Spatial Patterns Confound Experiments in Orchard Crops

Spatial autocorrelation of biological process has been shown to reduce experimental precision in field experiments, but horticultural researchers rarely consider this effect when designing their experiments. We tested how spatial autocorrelation affects the precision of orchard experiments, measured as Type I error rate, by simulating 40000 experiments based on a data-set gathered of 36000 individual tree yields over four years (4000 iterations of ten plot layouts in each year). Spatial autocorrelation varied among years and was a significant factor in the precision of experiments. Block designs were generally less effective than grid-based or random designs at controlling Type I error rate. Designs based on large blocks (twenty-seven trees) or small blocks (three trees) were robust against the effects of spatial autocorrelation. More moderately sized block designs (nine trees) and blocks based on rows of trees–designs common in tree crop research–performed the most poorly. These findings clearly demonstrated that spatial autocorrelation is a significant factor in the precision of tree crops and therefore might cause spurious experimental results if not explicitly taken into account