Rethinking local seed sourcing for the restoration of a foundational grass species in California

Restricting seed collecting to local populations is a common practice in restoration because it is assumed that most plants are adapted to local environmental conditions. However, there is still considerable debate about whether local seed collection should be the default seed‐provenancing strategy as the effects of climate change are increasingly considered in restoration planning. It is especially important to explore whether local seed provenance is necessary for dominant species used in restoration because the success of these projects often rests on these species. Stipa pulchra is one such species that is commonly used in grassland restoration projects in California. To explore how different seed‐provenancing strategies affected the establishment and growth of S. pulchra, we established three common garden experiments distributed across a latitudinal gradient in California. We collected seeds from seven wild populations, germinated seeds in a common greenhouse, then planted all populations at the three common gardens. We assessed growth and reproduction for 2 years. We found limited evidence that restricting seed sourcing to local populations increased the establishment of S. pulchra compared to seeds from more distant populations. Instead, we found evidence to support the use of regional admixture seed sourcing to increase resiliency to environmental variation. In particular, we found being sourced from a dry location during the dry 2018 season was a benefit, highlighting the importance of including dry‐adapted populations to increase climate resilience. Our experiment highlights the importance of exploring multiple seed‐provenancing methods when designing a project to improve long‐term success.


Introduction
The success of most habitat restoration efforts depends on access to high-quality seeds to ensure plant establishment and long-term persistence.In restoration, the use of local seeds has traditionally been advocated based on the assumption that local populations are adapted to local environmental conditions (Broadhurst et al. 2008;Bucharova et al. 2019;Walters et al. 2022).Local adaptation arises over generations as natural selection, in response to local environmental conditions, selects for genotypes that perform best in each environment.In the absence of gene flow (the immigration of seeds or pollen from populations potentially adapted to different conditions) or genetic drift, this process results in resident populations that exhibit a "home-site advantage," whereby locally adapted genotypes perform better than those sampled from populations adapted to different conditions (Kawecki & Ebert 2004;Savolainen et al. 2013;Hodgins & Moore 2016).In recent years provenancing for restoration projects has begun to turn toward alternative methods (e.g.predictive provenancing, climate-adjusted provenancing, and admixture provenancing) that aim to specifically incorporate concerns about climate change (Bucharova et al. 2019;Dupré La Tour et al. 2020;Woolridge et al. 2023), however, local provenancing continues to be used by many practitioners in the industry.
The practice of using local seeds ("local is best") has been common in restoration for decades (Herget et al. 2015;Bucharova et al. 2019;Gann et al. 2019;St. Clair et al. 2020) based on evidence that suggested local adaptation was present in many plant species (Linhart & Grant 1996;Leimu & Fischer 2008;Beierkuhnlein et al. 2011) and may impact plant fitness and restoration (Gordon & Rice 1998;Joshi et al. 2001;Hereford 2009).For example a meta-analysis evaluating local adaptation in 32 species (with 1,032 pairwise comparisons between local and non-local populations) found that local populations outperformed foreign populations in 71% of pairwise comparisons (Leimu & Fischer 2008).Local adaptation, however, has still not been confirmed for most species because it is time-and labor-intensive to detect it (McKay et al. 2005;Havens et al. 2015).Nonetheless, many advocates of this approach argue for using local seeds, because it is assumed that the pros of restricting to local material outweigh the potential cons including the introduction of poorly adapted genotypes, accelerating the loss of well-adapted local genotypes, and increasing the chance of local extirpation (Lesica & Allendorf 1999;Sackville Hamilton 2001;Hufford & Mazer 2003;Hancock et al. 2013).In practice, however, it is difficult to define what "local" means for a given species even when local adaptation has been unambiguously detected in one or more of its populations (Kawecki & Ebert 2004;Dupré La Tour et al. 2020).
