Intervention intensity predicts the quality and duration of prairie restoration outcomes

Restoration of grassland ecosystems is essential for mitigating global losses of biodiversity and is typically initiated to foster persistent, long‐term increases in biodiversity. Yet, evaluating long‐term impacts of restoration on biodiversity is rare, especially across sites restored using consistent methods. Evaluation of restoration outcomes is particularly important for increasing predictive capacity in restoration ecology to determine the level of restoration effort that is required to achieve both short‐ and long‐term restoration goals. We conducted a multisite study that explicitly compared the impacts of no intervention (“passive” or “natural” recovery), low intervention (seeding native plants), and moderate intervention (seeding native plants and using fire management) at 32 restored prairies differing in the age of restoration (3–23 years). Grasslands with natural recovery have equivalent native plant species richness compared to sites with low and moderate restoration intervention, however, they have significantly lower‐quality vegetation, as measured by Floristic Quality (mean C). We found that managing restored prairies with fire maintains native plant richness over time and is correlated with higher vegetation quality and presence of seeded species. Seed mixes with a high mean C score are positively correlated with plant community quality. However, seed mixes with more species are negatively correlated with the proportion of seeded species present. We found that while the degree of restoration intervention has no effect on the number of native plant species, greater levels of assisted recovery are required to produce restored prairies that resemble high‐quality remnant vegetation and, especially, to maintain these successes over the long term.


Introduction
Grasslands are globally important as they support high species diversity and provide critical ecosystem services such as erosion control, rainfall infiltration, and carbon sequestration (Grman et al. 2015).In the central United States, tallgrass prairie once dominated the landscape, covering more than 170 million acres, but agricultural conversion, development, and loss of fire have resulted in habitat loss with only 4-13% of the original extent remaining today (Samson & Knopf 1994;Samson et al. 2004).Remnant habitat occurs in small, isolated fragments that often do not receive the frequent fires that maintained this grassland ecosystem historically.The combined impacts of habitat loss, fragmentation, and altered disturbance regimes have led to increased plant extinction rates and declines in species diversity (Alstad et al. 2016).Tallgrass prairie survival, like the survival of many grasslands worldwide, requires prioritizing habitat restoration and recovery that goes beyond conserving remaining habitat alone (United Nations 2019).Restoring ecosystems is an essential step in preventing a continued loss of biodiversity and maintaining ecosystem function (Suding 2011;Zirbel et al. 2017;Denning & Foster 2018).
Despite international efforts to restore ecosystems that increase biodiversity and maintain services, restoration outcomes are extremely difficult to predict (Brudvig 2011(Brudvig , 2017;;Brudvig & Catano 2021).This is particularly true for long-term outcomes, as research evaluating restoration success over a long time frame is inherently rare (Lindenmayer 2020).Recent data on restoration outcomes have shown that as many as 20% of restoration actions are completely unsuccessful, while less than 15% are successful at meeting all restoration goals (Norland et al. 2018).As a result of this uncertainty, there has been a call to develop restoration ecology into a more predictive science by identifying the processes responsible for variation in success (Brudvig 2017).The common practice of comparing restoration outcomes to a reference (e.g. a remnant prairie) and examining variation between sites is insufficient for understanding why some restoration actions fail while others succeed.Predicting restoration outcomes requires determining the drivers of variable outcomes among restored sites (Brudvig 2017), which are often site-and year-specific and poorly understood (Wassenaar et al. 2007;Brudvig & Catano 2021).However, comparing restored prairies is difficult because it is rare to have historical data on restoration interventions, as these efforts are often completed by multiple organizations using different methods and data on restoration actions are often not stored.Without a deeper understanding of the factors that impact restoration outcomes across a wide variety of sites using consistent restoration practices, land managers may continue to invest money in restoration activities that do not improve biodiversity and ecosystem services, especially over the long term.
Here, we focus on tallgrass prairie restoration and three main drivers thought to affect restoration outcomes regardless of yearspecific variation: restoration intervention intensity, restoration age, and land-use legacies.

