Using climate‐driven adaptive evolution to guide seed sourcing for restoration in a diverse North American herb‐shrub species

Seeds zones are bounded geographic areas to guide seed sourcing for landscape scale restoration. Yet seed zones based on plant adaptative traits are lacking for most restoration species. Sulfur‐flower buckwheat (Eriogonum umbellatum Torr) includes herbs, subshrubs, and shrubs adapted to a range of ecological regions within North America. Its widespread occurrence and diversity make it a suitable model species to explore seed zone development based on evolutionary plant adaptations. Within a common garden, 69 populations from diverse seed sources and 13 taxonomic varieties were evaluated for 17 phenological, morphological, and production traits in 2016 and 2017. Analyses of variance showed taxonomic varieties and seed source locations differed for all plant traits. Linear correlation revealed source locations with warmer mean temperature and more precipitation generally had later phenology, larger umbels, more leaf area, higher leaf dry weight, and more seed and shoot dry weight production. Canonical correlation strongly linked seed source climates at source locations with plant traits evaluated in the common garden, suggesting climate‐driven adaptive evolution. Canonical variates 1 and 2, explaining 60% of the variation, were used to develop regression models that predicted their values from climate variables across the study area. Using geographic information technology these were mapped into 12 seed zones representing 1.31 million km2 in the Western United States. These zones were designed to provide guidance to practitioners when sourcing sulfur‐flower buckwheat for restoration projects. We expect this methodology can be successfully applied to other species to develop seed zones based on adaptive evolution.


Introduction
Overgrazing, exotic weeds, drought, and wildfire are significantly degrading habitat in many world regions with each intensified by climate change (Westerling et al. 2006;Davies et al. 2012). Often natural recovery on degraded areas is inadequate and active restoration is prescribed (Havens et al. 2015). Increasingly, native plant restoration utilizing "locally" sourced germplasm is preferred to promote adaptation, healthy ecological relationships, and appropriate ecosystem services (Hufford & Mazer 2003;Bucharova et al. 2017). The common practice of seeding selected ecotypes environmentally distant from a restoration site can lead to local genetic erosion (Hufford & Mazer 2003), outbreeding depression (Kramer & Havens 2009), and unfavorable interactions with other plant and animal species (Thompson & Cunningham 2002;Bucharova et al. 2017).
From the early work of Turesson (1922) environmental diversity across the landscape has been associated with phenotypic diversity (Baughman et al. 2019). In several species genetic variation in phenotypic traits has been linked with local climates suggesting climate-driven adaptive evolution (Cheplick 2015). These relationships have been used to develop empirical seed zones, bounded geographic regions to guide seed sourcing for restoration (see https://www.fs.usda.gov/wwetac/threat-map/ TRMSeedZoneData.php). Still, empirical seed zones are generally lacking for species considered of primary importance for ecosystem restoration in North America and other world regions.
Evolutionary ecology (Cheplick 2015), ecological genetics (Bower & Aitken 2008), and genecology (St Clair et al. 2013;Johnson et al. 2015) are overlapping disciplines that tap natural selection and evolutionary theory to assess adaptation. Field studies involving evolutionary ecology include reciprocal transplant studies, with each test population evaluated in their respective home and away environments, and common garden studies in which diverse populations are evaluated in single or multiple common environments (Cheplick 2015). In both cases, potential causes of population diversity can be explored if environmental differences in environments, such as climate, soils, or biotic agents, are known.
Reciprocal transplant studies directly assess adaptation, but with few exceptions (Wang et al. 2010), quickly become unwieldy for a large set of diverse populations. That is, to evaluate 50 populations in a reciprocal transplant study, all would need replication within each of the 50 environments of origin. Although common gardens are a less direct approach (Endler 1986;Cheplick 2015), they facilitate evaluation of many plant populations in a common environment and are thus assumed to reveal genetic variation in plant traits (Endler 1986;Kawecki & Ebert 2004;Cheplick 2015). If that variation is linked to seed source climates it provides evidence for natural selection and adaptative evolution (Endler 1986, Kawecki & Ebert 2004, Cheplick 2015. Evolutionary ecology studies of North American herbs and shrubs are generally lacking, and there are no such studies of wild buckwheat (Eriogonum sp.). With its roughly 250 species, it is among the most diverse genera in North America. Within the genus, E. umbellatum Torr (sulfur-flower buckwheat) is the most diverse species (Flora of North America, http://www. efloras.org/) and is represented by numerous taxonomic varieties occupying a wide range of niche environments from British Columbia to Mexico. Given its wide distribution, diversity, and importance in varied ecosystems, it is suitable as a model species to study evolutionary ecology and seed zone development in Eriogunum species and other North American herbs and shrubs.
In a common garden study, Fisk et al. (2019) found that among five sulfur-flower buckwheat populations, variation in cold de-acclimation and flowering phenology in the spring were related to plant survival, emphasizing the importance of plant response to climate when sourcing seed for restoration. They found later development for populations from warmer locations, pointing to the importance of plant and climate interactions. They recommended expanded common garden studies to evaluate a larger set of populations from diverse climates to establish adaptive based restoration protocols and seed zones. This was undertaken herein with source seeds originating from a wide range of climates and North American ecological regions as defined by Omernik and Griffith (2014) and the U.S. Forest Service (see https://www.fs.fed.us/land/pubs/ecoregions/ ecoregions.html).
Sulfur flower buckwheat taxonomic varieties and seed source locations were evaluated for variation in phenology, morphology, and production traits in a common garden. Our hypotheses centered on evolutionary theory based on natural selection and adaptation. We expected to find widespread variation in plant traits, including those related to fitness, among both taxonomic varieties and seed source locations.
Moreover, climates at seed source locations were expected to explain a high fraction of the variation in plant traits observed in the common garden, suggesting climate-driven adaptive evolution.
We also expected that populations from warmer locations would display later phenology (Fisk et al. 2019) and that other fundamental plant trait, climate associations would be revealed. Since adaptation entails an array of adaptive plant characteristics, a multivariate approach was used to model the plant trait response to climate and to develop seed zones. We expect methodologies applied here will be useful for seed zone development based on adaptive traits in North American herbs and shrubs and in other regions.

