Assisted gene flow in the context of large‐scale forest management in California,
 USA

Author(s): Young, DJN; Blush, TD; Landram, M; Wright, JW; Latimer, AM; Safford, HD | Abstract: © 2020 The Authors. As climate changes, locally adapted tree populations may become maladapted to the sites in which they presently occur. When natural adaptive processes are insufficient for populations to keep pace with changing climate, human-assisted relocation of genotypes (assisted gene flow) may be a useful tool for maintaining forest resilience. While existing empirical evidence provides insight into the potential outcomes and consequences of assisted gene flow, its applicability to large-scale plantings needs to be evaluated. We conducted a test of assisted gene flow in the context of operational postfire restoration plantings in three U.S. Department of Agriculture National Forests in California. Our experimental restoration plantings included seedling provenances representing both the local planting site and lower-elevation provenances that may be adapted to hotter and drier conditions. For the duration of the experiment, the planting sites experienced anomalously hot, dry conditions, offering a window into the potential outcomes of assisted gene flow in a future climate characterized by warmer temperatures and more frequent drought. In most cases, there was no significant difference in seedling growth or survival among provenances. However, in a few cases, lower-elevation provenances performed better than local provenances, suggesting a potential benefit of assisted gene flow as a management response to climate change. Our analyses accounted for spatial variation in shrub cover and detected a consistent and substantial negative association between shrub cover and seedling growth. In addition, our study revealed that the use of operational seed collections that are not geographically precise (and therefore also not climatically precise) can complicate selection of appropriate provenances and lead to unpredictable outcomes. Numerous other risks and uncertainties—including the fact that tree populations are often adapted to local site factors other than climate and that long-term outcomes may differ from short-term observations—complicate evaluations of the potential utility of assisted gene flow.


INTRODUCTION
When planning tree planting projects, managers select sources (e.g., collection locations or specific parent trees) for the tree seeds that are used. Based on a long history of studies that often identify strong local adaptation in tree popula

Seed source selection
Tree planting can be an important component of forest management and restoration, particu-larly following severe wildfires and intensive tions (Langlet 1971, Conkle and Critchfield 1988, Ying and Liang 1994, Howe et al. 2003, timber harvest (McDonald and Fiddler 2010. January 2020 ❖ Volume 11(1) ❖ Article e03001 YOUNG ET AL. Kitzmiller 2005, St Clair et al. 2005, Savolainen et al. 2007, Wright 2007, many forest managers have historically prioritized the use of seeds collected from near the planting site while also incorporating a reasonable range of genetic variation (Ledig and Kitzmiller 1992). This approach helps to ensure that most of the planted trees are well-adapted to the environmental conditions of the planting site and that planted populations maintain adaptive capacity in future generations (Savolainen et al. 2007, Alberto et al. 2013. Matching seed source and planting site environment often results in increased performance of planted trees relative to alternative approaches that do not carefully account for the provenance of planted trees (Langlet 1971, Aitken andBemmels 2016).
Appropriate seed source selection methods are less clear when environmental conditions are changing. Given a scenario of progressive warming and drying, for example, it may make sense to select seeds from source environments that are hotter and drier than the planting site, as genotypes from those environments may be better adapted to certain environmental conditions (e.g., frequent drought) that may eventually characterize the planting site (Aitken and Whitlock 2013). The approach of intentionally moving genotypes of a given species to new locations within the species range in order to track changing environmental conditions is referred to as assisted gene flow (Ledig andKitzmiller 1992, Aitken andWhitlock 2013). Assisted gene flow differs from assisted migration (also known as assisted colonization or managed relocation; McLachlan et al. 2007) in that the latter concepts include managed relocation of individuals or populations of a species beyond the species' historic geographic range limit as opposed to moving them within the existing species range. However, many of the same motivations, guiding principles, and cautions apply to both assisted gene flow and assisted colonization (Aitken and Whitlock 2013).

Empirical evidence of assisted gene flow outcomes
Under changing climatic conditions, local adaptation to current climate implies maladaptation to future climate (Aitken and Bemmels 2016), suggesting to some extent that populations may perform better if relocated to new sites. However, many of the same reciprocal transplant studies that identify local adaptation also highlight important potential negative consequences of assisted gene flow, particularly when populations are adapted to additional local factors beyond climate (Bucharova 2017; also see Discussion).
Despite the large body of literature on common garden and provenance tests, only a few empirical studies to date have directly evaluated the outcome of assisted gene flow-that is, specifically moving genotypes into sites where the environment has changed to more closely match the relocated genotype's historical source environment (or at least some component of it) (Bucharova 2017). One study tested assisted gene flow in a tree species (Populus tremuloides Michx.) that the authors suggest has experienced substantial adaptive lag over a long period of historical climate warming, such that most populations now occur in a climate substantially warmer than that to which they are adapted (Schreiber et al. 2013). That study observed substantial improvements in growth and survival in populations that were moved northward into colder sites, without accompanying indications of cold-related maladaptation, representing potential positive outcomes of assisted gene flow.
Other studies have taken an alternative approach, using periods of anomalously hot weather (Hancock andHughes 2014, Bucharova et al. 2016) to test applications of assisted gene flow. In contrast to expectations, these studies found that with only a few exceptions, individuals from local populations performed better than individuals from warmer provenances (i.e., candidate populations for relocation), even under anomalously warm conditions. This unexpected result may potentially be explained by the fact that the studies only evaluated adaptation to temperature (as opposed to other potentially important climatic factors such as moisture availability) and were conducted over a single growing season with plants that began <1 yr old, despite all study species being perennials.

