Waiting for main navigation ...

Improved evaluation of field experiments by accounting for inherent soil variability (2017)

Heil K., Schmidhalter U.

European Journal of Agronomy, 89 (), 1-15



Well-controlled field experiments are used to test agronomic management practices and evaluate the performance of cultivars in highly managed plots at experimental stations, in breeding nurseries or on-farm. However, the performance of crops and therefore the interpretation of experiments is affected by the inherent soil variability. To avoid large residual errors, replicate measurements or optimized designs are usually helpful but seldom adequately consider the unknown soil variability. The use of spatial covariates, such as proximally sensed data, in the statistical modelling of the target variable may provide a better estimate of such experimental residual variations (errors). Therefore, the purpose of this study was to determine whether the apparent soil electrical conductivity, topographical parameters and location information (expressed as Gauß-Krüger coordinates) could be used for an enhanced spatial and temporal characterization of the long-term and annual wheat yields within a static, long-term nitrogen fertilizer experiment that included six different forms of nitrogen and three levels of nitrogen fertilizer. Furthermore, this investigation aimed to propose statistical strategies for analysing this background variation by testing ANOVA (Analysis of variance) and ANCOVA (Analysis of covariance). ANCOVA with soil ECa, location information and topographic parameters as covariates improved the accuracy of the yield estimates of the multi-annual means for all treatments. Without these independent variables in ANOVA, the coefficient of determination (R2) was smaller and the root mean square difference (RMSD) was larger than those of ANCOVA (fertilized plots ANOVA: R2 = 0.19, RMSD = 3.26 dt ha−1; ANCOVA: R2 = 0.87, RMSD = 1.29 dt ha−1). In addition to the factor level of fertilization and form of nitrogen fertilizer, ECa was the dominant covariate for the averaged long-term and annual yields. The ECa was measured with different sensors and configurations and represented a significant independent variable. Of the topographic relief parameters, the predictor plancurvature was the dominant independent variable. The inclusion of plot-wise, time-invariant soil and relief parameters significantly improved the discrimination of testing the treatment performance within the long-term field trial. A further application of this approach to other experimental sites and breeding nurseries would likely be highly rewarding.

Intelligence for Soil (I4S)