There are two common methods that are often employed when practitioners are focused on using seeds adapted to local conditions for a restoration project.First, seeds, seedlings, or vegetative material for propagation may be collected from extant populations near the restoration site (Hancock et al. 2013;McKone & Hern andez 2021).Alternatively, seeds may be collected from habitats that are climatically or ecologically similar to that of the restoration site, regardless of their distance from the restoration site (McKay et al. 2005;Havens et al. 2015).Sourcing material from close to the restoration site is easier to implement, but there are significant problems associated with this approach.First, there are no clear guidelines regarding how close a population must be to be considered "local" (Kawecki & Ebert 2004;Dupré La Tour et al. 2020).This has resulted in practitioners having widely different interpretations concerning how far to source seeds for specific projects (Havens et al. 2015).Second, it is not clear that distance is useful for determining whether a population will be locally adapted (Dupré La Tour et al. 2020).In fact, a meta-analysis found that the distance between two sites was a poor predictor of whether plant material sampled from one site will perform well in another (Leimu & Fischer 2008).Sourcing plant material from a site that is climatically similar to a prospective restoration site, on the other hand, is more likely to identify genotypes that will perform well at that location (Lesica & Allendorf 1999;Havens et al. 2015;Doherty et al. 2017).This method, however, is much more complex than using geographic distance because the site and climate variables driving adaptation can differ between species (Leimu & Fischer 2008).
Restricting plant material to locally adapted material has been criticized, however, because not all species or populations exhibit local adaptation (McKay et al. 2005).Moreover, plants may be adapted to historic environmental conditions and lack the genetic variation to respond to a rapidly changing climate (Lesica & Allendorf 1999;Sgro et al. 2011;Bucharova et al. 2019;Woolridge et al. 2023).Another problem with using only local seed sources in restoration efforts is that most restoration sites are located within fragmented habitats, and the populations that exist within these landscapes are smaller and potentially more genetically inbred than historic populations (Leimu et al. 2006;Breed et al. 2013) with less additive genetic variation (Young et al. 1996;Hughes et al. 2008), both of which can reduce evolutionary potential and the ability of the population to respond to environmental change.Finally, some have pointed out that a focus on locally sourced plant material could lead to overharvesting seeds in small populations (Mortlock 2000).These issuesvariation in local adaptation, problems with guidelines for "local seeds," and detrimental impacts on local populations-suggest that a more nuanced approach to seed provenancing is needed.
Here, we describe an experiment in which we grew Stipa pulchra, a native perennial grass commonly used in grassland restoration efforts, from seven geographically distinct populations representing a range of climates in southern and central California in three climatically distinct common gardens.Grasslands in California are often targeted for restoration and management (Stromberg et al. 2007) because they have been extensively invaded by exotic annual grasses and forbs during a period of acute and chronic anthropogenic disturbances; this invasion drastically altered the species composition and structure of these communities (Stromberg & Griffin 1996;Stromberg et al. 2001;D'Antonio & Meyerson 2002).Although the exact composition of pre-European grasslands is unknown (Schiffman 2007), communities were thought to be once dominated by the long-lived perennial grass S. pulchra (Bartolome et al. 1986;Heady et al. 1988;Hamilton 1997).In addition, S. pulchra is often targeted for restoration because it provides the structural foundation for native perennial grasslands (Stromberg et al. 2007;Molinari & D'Antonio 2014).We focused on climate metrics related to aridity because precipitation is expected to be the largest driver of change in California grasslands in the future (Carter & Blair 2012) with extreme droughts predicted to increase in frequency.While grasses such as S. pulchra are adapted to seasonal droughts (Ehleringer & Mooney 1983;Vaughn et al. 2011) there are substantial differences between water availability at the different seed source locations (Table 1), which could lead to population-level differences in drought tolerance, growth, and establishment success.Our goal was to use S. pulchra to demonstrate how pilot studies can be used to create tailored seed-sourcing strategies and, more broadly, to inform seed collection for other species that are geographically widespread and commonly incorporated into restoration.
To accomplish this goal, we addressed the following questions.( 1) Do local populations consistently outperform non-local populations?(2) Do individuals sourced from populations closer to the common garden perform better than those sourced from further away (i.e. can geographic proximity be used to identify populations that are relatively well-adapted to a given location)?And, (3) does the climate at the site where seeds were collected impact how well they perform at a given common garden (i.e. can climate matching be used to identify local populations)?