Restoration Intervention Intensity
The intensity of restoration interventions lies on a continuum from no intervention (also termed natural recovery/passive restoration) to low, moderate, and high intervention (also termed active recovery/active restoration; Jones et al. 2018;Chazdon et al. 2021).Here, we define no intervention as the unassisted regenerative process that occurs after a degrading disturbance (e.g.agriculture, mining) has stopped, without human assistance or intervention (Suding 2011;Jones et al. 2018;Chazdon et al. 2021).The varying levels of assisted recovery, on the other hand, involve increasing degrees of human intervention intended to promote reassembly of a natural ecosystem through activities such as site preparation, seeding, establishment mowing, and prescribed fire management (definition modified from Suding 2011 andChazdon et al. 2021).We define low intervention as seeding with native plants species, moderate intervention as seeding with native species followed by prescribed fire management, and high intervention as restructuring of soil properties followed by native plant seeding and fire management (Fig. 1).The intervention intensity is often assumed to correlate with commonly used metrics of restoration success.For example, native plant species richness of a restored prairie might be predicted to increase with increasing species richness of the seed mix used in planting (Larson et al. 2018).
Fire is an integral part of the North American tallgrass prairie ecosystem and is often used in more intensive restoration practices with the goal of increasing native species diversity and associated ecosystem services (Zirbel et al. 2017).Historically, fires burned every 3-5 years in tallgrass prairie (Collins 2000), most likely due to indigenous management (Roos et al. 2018).Frequent fires promote vegetative growth by removing litter, allowing for warmer soil temperatures, increasing nitrogen, and increased light at the soil surface (Neary et al. 1999;Benson & Hartnett 2006).Fires act as a stabilizing force in grasslands by limiting shrub encroachment, which prevents conversion of prairies into savannas and eventually a closed-canopy forest (Van Auken 2000;Briggs et al. 2005).

Restoration Age
The number of years that have elapsed since restoration actions were completed (i.e. the age of restoration) and may also affect restoration outcomes.Previous research has shown a decline in seeded-species establishment and species diversity as restoration age increases (Grman et al. 2013;Barber et al. 2019;Blackburn et al. 2020).A key question is whether restoration intervention intensity can prevent species losses over time.In particular, if fire management continues to be used over time, it may be possible to prevent long-term species losses.On the other hand, there is evidence that fire can promote the over dominance of clonal C4 grasses and Canada goldenrod, making long-term outcomes uncertain even with an increase in restoration intervention intensity.

Land-use Legacies
Land-use legacies can have major impacts on plant community composition and diversity (Brudvig & Damschen 2011;Turley & Brudvig 2016), and therefore have the potential to affect restoration outcomes.Soil composition, organic matter, and nutrient levels are strongly influenced by agricultural landuse history due to tilling, fertilization, and irrigation (Compton & Boone 2000;Dupouey et al. 2002;McLauchlan 2006).Since the 1960s, there has been an approximated 700% increase in fertilizer use worldwide (Foley 2005), creating legacies of increased soil nutrients that remain for decades, and perhaps millennia, following abandonment (Compton & Boone 2000;Dupouey et al. 2002;McLauchlan 2006).These legacies affect both the total nutrients and the available nutrients in the soil.Available nutrients have the greatest impact on plant growth and development.Residual elevated phosphorus levels can limit native plant recruitment and establishment (Schelfhout et al. 2017).Native prairie plants, which evolved in ecosystems with lower nutrient levels, are unable to capitalize on the addition of nutrients as successfully as species that have evolved in resource-rich conditions (Suding et al. 2005;Wassen et al. 2005).While grassland restorations with agricultural histories have strong differences in plant community composition from sites without agricultural legacies (Brudvig & Damschen 2011;Turley & Brudvig 2016), it is not known whether the type or strength of altered soil nutrient properties from prior agricultural land use will require more intense restoration interventions to achieve restoration goals.

