Development of a decision-making application for optimum soybean and maize fertilization strategies in Mato Grosso

• Published in: Computers and electronics in agriculture 2022-02, Vol.193, p.106659, Article 106659
• Main Authors: Rotundo, José L., Rech, Rafael, Cardoso, Marcelo Moraes, Fang, Yinan, Tang, Tom, Olson, Nick, Pyrik, Benjamin, Conrad, Gabe, Borras, Lucas, Mihura, Eduardo, Messina, Carlos D.
• Format: Article
• Language: English
• Subjects: Agricultural production; Algorithms; Applications programs; Clay minerals; Clay soils; Corn; Crop yield; Decision making; Fertilization; Fertilization response; Grain; Management decisions; Mathematical analysis; Mixed models; Model accuracy; Nitrogen; Nutrient management; Optimization; Soil chemistry; Soybeans; Web application
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2021.106659

•Mato Grosso accounts for a high proportion of soybean and maize production in Brazil.•Soybean P and maize N fertilization guidelines are lacking in this region.•Statistical algorithms were developed to predict yield responses to fertilization.•Soil pH and P predicted soybean yield response, while clay and planting date predicted maize.•Fertilization management guidelines were incorporated in a user-friendly web application. Mato Grosso accounts for 25 and 31% of soybean and maize grain production in Brazil, respectively. Despite the importance of this region, there is limited information to optimize fertilization management decisions for these crops. Our objectives were to i) quantify soybean and maize yield response to fertilization, ii) develop prediction algorithms to prescribe recommendations based on site characteristics, and iii) incorporate these algorithms in a decision-making application. A total of 37 soybean and 27 maize field trials (year by site combinations) were conducted to test yield responses of soybean to phosphorus (P) and potassium (K), and maize responses to nitrogen (N) and K. Soybean and maize showed significant yield increments to applied P and N, respectively, but K fertilization showed minor yield effects in both crops (and was not considered further). Yield responses to P and N in soybean and maize, respectively, were highly variable across environments, and non-linear mixed model analysis was used to test co-variates to optimize yield response predictions. Including soil pH and initial soil P for soybean, and planting date and soil clay for maize, helped reduce a significant fraction of the site-to-site random variation. Soybean yield response to P increased in soils with higher pH values, agreeing with the concept of more mineral P retention in acidic soils. Maize showed lower N fertilization yield responses in later plantings dates, agreeing with higher water stress typical of late plantings. Out-of-sample relative root mean square error (rRMSE) of predictions were 15.9 and 19.5% for soybean and maize, respectively, showing the model has adequate accuracy to predict crop yield responses to nutrient fertilizations across the region. These algorithms were incorporated in a user-friendly web application to facilitate utilization, and dynamically calculates optimum fertilization rates including economic inputs such as grain and fertilizer prices. Our field results provide basic information to optimize fertilization rates for soy