Applying community assembly theory to restoration: overcoming dispersal and abiotic filters is key to diversifying California grassland

Ecologists have explored community assembly through the framework of ecological filters, which predicts that species must overcome a series of challenges (i.e. pass through “filters”) to successfully establish in a given community. In the context of restoration, these filters (dispersal, abiotic, and biotic) can be manipulated to alter the resulting plant community by favoring native species or disadvantaging non‐native invasive species. We conducted two studies manipulating assembly filters at two California grassland sites previously dominated by non‐native species. At Site 1, we explored how variations in sequential seeding of native grasses and forbs (to overcome dispersal and biotic filters caused by priority effects) influenced the resulting community. At Site 2, we explored how thatch removal (to overcome the abiotic filter of light limitation) and herbicide‐based weed control (to overcome the biotic filter of competition from non‐native species) influenced the addition of native forbs into a partially restored grassland. Native forbs at Site 1 did not suffer from arriving after grasses, but native grasses benefited when given 1 year priority over forbs. At Site 2, dethatching increased native forb cover in a high rainfall year. Herbicide application reduced non‐native grass cover in dethatched plots without negatively affecting native cover. Native forb and grass cover were significant predictors of non‐native grass cover. However, they accounted for only 29% of the variation observed, suggesting there are other influential factors not considered in this study. Our results suggest that forbs can be incorporated into established native grasslands more successfully after dethatching.


