Appropriate on-farm trial designs for precision farming

Most on-farm trials are based on classical experimental designs. Replication, blocking, and randomization provide sensitive and unbiased estimates of effects and comparisons in the presence of spatial correlation. The time and effort required by this methodology discourage producers hoping to conduct on-farm trials. Precision farming technology has reduced the cost of collecting on-farm data. Yield monitors and other sensors provide many observations per hectare cheaply with little disruption of field operations. But yield monitor and other sensor data typically violates classical assumptions because it is spatially correlated. The objectives of this paper are to describe a method for testing alternative experimental designs for dealing with spatially correlated data and to report on a preliminary test of this method. Using field trial data from Argentina, corn yields were simulated at two levels of spatial autocorrelation under an alternative experimental design with no treatment replications were compared to simulated yields under a design with five blocks with the three N treatments randomly positioned within each block. The results of 100 runs of this simulation show that both designs are quite conservative in their identification of spatial variability of response and profitability of VRN.