Methods
Focal Species: Stipa pulchra In California grasslands the perennial, wind-pollinated grass, S. pulchra is a dominant native species commonly planted in grassland restoration projects.The range of S. pulchra can be found across California from San Diego to Redding with seeds primarily dispersed via accidental and purposed movement by animals.It has been shown that traits of this species are influenced by climatic conditions, with plants sourced from wetter, northern California having taller more erect growth forms and producing fewer, heavier seeds than plants from drier sites in central and southern California (Knapp & Rice 1998).Studies have also found that there are population-level differences in genetic variation even in close proximity (Dyer & Rice 1997).However, there have been few common garden experiments conducted to determine whether these differences represent local adaptation and are important within a restoration context (i.e.adaptation to the local environment within a species results in non-local populations failing to establish at a restoration site).While there has been a reciprocal transplant study on S. pulchra that did find evidence for home-site advantage (Hufford & Mazer 2012), the study focused on only two relatively close populations in Santa Barbara County.Since S. pulchra is a widespread species and can be found across considerable topographic and climatic variation, it is not clear whether local adaptation is widespread throughout the species or just present within some limited areas.

Seed Sources
The seed was collected from seven S. pulchra populations distributed across a latitudinal gradient from the UC Santa Cruz Arboretum to the Santa Rosa Plateau Ecological Reserve (Fig. 1).The seven populations were chosen because they were distributed across central and southern California, and differed in climate variables (Table 1) and were located at sites that had not been restored (i.e.only local genotypes were present at collection sites).In 2017 between April and June, ripe seeds were collected from 25 to 30 individuals at each site.Plants were randomly selected within an area to represent the entire population and were at least 1 m apart.All selected plants appeared healthy and bore a minimum of five culms with ripe seeds.
Once collected, seeds were germinated and grown for all three common garden experiments at the Biology Greenhouse facility at the University of California Santa Barbara.Plants were germinated in a custom potting soil mixture made from widely available resources (4-parts Sun Gro Horticulture's Sunshine Mix #4: 1-part Coco coir: 1-part plaster washed sand) in  round plug flat trays that were 5.9 cm deep Â 4.8 cm wide and held 110 cm 3 of soil.The greenhouse was kept between 18 and 24 C with a relative humidity of 80%.Plants were grown for 60 days after germination.While in the greenhouse, they were watered three times a week with tap water.After 60 days, they were moved outdoors to harden for 2-3 weeks before being planted into the field sites.

Study Sites
We established three common gardens located throughout California: a northern site in Santa Cruz County (Santa Cruz garden), a central site (Sedgwick Reserve garden) near Los Olivos in Santa Barbara County, and a southern site near Calabasas (Stunt Ranch garden) in Los Angeles County (Table S1; Fig. 1).All sites are typical of California's Mediterranean climate but experience varying degrees of aridity (Table 1).The growing season occurs during the winter and spring months, with most precipitation occurring between October and April.To quantify the climate differences between the sites, all climatic parameters were downloaded from the Prism Climate Group (http://www.prism.oregonstate.edu/;23 January 2020).A growing season is defined as October-June and the summer is defined as July-September; soil classifications were obtained from USDA Natural Resource Conservation Service (https://websoilsurvey.sc.egov.usda.gov/).All climate variables described are 30-year normals (averaged between 1981 and 2010) unless otherwise noted.

Experimental Design
To determine whether and how seed origin affected the establishment, growth, and reproduction of S. pulchra populations cultivated under field conditions, we planted seedlings sourced from each of the seven S. pulchra populations in each common garden.Before planting we cleared all standing biomass and tilled soil with a rototiller to facilitate planting.To restrict plant interactions between individuals from the same populations, in January 2018 we planted 25 seedlings from each population in small subgroups (four replicate subgroups of each population per site) for a total of 100 individuals from each seed source.
Seedlings were planted by hand in holes that were dug to a depth of 15 cm using a hand trowel.The subgroups were randomly placed in a 7 Â 4 m grid with 0.5 m between subgroups to minimize root interaction between populations, but individuals were analyzed individually.All seedlings were hand planted in a single day and watered the day of planting.Plants were watered with 7.5 L of water a second time 1 week after planting.After this initial watering, plants were given additional water during the first growing season if more than 4 weeks elapsed between rain events.The plants did not receive supplemental water after the first year.To minimize competition from exotic species, plots were weeded periodically.This experimental design was replicated in each of the common gardens.

