Nitrogen economy in corn-soybean farming systems

Abstract

Nitrogen (N) is the most limiting nutrient for producing maize (Zea mays L.) and soybean [Glycine max (L.) Merr.] crops. The complex system governing the soil-plant N dynamics requires exploring multiple perspectives. Concomitantly, there is a marked need to deploy data-driven models that account for uncertainty in the processes of interest to provide improved N recommendations in both crops. Therefore, the objectives of this dissertation were: (i) to assess the contribution of environmental and crop management factors on the prediction of inherent maize productivity without N fertilizer; (ii) to identify the main drivers of both, expected values and uncertainties, of key components describing the process models for the maize yield response to N fertilizer; (iii) to summarize the impact of N and water management practices in maize grain quality; (iv) to study the residual effects of N management in maize on the following soybean crop; and, (v) to evaluate statistical techniques for the assessment of agreement between predictions and observations. In a joint effort between different academic and industry institutions in the US and Canada, a database with more than 1,200 maize N fertilization experiments (1999-2019) was built. Crop management factors such as previous crop and irrigation in combination with soil organic matter contributed to explain half of the variability of maize yield without N fertilization, while including spring weather variables (March-May) resulted in a similar performance than a framework including weather during the entire season. Crop management factors largely affected the prediction of the expected yield without N fertilizer, but just slightly impacted (<5%) the uncertainty of the response (and their components) of yield to N fertilizer. Conversely, weather variables were, undeniably, the most relevant factors and roughly contributing to 80% of the explained variance to predict the uncertainties on the yield response to N. On the other hand, a meta-analysis using a database of 92 site-years revealed that N fertilization not only increases yields but also shows a positive impact on the grain protein concentration, however, both starch and oil remained relatively constant under contrasting N fertilization levels. In contrast, water stress resulted in an erratic effect on all the evaluated grain quality components, possibly due to changes in the moment, severity, and extent of the stress. Evaluating two case studies under a maize-soybean rotation in Kansas, we documented that N fixation and soybean yields were marginally or not affected by the N management in the previous crop. Lastly, a novel and simple methodology on the use of linear regression to assess the prediction ability of simulation models is presented, also suggesting a derived decomposition of the prediction error into lack of accuracy and lack of precision along with the R-code to assist potential users. Forthcoming projects on N economy in maize and soybean farming systems should expand, provide incentives, and discuss standards in collaborative research, which represented a foundational component of this project. This dissertation highlights the advantages of deploying cutting-edge data analysis techniques for addressing research gaps on the N economy in maize-soybean farming systems. Machine learning, meta-analysis, and Bayesian statistics bring new horizons for improving forecast models as well as their interpretability. Future generations of predictive models in agriculture must be able to capture complex interactions as well as to emulate how farmers deal with uncertainties in the real world. Under this context, the awareness about uncertainties and their drivers should become one of the pillars of the dynamic N recommendations, which is crucial to convey wise information to stakeholders. Undoubtedly, we must move from static to dynamic crop models in order to design optimized GxM adaptation strategies under future climates.