Agricultural producers are tasked with managing crop yield responses to nitrogen (N) within systems that have high levels of spatial (biophysical), climatic, and price uncertainty. To date, the outcome of most variable rate application (VRA) research has focused on the spatial dimension, proposing optimal fertilizer prescription maps that can be applied year after year. However, temporally static prescriptions can result in suboptimal outcomes, particularly if they do not consider the impact and likelihood of alternative weather or price regimes that can drastically alter crop responses and net returns. Furthermore, most optimizations are built on the assumption of linear crop responses when non-linearity may be more biologically appropriate and could result in altered N prescriptions. 

     In this presentation, we outline our methodology to address these uncertainties using a non-linear spatiotemporal Bayesian updating framework. This strategy continually improves N optimizations, increases net returns and reduces uncertainty in the parameter estimates. The framework is able to quantify the probabilities of different net return outcomes, allowing the producer to choose their N management based on their particular level of risk adversity. It also enables the producer or researcher to assess the impacts of future scenarios such as prolonged drought or price fluctuations.

     This methodology was tested within a simulation to assess the number of years required for model convergence and enhanced net returns. It was then applied to the years 1980 – 1992 to hindcast the impact of extended drought in Montana during 1987-1991. Simulated crop responses incorporated realistic levels of residual variability based on ten years of observations from a dryland wheat farm located near Great Falls, Montana. For simplicity, the crop was assumed to respond non-linearly to variation in soil apparent electrical conductivity (EC_a), applied nitrogen (N), and precipitation. Historical wheat price data from this region also informed the model and served as an additional source of variability that impacted the net returns.   

     Parameter convergence and net returns higher than those of uniform fertilization were achieved after six to eight years, resulting in a spatial net return benefit of $23-25/hectare. After year six, the spatial random effects in the model

effectively eliminated the confounding influence of spatial autocorrelation on the crop response coefficients. Small experimental N rate treatments (0, 60, 120, 180 kg/ha) were randomly applied each year as a part of this framework to ensure that crop responses to N were explored under the full space of possible soil and precipitation conditions. These strip experiments reduced the time required for convergence of the parameter estimates. 

            During the late 1980s, the severe drought in Montana reduced hypothetical savings from a level of $450,000 in 1983 to below zero as early as 1988. The impacts on savings are mirrored in governmental data on farm bankruptcies during this period. Substantial variability remained around the estimates for the different fertilization scenarios; however the optimized fertilizer prescriptions consistently out-performed the uniform prescriptions on a field-wide basis. With a nominal level of governmental price support, producers spatially optimizing their N inputs would have survived the drought. Producers applying uniform levels of fertilizer would have increased levels of debt, especially under low and high input levels.

     This simulation study demonstrated a useful decision aid framework that can empower agricultural producers with site-specific management that accounts for the range of possible uncertainties producers must face. Decision support tools must be applicable across years rather than being optimal under only one set of climatic conditions. Decision support tools must use crop response functions that are biologically appropriate yet statistically tractable. Finally, the decision aid must acknowledge the variability not only in crop responses, but also the variability in crop prices that has a strong impact on net returns and management strategies. With the uncertainty associated with future climates, an approach for monitoring system agronomic and economic performance is crucial for maintaining resilient agro ecosystems. The framework developed here meets all of these requirements and can be easily adapted to incorporate additional driving variables or alternative crop response functions. By providing a flexible platform for progressively refining system parameters and optimizing spatial N prescriptions, this research provides a baseline tool that may be useful to producers across a wide range of crops and growing conditions.