Sixty years of community change in the prairie–savanna–forest mosaic of Wisconsin

Abstract Biodiversity loss is a global concern, and maintaining habitat complexity in naturally patchy landscapes can help retain regional diversity. A mosaic of prairie, savanna, and forest historically occurred across central North America but currently is highly fragmented due to human land conversion. It is unclear how each habitat type now contributes to regional diversity. Using legacy data, we resurveyed savanna plant communities originally surveyed in the 1950s to compare change in savannas to that in remnant forests and prairies. Savanna community structure and composition changed substantially over the past 60 years. Tree canopy density nearly doubled and many prairie and savanna specialist species were replaced by forest and non‐native species. All three habitats gained and lost many species since the 1950s, resulting in large changes in community composition from local colonizations and extinctions. Across all three habitats, regional species extinctions matched that of regional colonization resulting in no net change in regional species richness. Synthesis—Despite considerable species turnover within savannas, many species remain within the broader prairie–savanna–forest mosaic. Both regional extinctions and colonizations were high over the past 60 years, and maintaining the presence of all three community types—prairie, savanna and forest—on the landscape is critical to maintaining regional biodiversity.

The natural habitat complexity that supports biodiversity throughout the prairie-savanna-forest mosaic has become increasingly compromised as humans alter the landscape (Pogue & Schnell, 2001). Land conversion to agriculture and urban expansion following European settlement greatly decreased coverage of natural systems (Anderson, Fralish, & Baskin, 1999;Pogue & Schnell, 2001;Rhemtulla, Mladenoff, & Clayton, 2007). By the 1950s, prolonged, widespread livestock grazing helped make oak savannas with intact understory communities the rarest natural ecosystems in Wisconsin (Curtis, 1959). Furthermore, altered disturbance regimes, namely decreased fire frequency leading to mesification (Nowacki & Abrams, 2008) and increased deer herbivory (Wiegmann & Waller, 2006), caused substantial compositional shifts in recent decades (Rogers, Rooney, Olson, & Waller, 2008). Within forest understories, plant communities homogenized taxonomically over the past several decades (Rogers et al., 2008). Within prairies, loss of fire and increased isolation of remaining patches accelerated local extinctions and the spread of woody and weedy species at the cost of prairie specialist species (Alstad et al., 2016;Kraszewski & Waller, 2008;Leach & Givnish, 1996). Historically, savannas contained a mix of both forest and prairie species (Bray, 1960) and could play a critical role in maintaining biodiversity throughout the region. Given the substantial change to regional forests and prairies (Alstad & Damschen, 2015;Alstad et al., 2016;Rogers et al., 2008), savannas presumably also changed, but the change has not been evaluated.
Here, we use a unique legacy dataset from remnant savanna sites to ask: (Q1) How has the species composition of savanna communities changed over the past 60 years, and how does the amount of change compare to that of prairies and forests? (Q2) Have savannas acted as a refuge for prairie and forest species over the past 60 years to help maintain regional biodiversity? (Q3) How does the relative contributions of local diversity in prairies, savannas, and forests contribute to regional biodiversity?  (Curtis, 1959;Waller, Amatangelo, Johnson, & Rogers, 2012). Sites included both oak savannas and cedar glades that represented the best remaining native savannas in the state. Care was taken to select sites with minimal human disturbance, including limited logging and grazing and intact native understory communities not heavily invaded by non-native species (Bray, 1955(Bray, , 1960.