An operational forest management context
Despite growing recognition of the potential importance of assisted gene flow in forest management, existing studies of assisted gene flow have (1) been performed under highly controlled conditions that may not be representative of operational forest management and (2) yielded contradictory and unexpected results (see previous section). We sought to evaluate the potential outcomes of assisted gene flow applied in largescale postfire forest restoration plantings and gain insight into potential limitations or other considerations unique to implementation of assisted gene flow in an operational context. To this end, we studied postfire plantings of five different tree species conducted by the USDA Forest Service (USDA-FS) at three sites in California. The challenges faced by the USDA-FS in California (e.g., responsibility for managing a large and environmentally heterogeneous forested region given limited resources) are not unique, so we use the California plantings as a case study of assisted gene flow implementation and outcomes potentially relevant to any institution responsible for managing a large and climatically diverse forested region.
When replanting following severe wildfire, many management agencies, including the USDA-FS Pacific Southwest Region (which includes California), use seedlings grown from seed that was previously collected and then stored in a seed bank in anticipation of a future reforestation need. In the USDA-FS Pacific Southwest Region seed bank, an individual accession is referred to as a seed lot and is identified by the seed zone and 500 ft wide (~150 m wide) elevation band (e.g., 4000-to 4500-ft elevation) from which seeds were collected ( Fig. 1; Table 1; Appendix S1: Figs. S1-S6). Similar collection and cataloging systems are used by other large forest management agencies-for example, the California Department of Forestry and Fire Protection, or Cal Fire (Stewart McMorrow, Cal Fire Deputy Chief of Forestry Assistance, personal communication).
The California tree seed zones are variable in size, averaging roughly 30-50 km in latitude and longitude, and are intended to define regions with relatively consistent climatic and physiographic conditions (Buck et al. 1970). The reason for delineating seed lots based on seed zone and elevation band is to maintain separate collections of seed representative of specific local environmental (including climatic) conditions. A given USDA-FS seed lot may consist of seeds collected from one or more trees (generally 1-4) in each of one or more stands (usually 5 to 50) throughout the geographic region defined by the seed lot's seed zone and elevation band (Table 1). Collections have historically prioritized more stands vs. more trees per stand in order to maximize diversity (Arnaldo Ferreira, Geneticist, USDA-FS Pacific Southwest Region, personal communication).
USDA-FS seed lots are less geographically and climatically specific than the provenances typically used in research studies of local adaptation and assisted gene flow. However, they reflect the current reality of large-scale forest management. To gain realistic insight into the potential role of assisted gene flow in large-scale management practices (e.g., the management practices of the USDA-FS), it is thus essential to evaluate (1) the selection and performance of provenances as defined by management agencies (e.g., USDA-FS seed lots) and (2) the degree to which the use of geographically imprecise seed lots constrains the potential for implementation of assisted gene flow in an operational context.
In the present study, we evaluate the growth and survival of seedlings that originated from seed lots that are representative of the planting site, as well as seed lots representing lower elevations, at three sites that were planted following severe wildfire. For the duration of this experiment, California experienced a drought so extreme it had no historical precedent (Robeson 2015). The drought offers the potential opportunity to gain insight into expected outcomes of assisted gene flow in an overall hotter and more variable future climate. We additionally evaluate the limitations of relying on operational seed lots with imprecise collection location data, and we interpret our results (as well as implementation and expected outcomes of operational assisted gene flow in forests generally) in this context.

