Experimental Design
At each of our six sites, transplant locations were arranged in a nested design with three levels (plot, block, and grid), in order to systematically sample a large and representative variety of environmental conditions (Fig. S2). Three 5 x 5 m grids were placed within each site. Each grid contained 25 blocks that consisted of two, paired 0.09 m2 plots. Each paired plot received one of two treatments: (1) ‘abiotic only’, where the entire plot was cleared of all above ground vegetation as well the seedbank, or (2) ‘abiotic + biotic’, where plots were left intact. We cleared plots by clipping aboveground biomass, then disturbing the soil using trowels to remove an inch of topsoil and its associated seedbank (Germain et al.2017). We reduced cover by an average of 41% ± 0.89 SE, thus allowing for contrast between intact vs relaxed biotic interactions (Fig. S3, S4).
To test how persistence and occupancy varied across a productivity gradient, we measured ‘greenness’ of photos of every plot taken at peak biomass (May 2020) as a proxy for productivity at the block level (n=450). To do so, we used the Fiji package (Schindelin et al.2012) in Image J, choosing the ‘green band - red band’ vegetation index to measure productivity; this index best predicts aboveground biomass in low productivity systems, such as serpentine, on a global scale (Prabhakara et al.2015). Although the vegetation index spanned -22 to 10 , with harsher plots represented by higher numbers, for a more intuitive interpretation we multiplied our index by -1 so that more productive plots had higher numbers (Fig. S5). Although harvesting and weighing aboveground biomass would be a more direct measure of plot productivity, doing so was not possible as seeds were collected by local field assistants on emergency funds during the Covid-19 lockdowns.