The Dilemma of Reducing Nitrogen Fertilizer Rate for Corn: Data Analysis of On-Farm Trials.

On-farm evaluations of different nitrogen (N) management practices using precision agriculture technologies have become common in Iowa. However, analysis of data collected in such trials is often complicated because the trials are conducted without following formal controlled experimental designs, and because farmers use different rates, forms, and timing of application.  We tested a statistical method for identifying site-specific factors that can help farmers decide where they can decrease their normal N rates to corn (Zea mays L.) by about one-third.  Thirty five on-farm replicated strip trials were conducted in 2006 and 22 trials in 2007. Each trial compared a normal N rate alternated with a rate that had 56 kg N ha^-1 less. Yield reductions (YR) from the reduced N rate were calculated at 35-m quadrants using 3 to 8 treatment pairs within each trial. Hierarchical models with Bayesian approach were used to identify which field-level factors (previous crop, rainfall, N form, and timing of application) and within field variables (soil drainage class, soil organic matter level, and slope) affected YR from the reduced N rate within and across the trials. We simulated posterior predictive distributions to calculate the probability of observing a given mean YR in fields that have not been observed. The analysis revealed large spatial variability in YR, suggesting about potential effects of within field variables on YR. In 2007 (wet spring), the YR was tended to be larger in fields received >300 mm spring rainfall and within areas with low soil organic matter contain. In 2006 (dry spring and summer), the YR was economically significant in fields that received >80 mm rainfall in June. The tested statistical method can be applied to different types of on-farm trials, and results should help farmers make better management decisions using the best information from their fields.