Experimental design
We worked with three USDA-FS National Forests (Appendix S1: Table S1) to establish experimental tree plantations in the spring of 2011 during operational restoration planting in areas that had recently experienced high-severity wildfire. At each of the three planting sites (i.e., Fig. 1. Potential seed collection locations for the high-elevation (purple shading), mid-elevation (blue shading), and low-elevation (green shading) seed lots of incense cedar trees planted at the Klamath site. Potential collection locations are delineated based on the 500-ft source elevation band of each seed lot (Table 1) and additionally constrained by the seed zone (seed zone 301; orange outline) identified in USDA-FS seed lot records. The gray background shading depicts normal mean annual precipitation for the 1981-2010 reference period (PRISM Climate Group 2019), with lighter shades reflecting higher precipitation. The extent of the main map is outlined in red in the inset map of California. For maps of the other species and planting sites, see Appendix S1: Figs. S1-S6. National Forests), seedlings from multiple provenances of one or more species (Table 1) were planted in a randomized complete block design. The provenances planted were unique to each site; that is, each provenance was only planted at one of the three planting sites (Table 1). However, we used the same treatment group types (i.e., one local and one or two lower-elevation provenances) at each of the three planting sites. At each site, we established three experimental blocks, each containing all provenances of each species tested at the site. Because the sites were planted by operational tree planting crews, the use of fewer, larger blocks (as opposed to more, smaller blocks) was necessary to ensure accurate planting and tracking of the multiple species and provenances. Across all three sites, the species planted were as follows: Douglas fir (Pseudostuga menziesii [Mirb.] Franco), incense cedar (Calocedrus decurrens [Torr.] Florin), ponderosa pine (Pinus ponderosa Lawson & C. Lawson), Jeffrey pine (Pinus jeffreyi Balf.), and sugar pine (Pinus lambertiana Douglas). We used these species because they were present on each respective site prior to wildfire and they were the same species being used for operational postfire reforestation in the burned areas surrounding the experimental study plots.
In total, 147 USDA-FS nursery-grown seedlings (49 seedlings per block 9 three blocks) of each provenance of each species were planted at each site. Seedlings were planted in a contiguous spatial arrangement with 2-3 m between each tree. At this spacing, competition among these small trees should be minimal (Cole and Newton 1987). At the Plumas and San Bernardino sites, seedlings were planted in a separate 7 9 7 grid for each provenance and replicate. At the Klamath site, seedlings were planted in a wheel-and-spoke design (Nelder YOUNG ET AL. Notes: Seed lots contain seeds collected from one or more trees from one or more stands within the specified seed zone and elevation band. Seed lot ID: the USDA-FS seed lot identification code; Coll. stands: number of stands from which seeds in the seed lot were collected; Trees per stand: average number of trees per stand from which seeds in the seed lot were collected; Seed zone: California seed zone (Buck et al. 1970) from which seeds were collected; Source elev.: the 500-ft elevation band from which seeds were collected (in feet for consistency with USDA-FS delineations; see Appendix S1: Table S3 for elevations in meters); No. seedl.: the number of experimental seedlings that were followed in this study; Source precipitation and Source temperature: the range of precipitation (normal total annual precipitation over the 1981-2010 period) and temperature (normal mean annual temperature over the 1981-2010 period) within the source seed zone and elevation band of the seed lot. The highelevation seed lot of each species is from the same 500-ft elevation band as the planting site (except for ponderosa pine at the Klamath site, which is from the 500-ft elevation band below the planting site). The text "n/a" indicates records not available.
1962) with 12 spokes, a center tree, and four trees per spoke (12 9 4 + 1 = 49). This design was favored by the local manager because once the trees are much larger it will allow for evaluation of competition at a range of densities. At each site, the relative positions of each provenance cluster were randomized across blocks. Additional plantings were conducted at a fourth site (the Angeles National Forest), but one block of seedlings was inaccessible due to extremely high density of a hazardous plant (Eriodictyon parryi [A. Gray] Greene) so the site was excluded from this study. However, the plantings remain for future monitoring. Competing vegetation was manually removed from within at least 30 cm of each planted tree during planting at each site to reduce competitive/facilitative impacts. At the San Bernardino site, the existing vegetation (primarily shrubs) was masticated across the entire planting area and the masticated material was left on the site. In early 2015 at the Klamath site only, newly recruited competing vegetation was again manually removed from within at least 1 m of most planted trees.
Provenances were selected to represent scenarios of upslope assisted gene flow and status-quo management. At each site, one provenance of each species that was planted originated from the same 500-ft elevation band as the planting site (or in the specific case of ponderosa pine planted at the Klamath site, the 500-ft elevation band below the planting site; Table 1). The highelevation provenance was intended to reflect status-quo management in which seeds are sourced from as close to the planting site as possible. The remaining planted provenance(s) of each species originated from elevation band(s) 500 to 3500 ft (~150-1050 m) lower (Table 1), representing scenarios of assisted gene flow. January 2020 ❖ Volume 11(1) ❖ Article e03001 YOUNG ET AL.

