Where should they come from? Where should they go? Several measures of seed source locality fail to predict plant establishment in early prairie restorations

Abstract

During the “decade on restoration,” we must understand how to reliably re-establish native plant populations. When establishing populations through seed addition, practitioners often prioritize obtaining seed from locations geographically near the restoration site (i.e. “local seed sourcing”). They are assumed to be under similar environmental conditions to the restoration site and should establish more robust plant populations and preserve local biotic interactions better than seeds sourced from further away. However, this assumption remains virtually untested in realistic restoration settings and the importance of seed sourcing, relative to other factors such as seeding rate and management regimes, is unclear.To determine if seed sourcing decision impacts plant establishment, abundance and phenology, we developed a partnership between university-researchers and a native seed producer that kept records on where their seed was sourced from and where it was planted. At each site, we recorded the abundance and phenological stage of five commonly used tallgrass prairie restoration species seeded at 24 sites undergoing restoration across Michigan. We considered two measures of seed source locality: geographic distance (seeds were sourced from locations 6–750 km away from their respective restoration sites) and environmental distance. We also obtained data on the seeding rate and post-seeding management at each site.We found that no measure of seed source locality predicted the likelihood of plant establishment or abundance at restoration sites. However, sites sown with seed from further away, or from cooler and wetter climates, had a greater proportion of flowering individuals earlier in the season. Finally, sites with higher seeding rates had greater plant abundance, and post-seeding management of the restoration site increased the likelihood a species would establish by 36%.Overall, these results support that seed sourcing decisions did not impact plant establishment or abundance in our system. However, using less-local seed sources may alter flowering phenology.Our results suggest that tallgrass prairie restoration efforts should prioritize higher seeding rates, post-seeding management, and might expand the region seed sources are considered “local”, though this may impact flowering phenology. Future research leveraging native seed producer records can help answer critical questions about restoration seed sourcing.

INTRODUCTION

Almost 50% of the area once covered by natural ecosystems has been lost due to conversion of this land to cities, roadways, large-scale agriculture and other forms of human land use (Ellis et al., 2020; Vitousek et al., 1997). In response, the United Nations has declared the years 2021–2030 as “the decade on restoration,” with intentions to restore over 350 million hectares of land (UNEP—UN Environment Programme, 2019). To meet these goals, we must develop strategies to increase the success of restoration outcomes (which are notably unpredictable; Suding, 2011). This includes ensuring that native plant populations establish and persist, even under the degraded site conditions at the onset of restoration efforts (Benayas et al., 2009; Society for Ecological Restoration International Science & Policy Working Group, 2004). Although some target plant species may arrive at a restoration site on their own, species with limited dispersal capabilities may not (e.g. Hubbell et al., 1999; Pywell et al., 2002; reviewed in Holl & Aide, 2011), producing a community primarily composed of nuisance species that persist in the seedbank (e.g. Pyke et al., 2013). The re-establishment of these dispersal-limited species is primarily driven by active reintroduction, most commonly through seed addition (Kettenring & Tarsa, 2020; Kimball et al., 2015). Despite the importance of seed sowing, it is unclear which seed sources will best establish self-sustaining native plant populations (Hufford & Mazer, 2003; reviewed in Bucharova et al., 2019; Jones, 2013; McKay et al., 2005; Prober et al., 2015).

Current best practices for seed-based restoration support obtaining seed from one or more locations geographically near the restoration site, also known as local seed sourcing (Gustafson et al., 2005; McKay et al., 2005; Montalvo & Ellstrand, 2000). Since there are numerous studies identifying differentiation in plant populations caused by specific adaptations to their environment (i.e. local adaptation; Hereford, 2009; Turesson, 1922; Whitlock, 2015), practitioners assume local seed sources are under the same (or very similar) environmental conditions as the restoration site and will be better adapted to thrive under those conditions (Bower & Aitken, 2008; Lesica & Allendorf, 1999; Mortlock, 2000) compared to less local seed sources (Gustafson et al., 2005; reviewed in Gallagher & Wagenius, 2016; Vander Mijnsbrugge et al., 2010). For many species, systems and geographic regions, though, this assumption remains untested.

Seed sourcing decisions may impact restoration outcomes by affecting initial plant establishment or by introducing individuals with phenological traits that decrease population persistence or alter interactions with other species. Maladapted seed sources may have long-term consequences for focal population development: although initial establishment is not always indicative of the relative abundance of species in mature ecosystems, low initial establishment can increase population extinction risk at the restoration site due to low genetic diversity (Newman & Pilson, 1997) or demographic stochasticity (Purvis et al., 2000; Shriver et al., 2019). Low initial establishment can also cause the restoration to be dominated by non-sown species (Warren et al., 2002), especially at the onset of restoration after clearing the land of pre-existing vegetation (Society for Ecological Restoration International Science & Policy Working Group, 2004). If sown seed establishment is low, non-sown species can re-establish in their place.