Plant Materials
The genus Eriogonum is restricted to North America and likely arose in the Miocene as perennial shrubs with rapid evolutionary differentiation occurring in the late Miocene and early Pliocene (Shields & Reveal 1988). The repeated glaciation in the Pleistocene and the colder, warmer cycles likely caused upslope, down slope, and longitudinal migrations of Eriogonum species including sulfur-flowered buckwheat, resulting in continued diversification into the current Holocene period (Shields & Reveal 1988).
Optimal habitats for sulfur-flower buckwheat include relatively open areas in sage-lands, grasslands, woodlands, and forests from near sea level to 3,700 m above sea level (asl) (Young-Mathews 2012; Gucker & Shaw 2019). It colonizes harsh, disturbed sites, is relatively easy to establish, and grows in wide range of climates (Gucker & Shaw 2019). It is perennial with herb, subshrub, and shrub types (Flora of North America, http://www.efloras.org/). Varieties often form prostrate woody mats of varying size (Gucker & Shaw 2019). A well-developed taproot is consistent to its adaptation to xeric, rocky environments (Young-Mathews 2012), but it also occupies mesic, subalpine, and alpine environments.
Flowers form within umbels atop a bolting stem, often polygamodioecious and open, that attract a wide range of native pollinators (Flora of North America, http://www.efloras.org/). The presence of pollinators increases seed production (Young-Mathews 2012; Gucker & Shaw 2019), but the extent of selfpollination is undocumented. It provides nectar for many insects and is a larval host for many butterflies (Gucker & Shaw 2019). Blooming varies from June to September depending on variety, environmental adaptations, and local weather (Gucker & Shaw 2019). Seeds are light brown achenes 2-7 mm long that generally persist within the umbel for 2-3 weeks past maturity (Myer 2008) with seed dispersed by gravity and potentially enhanced by animals and moving water.
Sulfur-flower buckwheat is an important food source for a variety of bird species and serves as a cover for brooding sites in sharp-tailed grouse (Tympanuchus phasianellus [Linnaeus]) (Klott & Lindzey 1990;Bunnell et al. 2004) and the endangered sage-grouse (Centrocercus urophasianus [Bonaparte]) (Dunn & Braun 1986). Its leaves, flowers, or seeds are an adjunct food source for numerous small mammals and large ungulates in Western North America (Dayton 1960;Johnson et al. 1978;Austin & Urness 1986;Stevens et al. 1996).
Wildland seed collections were made using Seeds of Success  (Table 1). With such a large geographic area representing many individual plants within many populations used in this study, the occurrence of several sulfur-flower buckwheat varieties was all but inevitable. Varieties can be extremely variable within populations, and when proximate, intercrossing among varieties is common an intermediate forms may become stable and predominant (Gucker & Shaw 2019).

