Spatial Variability Effects on Precision and Power of Forage Yield Estimation

Abstract

Spatial analyses of yield trials allow adjustment of cultivar means for spatial variation, improving the statistical precision of yield estimation. While the relative efficiency of spatial analysis has been frequently reported in several yield trials, its application to long‐term Lolium spp. forage yield trials has not been characterized. The objective of this study was to evaluate the trend analysis, nearest‐neighbor analysis (NNA), and correlated error (CE) models for their ability to account for spatial variability in 138 Lolium spp. forage yield trials. This case study was performed on data from five locations and 11 yr (2001–2011) using randomized complete block design (RCBD) trials conducted by the Department of Agriculture, Food and Marine (DAFM) in Ireland. The relative efficiencies of trend, NNA, and CE models compared with RCBD models were 129, 143, and 193% for analysis by trial × year, and 121, 125, and 171% for analysis by trial, respectively. When the top one, two, three, four, or five cultivar(s) were compared between CE and RCBD models, the agreement between two models to find common cultivars varied from 66% for the top cultivar to 28% for the top five cultivars. Using CE models, four replicates were sufficient to detect mean yield differences between cultivars of 7% of the mean and 80% power. Spatial analysis should be added to the routine DAFM testing programs, not only to improve the precision of yield estimates, but also to reduce the risk of missing potential candidate cultivars, given the existence of spatial variation.

Improving Genomic Prediction in Cassava Field Experiments Using Spatial Analysis

“…A spatial model that is adequate for one scenario may not be suitable for another. This property of spatial variance points out the importance of an exploratory approach to identify the best model for a given dataset ( Richter and Kroschewski 2012 ; Richter et al 2015 ; Sripathi et al 2017 ). In our results, the Gaussian kernel, which is quite different from the Power kernel, best explained the underlying spatial variation in most scenarios.…”

Differential survival of alfalfa varieties interseeded into corn silage

“…Final stand density data collected at the 5.5‐ and 12‐mo sampling dates for ARL‐WI were transformed as ( Y + 0.5) 1/2 to obtain homogeneity of variance and then analyzed as a repeated measure using an AR(1) covariance structure. The RS‐PA and EL‐MI sites exhibited considerable spatial variability in final stand counts, which was accounted for by the use of correlated error models (Sripathi et al., 2017). The RS‐PA site exhibited similar patterns of spatial variability at the 5.5‐ and 12‐mo sampling dates, so final stand count data were averaged across dates and ultimately analyzed using a power covariance model.…”

Applications of spatial models to ordinal data

“…However, contrary to results indicating large plot sizes are better, evidence from a 28-year case study for optimizing experimental designs show that the relative efficiency of the Randomized Complete Block Design increased by 240% as the length of two-row plots decreased from 5.6 m to 1.4 m [76]. Similar results from a comparison of different spatial models among correlated error, nearest neighbor analysis, and autoregressive regression AR (1) indicates that smaller plot size is more efficient to capture spatial variation and thus increase the relative efficiency of the experimental design [77]. Both small plot size and ordinal IDC scores may be related to the high residual standard error we found in application of the P-spline model to data sets 2 and 3.…”