Additionally, seed sources from more distant locations may be adapted to different local climates, which may impact traits important for persistence at a restoration site (such as flowering time; Anderson et al., 2012; Neil & Wu, 2006). If populations established from less local seed sources flower at different times than local populations would, this could result in a pollinator mismatch: a reduction in the interactions between plants and their pollinators (reviewed in Bucharova, 2017). Low pollination rates can substantially reduce reproduction in primarily outcrossing species (e.g. Rafferty & Ives, 2012; Thomson, 2019), which could decrease long-term persistence for that species at a restoration site. A mismatch could also reduce resources available to local pollinators, affecting their population demographics (e.g. Kudo & Ida, 2013) and, beyond that, could decrease the biodiversity of the entire ecosystem (Ramos-Jiliberto et al., 2020).

Despite fears that less local seed will have reduced establishment and persistence at a restoration site, experimental results are mixed. While some studies have shown lower establishment of seed sourced from further away in single-species common-garden approaches (Montalvo & Ellstrand, 2000; Raabová et al., 2007) others see no difference in establishment (Galloway & Fenster, 2000; Smith et al., 2005) or find results differ by species (Carter & Blair, 2012). Further, studies identifying shifts in flowering phenology have largely been done in the context of climate change (e.g. Høye et al., 2013; Kudo & Ida, 2013; Ogilvie et al., 2017). Few field studies have quantified the phenology of less local plant populations in current climates nor how that may impact local pollinators. Of those few, one study found no difference in flowering phenology between the local population (Ohio) and plants from populations as far away as Texas (Selbo & Snow, 2005) while another did observe earlier flowering phenology in the most distant seed source, but this resulted in two times more interactions with pollinators than the local source (Bucharova et al., 2022). Given the mixed results, there is not strong evidence that local seed sources consistently establish better, nor better retain biotic interactions, than more distant seed sources under realistic restoration site conditions. Moreover, since there are substantial concerns that we will be unable to collect enough seed from local populations (Broadhurst et al., 2015; Nevill et al., 2016) and native seed producers (Ladouceur et al., 2018) to sustainably meet the high demand for restoration seed, using less local seed sources may become necessary; in some regions, it is even advocated for (Hancock et al., 2022).

One explanation for these mixed results is that predicting whether a seed source will be adapted to the climatic conditions at the restoration site is difficult. Several studies have shown that plants can be adapted to a wide variety of climatic variables ranging from annual temperatures (Baughman et al., 2019) to season-specific precipitation (Blumenthal et al., 2021). However, since the home environments of seed sources are rarely quantified, practitioners instead often use geographic distance (straight line distance between a seed source and a restoration site location) as a proxy for climate similarity (following the first law of geography, Tobler, 1970; reviewed in McKay et al., 2005). However, environmental conditions rarely change linearly over geographic space (Gerst et al., 2011; Tang et al., 2012). Therefore, geographic distance may be a poor proxy for environmental similarity and whether a seed source will result in high plant establishment and persistence at a restoration site (e.g. Wright et al., 2018). Instead, it may be worthwhile to quantify the differences in climates between source and site locations (i.e. “environmental distance”) to predict a seed source's suitability to a restoration site. Not only are measures of environmental distance more direct than geographic distance, but they also have the utility of including directionality. For example, although two plant populations may be equidistant from the restoration site, the population from drier conditions than the restoration site may outperform plants from wetter conditions (Midolo & Wellstein, 2020). While tools designed to describe the climate space of a particular location are increasingly available (e.g. PRISM Climate Group, 2018), and using them has become a focus in recent local adaptation literature (reviewed in Lortie & Hierro, 2022) their utility in the context of restoration seed sourcing is largely unknown.

Finally, few studies have measured the effects of seed sourcing on plant populations under realistic restoration field settings (reviewed in Gibson et al., 2016). This is a problem, as the results from common gardens may not translate to the complex nature of restorations (reviewed in Lau et al., 2019). At a restoration site, there are many other factors influencing the establishment of target populations aside from the source of the seed, including the age of the planting (Applestein et al., 2018), the seeding rate (Barr et al., 2017; Maron & Simms, 1997), post-seeding management decisions (Dowhower et al., 2021; Flory, 2010; Maron & Jefferies, 2001), the soil conditions of the restoration site (Bassett et al., 2005; Haan et al., 2012), and others (including biotic factors such as herbivory; Applestein et al., 2018). Given the many factors affecting population establishment and persistence, alongside the mixed results from controlled common-garden studies, it is unclear how important seed sourcing decisions are relative to other factors that affect restoration outcomes.

