Integrating aerial images for in-season nitrogen management in a corn field

Methods of determining in-season corn (Zea mays L.) nitrogen (N) requirements and yield estimates are needed for designing a resource-efficient corn production system that is both profitable and environmentally sustainable. The objectives of this study were to examine: (1) the role of spectral signatures of corn plants obtained by aerial images in examining the yield variability across various N treatments, (2) whether the images could be used to guide in-season N management decisions, and to predict in-season corn yield and corn yield loss, and (3) the influence of spatial resolution of imagery on the accuracy of corn yield prediction models. Twenty-four treatments evaluated were the combinations of eight fertilization times (at-planting (A), pre-planting (P)∗A, P∗A∗mid-season (M), P∗A∗late-season (L), PAML, AM, AL, and AML) and three at-planting N rates (11, 45, and 78 kg N ha⁻¹). Visual and thermal images were collected from manned aircraft and geo-corrected for the analyses. Vegetation indices and ratios were derived from three waveband combinations of visual images, and they were examined in relation to yield. Two linear regression models - model 1 (based solely on imagery) and model 2 (based on imagery and information about elevation and N fertilizer application rate), were tested on their performances (in terms of coefficient of determination (R²) and root mean square error (RMSE)) for in-season corn yield prediction at four spatial resolutions (0.35, 0.5, 1, and 2 m px⁻¹). Among individual wavebands, and vegetation indices and ratio, plant pigment ratio (PPR) at early growth stages were highly correlated to corn yield, particularly in the field that received limited N application. The correlation improved as the corn growth stage progressed, but weakened towards the end of the growing season. There were significant differences in PPR values between the treatments receiving the least and the most N application, and it was the amount of N applied at planting that created the most significant differences. The models for 0.35 to 1 m px⁻¹ spatial resolutions did not show significant improvements in R² over the lowest ground resolutions (2 m px⁻¹) (differences in R² ≤ 0.05). The model 2 showed higher R² (up to 0.64 at tasseling stage) and lower RMSE than model 1. These results indicate that the models developed integrating spectral and spatial information from aerial imagery with the information about elevation and N application rate help improve in-season corn yield estimates under different N management practices.