Growth Conditions
In April 2014 seeds from the 76 accessions were placed within plastic boxes (13.3 cm long, 12.7 wide, and 3.5 cm deep) containing water saturated vermiculate and stratified at 4 C for 4 weeks. After stratification, seeds were germinated at room temperature and seedlings planted in flats with 36 containers (5 Â 5 Â 5 cm) using Sunshine No. 5 plug mix (SunGro Horticulture, Bellevue, WA, U.S.A.). Plants were grown under greenhouse conditions with 12 hours "day" and "night" cycles at 20 and 15 C. As needed to promote optimal growth plants were watered and fertilized with pH-buffered liquid plant food (http:// aerogarden.com/).
In late August 2015 plants were moved to a partially covered outdoor structure or lathhouse located near the common garden site to promote plant acclimation prior to field transplanting at the Western Regional Plant Introduction Station farm in Pullman, WA (42.7224 N, 117.1351 W, 763 m asl). Transplanting was completed in mid-September 2015. Plots were randomized into 10 blocks each consisting of a single plant from each source location. Plants were spaced within rows 1 m apart with rows spaced 1.5 m apart.
In May 2016 species and variety taxonomy were assessed on all plants according to Reveal (Flora of North America, https:// efloras.org) utilizing a combination of morphological characteristics and geographical distribution patterns. Four accessions were identified as parsnip-flower buckwheat (E. heracleoides var. heracleoides), and three as either a mix of varieties or did not survive. As a result, those seven accessions were not included in the data analyses leaving 69 sulfur-flower buckwheat populations (Table 1). Although variety taxonomy for sulfur-flower buckwheat is challenging (Gucker & Shaw 2019), most were identified with acceptably high confidence (Table 1).

Data Collection
Sulfur-flower buckwheat populations were evaluated for 17 plant traits in 2016 and 2017 (Table S1). Plant survival was determined in spring prior to bolting each year. Plant phenology stages were recorded on each plant as the Julian day (day of year) for bolting (when a lead shoot started elongation), blooming (when 50% of flowers were open on a lead shoot), and maturity (when 50% of the flowers on a lead umbel were dry). On a lead shoot at blooming, umbel width and stem length were measured with stem length as the distance from the plant base to umbel base. Petiole length, leaf area (measured with a Li-Cor LI-3000A area meter and LI-3050A belt conveyor; LI-COR, Lincoln, NE, U.S.A.), and leaf dry weight were determined on a representative leaf between bolting and blooming stages, and specific leaf area calculated as cm 2 leaf area per g dry weight.
After maturity on each surviving plant, counts of umbels were made and dry weight of shoots, stems, umbels, seeds/ plant, seed weight/plant, and weight per seed were determined. Infructescences were initially hand-rubbed between ribbed rubber blocks to separate seeds. Whole seeds were then separated from plant debris and poorly formed seeds with a seed blower (Oregon Seed Blower, Hoffman Manufacturing, Corvallis, OR, U.S.A.). The denser, viable seeds are retained, and the debris is blown over into a "cup" and discarded. The efficiency of seed production (harvest index) was calculated as seed weight divided by shoot dry weight per plant. Basal areas were irregular in shape and estimates were derived by measuring the basal length and width at right angles to each other from the widest and longest points across the vegetation and the area calculated. These estimates were reproducible but overestimated the true area. To present basal area estimates closer to the actual area, we divided by two.