To understand the impacts of seed sourcing decisions on the establishment and phenology of native species in realistic restoration settings, we developed a collaborative research project between university researchers, a native seed producer, and 22 land stewards. Using the data kept by this native seed producer on where they purchased seeds from and where they were planted, we located 24 recently installed tallgrass prairie restoration efforts across Michigan, which sourced seed from locations 6–750 km away from the restoration site. At each site, we surveyed plant establishment, abundance and phenology of five commonly planted tallgrass prairie restoration species. Using these data, we answered the following research questions: (1) Do measures of seed source locality (either geographic or climatic) explain the variation in plant establishment, abundance, and phenology? (2) How important are other factors, such as seeding rate, post-seeding management, and soil conditions at the site, for predicting the establishment, abundance, and phenology of restored plant populations relative to measures of seed source locality? We predict that if local seed sourcing is important for plant populations, we would observe reduced establishment and abundance, and greater changes in phenology at sites where less local seed is sown. We also expect that direct measures of the environmental distance between the seed source and the restoration site will be better predictors of our response variables than geographic distance. Finally, we expect environmental and management factors to influence plant populations across restoration sites, although we make no predictions as to their importance relative to seed source locality.

MATERIALS AND METHODS Sites

We selected 24 newly installed prairies from across Michigan, ranging from 0.0003 to 0.03 km² to survey from mid-July to mid-August of 2021 (Figure 1a). Sites were seeded between 2017 and 2020 and differed in the land use immediately prior to restoration initiation, planting methods, land management, and site preparation (Table S1). The amount of seed per species added to the site varied (0.002–3.36 g/m²) and half of the sites had no post-seeding management while the other half had some form of management (e.g. burning, mowing, weed removal or a combination of these). Since management approaches varied substantially between sites, and there were not many site replicates for each type of approach, post-seeding management actions were coarsely grouped into a yes or no variable. The seed mix planted at each site varied in geographic origin and each mix contained at least two of the five focal species in this study. All sites were either on public land or on private land surveyed with the owner's permission.

FIGURE 1. Map of Michigan indicating the locations of the 24 restored prairies where surveys took place (a) and the 18 source locations for the five prairie species surveyed (b). Not all species were obtained from each source nor planted at each site.

Study species

The seed mixes for the restoration sites in this study were supplied by Native Connections seed farm (Kalamazoo, MI, USA), a native seed company that produces some prairie species locally and acquires others from other Midwest seed producers for prairie restorations throughout Michigan. For a set of these restorations, they have detailed records of where seeds were obtained from, as well as where those seeds were planted. We only surveyed sites where a single source of each species was planted (excluding any sites where overseeding of our focal species was done). Using those records, we selected five focal species that were commonly sown into prairie restorations in Michigan and obtained from multiple source locations: Schizachyrium scoparium (sown at 20 sites), Ratibida pinnata (21 sites), Rudbeckia triloba (10 sites), Symphyotrichum laeve (14 sites) and Solidago rigida (21 sites). Overall, we sourced seed from five seed producers, nine different states and 18 different locations (Figure 1b). All seeds were produced on native seed farms, originally established with seed from wild populations. Since seeds were not always sourced from areas near the production farm (although they often were), we considered the source location of the seed to be the wild population where the seed lot originated from. While most seed producers could provide a single county or township where their seed lot was sourced from (12 seed lots), three seed lots were composed of seed from multiple counties and another three could only be described generally (e.g. “from the southern half of Michigan”). In these cases, the most centroid point (between counties, or the center of a state) was recorded as the source location.

Establishment surveys

To quantify species establishment and abundance, we established a 25 m transect in a random direction at the approximate centre of each site. If a site could not fit a 25 m transect, the transect was laid out along the longest portion of the site and rounded down to the nearest 5 m. Five 5 × 5 m subplots, or as many as could fit given the length of the transect, were marked on either side of the transect (up to 10 subplots total, per transect), and we counted the number of individuals of each species in each subplot to quantify plant abundance. To expedite field surveys, once 20 individuals of any focal species were found in a subplot, we recorded the distance along the transect where the 20th individual was recorded and counting for that species ceased; this smaller area then comprised our subsample of the subplot and was used to estimate the number of individuals of that species at the 25 m² scale. Since several of these species are clonal, we assumed that clusters of stems at least 15.24 cm (0.5 ft) away from other clusters of stems were separate individuals. Occasionally, one or several focal species were not present in any of the subplots. Then, we constructed another 10 × 25 m plot directly adjacent to the first; if a species was found in this plot, it was recorded as “present” in the site, but not included in abundance measurements. If it was not found, it was recorded as “absent”. We used this presence/absence data to quantify the likelihood that a species would establish at a site.