Introduction
Community assembly theory is increasingly being used to direct ecological restoration (Funk 2021).One prediction from this theory is that species are sorted into a community from a regional species pool by passing through a series of ecological filters (dispersal, abiotic, and biotic) based on their traits (Weiher & Keddy 1999).The dispersal filter is a strong impediment in many restoration efforts as degraded lands frequently experience seed limitation of native species (e.g.Grman et al. 2015;Halassy et al. 2016;Török et al. 2018).Abiotic filters may exclude species that are not well adapted to, e.g.resource availability or temperature conditions in a restored site.Biotic filters may include the absence of necessary symbionts (e.g.Koziol & Bever 2017) or competition from the resident plant community.Many applications of community assembly theory have been applied to restoration of grasslands, which are some of the most heavily impacted ecosystems on the planet (Wilsey 2021).Grasslands provide critical ecosystem services and often support a diversity of plant and animal species even when highly degraded (Eviner 2016).While several studies have examined how to modify ecological filters to favor native species during and following restoration efforts (Cleland et al. 2013;Hulvey & Aigner 2014;Baer et al. 2020), identifying practical and cost-effective ways to restore grasslands remains challenging (Stromberg et al. 2007;Kimball et al. 2015;Werner et al. 2016).
Several abiotic factors can impede restoration efforts, including limitations in light, water, and nutrient availability.Litter buildup of dominant grasses (i.e.thatch) can lead to competition for light resources (Knapp & Seastedt 1986;Foster & Gross 1998;Hern andez et al. 2022) and declines in native species establishment and performance (e.g.Molinari & D'Antonio 2020;LaForgia 2021;Charles et al. 2022).Management techniques that remove thatch and increase light availability, such as burning, mowing, and grazing, can facilitate forb success by weakening this abiotic filter (Eviner 2016).Annual mowing in California grasslands has been shown to increase the abundance and richness of both native and non-native forbs from the seed bank or when seeded, with the added benefit of reducing non-native annual grass abundance (Maron & Jefferies 2001;Lulow 2008;Valliere et al. 2019).Similarly, tussock thinning and litter removal of a dominant native perennial grass prior to native forb seeding increased both the richness and abundance of sown forbs in an Australian temperate grassland (Johnson et al. 2018).By reducing the filter of light limitation on establishing native species, management techniques that decrease litter may promote a more diverse, invasion-resistant grassland.
The concept that the order of species arrival can affect community structure, or more specifically that early arriving species have an advantage over later-arriving species during community assembly, is broadly known as priority effects.Restoration ecologists have explored the potential for priority effects to facilitate the establishment of less competitive native species (e.g.Porensky et al. 2012;Werner et al. 2016;Stuble & Young 2020) and to improve invasion resistance by giving native species temporal priority over non-native species (e.g.Vaughn & Young 2015;Young et al. 2015Young et al. , 2017)), thus, dampening the biotic filter for native species.Several restoration techniques leverage temporal priority, including staggered seeding, planting plugs to simulate earlier seeding, weed control to delay non-native establishment, and aggregated or "mosaic" planting to delay competition between species (Young et al. 2017).A recent review found that 42 of 43 studies that experimentally tested priority effects by manipulating the timing of species arrival demonstrated priority effects via changes in community composition, diversity, and invasibility (Weidlich et al. 2021).Manipulating temporal priority can not only dampen the biotic filter for native species, it can also heighten the biotic filter for invasive species.Allowing sown native species to mature before invasive seed arrival or establishment can increase biotic resistance of a given site, as stand biomass is a competitive deterrent to invasion (Lulow 2006;Hess et al. 2020a).
Another way to strengthen invasion resistance may be to maximize functional overlap between restored native communities and potential invaders (e.g.Fargione et al. 2003;Emery 2007;Price & Pärtel 2013).The theory of limiting similarity predicts that successful invaders are functionally distinct from the species in an existing community and are therefore able to take advantage of an unfilled niche (Funk et al. 2008).Thus, niche overlap with established native species can function as a biotic filter for potential invaders.In the context of restoration, this concept could be implemented by planting native species that are functionally similar to potential invaders or, more broadly, by planting a diversity of native species to reduce available niches (Byun et al. 2018).For example, a restored site with high native grass cover might experience less non-native grass invasion as grasses can share many resource-use traits such as similar root depth and morphology (e.g.Larson et al. 2020).However, Hess et al. (2020b) argue that using limiting similarity to guide restoration oversimplifies the complexities of invasion in natural systems.Factors other than functional overlap may have a greater impact on invasion resistance, such as prioritizing early emerging, highly productive native species (Yannelli et al. 2018(Yannelli et al. , 2020) ) or selecting native species based on traits that confer high fitness in a given environment (Funk & Wolf 2016).
Only about 1% of California's native grassland remains today with most of these areas ranging from somewhat to entirely degraded (Eviner 2016).Despite being dominated by non-native plant species, these remaining grasslands are still considered a biodiversity hotspot where up to 90% of California's rare and endangered plant species can be found (Skinner & Pavlik 1994).Native perennial bunchgrasses are often ideal candidates for grassland restoration projects due to their drought tolerance, ability to grow in a variety of soil types, high seed production, and competitive ability with invasive species once established (e.g.Bartolome & Gemmill 1981;Lulow et al. 2007;Eviner & Malmstrom 2018).Alongside perennial grasses, California grasslands have historically included a rich diversity of native forbs (Stromberg et al. 2007;Werner et al. 2016).However, native perennial grasses, including California's state grass Stipa pulchra (purple needlegrass), are not always initially as successful when sown together with native forbs as when they are sown alone (Lulow 2004;Werner et al. 2016;Young et al. 2017).One solution is to separately seed the grass and forb mixes (i.e.giving grasses temporal priority), which also makes it possible to use broadleaf herbicide to control non-native forbs that tend to be longer-lived in the soil seed bank between rounds of seeding.However, restoration practitioners must consider tradeoffs when using sequential seeding to establish species-rich grasslands or when introducing a native forb component to a grassland.Forb establishment may be compromised by strong competition from established perennial grasses (Werner et al. 2016;Eddy & Van Auken 2019;Stuble & Young 2020).
Here, we follow the performance of native and non-native species at two grassland restoration sites in response to variations in sequential seeding of grasses and forbs, thatch removal, and herbicide-based weed control.Broadly, our goal was to explore how effective these methods are in alleviating dispersal, abiotic, and biotic filters in native species in order to establish a diverse, invasion-resistant native grassland.More specifically, we wanted to determine the optimal method in which to incorporate native forbs into grassland restoration projects.We established two different studies at two California grassland sites.At the first site, we addressed the following question: (1) Does timing of native forb seeding relative to native grass seeding (i.e.priority effects) influence community composition?At the second site, we asked: (2) How do dethatching and selective herbicide application affect the success of seeding forbs into established native grasslands?We then used the data collected across both sites to explore our final question: (3) Do native forb and perennial grass cover differentially promote invasion resistance?