Field measurements
We measured the experimental seedlings twice: once in September 2012 after two growing seasons in the field, and once in September 2015, three years later. In 2012, we measured all surviving planted seedlings at the Klamath and San Bernardino sites and a random sample of onethird of all surviving planted seedlings (due to time constraints) at the Plumas site (Table 1, No. seedl.). In 2015, we remeasured (or recorded mortality of; Table 1) all previously measured seedlings that (1) were not beneath an object such as a fallen log, (2) had no apparent severe mechanical damage (e.g., from a rolling log), (3) were not apparently killed by a burrowing mammal, and (4) could be confidently identified to a specific seedling measured in 2012. Seedlings not meeting these criteria were excluded from the dataset. We additionally tested excluding seedlings with signs of animal browsing, but we found that this did not qualitatively influence inferences or conclusions (results not shown). We therefore included browsed seedlings in the analyses we report here in order to reflect the reality of operational forest management.
In both surveys (initial and final), we measured the height and basal diameter of each seedling. In the final survey, we also visually estimated the percent cover by shrubs of the area within 1 m of each seedling. We estimated percent cover in categories (0%, 0-1%, 1-5%, 5-10%, 10-25%, 25-50%, 50-75%, 75-90%, and 90-100%) and used the middle value of each category for statistical analyses. This approach requires confident knowledge of the location of a seedling. We were unable to precisely locate many dead seedlings, so we could not confidently estimate shrub cover surrounding them. We therefore did not include shrub cover in survival analyses. We estimated the stem volume of each tree at each time point based on the tree's height and basal diameter, modeling the stem as a cone (Broncano et al. 1998). We only measured aboveground morphology, but incorporating basal diameter into the growth metric may additionally capture aspects of belowground morphology better than height alone (Marx et al. 1977, Dey andParker 1997).

Identifying provenance and planting site climates
Because specific collection locations within a given seed zone-elevation band combination are unknown, we present the climatic range of potential collection locations within each experimental provenance (USDA-FS seed lot) as a range (Table 1) and as a point cloud (Fig. 2) representing the climate across the region defined by the seed lot's seed zone and elevation band. To do so, we first defined the geographic region from which seeds in each seed lot may have been collected by using geospatial layers of seed zones (Eldorado National Forest, Geographic Information Services, personal communication), and elevation (USGS 2018; Fig. 1; Appendix S1: Figs. S1-S6). Within the resulting region, we placed a grid of points with 100 m-by-100 m spacing and computed temperature and precipitation values at each point using the gradient-inverse distancesquared (GIDS) statistical downscaling method (Nalder and Wein 1998) as modified by Flint and Flint (2012). We obtained normal mean annual temperature and normal total annual precipitation during the 1981-2010 reference period from the ~800-m resolution TopoWx dataset for temperature (Oyler et al. 2014) and the ~800-m resolution PRISM dataset for precipitation (PRISM Climate Group 2019). We chose to use these relatively simple climate variables because (1) they are easily obtained and used in management applications and (2) within small regions typical of our analysis (e.g., Fig. 1), spatial variation in annual temperature and precipitation is highly correlated with spatial variation in many other biologically relevant climate variables (De Clercq et al. 2015).
We also quantified the climates of the planting sites ( Fig. 2; Table 1). We identified the normal climate over the 1981-2010 period by extracting values at planting site locations using the modified GIDS downscaling method from the same TopoWx and PRISM layers used to define the climates of the seedling provenances. Additionally, to represent the climate during the 4-yr period that the seedlings were growing in the field, we computed mean annual temperature during the October 2011-September 2015 period from monthly ~800-m resolution TopoWx temperature layers (Oyler et al. 2014) and mean annual total precipitation over the same period from monthly ~4-km resolution PRISM layers (PRISM Climate Group 2019) and extracted values at planting site locations using the same modified GIDS downscaling method. YOUNG ET AL. Fig. 2. Potential climate space from which seeds in each seed lot may have been collected. Each point reflects a random location from within the region constrained by the seed zone and 500-foot elevation band that each seed lot represents. Each color depicts a different seed lot within each site and species combination. The gray star depicts the climate at the planting site over the 1981-2010 normal period. The orange star depicts the climate at the planting site during the four-year duration of the experiment (October 2011-September 2015). The high-elevation seed lot of each species is from the same 500-ft elevation band as the planting site, except for Ponderosa pine at the Klamath site, which was from the 500-ft elevation band below the planting site. Point clouds appear slightly different across seed lots sourced from the same seed zone and elevation band because the points are randomly drawn (see Methods).

