Monitoring Nitrogen Levels in the Cotton Canopy using Real-Time Active-Illumination Spectral Sensing

Managing nitrogen (N) fertilizer is fundamental to efficient cotton production. Traditional N management strategies often utilize N inefficiently through sub-optimal rate prescriptions and inappropriately timed applications. This leads to reduced production efficiency and increased environmental risk. Both deficiency and excess of N in cotton crop negatively affects lint yield and fiber quality. Thus, the aim is to monitor in-season cotton N levels in real-time at a growth stage where supplemental N can be applied. Research has shown high correlation of cotton leaf N concentrations with spectral reflectance of plants. The GreenSeeker® sensor is a ground-based active-light sensor developed to nondestructively evaluate N status in crops. However, the Normalized Difference Vegetation Index (NDVI) reported by the sensor is subject to influence by the soil background. The objective of this research was to develop an algorithm that improves a ground-based sensing system’s ability to discriminate between plant biomass and soil, allowing it to better estimate N status in cotton. Three cotton varieties, three seeding rates, and four N rates were established in a field experiment in Milan, TN. GreenSeeker readings and ultrasonic plant height data were collected and analyzed to investigate the influence of these crop management factors on NDVI. Strong positive correlation (r>0.72) between NDVI and plant height was confirmed. Seeding rate affected NDVI throughout the season, confirming an effect of soil background noise on NDVI values. To aid in algorithm creation, NDVI data were collected from a subset of plots, the plant population was thinned, and re-sensed. Difference in NDVI of these populations was minimized when data below a threshold was removed prior to index calculation. Two algorithms were identified that reduced vegetation indices difference to within the published error of the sensor. The reduction of plant population effect on NDVI was validated by post-processing a larger data set using both algorithms.