Flowering phenology

To gather data on differences in flowering phenology, we measured the abundance of flowering or seed-setting individuals (individuals with open flowers and/or dispersing seed) relative to the number of total individuals of each species in each 25 m² subplot. Since we conducted surveys once during the growing season due to time constraints, we assumed that sites with a greater proportion of flowering individuals during the survey had populations with an earlier flowering phenology (Lisi & Schindler, 2011).

Site environmental factors

We also collected data on the environmental conditions of the sites. We used soil water holding capacity to characterize site soils, as it is known to affect the abundance of sown prairie species (Grman et al., 2021; Zirbel & Brudvig, 2020). We collected 10 cm soil cores at 15 points along the entire 25 m² plot and pooled these together for each site. We dried soils in the lab and conducted soil water holding capacity measurements,  the proportionate difference between saturated wet weight and oven dry weight, following the same methods as Brudvig and Damschen (2011).

DATA ANALYSIS Source and site climates

All analyses were performed in R studio using R version 4.2.0 (R Core Team, 2022). We took a broad approach to defining the climate of our source and site locations, since plants may be adapted to a wide variety of correlated climatic variables ranging from annual temperatures (e.g. Baughman et al., 2019) to season-specific precipitation (Blumenthal et al., 2021). We generated 19 bioclimatic variables (Hijmans et al., 2005) from 800 m resolution PRISM (PRISM Climate Group, 2018) monthly climate data averaged across the years 2017–2021 (dismo package; version 1.3–5; Hijmans et al., 2021). These variables included temperature and precipitation metrics both annually and seasonally, which have been shown to influence the abundance of sown species in prairie restorations (Groves et al., 2020). Since many of these variables were highly correlated, we utilized a Principal Components Analysis (PCA) to summarize them (see Figure S1). The three highest loading axes explained 84% of the total variation (see Table S2 for the top nine loadings of each variable). We recorded the value along each PCA axis for each source and site location and subtracted the value for each source location from the value of each site where a given species was planted. This value was used in analyses to account for differences in climate between the source and site locations (named PC1 difference, PC2 difference, and PC3 difference).

Plant establishment and abundance

To test how seed sourcing and other management factors influenced the plant populations of our focal species, we focused on two metrics: whether a species established at a site (yes/no) and the number of individuals of that species per 25 m² area. We analyzed the likelihood of any given species establishing at a site using a generalized linear mixed model with a binomial distribution and a logit link with the following standardized predictor variables: geographic distance (km), PC1 difference, PC2 difference, PC3 difference, site age (years), seeding rate (oz/acre) for each species, whether a site was managed, and soil water holding capacity (%). Species identity was used as a random factor. Of all the variables used in analyses, only geographic distance and PC1 difference were marginally correlated (Pearson's R = 0.67; Figure S2). However, the VIF's of the model including all variables were < 2 indicating multicollinearity was low, so all variables were included. Rather than choosing the best-fitting model with only one or two predictors, excluding less powerful predictors despite their biologically relevant contributions, we used a model averaging approach using all subsetted model combinations (128 total models; Hoeting et al., 1999; Hooten & Hobbs, 2015). We estimated regression coefficients of each predictor by averaging the coefficients of each subset model weighted by each model's AIC (MuMIn package; version 1.4.6; Bartoń, 2022). Since each variable was used in the same number of models, the relative importance of each variable was quantified through adding the AIC weight of each variable in all model combinations (the sum of weights [SW]). Since there is no cut-off value for SW to indicate significance (Galipaud et al., 2014), we rank the relative importance of each variable by its SW (Burnham & Anderson, 2003). Using the best-fitting model (including any biologically relevant covariates), we visualized the conditional effects of the two most important variables (ggeffects package; Lüdecke, 2018) to understand their effect on each response variable. As the model averaging approach intentionally selects higher-fitting models, no p-values are reported in this manuscript.