Study Sites and Experimental Design
Study sites were located in the foothills of the Santa Ana Mountains within the Irvine Ranch Natural Landmarks located in Orange County, California, U.S.A. From the 1800s to the early 2000s, this area was used for cattle grazing.Soil textures range from sandy clay loam to clay.Long-term mean annual rainfall is 326.64 mm, over 90% (298.20 mm) of which falls from November to April (Tustin Irvine Ranch, California Climate Summary 19 January, 1902 to 30 June, 2003; Western Regional Climate Center, https://wrcc.dri.edu/).Precipitation varied throughout the experiment: 361.95 mm in 2016-2017, 72.39 mm in 2017-2018, and 395.48 mm in 2018-2019 (California Irrigation Management Information Systems, CIMIS Station #75 Irvine, https://cimis.water.ca.gov/).
We conducted two separate field trials at two grassland sites within Irvine Ranch Natural Landmarks-West Loma and Bee Flat Canyon, hereafter referred to as Site 1 and Site 2, respectively.The two sites are located approximately 4.83 km apart, with each site encompassing an area of about 0.14 km 2 .Both sites had approximately 95% cover of non-native grasses and forbs prior to site preparation, a 2-year process consisting of mowing followed by multiple glyphosate applications to reduce the non-native seed bank and expose bare mineral soil for seeding.At the time of project implementation in fall 2016, sites were at different stages of restoration.Site 1 had undergone site preparation but had not yet been seeded with native grasses or forbs.Site 2 had an established stand of Stipa pulchra covering much of the site, but no native forb component.The full experimental design for both sites is synthesized in Table 1.
Site 1 was drill seeded with S. pulchra in November 2016 at a rate of 9 pure live seed (PLS) lbs/acre.Experimental plots (2 m Â 2 m) were then selected across four areas within the site.In January 2017, all plots were sprayed with selective broadleaf herbicide (triclopyr butoxyethyl ester; Element 4; 0.5 oz./gal) in Table 1.Description and timeline for each of the four treatments associated with the grassland restoration experiments at Site 1 and Site 2. Note that site preparation at both sites consisted of mowing followed by multiple glyphosate applications throughout the 2 years prior to Stipa pulchra seeding.*Following drill seeding of S. pulchra at Site 2, all plots received applications of Element 4 consistent with large-scale maintenance of grassland restoration.These applications stopped prior to forb seeding in December 2016.a manner consistent with large-scale grassland maintenance to reduce competition from non-native forbs (primarily black mustard, Brassica nigra).In March 2017, all plots were handweeded of non-native grasses and not weeded again for the remainder of the study (Table 1).
To explore how the timing of native forb seeding relative to native grass seeding affected the resulting community (question 1), Site 1 included the following four treatments: (1) drill seeding S. pulchra in November 2016 with no forb component (hereafter G for grass only); (2) drill seeding S. pulchra immediately followed by hand seeding of forb mix in November 2016 (GF for grass/forbs concurrently); (3) drill seeding S. pulchra in November 2016 and hand seeding of forb mix in February 2017 after selective herbicide application (GtF for grass then forbs after herbicide; i.e. short-term grass priority); and (4) drill seeding S. pulchra in November 2016 and hand seeding of forb mix in November 2017 (GttF for grass then 1 year before forbs; i.e. long-term grass priority; Table 1).The native forb seed mix included 15 species sown at a total rate of 19.35 PLS lbs/acre (Table 2).We used a randomized block design with 16 replicate blocks (four replicate blocks in each of four areas of the project) for a total of 64 plots (16 replicate blocks Â 4 treatments; Fig. S1).Note that in 2017, data were collected in only 14 of the 16 blocks due to time constraints.
Site 2 was partially restored at the time of project implementation, having been drill seeded with S. pulchra in December 2014 at a rate of 9 PLS lbs/acre.In January 2016, 2 m Â 2 m experimental plots were established.Dethatched plots were mowed and raked in summer and fall of 2016.All plots were hand seeded with native forb seed mix at a rate of 16.1 PLS lbs/acre in December 2016 (Table 2) and broadleaf weeds were removed by hand from all plots in January and March 2017 (Table 1).Plots were not weeded again after the first growing season.At the time of forb seeding, S. pulchra plants had established and were at high cover (≥50%) across the majority of the site.Due to natural variation, however, there were some areas within Site 2 with low (<20%) S. pulchra cover.
To explore how thatch removal and herbicide application influence forb establishment in a partially restored grassland (question 2), Site 2 included the following four treatments: seeding forb mix into (1) low cover (<20%) S. pulchra stand (low control); (2) high-cover (≥50%) S. pulchra stand (high control); (3) mowed and raked high cover S. pulchra stand (high mow); and (4) mowed and raked high cover S. pulchra stand with a low-dose grassspecific fluazifop-P-butyl herbicide application (Fusilade II; 0.4 oz./gal) to target abundant non-native grass seedlings (high mow + herb; Table 1).We used a partially randomized complete block design with a total of 48 plots (3 blocks Â 4 treatments Â 4 replicates; Fig. S2) in 2017 and 2018.Data were only collected from two blocks in 2019, resulting in a total of 32 plots (2 blocks Â 4 treatments Â 4 replicates) that year.Treatments were randomly assigned to each plot, except for "low control" plots as areas with low cover S. pulchra were less common across Site 2.

