Spatial Variation Affects Precision of Perennial Cool‐Season Forage Grass Trials

Abstract

AbstractRandomized complete block (RCB) designs, although commonly used in testing forage grass cultivars, often are inefficient. The objective of this study was to evaluate the RCB design, the lattice design, trend analysis, and nearest‐neighbor analysis (NNA) for their ability to account for spatial variability in 27 perennial cool‐season forage grass cultivar evaluation trials conducted over a 15‐yr period at Arlington, WI. Each trial was made up of cultivars or experimental populations of one of six species or groups of species: orchardgrass (Dactylis glomerata L.), ryegrasses [a group comprising perennial ryegrass, Lolium perenne L.; intermediate ryegrass, L. hybridum Hausskn.; and festulolium, ✕ Festulolium braunii (K. Richt.) A. Camus], smooth bromegrass (Bromus inermis Leyss.), reed canarygrass (Phalaris arundinacea L.), tall fescue (Festuca arundinacea Schreb.), and timothy (Phleum pratense L.). Dry matter forage yield was determined for 3 yr on each trial and expressed, for each plot, as mean annual yield. Yield data for each trial were analyzed by the RCB or lattice model, trend analysis, and two NNA models (one or two covariates). Three incremental improvements in precision of entry means were observed: an average of 15% due to the RCB design, an additional average of 17% due to the lattice design, and an additional average of 22 to 30% due to trend analysis or NNA. Trials were highly variable in the relative efficiency of both their experimental design and the spatial analyses, but this variability was not related to species, planting years, or plot size. The only useful predictor of trial efficiency was number of entries: three of five trials with eight or fewer entries failed to show significant differences among entries, regardless of the analysis. Spatial analysis methods cannot overcome lack of true differences among entries or failure to detect differences due to low degrees of freedom. Trend analysis or NNA appear to be useful mechanisms to account for intrablock variability in fields where it is impossible or uneconomical to predict proper blocking patterns.