Statistical analyses
To test for differentiation in seedling growth and survival among the provenances (seed lots) of each species, we developed statistical models that use initial (2012) observations to establish a baseline that accounts for variation due to nursery and planting practices (Jacobs et al. 2005), and the change between measurements (2012-2015) to identify variation potentially attributable to provenance differences. Specifically, for seedling growth, we fit hierarchical Gaussian (normal) linear models to explain final (2015) stem volume using initial (2012) stem volume, shrub cover, and provenance. This modeling approach allowed us to evaluate the effect of seedling provenance on final seedling size, independent of the influence of initial seedling size and shrub cover. We treated provenance as a categorical variable, with 1 or 2 dummy variables for species with 2 or 3 provenances, respectively. We always considered the high-elevation (local) provenance as the baseline provenance represented by the intercept of the model (as opposed to a dummy variable), as this was the provenance representing status-quo management. We allowed the model's intercept and coefficient(s) for provenance dummy variable(s) to vary randomly across experimental blocks to account for block effects and for the existence of multiple experimental units (trees) within each block-by-provenance combination (Quinn and Keough 2002). The survival models thus effectively incorporate a random intercept for each data row (i.e., block-by-provenance combination of survival/mortality counts), which additionally serves to account for any overdispersion in the data (Elston et al. 2001, Agresti 2002. Prior to fitting models, we log-transformed initial and final stem volume values in order to satisfy assumptions of normality, and we standardized continuous predictor variables by subtracting the mean and dividing by the standard deviation (separately for each site-byspecies combination).
We fit a separate model for each site-by-species combination. The model specification given tree i within experimental block j for a site-by-species combination with two provenances is as follows: We repeated the same modeling procedure for survival between initial (2012) and final (2015) surveys, except that we used a binomial response distribution (with a logit link) and did not include shrub cover or initial stem volume as predictors. We excluded shrub cover because it was often impossible to precisely locate the planting locations (and thus quantify the surrounding shrub cover) of dead seedlings. It was not possible to fit survival models for ponderosa pine at the Klamath site or for Jeffrey and sugar pines at the Plumas site, either because survival was consistently very high (Plumas species; Table 1), or because one experimental block contained too few seedlings to achieve robust model fits (Klamath ponderosa pine).
To visualize differences in seed lot growth and survival among provenances, we used fitted models to predict final stem volume and survival for a hypothetical average experimental block, average shrub cover, and average initial stem volume. To do so, in making model predictions, we held shrub cover and initial stem volume explanatory variables constant at their means for each respective site-by-species combination, and we held all effects that vary by block at the overall mean. We randomly sampled 1000 sets of model fixed-effect coefficients using the fitted multivariate normal distribution of coefficients. For each set of coefficients, we predicted the response (stem volume or survival probability) for each provenance, and we then computed the median and 95% confidence interval of the responses across all 1000 sets of coefficients (Gelman andHill 2007, McElreath 2016). We additionally computed pairwise contrasts (the difference in predicted response value and associated 95% confidence interval) between all provenance pairs for each site-by-species combination using the same sets of sampled model coefficients. We performed all statistical analyses in R version 3.5.0 (R Core Team 2018) using the package lme4 (Bates et al. 2015) for fitting generalized linear mixed-effects models.

Provenance and planting site climates
The geographic region from which seeds of a given provenance were reported to have been collected (based on the recorded seed zone and 500-ft elevation band of the corresponding seed lot) was generally very broad (e.g., Fig. 1; Appendix S1: Figs. S1-S6). Mean climate generally varied widely across space within each of these regions (Fig. 2). For example, the high-elevation seed lots of Douglas fir and incense cedar planted at the Klamath site (defined by seed zone 301 and the 4500-to 5000-ft elevation band) encompassed a region with normal annual precipitation ranging from under 1200 mm to nearly 4000 mm and normal mean annual temperature ranging from 8.1° to 10.2°C (Fig. 2). Most other provenances evaluated in the experiment also originated from seed lots representing regions that encompass a >twofold range in mean annual precipitation and a 2°-4°C range in mean temperature (Fig. 2).
Source precipitation and temperature values also differed among provenances of a given species, to varying extents (Fig. 2). The amongprovenance differentiation is driven by geographic variation in climate, with cooler, wetter conditions generally more common at higher elevations and in more coastally influenced regions. However, the ranges of precipitation values overlapped substantially among provenances, despite the fact that the elevation limits of all provenances were separated by at least 500 ft (Table 1). For example, for Douglas fir planted at the Klamath site, the potential source precipitation range for the high-elevation seed lot was 1170-3986 mm, for the mid-elevation seed lot was 888-3732 mm, and for the low-elevation seed lot was 875-3184 mm (Table 1). Potential source temperature ranges were generally more clearly differentiated among provenances, though overlap did occur to some extent (Fig. 2). The overlap was generally greater for provenances from elevation bands that were closer in elevation (e.g., Plumas species and Klamath ponderosa pine; Table 1).
The normal climate of the planting sites (i.e., the mean annual temperature and total annual precipitation over the 1981-2010 reference period) generally fell within the climate space of potential seed collection sites of the high-elevation provenance for each species and site combination (Fig. 2). At each planting site, the climate during the 4-yr duration of the experiment was substantially hotter and drier than the long-term average climate (Fig. 2). As a result, in most cases the climate conditions at the planting site during the experiment fell within the climate space of potential seed collection sites of a lower-elevation provenance planted at the site.