Analyses for Variation in Plant Traits
Analyses of variance were completed on each plant trait to determine if varieties and locations differed within the common garden. The mixed procedure (Proc Mixed) in SAS/STAT software version 9.4 (https://support.sas.com/software/94/) was used according to the model: where the variance σ 2 for each factor was defined as P for total phenotypic, y for year, v for variety, loc (v) for source location within variety, b for blocks, and r for residual error. Locations were nested within varieties. Years, variety, and seed source locations were fixed effects, with blocks and residual error random. Differences among traits were declared significant at p < 0.01. Using previous methodology (Johnson et al. 2015), variety by year and location by year interactions were assessed through Pearson linear correlation between trait means from 2016 and 2017. If positive and strongly significant, the magnitude of the response between years differed but not the direction of the response. Thus, averaging over years was appropriate. For traits not positively correlated, or negatively correlated, year means were considered as separate traits. The assessment of plant traits was completed in a common environment within both years to control for plasticity. As a result, variation in plant traits within common gardens is generally assumed to have a strong genetic component (Cheplick 2015).

Seed Source Climate and Variation in Plant Traits
Latitude, longitude, and elevation were collected at each collection site. These were entered into ClimateNA version 6.21 software available at http://climatena.ca/, a stand-alone MS Windows application that downscales PRISM (Daly et al. 2008) monthly climate normal data (30-year averages). Averages of 22 climate variables were generated for each location for the 1981 to 2010 period based on methodology described by Wang et al. (2016). Pearson linear correlations among climate variables were often highly correlated. Climate variables were considered redundant when correlations exceeded r = 0.90, leaving 11 in total. The high r-value ensured that variables important in describing plant-climate relationships were retained.
The climate variables used from each seed source location were mean annual temperature, mean warmest month temperature, mean coldest month temperature, continentality, mean annual precipitation, mean annual summer precipitation, the annual heat:moisture index, frost-free days, precipitation as snow, the 30-year extreme minimum, and 30-year extreme maximum temperature (Table S2).
Given the constraints involved with presenting all relationships within and among the 17 plant traits, graphs of blooming day and shoot dry weight, were used to graphically illustrate fundamental responses. Correlations among plant traits and among climate variables were frequent. Blooming date correlated (p < 0.001) with all other plant traits except stem length, specific leaf weight (SLW), umbel count, and survival. For shoot dry weight, all but SLW and weight per seed correlated. Likewise, for the 11 climates variables, mean annual temperature and precipitation were used to illustrate relationships. Mean annual temperature correlated with all other climate variables used and mean annual precipitation with all except frostfree days.
Pearson linear correlation was completed between each plant trait and climate variable. With 17 plant traits and 11 climate variables there were 187 correlations. To control for false positives, we used the false discovery rate (FDR) procedure of Benjamini and Hochberg (1995) as outlined by McDonald (2014). The FDR was set at 0.05 and calculations were completed using the spreadsheet provided by McDonald (2014) resulting in significance at p = 0.02 level.
Canonical correlation was used to test plant trait-climate associations in multivariate dimensions and to develop composite Table 1. Taxonomic varieties of sulfur-flower buckwheat (Eriogonum umbellatum Torr) grown in a common garden at Pullman, WA, U.S.A., in 2016 and 2017. Included is the confidence of variety identification, the number of seed locations collected, observation number for plant trait means, and life form in the Western United States. a Average variety identification rated 1 (low confidence) to 10 (high confidence). b From Flora of North America (http://www. efloras.org/); plants are typically low growing (<12 dm) forming cespitose, compact or spreading mats with woody crowns and aerial flowering stems. Values with different letters are different for each plant trait based on the LSD at p < 0.05. variables based on information from all traits. Canonical correlation (Proc Cancorr; SAS/STAT software version 9.4) is a data reduction procedure in which correlation between two groups of variables is assessed and linear combinations or sets of canonical variates are produced (Manly 1986). The first variate explains as much variation as possible between groups and the remaining unexplained variation is then used to produce the second independent variate and so forth until all the variation is explained. If most of the variation between plant traits and climate variables is explained in the first one or two variates they are useful as composite plant traits that include information from all 17 plant traits.

