Location: Grassland Soil and Water Research Laboratory
Title: Understanding within-field variation in Nitrogen Use Efficiency (NUE) using proximal sensing, field observations, and machine learningAuthor
Submitted to: Meeting Abstract
Publication Type: Abstract Only Publication Acceptance Date: 2/5/2024 Publication Date: N/A Citation: N/A Interpretive Summary: Nitrogen (N) fertilizer input is a major decision farmers make to boost their crop yield. A better understanding of N use efficiency (NUE) across fields can help farmers optimize N fertilizer use to reduce N loss and off-site environmental impacts, and potentially increase farm income. We mapped NUE across nine corn fields in Riesel, Texas using grain yield, grain protein content, N-fertilization, and field soil observations. The fields were divided into four NUE zones (Low, Low-mid, Mid-high, and High NUE) that could be used for site-specific N-management decisions. Technical Abstract: Nitrogen (N) fertilizer input is a major decision farmers make to boost their crop yield. A better understanding of N use efficiency (NUE) can help farmers reduce N loss and off-site environmental impacts, and potentially increase farm income. The amount, timing, and application of N-fertilizers and their crop yield response have been studied but less has been understood about within-field NUE variation, and how proximal sensing and machine learning can help achieve this goal. Our objectives were (i) to test the usefulness of the NIR spectrometer to quantify grain protein content on the go, (ii) to calculate grain N and NUE using yield, grain protein, and field soil observations, (iii) to predict and map NUE using apparent electrical conductivity (ECa), topography, and machine learning, and (iv) to divide the field into NUE zones for precision management decisions. The study was conducted in nine corn fields in Texas Blackland Prairie soils in 2023. Corn yield and grain protein content were measured using a combine equipped with an NIR spectrometer to monitor grain protein content during harvest. NUE was calculated from grain yield and its N content, N-fertilization, and soil N observations. A random forest model was utilized to predict and map NUE where ECa, and terrain attributes were used as covariates, and the model was evaluated on 25% test data. Each field was divided into four NUE zones (Low, Low-mid, Mid-high, and High-efficiency zones) that could be used for site-specific N-management decisions. Corn yield and field gross margin (revenue – input cost) from the zones were compared across fields and zones. The ECa and topography explained up to 57% of variations in NUE. The NUE maps showed that almost all areas of the 6-12 field and >60% of the SW-16 field had Low NUE whereas, >75% area of the Y-10 had Mid-High to High NUE. We will test the usefulness of remote sensing data as predictors of NUE in the future. Results from this study will help farmers optimize N-applications for economic and environmental benefits. |