Trial designs for detecting spatial variability of treatment effects in on-farm precision experiments

Abstract

Precision agriculture involves the existence of spatial variability in crop response to input application. Field-scale experiments allow for exploring such variability. However, the interaction between the spatial variability of factors controlling crop response and the applied experimental design conditions the results. It is necessary to identify experimental designs that optimize the acquisition of reliable information on intra-field crop response. Experimental designs at field scale with different spatial resolutions were evaluated to estimate the spatial variability of crop response to input application. Spatial response patterns were simulated as an underlying process to generate yield maps. Geographically weighted regression (GWR) was used to estimate crop response patterns, which were compared with the underlying stochastic field. Designs with high spatial resolution better capture underlying spatial variability patterns across a wide range of considered spatial structures. Furthermore, checkerboard plot designs outperform strip designs as they enable detecting spatial variability in both directions. However, the agreement between GWR-estimated response maps and reference maps is sensitive to kernel selection and bandwidth.