Provenance performance
In several cases, the low-elevation provenances outperformed the high-elevation provenances in terms of growth and/or survival, but in most cases the differences were not individually significant for a given species at a given site (Figs. 3-4; Appendix S1: Table S2). Performance among provenances differed significantly only for incense cedar planted at the Klamath site and for Jeffrey pine planted at the San Bernardino site. For incense cedar, the low-elevation provenance had significantly larger model-predicted final stem volume (mean 30 cm 3 ) than both the midand high-elevation provenances (mean 14 and 16 cm 3 stem volumes, respectively), accounting for differences in starting size and shrub cover ( Fig. 3; Appendix S1: Table S2). Additionally, both the low-and high-elevation provenances of incense cedar had significantly higher modelfitted survival rates (mean 92% and 97%, respectively) than the mid-elevation provenance (73%; Fig. 4; Appendix S1: Table S2).
As with the incense cedar at the Klamath site, the low-elevation provenance of Jeffrey pine at the San Bernardino site had a significantly larger model-predicted final stem volume (mean 200 cm 3 ) than the high-elevation provenance (126 cm 3 ; Fig. 3; Appendix S1: Table S2). The low-elevation provenance also had a significantly higher model-predicted survival rate (mean 98%) than the high-elevation provenance (mean 92%; Fig. 4; Appendix S1: Table S2).
With few exceptions, in models that included shrub cover as a predictor of stem volume, shrub cover was significantly and substantially negatively associated with final stem volume ( Table 2). The only exceptions were the individual species-by-site models for two species at the Klamath site, where shrubs had recently been manually removed from within approximately 1 m surrounding each tree.

Differentiation in performance among some provenances
In two separate cases (Klamath incense cedar and San Bernardino Jeffrey pine), seedlings from a lower-elevation provenance exhibited greater growth and survival than the high-elevation provenance that was selected to reflect statusquo local seed sourcing (Figs. 3-4; Appendix S1: Table S2). Within a given seed zone, lower elevations are most often climatically hotter and drier than higher elevations (though there are many exceptions; Fig. 2), so the stronger growth of the low-elevation provenances (also seen in all other species and sites, but not significantly) may reflect adaptations to the anomalously hot, dry conditions that prevailed at the (higherelevation) planting site for the duration of the experiment. This observation aligns with other observations of local adaptation along climatic gradients in pine species in California (e.g., Kitzmiller 2005), including Jeffrey pine (Mart ınez-Berdeja et al. 2019); however, we note that there is a lack of existing data for incense cedar. Our observation also aligns with the conceptual notion that in a future with hotter temperatures and more frequent drought, assisted gene flow has the potential to yield increased tree growth and survival relative to status-quo management (Aitken and Whitlock 2013).
Faster growth and greater short-term survival relative to other provenances may not, however, necessarily reflect better adaptation to the conditions at a site. Plants from warmer provenances are often observed to grow faster than plants from cooler provenances, even when planted YOUNG ET AL. Fig. 3. Median model-fitted stem volume (and 95% confidence interval) of the seedlings of each seed lot at the end of the study, following three years of growth in the field. Fitted final stem volume accounts for (and holds constant across seed lots within each site-by-species combination) initial seedling stem volume and shrub competition (see Methods). The high-elevation seed lot of each species is from the same 500-ft elevation band as the planting site (except for Ponderosa pine at the Klamath site, which was from the 500-ft elevation band below the planting site), and the other seed lots were collected from 500 to 3500 ft (~150-1050 m) lower in elevation. Fig. 4. Median model-fitted survival probability (and 95% confidence interval) of the seedlings over three years of growth in the field. Survival of Ponderosa pine at the Klamath site and Jeffrey and sugar pine at the Plumas site is not shown because survival was either consistently very high (Plumas species; Table 1) or because one experimental block contained too few seedlings for robust model fits (Klamath ponderosa pine). The high-elevation seed lot of each species is from the same 500-ft elevation band as the planting site, and the other seed lots were collected from 500 to 3500 ft (~150-1050 m) lower in elevation. YOUNG ET AL.

0.69
Notes: Models were fitted using standardized predictor variables (see Methods) to facilitate comparison among coefficients and species. A separate model was fitted for each site-by-species combination. The source elevation coefficients correspond to dummy variables for seed lots (low elevation and mid-elevation), with high elevation as the base level incorporated into the model intercept. For pairwise contrasts among seed lots, see Appendix S1: Table S2. Coefficient estimates with 95% confidence intervals that exclude zero are bolded. Coefficients for survival models for Ponderosa pine at the Klamath site and Jeffrey and sugar pine at the Plumas site are not shown because survival was either consistently very high (Plumas species; Table 1) or because one experimental block contained too few seedlings for robust model fits (Klamath ponderosa pine). AUC: area under the receiver-operating characteristic curve. into cooler environments (Mangold andLibby 1978, Morgenstern 2011). This observation may be attributed to a tradeoff between growth rate and cold hardiness (Loehle 1998, Koehler et al. 2012, as trees that invest in cold hardiness (e.g., increased solute accumulation and delayed bud break) have fewer resources available for growth. Thus, although warmer-and wetter-provenance trees may grow faster than local trees, they may suffer more extensive damage in rare cold or dry events that are not captured in short-term studies like ours.
Notably, the mid-elevation seed lot of Klamath incense cedar showed significantly lower survival than either the low-elevation or high-elevation (local) incense cedar seed lots, thus providing an important counterexample to the idea that upslope assisted gene flow leads to stronger performance. This observation may be a consequence of the limited geographic specificity of seed lot collection locations: The potential climatic range from which the seeds may have been collected includes conditions very distinct from the planting site's climate (see Limited geographic specificity of seed collection locations, below).
It is also possible that the provenance differentiation we observed was driven by nongenetic factors such as differential treatment of the seedlings in the nursery (Jacobs et al. 2005) and maternal provisioning (Roach and Wulff 1987), but our model, which accounts for initial size and survival after one year of growth in the field, should help to rule out (but potentially not completely exclude) such effects. January 2020 ❖ Volume 11(1) ❖ Article e03001 YOUNG ET AL.