The average (or estimated) number of individuals was averaged across every subplot at a given site, to provide a single value for the average individuals in a 25 m² area at each site. The count data were overdispersed but not zero-inflated (ratio of expected to observed zeroes 1.01:1, p = 1), so we analyzed this variable with a negative binomial generalized mixed model (glmmTMB package version 1.1.3; Brooks et al., 2017) using the same predictors and model averaging strategy as above (128 total models; all other model assumptions were met). Additionally, we wanted to test the sensitivity of our choice to use a centroid point as the source location for the seed lots we could not track to a single county (n = 9), as well our inclusion of sources that were composed of seed from multiple counties (n = 3). We tested the former by creating new source coordinate variables for those datapoints: source coordinates that are the furthest away from the planted site, and coordinates that are the closest to the planted site, given the collection region provided by the seed producer. We tested the latter by removing those composite points and re-running our analyses. In both cases, we found no qualitative changes in the relative importance of variables for the likelihood of establishment nor average count analyses, but there were changes in flowering phenology (Figures S3 and S4). Therefore, we used the centroid point as the estimated source location in further analyses, as we think this minimizes the possible incorrectness of the source location we chose, but present the results of these alternative analyses for transparency. We also chose to retain the three composite sources in our analysis, as all of our source populations could have differing levels of genetic diversity caused by a multitude of factors including the size of the original source population (Newman & Pilson, 1997), cultivation practices (Dyer et al., 2016), years a source has been cultivated (Pizza et al., 2021), and many others. Since we lack information on these factors from all the sources, and make no attempts to quantify the effect of genetic diversity on our results, we do not feel excluding these datapoints is appropriate, but do present the results of removing these points in the supplementary materials for transparency.

Flowering phenology

To test if seed sourcing, or other management decisions, affected flowering phenology across the sites, we considered the number of flowering individuals relative to the total number of individuals of each species in each 25 m² subplot. Since we conducted surveys once during the growing season due to time constraints, we assumed that sites with a greater proportion of flowering individuals during the survey had populations with an earlier flowering phenology (Lisi & Schindler, 2011) while accounting for differences in site latitude, survey date, and site age. Since it is unknown whether vegetative individuals would flower that year or not, we created two flowering variables: one including those vegetative individuals in the total, and one not including vegetative individuals. Both were fit with a negative binomial mixed model, which was the only model of three (untransformed linear, square root transformed linear, and negative binomial) that met model assumptions given the high number of zeroes in the dataset. This model included the same predictors as the establishment models (geographic distance (km), PC1 difference, PC2 difference, PC3 difference, site age (yesrs), seeding rate (oz/acre) for each species, whether a site was managed, and soil water holding capacity) and additionally included restoration site latitude and the Julian survey date as covariates, the former to control for further north sites flowering later than further south sites and the latter to control for sites surveyed later in the year being further along phenologically (see Table S2 for regression coefficients for these covariates). Since there was no qualitative difference in the statistical output of our two models, we report only the first flowering variable (including vegetative individuals in the total) but report the results of our alternative analysis in the supplemental materials (Figure S5). All predictor regression coefficient values were averaged across 512 subset models.

RESULTS Climate data

The first PC axis explained 54% of the variation in our climate variables and was negatively associated with the mean temperature of the driest month, the minimum temperature of the coldest month and the temperature annual range (Figure S1; see Table S2 for loadings of the top nine variables for each axis). We interpreted higher positive PC1 differences to represent when seeds were sourced from environments with warmer and wetter winters than the site. PC axis 2 explained 18% of the variation in our climate variables and had negative associations with the mean temperature of the warmest quarter, maximum temperature of the warmest month, and precipitation in the wettest quarter. Higher positive PC2 differences represent when seeds were sourced from environments with warmer and wetter summers than the site. Finally, PC axis 3 explained 14% of the variation in our climate variables, with higher positive PC3 differences representing when seeds were sourced from hotter and drier environments than the site. Geographic distance was not significantly correlated with any of the climate variables (VIF < 2; Figure S1).

Plant establishment and abundance

The two most important predictors for whether a species would establish at a site were whether the site was managed post-seeding and the seeding rate of that species (Figure 2). Sites under any form of post-seeding management were 36% more likely to have any given species present than unmanaged sites (Figure 3; β = 1.01 ± 0.65, sw = 0.85). Sites with higher seeding rates also experienced more reliable establishment than sites with lower seeding rates (Figure 4; β = 0.39 ± 0.36, sw = 0.66). The most important predictor for plant abundance was seeding rate: adding 3× more seed than average to a site increased plant abundance by 10 individuals per 25 m² (Figure 5; β = 0.22 ± 0.12, sw = 0.72). Importantly, all predictors of seed source locality (geographic and environmental) had low importance in our models for plant establishment and abundance (Figure 2; Table S3). Site-specific factors, including soil water holding capacity and site age were also not important predictors (Figure 2; Table S3).

FIGURE 2. Standardized regression coefficients (± standard error) estimated through model averaging for three response variables to quantify the effects of seeding restorations with seed sourced from various degrees of locality. See Table S2 for all model statistics and sum of weights (SW).

FIGURE 3. Conditional effects of management on the likelihood a given species would arrive at a site, taking into account different seeding rates (n = 84). Bars indicate mean values calculated in emmeans and error bars show standard error. Estimated means were calculated using the best fit model in the model averaging output.

