Simulated photovoltaic solar panels alters the seed bank survival of desert annual plant species Supporting Information

Seed bank survival underpins plant population persistence but studies on seed bank trait-environment interactions are few. Changes in environmental conditions relevant to seed banks occur in desert ecosystems owing to solar energy development. We developed a conceptual model of seed bank survival to complement methodologies using in-situ seed bank packets. Using this framework, we quantified the seed bank survival of two closely related annual desert plant species, one rare (Eriophyllum mohavense) and one common (Eriophyllum wallacei) and the seed bank-environment interactions of these two species in the Mojave Desert within a system that emulates microhabitat variation associated with solar energy development. We tracked 4,860 seeds buried across 540 seed packets and found, averaged across both species, that seed bank survival was 21% and 6% for the first and second growing seasons, respectively. After two growing seasons, the rare annual had a significantly greater seed bank survival (10%) than the common annual (2%). Seed bank survival, across both species, was significantly greater in Shade (10%) microhabitats compared to Runoff (5%) microhabitats and Control microhabitats (3%). Our study confers insight into this early life-stage across rare and common congeners and their environmental interactions using a novel conceptual framework for seed bank survival.


Supporting Information -Methods
Supporting Methods I -Artificial photovoltaic installation. Thirty-two of these panels were installed in 2011 and were reallocated to this experiment in 2016, when we also installed four additional panels per site, for a total of twenty per site. We covered all panels with clear plastic sheeting (4 mm Coroplast, corrugatedplastics.net, New Jersey, USA) in summer 2016 to emulate the smooth surface of a PV panel and facilitate rainfall runoff. Within sites, plots were selected to minimize heterogeneity of substrate and slope; due to patchy distribution of annual species in shrub interspaces, plot locations were chosen non-randomly to contain threshold numbers of focal species, ensuring habitat conditions suitable for seed germination. All plots were established in areas where they would not be shaded by nearby shrubs or the infrastructure associated with nearby plots.
Supporting Methods II -Staining Assays. Formal assays were carried out during summer on seed recovered from packets collected the previous spring, with one exception: resource constraints delayed assay of the 2016 cohort collected in spring 2017 until the summer of 2018. However, staining results for this cohort do not suggest that additional storage time negatively affected seeds. Specifically, we found no differences in staining rate for E. mohavense cohorts recovered in 2017, and observed a higher staining rate for the 2016 E. wallacei cohort recovered in 2017. Before formal assays, intact seeds were imbibed in deionoized water for 24 hours. We prepared a 1% solution of 2,3,5-triphenyltetrazolium chloride and deionized water, and cut seeds longitudinally using a precision knife (Xacto #11 blade) to expose the embryo and pericarp. E. wallacei seeds were soaked in solution for 24 hours at 17º C, and E. mohavense seeds were soaked for 6 hours at 35º C. Within 1 h following soak, all exposed embryos were examined under a high-power stereoscope (SMZ800, Nikon Inc., Tokyo, Japan). The intensity and completeness of embryo staining varied among individuals as well as across species, so we classified seed according to presence or absence of stain. Individuals with completely white embryos were considered retained dead seed, and those exhibiting any stain were considered retained live seed (Fig. 2b). Effectiveness of seed viability assays may differ across species and thus similar methodological assessments should be performed to evaluate the accuracy of viability-based observations for individual plant species.
Supporting Methods III -Statistical Analysis. We built quasibinomial generalized linear models (GLMs) with logit link functions to evaluate retained seed pools and seed staining rates (version 1.2.5042, Rstudio, Boston, Massachusetts, USA). We used the Anova function in the car package (Fox and Weisberg 2011) to evaluate models and generate Type III p-values, and conducted post-hoc tests on estimated marginal means using the emmeans package (Lenth 2019).
In the GLM evaluating the retained seed pool (Fig. 2a, see C), the proportion of retained seed per packet was the response, and proportions were weighted by the number of seeds recovered from a given microhabitat and plot (combining seed of the same cohort where multiple packets were collected in the same location). Year, species, microhabitat, seed cohort, and all interactions were included as fixed effects; plot was not included as a blocking effect because blocks were incomplete. Although quasibinomial approaches are recommended to compensate for overdispersion (Carruthers et al., 2008), overdispersion could not be eliminated, so p-values should be regarded as approximate.
The GLM evaluating seed survival (Fig. 2a, see D) used stain presence on individual seeds as the response variable (stain present or absent). Fixed effects included year, species, microhabitat, seed cohort, and all interactions. To test for differences in seed bank survival by burial duration (2017 -one growing season, 2018 -two growing seasons) between the rare and common species (including differences across all treatments and within control plots only), we used a nonparametric Mann-Whitney U test on two medians using ranks of the sample data, as comparative datasets were not normal (e.g., W =0.52317, p-value = 0.00112, Shapiro-Wilk normality test). To test for differences in seed bank survival across microhabitats by burial duration (2017 -one growing season, 2018 -two growing seasons), we used a Kruskal-Wallis test (with Dunn's multiple comparison post hoc test) on the equality of medians, as these datasets were also not normal (Shapiro-Wilk normality test).

Literature Cited
Baskin   ) for the 2016 seed cohort. Data points overlaid on boxplots show the number of packets collected from each microhabitat, and the numbers above each boxplot show the total number of seeds recovered from collected packets. Where letters above boxplots differ, the percentages of retained seed recovered were significantly different at the p <0.05 level. Retained seed pools broken down by species, cohort, and microhabitat are provided in Table S3.  Table S5 for full seed bank survival calculations). Figure S4. The seed bank survival model showing empirical seed bank pools and types in the Control and Shade microhabitats for (a) E. mohavense and (b) E. wallacei (averaged across cohorts for each species) after two years of burial. We observed higher seed retention in the Shade compared to the other two microhabitats (we show only Shade and Control flows here; flows in the Runoff microhabitat are very similar to Control flows). We cannot confidently partition decayed seed (A) from germinated seed (B) in the expended seed pool (due to the delay between the winter annual germination period and collection of packets in spring), so we visualize these flows as equivalent in size. Flows exiting the staining assay (pink chevron) visualize the percentage of live seed for a subset of the retained seed pools (C) exposed to staining assays.     Table S5. Retained seed pools, staining rates, and calculated seed bank survival (%) from field data. (a) Empirical values by year and species (averaged across cohorts and microhabitats); (b) empirical values by year and microhabitat (averaged across species and cohorts). Retained seed pools and seed staining rates broken down by species, year, cohort, and microhabitat are provided in Tables S3 and S4.
Year collected

Retained seed pool
Staining rate Seed bank survival