Data Collection
Species richness and cover data were recorded for both sites in April 2017, 2018, and 2019 at the peak of each growing season.Cover data was collected by placing a 1 m Â 1 m gridded quadrat within each 4 m 2 plot, avoiding plot edges to minimize edge effects.Because the center of each plot was not marked, it is unlikely that the exact same 1 m 2 subplot was measured each year.The gridded quadrat was constructed with 25 intersecting points, crossing every 0.25 m starting from 0 m (Fig. S3).We recorded all plant species present at each of these points and calculated percent cover as the number of times each species was recorded out of 25 points.Each species present in the 1 m 2 sampling area was recorded to determine species richness.

Statistical Analyses
All statistical analyses were conducted in R (R Core Team 2022).To determine how timing of grass seeding relative to forb seeding affected the resulting restored community (question 1), we performed type III Analysis of Variance (ANOVA) using the Satterthwaite's approximation ("lmerTest" package) with treatment (G, GF, GtF, and GttF) and year (2017, 2018, and 2019) as fixed effects and block as a random effect.Similarly, to determine how dethatching and herbicide application of a partially restored area affected the success of forb seeding (question 2), we performed type III ANOVA using the Satterthwaite's approximation ("lmerTest" package) with treatment (low control, high control, high mow, and high mow + herb) and year (2017, 2018, and 2019) as fixed effect and block as a random effect.For both sites, we evaluated the effects of treatment, year, and their interaction on native forb cover, native forb richness, S. pulchra cover, and non-native grass cover.After visually inspecting residuals versus fits plots and normal quantile-quantile plots, outliers were removed and data were transformed where necessary to meet the assumptions of normality and homogeneity of variance for ANOVA (Table S1).We performed post hoc pairwise comparisons where applicable using Tukey honestly significant difference (HSD) tests ("emmeans" package).
We used multiple regression to evaluate the influence of S. pulchra cover and native forb cover on non-native invasive grass performance across both sites (question 3) and used Analysis of Covariance (ANCOVA; car package) to assess the significance of these relationships after controlling for the effects of year, treatment, and site.

Site 1
In 2017, plots where grasses and forbs were sown simultaneously (GF) and plots where grasses were given short-term priority over forbs (GtF) had significantly higher native forb cover than long-term grass priority plots (GttF) as forbs had not yet been sown in these plots (Fig. 1).However, native forb cover in these forbseeded plots (GF and GtF) was not significantly different compared to native forb cover in control plots sown only with native grasses (G; Fig. 1).In 2018 and 2019, native forb cover did not differ significantly between plots where forbs were sown (GF, GtF, and GttF) and plots where forbs were not sown (G; Fig. 1).
In 2017, native forb richness was significantly higher when forbs were sown later in the season after broadleaf herbicide application (GtF) than when forbs were sown concurrently with grasses (GF; Fig. 1).From 2017 to 2018, native forb richness in shortterm grass priority plots (GtF) decreased significantly (Fig. 1).Native forb richness in 2018 was similar in all forb-seeded plots (GF, GtF, and GttF) and significantly greater than in control plots (Fig. 1).In 2019, native forb richness was significantly higher in grass priority plots (GtF and GttF) than in control plots (G; Fig. 1).
There was a significant effect of treatment and year on Stipa pulchra cover at Site 1 (Table S1; Fig. 2).S. pulchra cover was higher in control (G) and long-term grass priority plots (GttF) than in short-term grass priority plots (GtF) and was lower in 2018 than in 2017 or 2019 (Fig. 2).
There was no effect of treatment on non-native grass cover, but there was an effect of year with non-native grass cover increasing significantly each year of the study (Table S1; Fig. 2).