Data Collection
In early June 2018 and 2019, we collected data on the survival, plant size, and reproductive output of individual plants.We measured the tussock basal circumference of each individual as a proxy for biomass in both years.Reproductive success was measured as reproductive output, which is the total seed weight produced per individual.This was calculated using the following equation: where R y is the reproductive yield, C is the total number of culms per plant, S is the mean number of seeds produced per culm, and W is the mean weight of a single seed.Individual seeds were measured to 0.01 mg.The number of culms was counted for each individual in 2018 and 2019.The mean number of seeds produced per individual was estimated by counting the number of seeds produced on three randomly selected culms.The mean seed weight for a plant was estimated from 10 randomly selected seeds.Unfortunately, many individuals died between the surveys in 2018 and 2019 (Table 2) due to herbivory by gophers, which forced us to alter how we estimated reproductive effort.In 2018, S and W were estimated for 12 individuals from each population at each common garden and then averaged for one S and W per population per common garden.In 2019, S and W were estimated for every individual, due to the reduced number of individuals, to calculate the average for each population per common garden.The individuals we harvested culms from were undisturbed by gophers and there was no evidence of culm loss due to herbivory.Rethinking seed sourcing for grassland restoration

Analysis
All statistical modeling and data manipulation were done in R (R v. 3.5.1,R Core Team, Vienna, Austria).To assess whether local populations performed better than non-local populations, we compared the basal circumference and reproductive output of the local population with each non-local population using Welch's t-test.We defined the local population as the seed source that was collected at the common garden location.We did not statistically compare how survival differed between local and non-local populations because we only had one survival rate for each population at each common garden.
To assess how distance from the common garden location affected S. pulchra performance, we estimated the linear distance of each seed source location from the three common garden locations using Google Earth (https://earth.google.com/web/).We then used linear regressions to evaluate relationships between the distance between the seed source and the garden (from here on referred to as distance) and the growth, reproductive output, and survival of each population at each common garden.Data from each of the common gardens were analyzed separately.We analyzed survival data only from the first year (2018) since the most plant mortality during the second year (2019) was from gopher activity and not related to the climate.We do, however, present the data for reference.
To assess how the climate from where seeds were collected influenced how well individuals performed in the common gardens, we collected long-term data (30-year normals between 1981 and 2010) on 10 different climate variables for each seed source location: maximum vapor pressure deficit (VPD) in January, March, and May, monthly precipitation totals for January-May, total annual precipitation, and the number of rain events during the growing season.All climate data was downloaded from the Prism Climate Group (http://www.prism.oregonstate.edu/;23 January 2020).Maximum VPD was used as a proxy for how dry the site was during the beginning (January), middle (March), and end (May) of the growing season.Monthly precipitation totals allowed us to explore how variation in rainfall during the growing season impacted individual performance.Total annual precipitation, on the other hand, encompasses the total amount of rainfall at each site that was available to the plants over the entire growing season.The number of rain events represents how that rainfall was distributed throughout the year, with a smaller number of rain events equating to fewer events with more dry days in between events.A rain event was defined as a continuous period of precipitation (<24 hours between precipitation totals) with at least 0.1 cm of rainfall.In addition to compiling data on individual climate variables, we also calculated an aridity index for each site using BioClim variables (Fick & Hijmans 2017) following Welles and Funk (2021), which is based on mean and maximum temperature, standard deviation of temperature, precipitation of the wettest month, and coefficient of variation for precipitation (Harouna & Carlson 1994).
We used linear regression to evaluate relationships between mean response metrics (basal circumference, reproductive output, and survival rate) for each seed source location and the long-term climate variables and aridity index from the corresponding seed source location.Each climate variable and the three common gardens were explored separately.Assumptions for linearity of the data, normality of the residuals, homoscedasticity, and independence of residuals for each linear regression were tested using the plot function in the base R package.All models were run using the lm function in the base package of R. Finally, within each common garden location, we used Akaike Information Criterion (AIC) to determine which climate variable best predicted each response metric (basal circumference, reproductive output, and survival).