Landscape Mapping for Seed Zones
Mapping of traits for developing seed zones was based on regressing the first and second canonical variates on climate variables. If highly significant (p < 0.01) models resulted, plant trait response to local climates were mapped and seed zones developed. The models were used to predict plant traits represented by canonical variates 1 and 2 using climate data provided by Clima-teNA (1981. The predicted values for each canonical variate were projected across the mapping area at the 30 arc-sec resolution (approximately 1 km 2 ). For seed zones, the range of predicted canonical scores were divided into four segments for variate 1 and three for variate 2 defining geographic areas that were mapped and then overlaid to produce 12 seed zones. The extent of the mapping was limited to Omernik level III ecoregions (Omernik & Griffith 2014), which are commonly used in North America to classify areas with similar biotic and abiotic characteristics. To prevent overextending models, regression predictions were kept within the range of the observed values for the canonical scores. All geographic analyses were performed using R (R Core Team 2013) and code for this process can be found at https://github.com/dcarver1/ERUM_seedZones over the Omernik level III ecoregions.
Regression models were selected using the r 2 option in Proc Reg using SAS/STAT software version 9.4. For each canonical variate, all climate variables used (Table S2) from each source location were initially included in the modeling process. The final model selected was the combination of climate variables that produced the highest r 2 with the lowest Akaike information criterion statistic (Akaike 1969). This approach maximizes model predictive capacity and minimizes over-parameterization (Draper & Smith 1998). Because linkages among variables are common with regression, they do not necessarily identify the most important, independent climate variables for a given trait (Manly 1986).

Variation for Plant Traits
Highly significant differences among taxonomic varieties and locations within varieties (p < 0.01) were observed for all plant traits. All traits differed between years or had significant year by variety interactions except for seed number/plant and seed weight/plant even though their interactions trended toward significance (Table S1), and mean differences for some varieties were observed between years for both seed number/plant and seed weight/plant but more frequently for weight per seed. Since correlations between years were strong and positive for all traits (p < 0.0001), year interactions resulted only from differences in the magnitude of a given plant trait response, not its direction, showing interactions had a relatively small effect on varieties and locations within varieties.
Among taxonomic varieties, mean differences were frequently observed among all 17 plant traits. Yet, as for blooming and shoot dry weight, varieties often did not differ for a given plant trait (Table 1). In most cases plant traits correlated positively with one another and were as such often interrelated (p < 0.02). Later blooming, for example, associated with wider umbels, longer petioles, more leaf area, greater basal area/plant, leaf weight, seeds/plant, seed weight/plant, and shoot dry weight (Fig. 1).