| Site relocation and vegetation surveys
In 2014, original survey locations of savannas were relocated and surveyed following the same methods from the 1950s surveys.
Resurvey effort focused on sites with species frequency data (as opposed to only species presence) for understory communities in 1950s and sites that were not converted to a different land-use type since the 1950s (e.g., pasture, tree plantation, golf course).
To relocate the original survey sites, detailed notes and handdrawn maps from the 1950s surveys were compared to aerial images F I G U R E 1 Survey locations of forest (triangles), savanna (diamonds), and prairie (circles) sites across Wisconsin and maps in ArcGIS (ESRI, Redlands, CA, USA). Specifically, we used historical aerial images obtained from the UW Geography Museum, current aerial images, Wisconsin Land Economic Inventory (Bordner Survey) maps from the 1920s (http://digital.library.wisc.edu/1711. dl/EcoNatRes.WILandInv), and topography from USGS (http://store. usgs.gov). Additionally, tree density in historic images was compared to tree density in survey records to further locate transect placement within a site. In most cases, current land use was detectable from contemporary aerial images, but for sites where land use was uncertain, additional land-use maps were obtained from the Southeastern Wisconsin Regional Planning Commission (www.sewrpc.org). A polygon was drawn around the extent of each intact site, and parallel transects were drawn within the site to guide plot location during field surveys. The ideal plot layout was a grid with sample points 10 m from one another, but site area and dimensions restricted transect configuration at most sites. To obtain land-use histories for the past 60 years, we talked with landowners and specifically asked whether and when the site had been grazed, burned, harvested, or experienced any other disturbance since the original survey. Signs of recent disturbances, including stumps and burn scars, were noted during resurveys.
In summer 2014, we surveyed the savanna communities following the same methods as the initial 1950s survey (Bray, 1960). We selected at least 20 sampling points >10 m from each other and surveyed the canopy, understory, and shrub vegetation at each point.
For canopy trees, the random pairs technique was used (Cottam & Curtis, 1949), and two trees were measured at each point, including the size (DBH), species identity, and distance between trees.
For understory vegetation, a 1 × 1 m quadrat was placed North and center of the sampling point and presence of all woody species less than 1 m tall and all herbaceous species rooted in the quadrat were recorded. First-year tree seedlings were recorded separately from older seedlings given the high mortality rate of first-year seedlings.
For the shrub layer, a straight 2-m-wide transect was positioned between the two surveyed trees. All shrubs >1 m tall and small trees (>1 m tall and <10 cm DBH) rooted within the transect were identified to species and counted.
Taxonomic resolution was kept consistent between the 1950s and contemporary surveys. Most plants were identified to species, but some were identified to genus (e.g., Carex spp.) as per the original sampling. Nomenclature followed the Wisconsin Flora (Chadde, 2013).
In total, sixteen savanna sites were resurveyed in 2014, four of which were previously resurveyed in 2004 (Mills, 2008). All sites experienced altered disturbance regimes, especially through the loss of fire. In recent decades, three sites experienced substantial management including canopy thinning and the return of prescribed fire. To compare changes in savannas relative to those in prairies and forests, savanna resurvey data were compared to forest and prairie resurvey data from similar historic datasets (Curtis, 1955;Rogers et al., 2008;Alstad & Damschen, 2015;Figure 1). Hardwood forests were resurveyed from 2002 to 2004 (Rogers et al., 2008), and prairies were resurveyed in 2012 (Alstad & Damschen, 2015). We also included resurvey data from eleven additional prairie sites that were initially sampled using identical methods to the surveys above (Whitford, 1958) and resurveyed in summer 2015.

| Data analysis
To examine changes in savanna plant communities over the past 60 years, we evaluated compositional change for both the canopy and understory. First, we compared changes in canopy tree density (tree per acre) between 1950s and 2014 with a paired t test.
Canopy composition at the two time points was compared with a nonmetric multidimensional scaling (NMDS) approach using Bray-Curtis dissimilarity. To test for a significant difference between canopy composition in the 1950s and 2014, we used the PERMANOVA function "adonis" in the "vegan" package in R.
Except where noted elsewise, all statistical analysis was run in R (R Foundation for Statistical Computing, Vienna, Austria, version 3.2.1). Second, changes in savanna understory communities were assessed by comparing rank abundance curves from the 1950s and 2014. To visualize changes through time, species were colorcoded based on their presence in savannas and the broader mosaic at both survey points. To determine the degree of change in savannas relative to prairies and forests, we used two separate NMDS analysis, one examining species presence-absence with a Jaccard distance metric and a second examining species abundance with Bray-Curtis dissimilarity. To test for significant differences among habitats types (prairie, savanna, forest), time periods (1950s, 2010s), and their interaction, we ran repeated-measures PERMANOVAs in PRIMER (Clark & Gorley, 2015). To test whether sites within each habitat type became more similar to one another over time, we ran a multivariate homogeneity of group dispersions (variances) analysis with the function "betadisper" in the "vegan" package to compare the pairwise distance among all sites for each time period. The test is sensitive to sample size, which varied between habitat types, so a separate test was run for each habitat type to individually evaluate changes in homogeneity through time.
To further examine changes in regional diversity throughout the mosaic, we first compared changes in species abundance in savannas relative to neighboring communities and then examined the relative influence of local diversity of each habitat on regional diversity. To determine whether savannas acted as refuges for prairie or forest species, we assessed the ability of savanna sites to retain or gain species that were declining in prairies and forests. To do this, we calculated an index of change for each species found in savannas, specifically: where "f" is the count of sites a species occupied during the survey time noted in subscript. If the species decreased between 1950 and 2010, the index was multiplied by -1 to differentiate increasing indicated that overall savannas acted as a refuge, while an intercept <0 indicated that either prairies or forests acted as a refuge for savanna species. Next, to understand how relative contributions of local diversity in prairies, savannas, and forests contributed to regional diversity we examined the number of shared and unique species among the three habitat types in both the 1950s and 2010s. Specifically, we examined how many species entered (colonization) and left (extinction) over the past 60 years and whether colonizations and extinctions were limited to one habitat but not the whole mosaic (local) or pertained collectively to all three habitat types (regional). the region. Across all habitats, the number of regional extinctions and colonizations was very similar (Figure 6c), leading to no overall change in regional species richness between the 1950s and 2010s

| RE SULTS
( Figure 5). Importantly, our measures of species richness are more comprehensive in forests, as sampling intensity was much greater in forests (85 sites) than in savannas or prairies (16 and 17 sites, respectively).