Do Local Populations Outperform Non-local Populations?
In year 1 (2018), in all three common gardens, plants representing the local population had significantly larger mean basal circumference than some, but not all, of the other seed source populations (Table S2; Figs.2-4).However, in 2019 this effect was almost entirely lost with only the local population in the Stunt Ranch common garden growing larger than the Vandenberg seed source (Fig. 2B).Local populations also exhibited greater reproductive effort than some, but not all, of the other seed sources in the Sedgwick (Fig. 3C & 3D) and Santa Cruz (Fig. 4C & 4D) common gardens in 2018 and 2019.At Stunt Ranch, however, the local population reproduced less than some of the other seed sources in 2018 (Fig. 2A) and there were no differences among populations with respect to reproductive output in 2019 at this site (Fig. 2D).

Do Seeds Sourced From Locations Closer to Common Gardens Perform Better?
In 2018, among populations observed at Stunt Ranch, there was a negative relationship between survival and population distance from the common garden (Fig. 5G; F [5,7] = 7.69, p = 0.04, r 2 = 0.53), but there was no significant relationship between these variables in the other two gardens (Fig. 5H &  5I).The distance between the seed source location and the common gardens did not have a significant effect on the growth of Stipa pulchra in 2018 or 2019 in any common garden (Fig. 5A-C).Reproductive output was not correlated with distance in any common garden in 2018.However, there was a significant negative relationship between reproductive output and population distance in the Stunt Ranch common garden in 2019 (Fig. 5D; F 5,7 = 7.36, p = 0.04, r 2 = 0.51).It is important to note that for the Sedgwick common garden, the distance between the garden and seed source locations was more constrained than the other two gardens given its more centralized location.We found there were gene-driven climate differences by plant population (F = 4.46, p < 0.001), environmental differences (F = 13.5, p < 0.001), and gene Â environment differences (F = 2.26, p = 0.009) for basal circumference and reproductive effort (F = 7.60, p < 0.001; F = 49.1, p < 0.001; F = 2.06, p = 0.019).
Rethinking seed sourcing for grassland restoration Does the Climate Where Seeds Were Collected Impact How Well They Perform in a Given Common Garden?
For both years of the study, linear regressions showed no consistent relationships between climate variables and the growth, reproduction, or survival of S. pulchra across the three gardens (Tables 3 & 4), though we did identify somewhat idiosyncratic relationships between climate parameters and plant performance in some instances.For example in 2018 (the drier of the 2 years of the study) plants from drier locations showed higher survival rates in the Santa Cruz and Stunt Ranch common gardens, while plants from less arid locations were larger in the Santa Cruz common garden (the least arid of the gardens).Specifically, plants sourced from areas that were more arid in the middle of the 2018 growing season (higher maximum VPD in March) survived at high rates than plants from less arid areas (lower maximum VPD in March 2018) in the Santa Cruz common garden (F [5,7] = 5.69, p = 0.06, r 2 = 0.44).In the Stunt Ranch garden, populations sourced from areas that received less total April precipitation in 2018 survived at higher rates than those sourced from areas that received more precipitation in April 2018 (F [5,7] = 11.81,p = 0.02, r 2 = 0.64).Plants sourced from areas that had more precipitation in May 2018 grew larger in the Santa Cruz common garden than plants from areas that received less rain in May 2018 (F [5,7] = 4.48, p = 0.09, r 2 = 0.37).In contrast, none of the climate variables showed significant relationships with growth and reproduction in 2019 (Table 4) with one exception; plants sourced from areas that were more arid in the early part of 2019 (higher maximum VPD in January) had higher mean reproductive output in the Stunt Ranch garden than those sourced from areas that had lower maximum VPD in January 2019 (F [5,7] = 4.35, p = 0.09, r 2 = 0.36).The aridity index we calculated for each population, which represents an integrated measure of temperature and precipitation, showed no relationship with plant performance across years and sites (Table S3).