Advancing Restoration as a Predictive Science
One of the most challenging aspects of making restoration ecology a predictive science is contending with multiple drivers simultaneously, including interactions between management practice decisions (i.e.seeding and fire management), the magnitude of past land-use legacies, and how restoration outcomes will change over time.Brudvig et al. (2017), argue this variability must be explicitly confronted by comparing how conditions differ among restored sites, and not only to reference conditions.It is especially difficult to compare restoration outcomes because information about the restoration interventions and initial site conditions including past land use is often unrecorded or unavailable.Additionally, no intervention (i.e.passive restoration) is an uncommon form of grassland restoration in the United States, likely because dispersal limitation can be problematic in post-agricultural ecosystems (Turley et al. 2017).We do not know how outcomes for sites restored with increasing intervention intensity compared to sites that have naturally recovered.
Here, we use a rare dataset from a large number of tallgrass prairie restorations that have documented the intensity of interventions, the age of restored sites, and the degree of agricultural land-use legacies to evaluate how the restoration intervention intensity, restoration age, and the intensity of agricultural land-use legacies affect restoration outcomes.In particular, we ask: Q1: Does low and moderate intervention in prairie restoration efforts result in improved restoration outcomes compared to sites that naturally recover following agricultural abandonment?Q2: Does restoration intervention intensity, from low intervention (seeding native species) to moderate intervention (seeding native species and managing with fire), correspond to improved restoration outcomes for prairie plant communities?Q3: Do altered soil characteristics resulting from past agricultural activities affect restoration outcomes?We hypothesized that: (1) Prairies restored with low and moderate intervention intensity will have greater native plant species richness and Floristic Quality (i.e.mean C) than sites that naturally regenerate following agricultural abandonment.(2) Sites with moderate intervention intensity will have increased native plant species richness, Floristic Quality, and success of seeded species compared to sites with low intervention.(3) Restored prairies with higher soil phosphorus will have lower native species richness, vegetation quality, and seeded species establishment than sites with lower soil phosphorus.

Study System and Site Selection
This study aimed to determine whether restoration intervention intensity and land-use legacies can predict tallgrass prairie restoration outcomes.We used data from prairie restoration sites that occurred across a range of environmental conditions and that had excellent management records, which have been maintained by the Natural Resources Conservation Service (NRCS).Management records included information on seeding lists, age of restoration, restoration size, and fire history.Our study took place on restored tallgrass prairie that is enrolled in the NRCS's Agricultural Conservation Easement Program (ACEP) Wetland Reserve Easements (WRE).We restricted study sites to easements found within a 161 km (100 mile) radius from Madison, Wisconsin, and that had at minimum 0.8 ha (2 acres) of restored upland.All sites enrolled in this program were previously used for row crop agriculture.
To determine restoration methods utilized at each site, we used data collected from the NRCS State Office management records to divide the WRE easements into three main categories: no intervention sites (no seed, n = 6), low intervention sites that were seeded with native prairie species (seed, n = 13) and moderate intervention sites that were seeded and managed with fire (seed + fire, n = 13).From all possible sites that fit our study criteria, we randomly selected up to 13 from each category, for a total of 32 sites throughout southern Wisconsin (Figure S1).Half of our no intervention sites were dropped, as the soils were too wet to fit our study requirements.After site selection, we verified the management history with two additional methods.First, we sent a survey to NRCS Field Office staff who worked with our selected easements so they could verify restoration and management history with files kept in their respective Field Offices.Second, we sent surveys to all landowners.If data obtained from Field Offices and landowners conflicted with prior records, we used data from official documents found at NRCS.This robust three-pronged approach allowed us to confidently assign historical management actions so that it was possible to evaluate the impact of multiple drivers on restoration outcomes.It also allowed us to determine the number of years since restoration was initiated.Restoration age varied from 3 to 22 years old.

Plot Location
For each site, we overlaid seeding maps obtained from NRCS records onto aerial photographs of the study sites using ArcGIS Desktop 10.5.Within the seeded portion of the site, we randomly placed a 20 Â 50-m study plot polygon in an area of the site that visually appeared to have upland grassland habitat and was a minimum of 15 m from the edge of the seeding treatment and 30 m from any wetland.No plots were placed on filled ditches or embankments or in areas with saturated soils or standing water.Study plot placement was verified in the field to ensure that the study plot met the above criteria.Using these criteria allowed us to evaluate the impacts of restoration intervention intensity within sites that had consistent environmental conditions.

