A better understanding of processes involved in crop production as well as the use of models as tools for managing agricultural systems are main reasons for the development of growth models. During the last three decades, the complexity of models increased from simple input-output relationships to highly sophisticated mechanistic process oriented models. Nevertheless, today, the simpler crop yield models have received a wider application, e.g., in management oriented decisions for plant production and in the evaluation of field experiments at multiple scales. The spatial variability considerably influences the experimental error variance which is particularly relevant in highly managed production experiments and in breeding nurseries. Therefore, the purpose of this study was to predict the yield of barley based on regressions and analysis of variance/covariance (ANOVA, ANCOVA) in a long-term static nitrogen fertilizer experiment receiving six different forms of nitrogen and three levels of nitrogen. The factors were the level of fertilization and the applied N-fertilizer forms. As covariates, the apparent soil electrical conductivity (ECa), relief parameters, and location coordinates were used. These covariates were served as proxies for site conditions. The apparent conductivity was measured with different sensors (EM38, EM38-MK2) and in all possible configurations. The ANOVA indicated smaller R2-values and higher root mean square differences (RMSD) in comparison to the ANCOVA ((fertilized plots ANOVA: R2 = 0.21, RMSD = 5.23 dt ha−1; ANCOVA: R2 = 0.845, RMSD = 2.30 dt ha−1). Besides the form and level of fertilization, conductivity readings were the most important covariates accounting for the averaged long-term yields. Looking at the individual years, ECa occurred in the ANCOVA models in five of 11 years, whereas topographic covariates (mainly variables with the parameter slope in the derivation) appeared in eight years in the models. The introduction of plot-wise, time-invariant soil and relief parameters improved the discrimination of testing and modelling the treatment performance within the long-term field trial. The high degree of explanatory power in yield prediction delivered by static soil parameters makes them highly attractive and can contribute to an improved evaluation of production oriented field experiments.
Read full abstract