Limited Benefit of Restricting Seed Collection to Local Populations for Stipa pulchra
Our experiment only found a limited benefit to using a local seeding provenancing strategy to restore S. pulchra populations, regardless of how local was defined.There was no strong or consistent home-site advantage or benefit to being sourced from or geographically closer to the planting site.These findings are consistent with a previous common garden experiment on S. pulchra (Hufford & Mazer 2012) and other studies on local adaptation in grassland species (e.g.Carter & Blair 2013), which found that local populations only exhibited home-site advantage in some years and locations.We also did not find strong evidence for using climate matching to improve S. pulchra performance.While we found that populations sourced from areas that were more mesic grew larger in the Santa Cruz garden, which was the most mesic common garden, and populations sourced from more arid populations reproduced more in the Stunt Ranch garden, which was one of the two more arid gardens, there was not a consistent benefit to being sourced from a similar climate.Both of the significant effects we found only occurred in 1 year and were not consistent across the gardens (i.e.mesic populations did not always outperform more arid populations in the Santa Cruz garden across all three metrics).Therefore, while we did find that populations sourced from geographically close to the planting site and populations sourced from similar climates occasionally performed better than non-local populations, there was not strong evidence for local populations (or populations from similar climates) consistently outperforming non-local populations (or populations from dissimilar climates).
Rethinking Seed Sourcing in the Context of Climate Change: Support for Regional Admixture Seed Sourcing In contrast to local seed provenancing, we propose that our results support the idea of regional admixture provenancing, defined as collecting seeds for a restoration project from multiple distinct populations that exist within a defined regional area (Bucharova et al. 2019).Unlike traditional local provenancing this approach purposely mixes propagules from multiple populations to increase genetic diversity, but does so within a defined region that experiences similar environmental conditions (Bucharova et al. 2019;St. Clair et al. 2020).The benefit of this approach is that increasing genetic diversity is expected to help populations be successful over a wider range of environmental conditions (Bradshaw 1984;Montalvo et al. 1997;Gustafson et al. 2004), such as those expected to occur as climate change progresses.We believe that regional admixture provenancing is particularly relevant in a region like southern California, which has both wide variation in weather year-to-year, as well as distinct microclimates due to topographic and maritime influences.Populations within these microclimates are likely genetically distinct and adapted to a range of environmental conditions (Bucharova et al. 2019).By incorporating propagules from a variety of populations within this region, a practitioner would be increasing the chance that there are plants adapted to whatever weather conditions occur at a site both now as well as in the future.For example, we found that populations sourced from more arid climates did well in all the common gardens in 2018, which was a dry season overall.We also found that the identity of the populations that grew the largest and reproduced the most differed between the 2 years in all three gardens.We suspect this was due to the two seasons having very different climates; with the Santa Cruz garden receiving twice as much rain and the Sedgwick and Stunt Ranch gardens receiving over three times as much rain in 2019 compared to the 2018 growing season.This suggests that while the S. pulchra populations in our experiment were not strictly adapted to local conditions, the different populations were adapted to distinct climates.
This finding also highlights one of the potential negative consequences of restricting provenancing to a single source; the unintentional reduction in genetic diversity (Sackville Hamilton 2001;Wilkinson 2001;McKay et al. 2005;Havens et al. 2015;St. Clair et al. 2020), and lowered ability of restored populations to respond to and adapt to rapidly changing climates (Rice & Emery 2003;Harris et al. 2006;Beierkuhnlein et al. 2011).When a site is located in an area that is expected to experience an increase in climate variability, increasing the genetic diversity of restored populations, by including populations sourced from a different climate, would be a proactive way to ensure that they can persist: a common goal of most restoration projects (Bradshaw 1987;Montalvo et al. 1997).For example our results highlight the importance of including populations from more arid locations to increase the resiliency of the restored population during dry years.However, at the same time, restricting provenancing to defined regions, as opposed to across the range of the species, balances the benefits of increasing genetic diversity with the potential risk of choosing maladapted populations and losing historical patterns of genetic diversity (Bucharova et al. 2019).Taken together, our experiment highlights the value of a provenancing method that promotes genetic diversity instead of attempting to identify the best-adapted population, whether that be a local one or one from a climatically similar location.