Vegetation Data Collection
To assess restoration outcomes, we measured plant community responses using a modified Carolina Vegetation Survey protocol (Peet et al. 1998), which records plant occupancy and composition data across spatial scales.Each 20 Â 50-m plot contains ten 10 Â 10-m modules.In four of these modules, two 1 Â 1-m quadrats were placed in preassigned corners.We recorded all taxa present in each of these four modules to species if possible, or to genus if we could not confidently identify to species.We collected specimens and took photographs of individuals that could not be identified in the field for later identification in the lab.After recording all species in the four main modules, new species were recorded consecutively in each of the six remaining modules.This permits estimates of species richness at a range of scales from 10 cm 2 to 100 m 2 .Percent cover for each species was recorded for the eight 1 Â 1-m subplots.Vegetation sampling was conducted over 2 years.In 2018, data were collected at 11 sites between July 18 and September 5.In 2019, data were collected at 21 sites, between July 3 and August 31, for a total of 32 easements.
To assess the degree to which restoration outcomes resulted in high quality vegetation, species were assigned a value based on geographic origin (native and non-native) and on regional coefficient of conservatism (C) scores.Origin was determined from the United States Department of Agriculture (USDA) PLANTS database (USDA, NRCS 2002) and C scores were obtained from Chadde (2019); if Chadde (2019) was missing a particular species, C scores were obtained from Wilhelm (2017), which gives C scores for species in the Chicago region.
Coefficient of conservatism scores range from 0 to 10 and are qualitatively assigned to each species by regional botanical experts (Watermolen 2003), where 0 represents species of low conservation value that are widely distributed and grow readily in degraded habitats, and 10 represents species of high conservation value that are restricted to undegraded, remnant habitat (Spyreas 2019).

Soil Data Collection
To evaluate how soil properties affected restoration outcomes, we collected three soil cores (2.5 cm diameter Â 15 cm deep) from five consistent locations within each plot at all study sites.Soil samples from each plot were homogenized, dried at room temperature, and processed at the USDA Kellogg Soil Survey Lab, in Lincoln, Nebraska, U.S.A.All soils were analyzed for texture, total carbon, total nitrogen, total sulfur, organic carbon, carbon-nitrogen ratio, aluminum-iron oxalate, iron oxalate, magnesium oxalate, electrical conductance, oxalate extractable phosphorus, New Zealand phosphorus retention, water-soluble phosphorus, and pH.The phosphorus measurements were of particular interest as most phosphorus is immobile in the soil and is a good indicator of the intensity of agricultural legacies (McLauchlan 2006;Bizzari et al. 2015).

Analyses
To evaluate restoration outcomes and success for each site, we used presence-absence data in the 20 Â 50-m plots to calculate native plant species richness, average herbaceous plant community mean C, and seeded-species establishment success (i.e.proportion of seeded species detected).We focused on the herbaceous community, because woody species are not a dominant part of grassland ecosystems.We conducted supplementary analyses and calculated native plant cover and cover-weighted mean C calculated from abundance data within the 1 Â 1-m plots.All data were checked for normality using the Shapiro-Wilk normality test and all data, except for native plant cover, were normally distributed.Appropriate models were chosen to account for the non-normality of the native plant cover data.
We used univariate analyses to evaluate how restoration intervention intensity, restoration age, and agricultural legacies influenced tallgrass prairie restoration outcomes.All analyses were conducted in R version 4.2.0 (R Core Team 2022).To evaluate our first hypothesis that low and moderate interventions will result in grasslands that have greater native plant species richness and higher-quality plant communities than no intervention sites, we tested for significant differences in our vegetation metrics between natural recovery (no intervention) and assisted recovery (low and moderate intervention), using the lm function.Previous research has shown that mean C does not always have straight linear trends and that a quadratic covariate (e.g.age) may be the most appropriate (Spyreas et al. 2012).Here, we compared linear models with quadratic models for native richness and mean C and found that linear models had lower Akaike information criterion (AIC) values, so used linear models for these analyses.Given the non-normality of the native plant cover data, we used the glm function with a quasi-binomial distribution for this model.We considered restoration age and soil properties as covariates in our models.
To investigate the impact of the level of intervention intensity among assisted-recovery sites on plant communities, we tested for significant differences in native plant species richness, community mean C, and seeded-species establishment between low intervention sites (i.e.seeded) and moderate intervention sites (i.e.seeded and managed with fire), using the lm function.To account for non-normal data, we used the glm function with a quasi-binomial distribution to investigate the effects on native plant cover.We considered seed mix characteristics, time since restoration, and agricultural land-use legacies (i.e.soil properties) as covariates in our models.Seed mix composition was site specific, so we calculated the species richness and average coefficient of conservatism score for each site's seed mix.Due to high correlation between seed mix richness and seed mix mean C (r = 0.688), we included only one of these variables in our models, depending on the response variable being investigated.Agricultural land-use legacies were evaluated with a principal components analysis (PCA) of the soil characteristics, using the vegan package.We then extracted the first two axes of the ordination to use as model parameters in our linear models.
Given that our models had many potential parameters and interactions, we chose four models with increasing complexity that were consistent with our hypotheses (Table 1) and selected the best fit model using the AICcmodav package, version 2.3-1 (Mazerolle 2020).We considered the best fit model to be the model with the lowest significant AICc value, which was defined as AICc > 2. The final models were always one of the parsimonious models, and soil properties were never included in the best fit models (Table 1).The final linear models used to Table 1.Model selection summary for each response variable.Best models (highlighted in bold) were selected based on the model with the lowest significant AICc value (i.e.ΔAIC was at least 2 less than other models).For each response variable, four key models were compared to determine best fit.Assisted-recovery models include low intervention (seeded) and moderate intervention (seeded + burned) site categories.All site models include no intervention (natural recovery), low intervention, and moderate intervention.In model comparisons that had AICc values that did not differ by at least 2, we selected the least complicated model.Finally, we tested our hypothesis that more intense agricultural legacies, as measured by elevated phosphorus, limit native plant species richness, average coefficient of conservatism, and seeded-species establishment in restored prairies.We had three different soil phosphorus measurements (i.e.water-soluble P, New Zealand P, and oxalate P) that we used in nine different linear models.Three of the models examined the correlation of each soil phosphorus measurement to native plant species richness, three models examined how these phosphorus measurements were correlated with mean C, and three examined the correlation of each soil phosphorus measurement with seededspecies establishment.