FIGURE 4. Conditional effects of seeding rate (scaled [mean/standard deviation] for easier visualization) on the likelihood, a given species would arrive at a site, taking into account different management regimes (n = 84). Dark line indicates regression of two variables, and shaded area shows standard error of the regression line. Regressions were created using the best fit model in the model averaging output.

FIGURE 5. Conditional effects of increased seed addition (scaled [mean/standard deviation] for easier visualization) on the number of individuals of a given species in a 25 m2 area, while accounting for differences in management (n = 84). Dark line indicates regression of two variables, and shaded area shows standard error of the regression line. Regressions were created using the best fit model in the model averaging output.

Plant phenology

The two most important predictors of plant flowering phenology were PC3 difference and geographic distance (Figure 2). There appears to be no effect on flowering phenology when sourcing seeds from warmer and drier locations than the restoration site (when PC3 difference <0), but there were more flowering individuals when seeds were sourced from cooler and wetter locations than the restoration site (when PC3 difference >0; Figure 6; β = 0.57 ± 0.47, sw = 0.73). Restoration sites sown with seed from locations geographically far from the restoration site tended to have more advanced flowering phenology (β = 0.65 ± 0.55, sw = 0.71), although this relationship is not linear (Figure 7). Other analyzed factors including soil water holding capacity, site age, management, and seeding rate were not important predictors (Figure 2; Table S3).

FIGURE 6. Conditional effects of increasing environmental distance (PC3) on the percentage of individuals of a given species at a site that are flowering taking into account any differences attributed to geographic distance, survey date, and site latitude (n = 84). PC3 difference values [less than]0 indicate when seed was sown from locations cooler and wetter than the site, and values >0 indicate when seed was sourced from locations warmer and drier than the site. Dark line indicates regression of two variables, and shaded area shows standard error of the regression line. Regressions were created using the best fit model in the model averaging output, with the addition of relevant covariates.

FIGURE 7. Conditional effects of increasing geographic distance on the percentage of individuals of a given species at a site that are flowering, taking into account any differences attributed to differences in climate, survey date, and site latitude (n = 84). Dark line indicates regression of two variables, and shaded area shows standard error of the regression line. Regressions were created using the best fit model in the model averaging output, with the addition of relevant covariates.

DISCUSSION

Although it is widely believed that local seed sources will result in the highest initial establishment at restoration plantings (reviewed in McKay et al., 2005; but see Hancock et al., 2022), our results do not substantiate that claim. No metrics of seed source locality were good predictors of plant establishment or abundance at our newly installed restoration sites. Instead, seeding rate and post-seeding management were the only measured factors that reliably increased plant establishment and abundance. However, plants from less local seed sources did flower earlier than local seed sources, which could affect population dynamics and biotic interactions at restoration sites.

Plant establishment & abundance

A key finding in our study was that neither the geographic distance nor climate similarity between seed sources and restoration plantings predicted initial plant establishment or abundance. Our results add to a small, but growing, body of evidence that the importance of seed source locality for plant establishment in restoration efforts is outweighed by other factors. Research on the early establishment of four Andropogon gerardii ecotypes in a multispecies community study showed no difference in percent cover for that species (a proxy for plant abundance) during the first 2 years of establishment (Galliart et al., 2019), although differences did emerge after that time. Another study did observe differences in establishment success between different ecotypes of four plant species, but these differences were not explained by geographic or environmental distance (Bischoff et al., 2010). Importantly, both studies were done under natural field conditions and are most comparable with our study design, emphasizing that although seed sourcing may have a measurable impact on plant establishment in less natural conditions (e.g. Montalvo & Ellstrand, 2000) its impact may disappear when establishment is influenced by multiple other factors. Additionally, in our study differences in the environments between each source and the sites at which it was planted did not appear to be geographically structured, indicating that geographic distance is a poor proxy for climate similarity. This could be explained by seed lots adapting to the conditions on the farm, especially those that were grown on the farm for several generations (Pizza et al., 2021) or if those conditions differ substantially from the source environment. However, since the number of generations each seed lot was grown on the farm was not available, we are unable to test that hypothesis. It is also important to acknowledge that our metrics of locality may have been too coarse to capture environmental variation that causes local adaptation (which can occur at very fine scales, Knight & Miller, 2004). Although other measures of environmental similarity such as soil composition (Macel et al., 2007) may have captured that variation, they may not be as readily available to restoration practitioners or native seed producers. Given currently available metrics, our results support that selecting the nearest seed source does not seem to impact sown species establishment. Instead, it provides evidence that there are likely multiple suitable sources of restoration seed available for each project, and that using seed sources <700 km away from the restoration site is likely to reduce the probability of introducing maladapted seed, as no differences in establishment or abundance were seen in that range in our study.