| D ISCUSS I ON
A mix of prairie, forest, and savanna specialist species once cooccurred in the savannas of southern Wisconsin but over the past 60 years, sites lost many prairie and savanna species and now more closely resemble forests. Over half the species historically found in both savannas and prairies decreased in abundance over the past 60 years (Figure 4). During this time, fire was largely absent from both savannas and prairies (Alstad & Damschen, 2015; Ladwig personal comm.). Species diversity in savannas relates to fire frequency (Peterson & Reich, 2008;Weiher, 2003), and as fire frequency decreased so did the diversity of prairie species. Loss of fire also influenced canopy dynamics, as the most visually apparent change in savannas was a near doubling of tree density since the 1950s (Supporting information Figure S1). Prior to the original surveys, tree density had already increased in savannas since pre-European settlement (Cottam, 1949) and this trend continued over the past 60 years. One factor contributing to the continued increase in tree density, particularly of mesic tree species, may be the loss of historically routine, low-intensity fires that maintained savanna ecosystems (Wolf, 2004). In the absence of fire, trees and shrubs can more easily establish and succession progresses (Nowacki & Abrams, 2008;Wolf, 2006). Although savannas changed substantially, the observed changes were expected given the loss of fire and increase in woody plant cover. An increase in woody plant cover is not only a measure of change but also a driver of change (Briggs et al., 2005). Across North American grasslands and savannas, plant diversity decreases as woody plant cover increases (Rataczjak, Nippert, & Collins, 2012 (Leach & Givnish, 1999), but this benefit for biodiversity is now missing in the region.
All three habitats within the prairie-savanna-forest mosaic experienced large ecological shifts over the past 60 years (Rogers et al., 2008;Alstad et al., 2016; Figure 3). Woody and non-native species increased in both forest understories and prairies (Alstad & Damschen, 2015;Rogers et al., 2008) and similar changes occurred in savannas (Supporting information Figures S1 & S2). Meanwhile, many prairie and savanna specialist species decreased or went locally extinct over the past 60 years (Alstad et al., 2016;Figures 2 and 6). At a regional scale, the number of species colonizations and extinctions was roughly equal (Figure 6c), leading to no overall change in regional species richness ( Figure 5). In a previous study of 47 prairies in the region, community composition also changed F I G U R E 5 Species richness partitioned among habitat types in the prairie-savanna-forest mosaic in the 1950s and 2010s. Sampling intensity was consistent between times but varied among habitats, as more forest sites (85) were surveyed than prairie (17) or savanna (16) sites F I G U R E 4 Change in frequency (∆ index; see methods for calculation) of species in prairies (left) and forests (right) relative to savannas between the 1950s and 2010s. Positive changes indicate increased abundance through time while negative changes indicated decreasing abundance. Points within the colored square outlines indicate species that decreased in one habitat but increased in a neighboring, or refuge, habitat The prairie-savanna-forest mosaic has been dynamic in the past (Davis, 1977) and will likely continue to shift in the future. The absence of fire may intensify mesification (Nowacki & Abrams, 2008) and promote woody encroachment in open areas (Heisler, Briggs, & Knapp, 2003;Van Auken, 2000), favoring forests over prairies.
Alternatively, larger droughts that stress trees may favor prairies and savannification of forested areas (Allen, Breshears, & McDowell, 2015;Brzostek et al., 2014;Frelich & Reich, 2010;Gustafson & Sturtevant, 2013). Maintaining the full variety of habitats within the prairie-savanna-forest mosaic could allow for future retention of species on the landscape as species continue responding to global change.
The gradient of community types once present throughout the prairie-savanna-forest mosaic is disappearing, but not yet gone. In the 1950s, savannas were already rare on the landscape (Curtis, 1959) and contained a mix of prairie to forest species (Bray, 1960; Figure 3). Sixty years later, the occurrence of savanna specialist and prairie species in savannas is rare, as understories now more closely resemble forests (Figure 3). Although the abundance of prairie and savanna specialist species has greatly decreased in recent decades, many species still remain in savannas but at much lower densities (e.g., occurring in one quadrat at one site), and some of these species could act as indicator species to predict restoration success (González, Rochefort, Boudreau, & Poulin, 2014).
Yet it remains unknown how long savanna understory species can persist without frequent fire, making it urgent to restore remnant savannas. Returning historic disturbance regimes (e.g., periodic low fire; Peterson & Reich, 2001, 2008Weiher, 2003;Weiher & Howe, 2003) and initiating management techniques to reduce canopy cover

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
LML, EID, and DAR conceived the ideas, designed the methodology, and collected data; LML analyzed the data; LML, EID, and DAR contributed critically to the drafts.

DATA ACCE SS I B I LIT Y
Data will be available through the Dryad Digital Repository (datadryad.org) and historical data are also available through the UW Plant Ecology Laboratory (botany.wisc.edu/PEL/data.html).
F I G U R E 6 Local (a, b) and regional (c) species colonizations and extinctions in the prairie-savanna-forest mosaic of southern Wisconsin over the past 60 years. The number of sites surveyed (n) varies among habitat types (a) (b) (c)