Limitations and Considerations
While we did not find strong evidence to justify restricting seed sourcing to local populations for S. pulchra, it is important to note that the duration of our experiment was much shorter than the lifespan of this long-lived perennial grass.It is possible that over a longer period, local populations would begin to consistently outperform non-local ones.For example previous common garden experiments on S. pulchra did not detect a home-site advantage until the third year of the study (Knapp & Rice 2011;Hufford & Mazer 2012).While local adaptation could become more apparent with more mature populations of S. pulchra, this is less likely to occur in areas where the climate is changing rapidly.In addition, selection could also be occurring on the germination or establishment life history stages of these plants.In particular, low recruitment rates are often cited as the reason for low success rates when seeds are direct broadcast during restoration (Larson et al. 2015).However, poor recruitment from individuals that were planted during restoration could also lead to conversion back to exoticdominated communities over time.Thus, there could be both, short-term impacts, when plants are reintroduced from seed, and long-term impacts, when seedlings are planted, on the viability of the restored population.It is also worth noting that we may have observed more distinct responses among populations if our study sides had included a broader geographic range encompassing greater climatic variability.Finally, a number of other biotic and abiotic environmental factors (e.g.soil conditions) not examined in this study could play an important role in local adaptation and restoration success (Rúa et al. 2016).Understanding the complexity of such factors in driving adaptation would undoubtedly represent an even greater challenge to restoration practitioners than the approaches tested here, and this is an important avenue for future research.As restoration practitioners grapple with how to establish populations that are resilient to future climate change, it is increasingly important to consider expanding seed sourcing beyond the "local is best" paradigm.The current study is consistent with other studies that have explored local adaptation in other perennial bunchgrasses that have found no clear benefit to using local seed sources to establish native populations (Wilsey 2010;Baer et al. 2014).These results demonstrate the importance of exploring alternative methods, especially for widespread, foundational species such as S. pulchra.Understanding local adaptation in foundational species used in restoration is particularly important as the success of projects often depends on the establishment of these species.If a dominant species displays local adaptation that leads to higher establishment, growth, and reproductive rates, then using locally sourced material would increase the likelihood of creating self-sustaining populations.For example Uselman et al. (2018) found that shrubs planted from locally sourced material exhibited population densities that were approximately 4.5 times higher than commercial cultivars when planted in former agricultural fields in the Great Basin Desert, USA.The authors propose that the higher emergence rates of seeds sourced from local populations indicate that the use of local seed sources would be warranted in this system (Uselman et al. 2018).
In the absence of evidence for local adaptation, however, the restriction of seed sources to local populations may be unnecessary and, in some instances, counterproductive.For example studies in Iowa and Illinois on dominant native prairie grasses found no benefit to the use of locally sourced seed when sowing them into abandoned agricultural fields (Wilsey 2010;Baer et al. 2014).These results suggest that locally sourced populations in these systems do not necessarily lead to superior restoration outcomes, and the use of only locally collected seeds could unintentionally create populations that are less likely to have the capacity to adapt to  (Breed et al. 2013).Thus, while there are clear benefits to strict local seed sourcing for some species, there are also potential negative consequences of using this method when local adaptation is absent (Havens et al. 2015).Following this logic, we argue that, for widespread species such as S. pulchra, a strictly local seed-sourcing strategy might be unnecessary, especially when considering how climate change will alter local environmental conditions.In addition, sourcing plant material is already one of the most expensive costs in restoration (Brancalion et al. 2019) and the use of unnecessary local seed would further increase project costs.
Our results also broadly highlight the importance of evaluating local adaptation in species that are key to restoration success, such as foundational perennial bunchgrasses in grassland restoration.Pilot studies could be conducted by scientists for a subset of species, such as foundational species or species of concern, to identify whether local adaptation exists and whether there are negative impacts that could arise from only using local seed sources.This in turn could be used to create a tailored seed sourcing strategy for these key species while a more general strategy, such as only using local populations, could continue to be used for the other species being reintroduced.This more nuanced approach to seed sourcing would allow practitioners and scientists to focus on creating seedsourcing guidance for species whose persistence at a site is needed for restoration success.

Figure 1 .
Figure 1.A map of California showing the approximate locations of the seven different populations of Stipa pulchra collected within central and southern California and the three common gardens (identified by a star).