The most important predictors for whether a plant would establish or not were post-seeding management and seeding rate. Since few sites performed identical management strategies, our management variable encompassed any form of post-seeding intervention to promote sown species establishment. However, all managed sites experienced some amount of mowing. This is consistent with other literature, showing that mowing can increase the abundance of sown forbs by reducing shade competition from larger established grasses (e.g. Dowhower et al., 2021; Maron & Jefferies, 2001; Williams et al., 2007). This was further supported in our study, as unmanaged sites were 36% less likely to have our studied species establish than managed sites. Although the importance of mowing restoration sites is widely understood by restoration practitioners as a method to increase the density of sown forbs (Rowe, 2010), half of our sites were unmanaged. This was often due to financial constraints which prevented land stewards from monitoring the site for management needs (R. Pizza pers. obs.; Barak et al., 2021; Phillips-Mao et al., 2015). Thus, the positive effects of any form of management on plant establishment demonstrated in our study can be used by land stewards to justify funding these post-seeding interventions, especially mowing.

The other factor that increased the likelihood of establishment is increased seeding rate: mirroring other studies in prairie systems (Applestein et al., 2018; Barr et al., 2017). Although the increase in establishment (~15%), and abundance (~10 individuals per 25 m²) after tripling the amount of seed sown observed in our study may seem small, it could have long-term demographic consequences: populations with more individuals tend to have greater genetic diversity (Ellegren & Galtier, 2016), and are less likely to go extinct (Newman & Pilson, 1997; Purvis et al., 2000). Thus, increasing the number of individuals present in the early establishment stage may increase the likelihood that a population will persist into the future. Higher seeding rates can also decrease the establishment of nuisance species at the restoration site (Pyke et al., 2013). Therefore, although seeds can be the most expensive part of a restoration project (Phillips-Mao et al., 2015), sowing seeds at greater densities may reduce the necessity of costly invasive species removal later in the restoration process.

Importantly, while our data do not show an overall trend of local seed sources establishing best, the relative importance of seed sourcing likely varies between species due to the range of environmental conditions it can inhabit (e.g. Macel et al., 2007). Species that have populations in a wide range of climates (e.g. Andropogon gerardii) may have more differentiated phenotypes than species that persist in a narrower range of climates (Galliart et al., 2019). Additionally, in more extreme environments such as the arid Great Basin, local adaptation may be more important for plant establishment (Baughman et al., 2019) than in less climatically extreme environments like the Midwest (reviewed in Hereford, 2009). It is also possible that planting year weather may be more important for plant establishment than overall climate averages (e.g. Groves et al., 2020). Finally, our study only focused on the initial establishment stage, so it is possible that seed sourcing may impact restored populations at later successional stages (e.g. Galliart et al., 2019). Since the biotic context of a site changes as the community develops, traits that allow persistence during early establishment may not confer persistence at later successional stages. Future studies focusing on the importance of seed sourcing for individual species, especially those commonly used in restoration, and at later successional stages, should be conducted to understand the species-specific and long-term consequences of seed sourcing decisions.

Plant phenology

An important finding in this study was that less local seed sources, both geographically and environmentally, flowered earlier than more local sources. This was also the only analysis where a measure of locality was an important predictor, indicating that even coarse measures of locality may account for changes in phenology. This is likely caused by flowering phenology being strongly determined by temperature (Hülber et al., 2010) and latitude (Debieu et al., 2013; Rushing et al., 2021) whereas establishment is influenced by many other factors (e.g. Zirbel & Brudvig, 2020). Our results contrasted with previous studies on the impacts of seed sourcing decisions on plant phenology (Bucharova et al., 2022; Selbo & Snow, 2005). Selbo and Snow (2005) observed no difference in flowering phenology between ecotypes across a large geographic gradient in an experimental setup that more closely resembles ours, whereas Bucharova et al. (2022) observed significant earlier phenology, but in ecotypes <400 km away from one another when plants were grown in pots in a common garden. Given these conflicting results, we conclude that seed sourcing decisions can impact plant flowering phenology, but the mechanisms behind that shift are not consistent. Importantly, this analysis appears to be sensitive to changes in individual datapoints (see Figures S4 and S5). Since we did not observe this sensitivity in the establishment or abundance analyses, future studies including a larger number of sites to survey should be done to confirm the trends observed in this study. Future research to understand the mechanisms behind these shifts in flowering phenology may help practitioners understand when using less local seed sources could affect biotic interactions at a restoration site.