Site 2
Native forb cover was higher in dethatched plots (high mow and high mow + herb) than in control plots (low control and high control) in 2017 (Fig. 3).In 2018, native forb cover was higher than control plots only in dethatched plots treated with Fusilade II (high mow + herb; Fig. 3).The following year, native forb cover was higher only in dethatched plots (high mow) relative to high cover control plots (high control; Fig. 3).
From 2017 to 2018, native forb cover decreased significantly in dethatched plots (high mow) and Fusilade II-treated dethatched plots (high mow + herb; Fig. 3).From 2018 to 2019, however, native forb cover in dethatched plots (high mow) increased significantly along with native forb cover in control plots (low control and high control; Fig. 3).
Dethatched plots (high mow and high mow + herb) had higher native forb richness than high cover control plots (high control) in 2017 and 2018 (Fig. 3).Native forb richness did not differ significantly across treatments in 2019 (Fig. 3).From 2018 to 2019, native forb richness in high control plots increased significantly (Fig. 3).
S. pulchra cover did not differ among high cover plots (high control, high mow, and high mow + herb) in 2017 and 2018 and dethatched plots (high mow) maintained significantly higher S. pulchra cover than low cover control plots (low control) during these years (Fig. 4).In 2019, S. pulchra cover was significantly lower in Fusilade II-treated dethatched plots (high mow + herb) than in high cover control plots, having decreased significantly since 2018 (Fig. 4).
Non-native grass cover was higher in control plots (low control and high control) than in dethatched plots (high mow), whose non-native grass cover was higher than in dethatched plots treated with Fusilade II (high mow + herb; Fig. 4).Nonnative grass cover was significantly lower in 2019 than it was in 2017 or 2018 (Fig. 4).
Treatment was not a significant predictor of non-native grass cover and was therefore removed from the multiple regression model.After controlling for the effects of site and year, native forb cover ( p < 0.001) and S. pulchra cover ( p < 0.001) contributed significantly to the model, which explained 29% of the variance in non-native grass cover.There was a significant negative correlation between S. pulchra cover and nonnative grass cover, and native forb cover and non-native grass cover (Fig. 5).

Discussion
The results of our first study, in which timing of native grass and forb seeding was manipulated during grassland restoration, showed that the success of sown native forbs did not change significantly whether they were sown concurrently, 3 months after, or 1 year after the native grass species Stipa pulchra.These results contrast with other studies of restored grassland communities in which native grasses given priority limited the success of native forbs (Werner et al. 2016;Stuble & Young 2020).However, our finding that thatch removal increased native forb establishment supports the results of previous studies and points Facilitating forbs in grassland restoration to the importance of overcoming light limitation as an abiotic filter (Hulvey & Aigner 2014;Johnson et al. 2018;Valliere et al. 2019).