Results
Native plant species richness in our study plots ranged from 9 to 48 species present, with a mean of 28 species.Interestingly, prairies with assisted recovery, which have been seeded with native species and sometimes managed with fire, do not support greater native plant species richness relative to no intervention sites that are naturally recovering (F [2,28] = 2.05, p = 0.15, Fig. 2A).Instead, restoration age was the only significant variable in models that compared no intervention to low and moderate intervention, with a negative effect on native plant species richness (F [1,28] = 5.79, p = 0.02, Fig. 2B).Among assistedrecovery sites, there is modest support that the intervention intensity influences native richness (F [1,17] = 3.02, p = 0.10, Fig. 2A), while restoration age has a significant negative effect (F [1,17] = 12.4,p = 0.003, Fig. 2B).Additionally, there is a significant interaction between intervention intensity and restoration age (F [1,17] = 7.56, p = 0.01, Fig. 2C) on native richness.
Average plot coefficient of conservatism (mean C) scores for herbaceous species ranged from 0.93 to 3.78, with a mean of 2.51.Assisted recovery resulted in grasslands with greater mean C scores compared to sites that have natural recovery (F [2,28] = 6.73, p = 0.004, Fig. 3A); age of restoration has only a modest effect in this model (F [1,28] = 3.42, p = 0.08, Fig. 3B).Models investigating mean C among restored sites, show that intervention intensity has a significant effect, with higher community mean C at sites with increased intervention (F [1,19] = 9.14, p = 0.007, Fig. 3A).Restoration age has a negative impact on plant community mean C (F [1,19] = 4.40, p = 0.05, Fig. 3B), while seed mix mean C has a positive correlation with plant community mean C (F [1,19] = 7.53, p = 0.01, Fig. 3C).
The establishment success of seeded species was very low at some sites, with the proportion of seeded species present in our study plots ranging from 6 to 100%, with a mean of 63.8%.Sites restored with moderate intervention, that are managed with fire, support a higher proportion of seeded species than low intervention, unburned, sites (F [1,18] = 10.43,p = 0.005, Fig. 4A).Again, restoration age has a significant negative effect (F [1,18] = 5.61, p = 0.03, Fig. 4B), though it appears to have less of an effect on sites managed with fire than those that have not been burned.Seed mix richness also has a significant negative effect on seeded-species establishment (F [1,18] = 9.86, p = 0.006, Fig. 4C).

