In this study, we developed and validated an optimized phenotypic analysis method using time-series data collected throughout the full growth cycle of 96 rice cultivars. Height growth curves were compared across three major phenotyping tools—ImageJ, OpenCV, and PlantCV—each showing unique performance characteristics at different stages. ImageJ displayed significant variability in early growth, while OpenCV suffered from decreased accuracy during later stages. In contrast, PlantCV provided stable and consistent results across all growth stages, identifying it as the most reliable tool for phenotypic analysis in this context. The study further examined the effects of replicate sampling, camera angle selection, and outlier removal on data variability and error rates. Results indicated that replicate sampling alone was insufficient to control variability; however, combining optimized processing techniques, particularly angle selection and outlier exclusion, substantially improved data reliability and precision. Outlier removal, alongside selecting maximum angle values, contributed to smoother growth curves with reduced variability, enhancing data robustness. The reliability of these phenotypic traits was further validated through genome-wide association studies (GWAS), leveraging 12,127 SNPs identified via a deep learning-based GBS (Genotyping-by-Sequencing) pipeline. GWAS analysis highlighted significant SNPs associated with rice height on chromosomes 3, 6, and 9. Notably, genes OsBIG, OsGH9B3, OsSTRL2, and OsCCS52A were identified, confirming prior findings and linking these genes to growth hormone biosynthesis pathways. This optimized phenotypic analysis method, therefore, proves valuable in accurately identifying trait-associated markers, demonstrating its potential utility not only in rice studies but also in other crop research.

Keywords: Phenotyping, Time series, Data optimization, Rice, Trait analysis