Strip differences account for 47% of the richness slope in relative biomass variance in 154 Cedar Creek plots, 2011–2021

We tested how spatial samples affect the relation between plant diversity and biomass variation at Cedar Creek, Minnesota. We used two strips in each of 154 plots with 1, 2, 4, 8 or 16 planted species in 2011–2021. An exact identity separates temporal variance of the strip mean into cross-strip covariance and one quarter of the variance of strip differences. We divided each component by the square of mean biomass. The observed richness slope was −0.01326 per doubling of planted species. Strip differences supplied −0.00628, or 47% of that slope. The shared component supplied −0.00698, with a 95% plot bootstrap interval of −0.01449 to +0.00017. The relation thus became smaller and less certain. However, the shared slope remained clearly negative after removal of linear trends and for adjacent-year squared changes. An independent CSV calculation matched all 3,388 strip totals and the primary variance estimates. Known-model simulations showed that covariance can remove independent sample variation, but shared sample errors can retain a false temporal relation. Strip differences include real spatial variation and sample processing differences. Their removal estimates whole-plot temporal variance only under error assumptions which two strips cannot identify. The result does not show that diversity has no stability effect. It provides a fixed-data audit and a method for separating components of measured variation. The package contains source data, provenance, code, all results, bootstrap draws, figures, and reproduction instructions.

biodiversity · Cedar Creek · grassland · biomass · temporal stability · spatial variation · sampling error · paired samples · variance decomposition · reproducibility