Determination of optimum resolution for predicting corn grain yield using sensor measurements

Abstract

Identifying the optimum resolution where differences in corn (Zea mays L.) grain yields are detectable could theoretically improve nitrogen (N) management, thereby resulting in economic and environmental benefits for producers and the public at large. The objective of this study was to determine the optimum resolution for prediction of corn grain yield using indirect sensor measurements. Corn rows, 15–30 m long, were randomly selected at three locations where the exact location of each plant was determined. In 2005 and 2006, four of eight rows at each location were fertilized with 150 kg N ha−1 as urea ammonium nitrate (28% N). A GreenSeeker™ optical sensor was used to determine average Normalized Difference Vegetation Index (NDVI) across a range of plants and over fixed distances (20, 40, 45.7, 60, 80, 91.4, 100, 120, 140, 160, 180, 200, 220, and 240 cm). Individual corn plants were harvested and grain yield was determined. Correlation of corn grain yield versus NDVI was evaluated over both increasing distances and increasing number of corn plants. Then, the squared correlation coefficients (rcc 2) from each plot (used as data) were fitted to a linear plateau model for each resolution treatment (fixed distance and number of corn plants). The linear-plateau model coefficient of determination (rlp 2) was maximized when averaged over every four plants in 2004 and 2006, and over 11 plants in 2005. Likewise, rlp 2 was maximized at a fixed distance of 95, 141, and 87 cm in 2004, 2005, and 2006, respectively. Averaged over sites and years, results from this study suggest that in order to treat spatial variability at the correct scale, the linear fixed distances should likely be <87 cm or <4 plants as an optimum resolution for detecting early-season differences in yield potential and making management decisions based on this resolution.