Discussion
We found that, although native plant species richness does not significantly differ between naturally regenerating sites and those receiving assisted recovery, Floristic Quality (mean C) does.Additionally, our results indicate that fire management further increases mean C and increases seeded-species establishment at our study sites.Together, this suggests that more intense restoration interventions lead to greater improvements in restoration outcomes.
A recent meta-analysis by Jones et al. (2018), suggests that more intense restoration interventions (i.e.native seeding and fire management), do not always result in greater recovery than naturally regenerated sites.Indeed, in our study, native plant species richness did not significantly increase with greater restoration intensity, even though it is a metric that is frequently used to evaluate restoration success (Reemts & Eidson 2019).Our two other metrics, mean C and proportion of seeded species present, however, did show improved restoration success with more intense restoration interventions.This highlights that there can be discordance among metrics that are commonly used to monitor restoration outcomes and the importance of choosing multiple metrics that align with all restoration goals.

Assisted Recovery Versus Natural Regeneration
We found that assisted recovery, regardless of the level of intensity (seeding native species or seeding and fire management), does not significantly increase native species richness beyond naturally regenerating sites.This is likely due to the high abundance of common, early successional native species that successfully recolonize naturally regenerating sites.Indeed, Solidago spp., Symphyotrichum spp., Asclepias syriaca, Erigeron spp., Achillea millefolium, Urtica dioica, Ambrosia trifida, and Lactuca canadensis, are native species that were very common in our study plots which have C scores of 2 or less.Despite similar levels of species richness, sites that were restored with more intense interventions (e.g.seed mixes with species with higher mean C) had many more prairie-specialist species.It is likely that the seeded species are driving the higher mean C at assistedrecovery sites, as the seed mix mean C for these sites ranged from 4.1 to 5.6.

Restoration Intervention Intensity
For sites that utilize assisted-recovery methods, increasing the intensity of intervention effort through continued fire management or tailored seed mixes improves restoration outcomes.Native plant species richness, plant community quality (i.e.mean C), and seeded-species establishment all increased with active fire management and seed mixes that have a greater abundance of high-C species.Interestingly, restoration intervention intensity interacted with restoration age such that fire management prevents declines in native plant species richness that otherwise occur over time in the absence of fire.Species richness declines may be due to competition with woody and non-native species that are poorly adapted to fire, both groups of which are known to increase in prairies that are not regularly burned (Briggs et al. 2005;Ratajczak et al. 2012).Likewise, our results indicate that increasing restoration age and lack of fire management both result in a lower quality plant community via the loss of high-quality prairie species.We also find that seed mix mean C positively contributes to higher community mean C, as expected.Similar patterns can be seen in seeded-species establishment; however, seed mix richness has a negative effect on establishment.This surprising result could occur for a variety of reasons.For instance, seeding rates for certain species may be too low for consistent establishment or species-rich seed mixes may include species with lower establishment rates (Ficken & Rooney 2020), as is typical for high-C species that are often left out of low-diversity mixes.Additionally, if selected species were not best suited for the site, habitat filtering may reduce the diversity of the seed mix to just the species that can persist at that exact location (Myers & Harms 2011).

Land-use Legacies
Former land use can leave a number of legacies on ecosystems (Foster et al. 2003).One aspect of agricultural land-use legacies that is particularly important in grassland and savanna ecosystems is soil composition and nutrient availability (Bizzari et al. 2015).We found that soil texture and nutrients did not predict restoration success in our study system.While we hypothesized that increased nutrient levels, particularly elevated phosphorus levels, resulting from agricultural land-use history would have a negative effect on restoration outcomes.Most of our phosphorous levels were not predictive, yet our data suggest that high levels of water-soluble phosphorus may limit native plant species richness, mean C, and establishment of seeded species, which confirms previous research (Wassen et al. 2005;Ohno et al. 2007).However, these results should be interpreted with caution, as most of our sites fall within a narrow range of water-soluble phosphorus and these patterns are driven by just a few sites.Other studies evaluating restoration outcomes have found that land-use legacies do not always play a significant role (Grman et al. 2013).Similar to the findings of Grman et al. (2013), all of our study sites have intensive crop land-use legacy.It is likely that if we had included different land-use histories among our study sites (e.g.some post-agricultural sites, some remnant sites, and some post-residential sites), we may have observed a significant difference between these categories.Additionally, all our study sites are part of the NRCS's Wetland Reserve Easement Program and are, on the scale of all potential prairie soils, relatively similar in important soil qualities.Although our study sites have a range of soil textures, they are all located near wetlands and are likely not water limited, which is a key driver of plant diversity in habitats like ours (Wyckoff 1973).

