Seeds of Success (SOS) is a national seed collection program led by the Bureau of Land Management. SOS represents the most comprehensive native seed repository in the US, supporting native plant restoration, management, and research. Since inception in 2000, SOS has collected seeds from over 24,400 native plant populations from ~5,600 taxa from 43 states. Collections include species important to wildlife, pollinators, and indigenous people, and over 10,000 collections have been shared for restoration and research use. We asked how many SOS sites have burned since collection, and identified 662 fires at 631 sites. If fire continues at the pace observed since 2011, an estimated 14% of collection sites will burn by 2050 and over 24% by 2080, putting genetic diversity at risk in areas where fire is linked with invasion. Analysis of 14 native forb species from the western US found that many collections were from the warmest and driest portions of their range, areas at the highest risk of wildfire, subsequent invasion, and local extinction. SOS provides an opportunity to understand change in natural populations, and represents a critical repository of native plant genetic resources for conservation and future use.

Seeds of Success programmatic information

SOS collection data from 2000-2018 were compiled from 24,462 records. We tabulated the number of taxa, genera, and plant families collected, the number of collection interns since program inception, the number of US states with SOS collections, and the number of seed collections that have been distributed for restoration and research use, through 2017. We also tabulated the number of collections per year, the number of plants sampled for each collection, and the estimated population size of each collection, and compared these metrics from the first years of the SOS program (2000-2008) to the most recent years (2009-2017) using general linear models.

Quantifying fires at Seeds of Success collection sites

We obtained geospatial fire records for fires occurring across the United States, spanning the years 2000-2018, from GeoMAC (Geospatial Multi-Agency Coordination - Wildland Fire Support accessed: 21 Nov 2018 (2000-2017), 16 May 2019 (2018), https://www.geomac.gov/). We used these data to determine how many of a subset of 18,202 populations that have been georeferenced to-date have been impacted by fire since the date of collection.

Case study of western forbs

We performed in-depth analyses on 14 focal forb species known to be specifically or generally important for providing resources for sagebrush obligate wildlife in western states (Connelly et al. 2011; Dumroese et al. 2015; Wood et al. 2015; Dumroese et al. 2016), and for their current and potential use in seed increase efforts for restoration purposes. Species included: Amsinckia tessellata, Astragalus filipes, Balsamorhiza sagittata, Chaenactis douglasii, Cleome lutea, Crepis acuminata, Erigeron pumilus, Erigeron speciosus, Heliomeris multiflora, Lomatium dissectum, Machaeranthera canescens, Penstemon speciosus, Phacelia hastata, and Sphaeralcea grossulariifolia.

Estimating species distributions

We created binary species distribution maps for each focal species using a Maxent modeling approach, as outlined in Barga et al. (2018). Briefly, we gathered occurrence points from georeferenced herbarium data and researcher-documented point locations, and gathered environmental variables that were created using a Thornwaite water balance approach, based on PRISM temperature and precipitation data (Daly et al. 2008). Next, we separated the presence data for each species into test and training groups and created an iterative series of Maxent models that varied in both feature type(s) and regularization parameter. The best model for predicting the suitable climate for each species was selected, based on Akaike’s information criterion (AIC) values created using ENMTools (Warren, Glor, & Turelli 2010), and used to estimate the geographic distribution for each species. For more information on the calculation of environmental variables, model background selection, model optimization, and model selection, see Barga et al. (2018).

Suitable habitat, fire history, and seed collection locations

To ask how collection locations were distributed across each species’ potentially suitable habitat, we used binary species distribution maps for each of our focal species to select 500 random points across their estimated area of occupancy using ArcMap 10.5 (ESRI, 2017). We then extracted values for four environmental variables (annual precipitation (mm), summer (June, July, August) precipitation (mm), annual minimum and maximum temperature (ᵒC)) from both the SOS collection locations for each species and at the random points across each species’ distribution. These variables were selected because previous work indicated that they were highly predictive of suitable climate for multiple Great Basin forbs (Barga et al. 2018). We analyzed differences in environmental variables between SOS collection sites and random points using general linear models. We also asked what proportion of the potential habitat of our focal species was affected by fire across this 18 year timeframe based on the binary species distribution maps.

Conditions across seed collection locations vs areas of restoration seeding

We acquired geospatial records for all of the implemented or planned restoration seedings on public lands that were available from the Land Treatment Data Library, for the timeframe of 2000-2017 (Pilliod & Welty 2013). We used ArcMap 10.5 to extract values for the same four temperature and precipitation values (described above) from 1500 random points distributed across the area of the restoration seedings. We then determined significant differences between the environmental values found at the SOS collection locations and the random points across restoration seedings using general linear models.

Analyses were conducted in Program R (R Development Core Team 2016) using the car package (Fox & Weisberg 2011) and Agricolae package (de Mendiburu 2016) and a P-value of 0.05 was used to determine statistical significance.