In our study, the only measure of environmental distance that had a positive relationship with flowering plant abundance (PC3; sourcing plants from cooler and wetter locations and planting them in warmer and drier ones) is not correlated with geographic distance. This further solidifies that geographic distance may be a poor proxy for climate similarity, and we suggest considering the difference in temperature and precipitation, especially during the winter and summer season (the primary variables that categorize this PC axis) to predict whether plants from sources further away will have a different flowering time than more locally sourced individuals. While plants from cooler and wetter environments may be adapted to shorter growing seasons, and therefore flower earlier than local sources (Haggerty & Galloway, 2011; but see Bradley St. Clair et al., 2013), the mechanism explaining the relationship between geographic distance and flowering time is unclear. Although there is evidence that flowering phenology can be driven by differences in temperatures at different latitudes (Debieu et al., 2013) and precipitation at different longitudes (Samis et al., 2012) neither source latitude or longitude were good predictors of flowering phenology (Figure S3). Another hypothesis is that since most seeds sown to the sites in our study originated from native seed production farms, cultivation practices including supplemental watering or early harvesting times could have unintentionally selected for early flowering plants (e.g. Dyer et al., 2016). Future work comparing restorations sown with seed from different producers in similar climates could parse out this relationship.

This shift in phenology may indicate that there could be a mismatch between plant pollen resources and pollinator abundance if pollinators emerge later in the season, which could affect both the plant populations through decreased reproduction, and the pollinator communities through decreased floral resources. Additionally, since earlier flowering phenology can be related to earlier phenology in other developmental stages (such as green-up time; Delbart et al., 2015), less local seed sources could have altered relationships with competitor plants, altering community composition at these sites (Wilsey et al., 2011). Since our study does not measure the duration of flowering, early flowering plants may have continued flowering throughout the season, providing even more resources to pollinators (e.g. Bucharova et al., 2022). Finally, since we did not measure seed set, it is unclear if, or how, an earlier flowering time can affect plant reproductive potential, or consequences for the timing and amount of fruits and seeds available to consumers. Given these questions, the mixed results of studies like this in the past, and the sensitivity of our analysis to individual datapoints, our results point to the need for further research on how seed sourcing decisions affect plant phenology.

Importance of co-designed research projects

This study exemplifies the utility of co-designed research projects, both for data availability and research potential. Due to the increased interest in using local genotypes for restoration, Native Connections seed farm had kept detailed records about where their seed for all species was procured from, and where it was planted, for the last 5 years. Without these detailed records, it would be difficult to parse where seed for each project was sourced from, making a large-scale project like this either impossible, or incredibly costly. Importantly, Native Connections was not the only seed producer with detailed records: despite no seed certification programs in the United States requiring that native seed producers report the source of the seeds they sell (Pedrini & Dixon, 2020), five of the six native seed producers could confirm the county that their seed was collected from for some (if not all) of their species, and all could report at least what state the seed came from. We suspect that records like these are available more broadly, both geographically and across ecosystems. If so, utilizing them can generate countless more sites to use to further understand the implications of seed sourcing decisions in different ecosystems across the world. Given the importance of their contributions, seed producers' participation in these co-designed research projects should be formally recognized.

CONCLUSIONS

Our results suggest that seed source locality, at the geographic and environmental scales measured, does not predict whether plants will establish, nor in what abundance, at restoration sites. This contributes to growing evidence that, under realistic restoration settings, local seed sources do not always establish better than less local sources. The results of this research may help expand the region(s) which practitioners might consider local, and further emphasize the importance of pre- and post-seeding management on ensuring population establishment. Future studies should test these same ideas at restoration sites in a larger geographic range, and in systems other than tallgrass prairies, to see if these trends translate to other geographic areas and ecosystems. Although much research still needs to be done, this project exemplifies the importance of collaborative research and challenging paradigms.

AUTHOR CONTRIBUTIONS

Riley B. Pizza and Lars A. Brudvig conceived the ideas; Riley B. Pizza, Lars A. Brudvig, and Jared Foster designed the methodology. Riley B. Pizza and Jared Foster collected the data; Riley B. Pizza analyzed the data and led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

ACKNOWLEDGEMENTS

The authors thank C. Catano for invaluable help analyzing this data. We also thank members of the Brudvig lab, D. Lowry, M. Weber and P. Zarnetske for feedback on this manuscript. Finally, we thank the 22 landowners and stewards, including Consumers Energy, who provided crucial knowledge on site conditions, and showed a commitment to restoration. Funding for this project was supported by the Kellogg Biological Station Graduate Research Fellowship and L.A.B. was supported by NSF DEB-1552197. This is Kellogg Biological Station Contribution #2341.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to disclose.

PEER REVIEW

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/2688-8319.12223.

DATA AVAILABILITY STATEMENT

Data and code used for analysis for this manuscript is openly available at the Dryad Digital Repository https://doi.org/10.5061/dryad.n5tb2rc0j (Pizza et al., 2023).