Advancing Restoration as a Predictive Science
Our best models explained between 26 and 48% of the variation in our data, which while useful for understanding, makes predicting restoration outcomes difficult.It is evident that seed addition, long-term fire management, and seed mix diversity and quality are effective tools for improving restoration outcomes.Our findings indicate that, although naturally regenerating prairies may meet some restoration goals (in our system, native species richness), providing some level of assisted recovery is important for creating and sustaining native prairie plant communities in post-agricultural restoration settings.Therefore, it is imperative for prairie restoration to be a long-term investment, with regular fire management to maintain high native plant species richness, high-quality vegetation, and high seeded-species establishment success.Regular, ongoing monitoring of restored prairie can help determine what management is needed for each site, as there is not one recipe to fit all restoration management needs.Given that native plant species richness and vegetation quality decrease over time, especially in naturally regenerating sites or sites far from seed sources, it may become necessary to periodically reseed easements.Generally, despite naturally regenerating and assisted-recovery restored sites having similar species richness, more intense interventions and continuous long-term management is required to produce and maintain high-quality restored prairies.

ACKNOWLEDGMENTS
This research was supported by the USDA NRCS award #68-5F48-17-023.We thank G. Kidd, S. Olson, and the many other NRCS staff who helped to navigate NRCS management history files, the landowners who permitted us to conduct this research on their properties, and A. Paolucci, who helped with soil analysis preparation.We are indebted to the numerous students who participated in data collection and to the Damschen Lab who provided feedback, especially C. Warneke, who provided edits on several versions of this manuscript.We thank my committee members, particularly, J. Orrock, who provided in-depth feedback on the manuscript.Additionally, we wish to acknowledge the indigenous peoples, primarily the Ho-Chunk, Potawatomi, and Menominee nations, that historically and presently inhabit the lands upon which this data collection took place.For millennia, these peoples managed the tallgrass prairie that we attempt to restore to its historical condition.

Figure 1 .
Figure 1.Typical grassland restoration activities lie on a continuum from no intervention to high intervention, with intermediate steps of low and moderate intervention.Definitions presented here are modified from Chazdon et al. (2021) and are specific to grassland ecosystems.

Figure 2 .
Figure2.There is not a significant difference in native plant species richness between no intervention sites and sites restored with low and moderate intervention intensity (F [2,28] = 2.05, p = 0.15, panel A).However, restoration age has a significant negative affect on native plant species richness in these models (F [1,28] = 5.79, p = 0.02, panel B).Models evaluating variation in native plant species richness between low intervention and moderate intervention sites indicate that managing with fire has a modest effect on native plant species richness (F [1,17] = 3.02, p = 0.10, panel A) and age has a negative effect (F [1,17] = 12.4,p = 0.003, panel B).There was a significant interaction between restoration age and intervention intensity, indicating that overtime, unburned sites have significantly less native richness than burned sites (F [1,17] = 7.56, p = 0.01, panel C).Seed mix richness does not appear to affect native richness at our study sites (F [1,17] = 0.00, p = 0.98, panel D).
moderate intervention) and the age of the restoration.The covariates in the best fit models used to investigate how intervention intensity among assisted-recovery sites (low vs. moderate intervention) affects native plant species richness, vegetation quality (mean C), and seeded-species establishment success varied slightly.To evaluate the differences in native plant richness between sites with low and moderate intervention, the best fit model included level of restoration intensity, age of restoration, seed mix richness, and the interaction between intervention intensity and restoration age in our models.To evaluate differences in vegetation quality between sites with low and moderate intervention, the best fit model included intervention intensity, age of restoration, and seed mix mean C.And finally, the best fit model to evaluate differences in the proportion of seeded species found at sites with low and moderate intervention included intervention intensity, restoration age, and seed mix richness as covariates.Supplementary models that considered relative native plant cover and cover-weighted plant community mean C included intervention intensity, restoration age, and either seed mix richness (for native plant cover) or seed mix mean C (for community mean C) in the best fit models.