Limited differentiation in performance among other provenances
Growth and survival were not significantly different among provenances for the remaining site-by-species combinations that we tested, although most combinations trended toward greater growth in low-elevation seed lots. This limited differentiation was unexpected given that local adaptation along environmental gradients, even at relatively fine scales, is often observed for many of the species in our study, including Douglas fir (Campbell 1979, St Clair et al. 2005, ponderosa pine (Kitzmiller 2005), and sugar pine (Eckert et al. 2015).
The lack of significant differentiation in our study could be due to several factors. First, we may simply not have had enough statistical power to detect these differences. The planting sites reflect actual postfire conditions in which tree planting is conducted, and these conditions are variable. Our models detected and accounted for a generally strong negative influence of shrub cover on seedling growth (Table 2), consistent with results of manipulative studies in this system that identify strong competitive effects of shrubs both aboveground (i.e., for light) and belowground (i.e., for water and/or nutrients; e.g., Conard and Radosevich 1982). However, there are numerous other sources of potentially important environmental variation, including shrub height and species identity, soil characteristics, and microenvironments created by logs and other objects Spies 1997, Gray et al. 2005). This variability reflects the reality of managed landscapes, and in this regard our results reflect the outcomes that may be expected in a management context. Progressively increasing the sample size would eventually almost always achieve significance (Quinn and Keough 2002), but our sample sizes are already sufficient for detecting differences that are large enough in magnitude to be relevant in the context of postfire management (Table 2).

