitive to fluctuating trends in precipitation and temperature, where extended droughts have caused high rates of tree mortality and canopy loss in transitional oak woodlands and forests (Albertson & Weaver, 1945;Rice & Penfound, 1959;Schwantes et al., 2017).  1966-19722014NARCCAP, 2012). The southern Great Plains is projected to become drier as well. In Oklahoma, summer precipitation is projected to decrease and the number of dry days is projected to increase (Kunkel et al., 2014).
Cause and effect relationships are difficult to determine and both climate and land cover change drive distributional changes for many bird species (Clement et al., 2019). We examined changes in potential distributions of eastern forest songbirds based on projections of climatic niche models onto historical climate conditions.
We did not evaluate contributions of land cover change to species distributional shifts. We also could not disentangle the confounding effects of increased precipitation and fire suppression on forest cover and structure (Wine & Zhou, 2012 Although the temperature is often seen as the primary driver of climatic change effects on wildlife, in our study the low predictive power of this driver in regional and individual species' responses highlights the need for more case-by-case research than for broad generalities (Currie & Venne, 2017). For forest-breeding passerines at the edge of their distribution within the transitional ecotone between the southern Great Plains and eastern temperate forests, our results suggest that dynamic trends in precipitation will likely induce greater variability in longitudinal shifts in distribution than temperature increases will affect shifts in latitude. Thus, important information need moving forward will be regional climatic vulnerabilities and the degree to which changes in distributional limits will affect the trajectory of total populations of eastern forest passerines.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

O PE N R E S E A RCH BA D G E S
This article has earned an Open Data, Open Materials Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https:// doi.org/10.5061/dryad.dfn2z 3525; http://biodi versi tyinf ormat ics.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data used in these climate niche models are publicly available.