Priority Effects
After 3 years, native forbs at Site 1 achieved similar cover and richness whether sown alongside or after native grasses, while native grasses were more successful when sown alone or given 1 year priority over forbs.Due to competition from established perennial grasses, we expected to see lower forb cover in grass priority plots (GtF, GttF;Werner et al. 2016;Stuble & Young 2020).Instead, we found that after 3 years forb abundance did not differ significantly between sown (GF, GtF, and GttF) and unsown plots (G), regardless of priority.However, native forbs achieved higher richness in sown plots.Because our primary goal was to diversify restored grasslands, our findings suggest that the timing of native forb seeding may not matter as much as the act of sowing them in the first place (i.e.overcoming the dispersal filter).
At Site 1, selective broadleaf herbicide was applied across all plots in January 2017 to target non-native forbs in a manner consistent with large-scale restored grassland maintenance.By this time, native forbs had been sown (November 2016) and likely begun to germinate in GF plots.While we cannot say definitively, the attempt to eliminate germinating non-native forbs may have also eliminated some sown native forbs, leading to similar native forb covers in grass only (G) and grass/forb (GF) plots.Further, weed removal in short-term (GtF) and long-term (GttF) grass priority plots may have augmented available resources (e.g.light, water, nutrients, and space), increasing the likelihood of success for sown forbs.
We predicted that S. pulchra cover would be lower when seeded with forbs (GF) relative to being seeded alone (G), and higher when sown a year before native forbs (GttF) than when sown 3 months before native forbs (GtF).We expected the longer interval between sowings to dampen the biotic filter for S. pulchra, with lack of forb competition benefitting the  1.Where applicable, post hoc results of significant treatment effects are denoted in the upper left corner; commas denote no significant difference between treatments and ">" symbols denote significant differences between treatments and how mean values compare.For example, S. pulchra cover was greater in grass-only (G) and long-term grass priority plots (GttF), which did not differ significantly from each other, than in short-term grass priority plots (GtF).
earlier-planted S. pulchra through priority effects.Indeed, S. pulchra cover was higher in long-term (GttF) than in shortterm grass priority plots (GtF).We did not, however, find a statistically significant difference in S. pulchra cover when it was sown alone compared to when it was sown alongside forbs.One possible explanation is that priority effects can be site dependent.Young et al. (2017) found that relative to more productive sites, sites with low to moderate native grass establishment in control (i.e.grass only) plots showed little to no evidence of priority effects when grasses were given a 1-year priority over forbs than when seeded together.At Site 1, the initial S. pulchra cover in control (G) plots was 56.9 AE 8.8%.While we considered areas greater than 50% S. pulchra "high cover," it is possible we may have seen stronger priority effect had our site been more productive (higher plant cover in grass-only plots).S. pulchra cover at Site 1 also changed significantly across years.We observed a decline in S. pulchra cover following the lowest precipitation season of our study (2017-2018), consistent with previous findings on this species' response to drought (Kimball et al. 2017).

Management to Promote Forb Establishment
Our results from Site 2 suggest that thatch removal may be critical to forb establishment in a partially restored grassland, and this aligns with other studies from a range of grassland systems (e.g.Maron & Jefferies 2001;Hulvey & Aigner 2014;Johnson et al. 2018).In 2017, native forb cover was significantly higher at thatch removal sites (high mow and high mow + herb) than in control plots.Our findings became slightly more nuanced in 2018 with native forb cover significantly higher than control plots only at thatch removal sites treated with herbicide (high mow + herb).In 2019, the only significant difference in native forb cover was observed between thatch removal plots (high mow) and high cover control plots (high control).
Taken together, these results suggest that mowing and biomass removal may have helped curb the abiotic filter of light Where applicable, post hoc results of significant treatment effects are denoted in the upper left corner; commas denote no significant difference between treatments and ">" symbols denote significant differences between treatments and how mean values compare.limitation, allowing for successful emergence of native forb seedlings.The addition of grass-specific herbicide to dethatched plots (high mow + herb) had no effect on native forb cover or richness up to 3 years after its application, but it significantly reduced non-native grass cover compared to untreated dethatched plots (high mow).While we expected low-dose grass-specific herbicide to lead to declines in S. pulchra cover, possibly reducing its utility as a management tool for restored grasslands, we found no statistical evidence for this.
Site 2 was seeded with forbs during a growing season with higher-than-average rainfall.Precipitation and other conditions during planting year can have significant and lasting effects on the success of grassland restoration (Bakker et al. 2003;Groves & Brudvig 2019;Werner et al. 2020), making it difficult to draw broad conclusions from a given study.Stuble et al. (2017) established identical grassland restoration plots across three sites and four starting years and found significant differences among the resulting plant communities, with temperature and precipitation among the main drivers of community composition.Thus, had we sown forbs at Site 2 during a lower rainfall season, we may not have observed as strong a response to thatch removal.Although average native forb cover dropped below 25% in mowed and raked plots in 2018, cover returned to over 90% in 2019, which received over five times the amount of rainfall as the previous year.This suggests that the increase in native forb cover due to thatch removal can be recovered following drought in response to high rainfall as long as there are forbs present in the seedbank (Eviner & Halbur 2021).