Limited geographic specificity of seed collection locations
Another potential explanation for the lack of substantial provenance differentiation is the uncertainty about the exact collection location and therefore the source climate of each provenance. USDA-FS seed lot designations (i.e., the intersection of the recorded seed zone and 500-ft elevation band) can cover wide geographic areas and a large range of climates ( Fig. 1; Appendix S1: Figs. S1-S6). Without any further information, the seeds of a given seed lot may have been collected from one extreme of the potential geographic/climatic space, the opposite extreme, or anywhere in between-each leading to different expectations of relative provenance performance. Further, seed lots often consist of seed collections from multiple locations (stands) within a seed zone and elevation band (Table 1). In all of the cases in which our study identified seed lot performance differentiation (growth and survival of Klamath incense cedar and San Bernardino Jeffrey pine), the potential source climate range of the best-performing provenance does not even include the planting site climate. Further, in the case of Klamath incense cedar survival, the potential source climate range of the worst-performing (mid-elevation) provenance does contain the planting site climate. Given the large range of potential source climate of each seed lot, it may be the case that seeds from the poorly performing seed lots were actually collected from a region of their potential climate space that was more dissimilar to the planting site than the collection locations of the other seed lots, potentially explaining the stronger performance of the other seed lots.
Our analyses further demonstrate that given limited information about collection locations, common assumptions (e.g., that lower-elevation provenance climates are warmer and drier) are not necessarily valid. All else equal, lower elevations are, indeed, generally hotter and drier than higher elevations. However, when comparing low-and high-elevation points at different locations within a seed zone, this pattern does not always hold. In fact, in comparing two 500-ft elevation bands separated by even 1000 ft or more of elevation, some sites in the higher-elevation band can actually be hotter and drier than some sites in the lower-elevation band (Fig. 2). This issue is exacerbated when comparing elevation bands from disparate seed zones.
Dependence on seed lots collected and/or cataloged with limited geographic specificity is a challenge common to many large land management institutions beyond the USDA-FS Pacific Southwest Region (e.g., Cal Fire; Stewart McMorrow, Cal Fire Deputy Chief of Forestry Assistance, personal communication). Given current staffing and financial resources, it is infeasible to store, track, and manage seed collections from individual source trees separately (Sara Wilson, former Seed Bank Manager, USDA-FS Pacific Southwest Region, personal communication). This reality highlights the fact that many principles of seed lot selection and assisted gene flow that are proposed and evaluated in the academic literature may not be directly applicable to large-scale management scenarios. It also highlights the importance of updating management practices to address the challenges of climate change.
Options for revising management practices include requiring greater geographic specificity when collecting and cataloging seeds (e.g., recording geographic coordinates of each parent tree or collection site), maintaining individual seed collections separately rather than pooling them (even when they originate from the same seed zone and elevation band), and developing quantitative seed transfer guidelines that incorporate empirical data on the climatic (and other environmental) tolerances of species and populations as well as expectations of climate change and associated uncertainty. Management agencies including the USDA-FS; Cal Fire; and the British Columbia, Canada, Ministry of Forests are already prioritizing many such revisions (Arnaldo Ferreira, Geneticist, and Sara Wilson, former Seed Bank Manager, USDA-FS Pacific Southwest Region, personal communications; Stewart McMorrow, Cal Fire Deputy Chief of Forestry Assistance, personal communication; Snetsinger 2004, O'Neill et al. 2017). Tools to facilitate climate-based seed transfer of precisely located seed collections have recently been developed (e.g., the Seedlot Selection Tool; seedlotselectiontool.org).
Nuances, challenges, and risks of implementing assisted gene flow It is possible that even small provenance differentiation will affect performance in the long term despite being less apparent in early years (Schuler 1994, Rice andKnapp 2008), when growth rates are relatively low and microenvironments can have a large influence (Gray and Spies 1997). This interpretation is supported by our observation that the species trial exhibiting the most growth by far (Jeffrey pine at the San Bernardino site) also showed significant provenance differentiation. Adaptive differentiation among provenances when trees are younger or smaller may be realized in phenotypic attributes that we did not measure, including phenology, freezing tolerance, xylem density, and rooting depth (Aitken and Adams 1997, Oleksyn et al. 1999, St Clair et al. 2005. While the seedling stage is often considered the most sensitive to environmental stress (Grubb 1977)-and thus potentially the most likely to exhibit performance differentiation -it is possible that any signals of environmental maladaptation were dampened by planting vigorous nursery-grown seedlings (Ledig and Kitzmiller 1992). To the extent that this is the case, it highlights the fact that planting tree seedlings can exclude opportunities for natural selection, and it emphasizes the importance of carefully selecting the genotypes used for seedling planting projects (Millar and Libby 1989).
An additional challenge in implementing assisted gene flow is that we often hope for the trees we plant today to persist on a site for decades to centuries, but climate change is expected to present a continuously moving target for at least decades to come (Aitken and Bemmels 2016). Thus, assisted gene flow may require choosing between (1) planting trees that may be well-adapted in the future but are not at present or (2) vice versa. Given the potential for limited cold hardiness in heat-and drought-adapted populations (Loehle 1998, Koehler et al. 2012, it may be advantageous to wait to move hot-and dry-provenance genotypes into historically cooler, wetter sites in response to (not in anticipation of) climate change, when rare cold events may be less common. This approach would only be effective, however, to the extent that trees can persist through some amount of climate change in their present sites.
Existing provenance studies-particularly those that involve planting trees of a given provenance into multiple environments-can also help to identify the extent to which provenances can be safely transferred in anticipation of future warming (Hufford and Mazer 2003). For example, a transplant study in Populus fremontii S. Watson determined that the spatial seed transfers necessary so that trees planted today are well-adapted in 100 yr would result in substantially reduced performance today (Grady et al. 2015). Ironically, the trees that can tolerate the largest transfers today in anticipation of future climate change (i.e., those that are the least climatically sensitive) may be the most able to tolerate climate change in the absence of assisted gene flow (Wang et al. 2006).
Additional risks surround implementation of assisted gene flow. Introgression between local and introduced populations could result in outbreeding depression (Weeks et al. 2011, Aitken andWhitlock 2013); alternatively, reproductive phenological mismatches between local and introduced genotypes could prevent desirable introduced alleles from establishing in the local population (Wadgymar and Weis 2017). Populations of tree species, including some of the species we studied, often exhibit local adaptation to factors that do not vary with climate (e.g., soil properties, photoperiod, pathogens, and mutualists; Wright 2007, der Putten 2012, Kranabetter et al. 2012, Way and Montgomery 2015, Grady et al. 2015), likely impacting the success of populations relocated specifically to track climate (Schiffers et al. 2013, Bucharova 2017. The existence of these other adaptations may provide an alternative explanation for our unexpected observations. Existing common garden studies designed to evaluate the extent of local adaptation to individual environmental attributes (e.g., climate or soil type) provide an important foundation for predicting outcomes of assisted gene flow. However, the most directly relevant information comes from studies that explicitly compare performance of relocated vs. local genotypes under altered climate. Such studies are relatively rare (Schreiber et al. 2013, Hancock and Hughes 2014, Bucharova et al. 2016. While results of existing and future research can help guide assisted gene flow decisions, there will always exist some uncertainty regarding the best tree provenance(s) for any given application. Uncertainty exists due to imperfect information regarding many important factors, including future climate, likelihood of introgression and outbreeding depression, and extent of local adaptation to non-climatic factors. Given this reality, assisted gene flow programs could be designed around a composite provenancing approach (Broadhurst et al. 2008) in which multiple provenances-each of which might potentially be appropriate under different assumptions-are combined. Such approaches, however, should account for the possibility of increased mortality (due to trees that prove to be maladapted; Ledig and Kitzmiller 1992) in determining planting densities and future follow-up management. Additional research into the potential outcomes and consequences of assisted gene flow-including studies of the climatic tolerances of tree populations and the strength of adaptation to non-climatic biophysical factors that trees may experience differently if relocated-could meaningfully inform seed selection decisions.