Seed Source Climate and Plant Trait Associations
Linear correlation between taxonomic variety means of plant traits with their respective mean climate variables were not significant (p = 0.02), so links between varieties and climates were not established. This was because source locations within varieties were found in a range of climatic niches as illustrated with mean annual temperature and precipitation combinations (Fig. 2). For example, a population of the variety dichrocephalum had mean annual temperature and precipitation values of 6.8 C and 320 mm, and another with values of 2.1 C and 1,202 mm (Fig. 2). For variety ellipticum, with the most source locations, populations ranged from a relatively warm, dry climate (10.8 C and 335 mm) to a more mesic climate (5.7 C to 896 mm). Similar patterns were found within other varieties (Fig. 2). With varietal populations occupying a range of climatic niches they were unsuitable as categories to aid seed zone development based on plant characteristic and climatic linkages.
For source locations, however, significant linear correlations between plant traits and temperature-based climate variables were found in 11 of the 17 traits (p < 0.02) (Table S2). And except for specific leaf area, all correlations were positive. Thus, garden plants from source locations with relatively higher temperatures tended to bloom and mature later, have wider umbels, longer petioles, larger leaf area and leaf dry weight, more umbels, more seeds per plant, more shoot weight, and higher harvest indices.
Except for precipitation as snow, which associates with lower temperatures, correlations between plant characteristics and the precipitation related climate variables (mean average precipitation, mean summer precipitation, and annual heat:moisture index) were not significant (p = 0.02) (Table S2). This was so even though source locations mean annual precipitation ranged widely, from 194 to 1,374 mm. However, the four locations with the highest mean annual precipitation, ranging from 929 to 1,374 mm, also had among the lowest mean temperatures, ranging from À0.6 to 2.8 C. For those locations, production variables such as shoot dry weight were very low in the common garden, averaging 14 g per plant compared to 108 g per plant for all locations. The result was a lack of significant linear correlation between precipitation related climate variables and plant traits. When the outlier locations were removed, correlations between mean annual precipitation became positive and significant for 10 of 17 plant traits (p < 0.02, FDR = 0.05) (Table S2). Mean annual temperature and mean annual precipitation correlated positively with blooming, maturity, umbel width, leaf area and leaf dry weight, seeds/plant, and shoot dry weight (Table S2).
Canonical correlation analysis consolidated and verified the association among plant traits and the differing climates. Canonical variates 1 and 2 were highly significant (p < 0.0001) accounting for 60% of the variation between the 17 plant traits and 11 climate variables.

Landscape Mapping for Seed Zones
Regression of the canonical variates 1 and 2 on climate variables resulted in strong models suitable for mapping climate-driven variation in plant traits across the landscape (Table S3). Twelve seed zones representing more than 1.3 million km 2 were produced (Fig. 3).
The zones were highly variable in size (Table S4) reflecting a range of climates from the relatively wet, cold zone 9 (dark green) to the warm, dry conditions of zones 7 (red) and 8 (pink) (Fig. 3). Mean temperature and precipitation averages all differed among seed zones (Table S4). As expected, the positive relationship between blooming and shoot dry weight in the common garden (Fig. 1) was also observed among seed zones. For example, source locations in seed zone 8 had the highest shoot dry weight (254 g/plant) and had relatively late blooming (day 177). Locations in seed zone 9, however, had the least average dry weight (31 g/plant) and relatively early blooming (day 142).

Variation for Plant Traits
The strong varietal differences in the common garden and for locations within varieties revealed variation in all 17 plant traits evaluated. Thus, our hypotheses that taxonomic varieties and locations within varieties would differ for plant traits was verified.
Initially we thought taxonomic varieties might form independent categories based on plant traits and would be useful for seed zone development. Although some varieties for given trait differed from all others, such as bloom day and shoot dry weight for ellipticum, more often mean differences in plant traits were from overlapping, statistical groups of two or more varieties and did not form independent categories. Among locations most plant traits correlated with one another. This was expected for traits among phenology and production categories, but it also occurred for traits from different categories, as shown in the positive association between blooming day and shoot dry weight. Thus, later phenology associated with higher plant production. As blooming was strongly associated with the bolting to maturity period (r = 0.95), the association between later blooming and stem dry weight in the common garden appeared related to a predisposition for longer developmental periods over the entire reproductive cycle.

