Vaishali Swaminathan, J. Alex Thomasson, Nithya Rajan, Amrit Shrestha, Jeffrey Siegfried Karem Meza Capcha, Xiongzhe Han Kangwon
Abstract
Nitrogen (N) availability and uptake, during the early growth stages in cotton, affect chlorophyll health, canopy development, maturity rate, and yield potential. Some of the visible symptoms of N stress include leaf chlorosis, stunted vertical growth, low canopy area, and early onset of senescence. Therefore, early detection of N stress is important for effective in-season N management, achieving yield goals and minimizing losses. The objectives of this study were (1) early prediction of N stress in cotton, and (2) yield prediction by tracking maturity rates. In this study, unmanned aerial vehicle (UAV) based multi-spectral images were used to capture spatiotemporal variabilities in a trial with four N application rates (0 lb./ac, 50 lb./ac, 100 lb./ac, 150 lb./ac). Normalized difference vegetation index (NDVI), normalized difference red edge index (NDRE), leaf area index (LAI), and canopy height were derived from the orthomosaic and digital elevation models. Early in the season, NDRE had the strongest correlation with total biomass N (R2 = 0.80), followed by LAI and NDVI (R2 = -0.46). Results for early detection of N stress indicated that K-Means classification based on weekly changes in two parameters (∆weekly NDRE and ∆weekly LAI) was able to distinctly separate the high and low treatment zones (silhouette score = 0.56), offering better clustering than absolute daily values of the parameters (silhouette score = 0.37). The harvested yield had a linear relationship with ∆weekly NDRE (R2 = 0.86), which was obtained around the time of flowering. The clustering algorithm based on ∆weekly NDRE and ∆weekly LAI during flowering stage was able to distinctly identify the different yield groups. Lastly, the K-Means classification algorithm was tested on an image dataset obtained at early boll opening to identify individual bolls visible on the canopy. The derived boll counts had a linear relationship with harvested yield, indicating the potential of early boll counts in predicting yield. The results showed that transforming the observed parameters to indicate growth rate (∆weekly) improved early season N stress prediction, as N availability is crucial for canopy expansion. Also, tracking maturity based on parameters obtained at the time of flowering improved yield prediction.
Comentarios