Analyzing Phenological Progression in Wheat Genotypes Through UAV Multispectral Imagery

Efficient crop management necessitates synchronizing irrigation, fertilization, and pest control with specific growth stages to optimize resource use and reduce losses. Monitoring wheat phenology is crucial for improving crop management and increasing yield. This study employs UAV multispectral data, enhanced with AI models, to conduct a temporal analysis of wheat phenological stages throughout the growing season. We extract Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge (NDRE), and Normalized Difference Water Index (NDWI) from 468 wheat plots over 18 different timestamps. Using zonal statistics (mean, minimum, maximum, standard deviation, variance, and range) for each plot, we achieve an overall peak accuracy of 92% with Random Forest and K-Nearest Neighbors in identifying wheat phenological stages, outperforming Decision Tree (89%), Support Vector Machine (82%), and Naive Bayes (80%). A generalizable classification model is developed that identifies these stages based on the temporal dynamics of vegetation indices across different wheat genotypes. We analyze spatial variability in phenological stages by creating a plot-level grid and examining temporal patterns and zonal statistics of NDVI, NDRE, and NDWI. This allows us to identify phenological transitions and generate stage maps for visualizing growth progression from planting to harvest. Stage classification model is further refined by incorporating environmental data and soil moisture levels. Development of a dynamic wheat phenology monitoring system based on these findings can benefit both farmers and researchers, improving wheat management and yield in stressed agricultural sectors.