Seed Source Climate and Plant Trait Associations
Since locations within varieties were found in diverse climates, variety averages for plant traits did not correlate with their corresponding climate averages. Some varieties are known to occupy more environmentally specific areas (Flora of North America, https://eflora.org/) but these did not appear unique to varieties in general. For example, porteri is typically found at higher elevations (Flora of North America, http://eflora.org/) but is also considered a high-elevation counterpart of aureum. The variety polyanthum is found on serpentine soils although not exclusively, and subaridum is found in Western North America saltbrush communities among others (Flora of North America, http://eflora.org/). As a result, our hypothesis that varieties would occupy different climatic niches was rejected. That and the absence of categorical differences in plant traits among varieties made varieties unsuitable for seed zone development based on plant trait and climate associations.
Unlike taxonomic varieties, plant traits in the common garden and climate at seed source locations frequently correlated, proving our hypothesis that source location climates and variation in plant traits were associated, revealing apparent climate-driven local plant adaptations related to temperature associated variables, and for precipitation variables with outliers removed. Those correlations were almost all positive showing a predisposition for longer, later reproductive phenology in warmer climates associated with greater plant production. Those correlations showing later development in warmer climates are opposite the expected response in situ. Warmer temperature, if not extreme, advances development owing to the high plasticity of phenological variables (Espeland et al. 2018). This is the basis for predicting phenology with metrics such as growing degree days. Still, genetic factors, the capacity for plasticity, and weather each year all combine to determine the timing of plant development. However, within the common garden an underlying predisposition for later development at warmer source locations was revealed with adaptive implications.
If stress factors are moderate later bolting would allow more time for growth before reproductive development, favoring more vegetative production and the potential for more shoots, umbels, and seeds. The role of developmental time periods was verified by the positive correlation between days from bolting to maturity with source location annual mean temperature (r = 0.62, p < 0.0001), ranging from 42 to 137 days.
Moreover, earlier blooming and lower shoot dry weight were most often from source locations from colder or drier climates. In those higher stress climates, the predisposition to earlier development should have an advantage even though the stress factors, low temperature and low precipitation, were different. In colder locations the developmental window is narrower so earlier phenology is advantageous as time for growth and seed production is limited. In dry climates earliness would promote growth when temperatures are lower, and moisture use would be more efficient. The earlier growth and development promotes avoidance of more severe heat and drought occurring later. A lower potential for production is offset by better overall adaptation to dry climates. This is consistent with selecting smaller plants with greater root mass as a path to improving establishment and restoration success in arid environments (Scot et al. 2015).