Invasion Resistance
Our restoration goals were not only to increase native species diversity but also to assemble native communities that could persist long term without regressing to a state of dominance by non-native invasive species.Across both sites we found that increased cover of S. pulchra and native forbs reduced nonnative grass cover.Together, however, native forb and grass cover explained only 29% of the variation in non-native grass cover, suggesting that there are other factors affecting invasive species success that we did not account for in the context of this study.
At Site 1, we expected cover of invasive grasses to be lower in grass priority treatments (GtF and GttF) than plots were grasses and forbs were sown together (GF) as S. pulchra took up resources then unavailable to invaders.In contrast to S. pulchra cover, non-native grass cover at Site 1 was not measurably affected by timing of forb relative to grass seeding; instead, non-native grass cover increased significantly each year of data collection.Consistent weed management could be beneficial in the first few years of restoration, both to allow sown native species to establish with less competition and more available resources, and to reduce invader seed input into the soil.
At Site 2, we expected lower invasion in mowed plots to accompany the increase in forb success consistent with previous studies (e.g.Maron & Jefferies 2001;Valliere et al. 2019).As anticipated, non-native grass cover showed significant treatment effects, with mowed plots (high mow, hight mow + herb) maintaining lower non-native grass cover than either of the controls (low control and high control).Based on our findings, the benefits of mowing as a restoration tool are clear and appear to persist for at least 3 years post-mowing.A continued schedule of mowing might help maintain grassland diversity by inhibiting the abiotic filter of litter on native forbs in the seed bank and preventing non-native annual grasses from dominating the site as sown native species establish.

Figure 1 .
Figure 1.Boxplots displaying the distribution of native forb cover (%) and native forb richness (number of native forb species per m 2 ) across four seeding treatments and 3 years at Site 1.To explore how variations in sequential seeding of native grasses and forbs affected the resulting community, treatments at Site 1 included: seeding Stipa pulchra in November 2016 with no forb component (G), seeding both S. pulchra and forb mix in November 2016 (GF), seeding S. pulchra in November 2016 and forb mix in January 2017 after herbicide application (GtF), and seeding S. pulchra in November 2016 and forb mix 1 year later in November 2017 (GttF).Significant effects of treatment, year, and treatment Â year are denoted in the upper right corner (***p < 0.001, **p < 0.01, *p < 0.05, NS, not significant).Open circles denote outliers and closed black circles denote mean values.Means not sharing letters are significantly different (p < 0.05).

Figure 2 .
Figure 2. Boxplots displaying the distribution of Stipa pulchra and non-native grass cover (%) across four seeding treatments and 3 years at Site 1. Open circles denote outliers and closed black circles denote mean values.Treatments and statistical notations as in Figure1.Where applicable, post hoc results of significant treatment effects are denoted in the upper left corner; commas denote no significant difference between treatments and ">" symbols denote significant differences between treatments and how mean values compare.For example, S. pulchra cover was greater in grass-only (G) and long-term grass priority plots (GttF), which did not differ significantly from each other, than in short-term grass priority plots (GtF).

Figure 3 .
Figure3.Boxplots displaying the distribution of native forb cover (%) and native forb richness (number of native forb species per m 2 ) across four maintenance treatments and 3 years at Site 2. To explore how thatch removal and herbicide application for weed control influenced the addition of native forbs to a partially restored grassland, treatments at Site 2 included seeding forb mix into: Stipa pulchra stand with low (<20%) cover (low control), S. pulchra stand with high (≥50%) cover (high control), dethatched high cover S. pulchra stand (high mow), and dethatched high cover S. pulchra stand with grass-specific herbicide application (high mow + herb).Significant effects of treatment, year, and treatment Â year are denoted in the upper right corner (***p < 0.001, **p < 0.01, *p < 0.05, NS, not significant).Open circles denote outliers and closed black circles denote mean values.Means not sharing letters are significantly different ( p < 0.05).

Figure 4 .
Figure 4. Boxplots displaying the distribution of Stipa pulchra and non-native grass cover (%) across four maintenance treatments and 3 years at Site 2. Open circles denote outliers and closed black circles denote mean values.Treatments and statistical notations as in Figure 3 (***p < 0.001, **p < 0.01, *p < 0.05, NS, not significant).Where applicable, post hoc results of significant treatment effects are denoted in the upper left corner; commas denote no significant difference between treatments and ">" symbols denote significant differences between treatments and how mean values compare.

Figure 5 .
Figure 5. Scatterplot and the corresponding regression line (blue) for the relationship between Stipa pulchra cover and non-native grass cover (left panel) and native forb cover and non-native grass cover (right panel) at Site 1 and Site 2 between 2017 and 2019.

Table 2 .
Seeding list for grassland restoration experiments at Site 1 and Site 2. *Due to insufficient seed resources, Castilleja exserta and Lupinus bicolor were not included in the seed mix for Site 2.