On-farm evaluation of the potential of precision sulphur management of corn in Minnesota.

Abstract

Sulfur (S) is an essential nutrient for corn (Zea mays L.) production, playing a critical role in plant metabolism, enzyme activation, and protein synthesis. However, reductions in atmospheric S deposition due to stricter air pollution regulations have led to an increasing prevalence of S deficiencies in agricultural soils. This shift has heightened the importance of optimizing S management strategies to maintain high crop productivity while minimizing environmental impacts. Precision S management (PSM) seeks to address this challenge by aligning S applications with the spatial and temporal variability of crop demand and soil nutrient availability. By improving S use efficiency (SUE), PSM has the potential to enhance crop yield, economic returns, and environmental sustainability; however, little has been reported on the benefits of PSM under on-farm conditions for corn, especially in Minnesota.The objectives of this thesis were to 1) evaluate the potential agronomic, economic, and sustainability benefits of PSM strategies for corn under on-farm conditions in Minnesota and 2) identify the key factors influencing within-field variability in corn yield and responses to S applications using machine learning for potential management zone delineation. Field trials conducted across Minnesota in 2022 and 2023 examined corn responses to S application under five levels (0, 11.2, 22.4, 33.6, and 44.8 kg S ha⁻¹). Results revealed significant variability in agronomic optimum S rate (AOSR) and maximum return to S rate (MRSR), driven by field-specific conditions such as soil organic matter (SOM) content, soil texture, and topography. For fields with higher SOM and finer textures, optimal yields were achieved with minimal or no S fertilization, while sandy, low-SOM fields demonstrated greater responses to S application. Field-specific S management (FSSM) improved yields and economic returns compared to traditional farmer S practices (FSP), but in highly heterogeneous fields, site-specific S management (SSSM) further optimized outcomes by tailoring S applications to individual field zones. Two fields were selected for identifying key factors influencing yield variability under different rates of S applications. This foundational knowledge allows for the delineation of management zones based on limiting factors, paving the way for better-targeted strategies that optimize farmer outcomes. A stacking ensemble machine learning model was employed to predict corn yield using S application rates and soil- landscape variables. The stacking model, integrating Random Forest, XGBoost, Gradient Boosting, and Support Vector Regression (SVR), outperformed individual models in predictive accuracy, achieving R² values of up to 0.98. Critical factors such as spatial yield trends, relative elevation, slope, SOM, and cation exchange capacity (CEC) were identified as the most influential factors. These findings underscore the value of advanced machine learning techniques in not only predicting yield but also understanding the complex interplay of factors shaping S responses. By leveraging these insights, site-specific management zones can be developed, enabling more effective and sustainable nutrient management strategies tailored to the unique conditions of each field. By combining field experiments with data science approaches, this research underscores the potential of PSM to improve SUE, optimize crop productivity, and enhance economic returns while promoting environmental stewardship. The findings highlight the need for practical tools and decision-support systems to facilitate the implementation of PSM strategies and address barriers to on-farm adoption. This thesis contributes to the advancement of sustainable agriculture by providing actionable insights into optimizing nutrient management at both field-specific and site-specific scales, paving the way for innovative, data-driven solutions to meet the challenges of modern agriculture.