Figure 2 .
Figure 2. Local populations of Stipa pulchra did not consistently outperform non-local populations in the Stunt Ranch common garden.Growth measured as basal circumference (A, B) and reproduction (C, D) of S. pulchra grown in the Stunt Ranch common garden in 2018 (A, C) and 2019 (B, D).The seven source populations from south to north are the Santa Rosa Plateau Ecological Preserve (Santa Rosa), Stunt Ranch Santa Monica Mountains Reserve (Stunt), Coal Oil Point Reserve (Coal Oil), Sedgwick Reserve (Sedgwick), Vandenberg Air Force Bases (Vandenberg), Kenneth S. Norris Rancho Marino Reserve (Cambria), Santa Cruz Arboretum (Santa Cruz).The bar outlined in blue is the population that is local to the common garden.An asterisk represents a significant difference calculated from a Welch's t-test ( p < 0.05) and represents a comparison between each seed source and the common garden population.All error bars represent AE SE.

Figure 3 .
Figure 3. Local populations of Stipa pulchra did not consistently outperform non-local populations in the Sedgwick common garden.Growth measured as basal circumference (A, B) and reproduction (C, D) of S. pulchra grown in the Sedgwick common garden in 2018 (A, C) and 2019 (B, D).The seven source populations from south to north are the Santa Rosa Plateau Ecological Preserve (Santa Rosa), Stunt Ranch Santa Monica Mountains Reserve (Stunt), Coal Oil Point Reserve (Coal Oil), Sedgwick Reserve (Sedgwick), Vandenberg Air Force Bases (Vandenberg), Kenneth S. Norris Rancho Marino Reserve (Cambria), Santa Cruz Arboretum (Santa Cruz).The bar outlined in blue is the population that is local to the common garden.An asterisk represents a significant difference calculated from a Welch t-test (p < 0.05) and represents a comparison between each seed source and the common garden population.All error bars represent AE SE.

Figure 4 .
Figure 4. Local populations of Stipa pulchra did not consistently outperform non-local populations in the Santa Cruz common garden.Growth measured as basal circumference (A, B) and reproduction (C, D) of S. pulchra grown in the Santa Cruz common garden in 2018 (A, C) and 2019 (B, D).The seven source populations from south to north are the Santa Rosa Plateau Ecological Preserve (Santa Rosa), Stunt Ranch Santa Monica Mountains Reserve (Stunt), Coal Oil Point Reserve (Coal Oil), Sedgwick Reserve (Sedgwick), Vandenberg Air Force Bases (Vandenberg), Kenneth S. Norris Rancho Marino Reserve (Cambria), Santa Cruz Arboretum (Santa Cruz).The bar outlined in blue is the population that is local to the common garden.An asterisk represents a significant difference calculated from a Welch t-test ( p < 0.05) and represents a comparison between each seed source and the common garden population.All error bars represent AE SE.

Figure 5 .
Figure 5. Sourcing Stipa pulchra from a population close to the common garden did not improve performance of the plants.Growth (A-C), reproduction (D-F), and survival (G-I) of the S. pulchra from each seed source population in 2018 (circles) and 2019 (squares) as a function of the distance of the seed source from the common garden.The population that is local to the common garden is outlined in blue.The seven source populations from south to north are the Santa Rosa Plateau Ecological Preserve (Santa Rosa), Stunt Ranch Santa Monica Mountains Reserve (Stunt), Coal Oil Point Reserve (Coal Oil), Sedgwick Reserve (Sedgwick), Vandenberg Air Force Bases (Vandenberg), Kenneth S. Norris Rancho Marino Reserve (Cambria), Santa Cruz Arboretum (Santa Cruz).Statistical information only presented for significant regressions.

Table 1 .
Location information (latitude and longitude), aridity index, and 30-year normal climate data (average total annual precipitation, average number of rain events annually, and average maximum vapor pressure deficit[VPD]in January, March, and May) for the seven Stipa pulchra seed source locations during the growing season.

Table 2 .
Plant survival (%) for each seed source location at the three common garden locations.

Table 3 .
Linear regressions comparing the growth (basal circumference), reproductive output and survival of Stipa pulchra plants with different climate variables (of the seed source locations) in 2018.Degrees of freedom (df) for all models is 5 and sample size (n) is 7. r 2 Values only reported for significant models and bolded values indicate most significant model according to AIC.

Table 4 .
Linear regressions comparing the growth (basal circumference), reproductive output and survival of Stipa pulchra plants with different climate variables (of the seed source locations) in 2019.Degrees of freedom (df) for all models is 5 and sample size (n) is 7. r 2 Values only reported for significant models and bolded values indicate most significant model according to AIC.