Predicting Risk from Reducing Nitrogen Fertilization Using Hierarchical Models and On-Farm Data

Published in Agron. J. 105:85–94 (2013) doi:10.2134/agronj2012.0218 Available freely online through the author-supported open access option. Copyright © 2013 by the American Society of Agronomy, 5585 Guilford Road, Madison, WI 53711. All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. N management of corn under rainfed conditions in the U.S. Midwest has been studied extensively during the last 50 yr but has become a subject of intensive scientifi c and public debate only during the last two decades. Th e recent focus on N management is attributed to environmental concerns about pollution of water bodies by NO3 escaping from corn fi elds, emission of greenhouse gases (i.e., N2O) to the atmosphere, and the large amount of petroleum-based energy required to produce, transport, and apply N. Historically, N fertilizer recommendations for corn have been based on a simplifi ed empirical formula called the yield goal (Hoeft et al., 2000; Stanford, 1973). Th e major premise of yield goal recommendations is that corn N requirements or optimal N rates are proportional to corn yields, with a constant multiplier of 21.4 kg N Mg–1 corn grain. Th ese calculations were based on the assumption of a constant supply of N from the soil under a wide range of soil and weather conditions. While the yield goal recommendations were based on a mass balance approach (N rates should approximate N removed by grain plus adjustments for N losses and N supplied by the soil), several studies have shown a low correlation of corn yields and optimal N rates (Scharf et al., 2006; Vanotti and Bundy, 1994). Th e low correlation is oft en attributed to large variability in the N supply from the soil and variable N losses by diff erent mechanisms such as leaching, volatilization, or denitrifi cation. Another method for estimating N fertilizer needs for corn was developed in the late 1950s based on conducting the so-called yield response trials. Th is method considered applying a wide range of N rates in small-size plots, measuring the yield at each applied N rate, and fi tting a model (i.e., fi tting regression curves) to the yield values to calculate the rates at which the marginal increase in grain value would equal the marginal N fertilizer cost (Heady et al., 1955, p. 292–332; Voss, 1975). Th is calculation produced economically optimum N rates (EONR) that would maximize, aft er the fact, the return to N per unit of area. Unlike the yield goal approach, the economic optimization method indirectly considered the variability in N supplied from the soil and fl uctuations in the prices of N fertilizer and corn with time. Except when calibrating soil and plant tissue tests in soil fertility studies, the EONR method has seldom been used as the basis for N fertilizer recommendations in production agriculture. Recently, combined eff orts involving several land grant universities led to the creation of a multistate database of yield response trials to estimate the EONR for corn (Sawyer et al., 2006). Th ese data enable researchers to partially address large spatial and temporal variability in the EONR as well as fl uctuations in the prices of N fertilizer and corn grain. At least two issues remain to be resolved, however. Nitrogen yield ABSTRACT Current systems for developing N recommendations for corn (Zea mays L.) lack methods to quantify the eff ects of factors infl uencing yield responses to N and quantify the uncertainty in N recommendations. We utilized hierarchical modeling and Bayesian analysis to quantify the risk from reducing N to corn using on-farm observations. Across Iowa, farmers conducted 34 trials in 2006 and 22 trials in 2007. Each trial had a farmer’s normal N rate alternating with a reduced rate (by about 30% less) in three or more replications. Yield losses (YLs) from reduced N were calculated at 35-m intervals. Posterior distributions were used to identify across-fi eld and within-fi eld factors aff ecting YL and to quantify the risk of economic YL (>0.31 Mg ha–1) from reducing N in unobserved fi elds. In 2006 (dry May and June), the economic YL for corn aft er soybean (C-S) was predicted to be 20% larger than that for corn aft er corn. Also in 2006, C-S fi elds with above-normal June rainfall had economic YLs 35% larger than those with below-normal June rainfall, and sidedress applications were about 20% riskier than spring applications. In 2007 for C-S, N reductions with above-normal spring rainfall were riskier than with below-normal spring rainfall. Areas with higher soil organic matter (SOM) had economic YLs about 20% smaller than those with lower SOM. Many on-farm trials can be conducted across the state and the use of the proposed statistical methodology can improve decisions on where to reduce N applications across and within fi elds.