Accounting for Soil Spatial Autocorrelation in the Design of Experimental Trials

Abstract

Soil heterogeneity complicates the design and analysis of field experiments. Block designs were developed for this purpose. However, the analysis of experimental results supposes that the residuals from the treatment are spatially independent and that within block variation is random. Experience indicates that this rarely is the case in field experiments, because of the strong spatial autocorrelation of soil properties. This paper applies geostatistical tools, such as variogram analysis and conditional stochastic simulation, to investigate the optimal experimental plot size and shape and to decide which experimental design is to be preferred. The methodology is illustrated using a case study of water‐use efficiency under semiarid conditions in Morocco. It was found that under these conditions an experiment with 16 treatments would use best a plot size of 4 by 8 m oriented north‐south, configured according to an incomplete block design with 8 plots per block oriented in two rows in the east‐west direction.