Field-specific fertilizer rate optimization is known to be beneficial for improving farming profit, and profits can be further improved by dividing the field into smaller plots and applying site-specific rates across the field. Finding optimal rates for these plots is often based on data gathered from said plots, which is used to determine a yield response curve, telling us how much fertilizer needs to be applied to maximize yield. In related work, we use a Convolutional Neural Network, known as Hyper3DNetReg, to create plot-specific non-parametric yield response curves. In this research, we then use these curves to determine the optimal amount of fertilizer to be applied to specific plots. However, there are additional issues that need to be taken into account when designing optimal prescription maps. In addition to optimizing yield, we want to reduce strain on farming equipment by minimizing rate jumps between consecutive cells. This helps machines run more efficiently and last longer, thus reducing waste. Furthermore, when creating these optimized prescription maps we also aim to improve environmental impact by reducing the overall fertilizer applied, as excess nitrogen seeps into the soil and drains into our waterways, negatively affecting water quality. In previous work, we found that it is possible to reduce overall fertilizer applied by 5 to 10% when creating experimental prescription maps without significantly impacting yield. Therefore, we hypothesize this will hold true for optimized prescription maps as well. We address these three separate, competing objectives using an adjusted genetic algorithm, known as Non-Dominated Sorting Genetic Algorithm II (NSGA-II), which finds a set of potential solutions that are optimal for the combined objectives. Such solutions are known as Pareto optimal, where one of the objectives cannot be improved without negatively impacting at least one other objective. We further adjust NSGA-II to use the Factored Evolutionary Algorithm (FEA) framework, which decomposes the variables into separate, overlapping groups to increase exploration of the search space, as well as enabling the ability to parallelize computation.