Landscape Mapping for Seed Zones
Seed zones employ the "local is best" paradigm, which is generally supported by research (Leimu & Fischer 2008;Hereford 2009;Bucharova et al. 2017;Baughman et al. 2019) but is also far from absolute. Hancock et al. (2013) found local populations of several species in Australia had limited advantage in establishment and survival, suggesting climate change and habitat fragmentation as possible contributing factors. Still, based on survival and fecundity at home and away sites, Leimu and Fischer (2008) and Hereford (2009) found locally sourced germplasm had an adaptive advantage in 71% of the cases they reviewed, but an important fraction had no apparent advantage. In the Baughman et al. (2019) review focusing on the U.S. Great Basin, an important region in our study, increased survival of more locally derived populations was observed in 64% of 24 reciprocal transplant studies, and a local advantage among populations and species was observed in 9 of 10 studies measuring fitness. Temperature and precipitation factors were associated with differential adaptation, as they were here. Our results  (Omernik & Griffith 2014). Regression of each variate on climate variables at seed source locations produced models used to map plant variation in response to climate using geographic information technology. The maps were then overlaid to produce the 12 seed zones. support the "local is best" paradigm by showing the role of local climate in shaping differences in plant traits for adaptation.
Our focus was on climate, but other environmental factors, such as soils, can also play an important role in seed zone development and adaptive evolution (Gibson et al. 2019). Although most of the variation between plant traits was explained by climate variables in this study, there was also unexplained variation leaving room for influence by other environmental factors such as soils, abiotic factors, and microclimate effects.
The process of developing seed zones has an arbitrary component; larger or smaller zones can be mapped depending on the overlay sections chosen, in our case four intervals for canonical variate 1 and 3 for variate 2, yielding 12 zones. Obviously, variation in plant traits associated with microclimates will exist within zones, as illustrated by the response of four outliers derived from extreme cold-wet climates. The source locations were not uniformly distributed across the mapped areas in two-dimensional space but represented a wide range of climatic variability even in the absence of the outliers and strong regression models between plant traits and climate variables were obtained for seed zone mapping.
For seed zone development we sought a balance between a high number of zonal divisions and the practical number of seed zones from which seed could be collected, increased, purchased, and utilized by practitioners. As an offset to practicality, enough detail is needed to ensure that the adaptation to climates is reasonably represented, and that problems such as genetic erosion or maladaptation are minimal (Hufford & Mazer 2003). For most restoration species, including sulfur-flower buckwheat, availability of seed is a limiting factor, especially given current high demand arising from disasters such as wildfires (Harrison et al. 2021). This severely limits the choices of zone-based seed available to practitioners and thwarts beneficial application of seed zone protocols. Ecoregions, such as Omerick level III commonly used in North America, have been suggested as administrative units for seed sourcing. They consider an array of factors including biotic, abiotic, climate, geology, and others (Omernik & Griffith 2014). As seen in our map different seed zones occurred within different level III ecoregions, and often overlap from one to another. The assumption that level III ecoregions could represent seed zones derived for a given species from climateplant trait associations, as done here, proved lacking. This is consistent with other evolutionary ecology studies Johnson et al. 2015;Johnson & Vance-Borland 2016).
The mapped seed zones can be used interactively with shapefiles provided at https://www.fs.usda.gov/wwetac/threat-map/ TRMSeedZoneData.php. Alternatively, the shapefiles can be requested from corresponding author via email. [Correction added on 20 April 2023, after first online publication: The preceding sentence was added to provide additional details about obtaining files.] Those files can be used for assigning seed sources to specific zones and other planning activities related to restoration projects that dovetail with practitioner knowledge and experience. Certainly, the specter of rapid climate change adds more complexity to utilizing "local" germplasm. This should energize comprehensive collection and conservation of restoration species (Havens et al. 2015); only then we will have the agility needed to respond to future needs.
Recognition that climate often drives local adaptive evolution has led to suggestions that seed zones based on climate only may be a useful surrogate for seed zones developed from adaptive responses associated with plant traits (Bower et al. 2014;Doherty et al. 2017), including those that augment climate-only zones with molecular genetics (Massatti et al. 2020). Seed sourcing with zones based only on climate, edaphic, and other attributes can be effective in promoting adaptation and favorable ecological relationship (Bucharova et al. 2017). And their application to all species within zones simplifies their implementation compared to the more complex species by species approach. Still, adaptation must be assumed for a given species until tested (Bucharova et al. 2017).
A problem with environment only seed zones is that different species demonstrate different patterns of adaptive responses to climate and other environmental factors. Some show wide adaptation and movement of populations to different areas carries less risk of maladaptation, whereas others are much more narrowly adapted to their source environment, leading to much higher risk of maladaptation when displaced . These different responses to environment can be revealed in evolutionary ecology and studies such as ours but go undetected in generalized or environment only seed zones. It also implies that application of seed zones for a given species to other, related species, cannot be assumed.
Producing seed zones based on adaptive plant traits driven by local environments for all restoration species in a region is a challenging yet rewarding goal. Seed zones based on adaptive plant traits may be pursued initially for the most critically needed species. For many species, seed zones based on environment, or those augmented by molecular genetics, may be required to advance seed sourcing protocols in the immediate future. Regardless, all approaches to seed sourcing should be verified through testing and monitoring of restoration success along with direct assessments of different sourcing methods.

Supporting Information
The following information may be found in the online version of this article: Table S1. Significance levels from analyses of variance of sulfur-flower buckwheat (Eriogonum umbellatum Torr) populations collected at diverse locations in the Western United States and growing in a common garden at Pullman, WA in 2016 and 2017. Table S2. Linear correlations coefficients between plant characteristics measured in a common garden and climate variables at source locations for sulfur-flower buckwheat (Eriogonum umbellatum Torr) growing in 2016 and 2017, Pullman, WA (n = 69). Table S3. Models based on 17 plant characteristics measured on 69 populations of sulfur-flower buckwheat (Eriogonum umbellatum Torr) in a common garden at Pullman, WA in 2016 and 2017, and regressed on 11 climate variables at 69 seed source locations. Table S4. Sulfur-flower buckwheat (Eriogonum umbellatum Torr) seed zone areas, percent of total area, mean average precipitation (MAP), and temperature (MAT).