Título:	Temporal phenomic of transgenic maize events under drought: different prediction models and cross-validations scenarios.
Autoria:	PEREIRA, H. D. NONATO, J. MOLTOCARO, R. C. R. YASSITEPE, J.
Afiliação:	HELCIO D. PEREIRA; JULIANA NONATO, UNIVERSIDADE ESTADUAL DE CAMPINAS; RAFAELA C RANGNI MOLTOCARO DUARTE, CNPMA; JULIANA YASSITEPE, UNIVERSIDADE ESTADUAL DE CAMPINAS.
Ano de publicação:	2025
Referência:	In: ANNUAL MAIZE GENETICS MEETING, 67., 2025, St. Louis. Program and abstracts. Beltsville: USDA, 2025.
Páginas:	p. 52.
Conteúdo:	Due to their importance and frequency, climate changes have guided breeding programs to develop tolerant cultivars, especially for water stress. Evaluating drought tolerance requires multiple assessments over time, which demands a significant investment in both time and labor. This study aimed to assess the effectiveness of using unmanned aerial systems (UAS) for predicting key phenotypic traits in maize and capturing the plant response to drought. Two trials – one irrigated and one under drought conditions - were carried out in the dry season of 2023 to evaluate different sensors, machine learning models, and prediction scenarios through temporal assessments of vegetation indices. Daily flights were performed at 12m heigh, using a UAS equipped with RGB and multispectral sensor. From these flights, 18 RGB and 13 multispectral orthomosaics were selected at various stages of the crop cycle for prediction purposes. Temporal best linear unbiased predictions (BLUPs) of 35 and 54 vegetation indices from RGB and multispectral sensors, respectively, were used across eight machine-learning models for predictive analysis. Genotypes were divided into training (80%) and validation (20%) datasets, utilizing 500 bootstrap resampling. The models were trained on data from the irrigated trial and tested on data from the drought trial. The results showed that heritability and R2 values for all traits under both irrigated and drought conditions were similar, indicating consistent genotypic contributions to trait variation across conditions. Despite this consistency, the trials yielded somewhat different results. The study underscored the importance of evaluating a wide range of vegetation indices, as correlations between and within these indices over different flight dates were found to be low to moderate. The RGB sensor proved to be more accurate for prediction compared to the multispectral sensor. Prediction accuracies in untested genotype and environment scenarios ranged from 0.40 to 0.70 for grain yield, 0.43 to 0.69 for days to anthesis, 0.51 to 0.67 for days to silking, and 0.35 to 0.57 for plant height. The ridge and random forest models consistently provided the most accurate predictions across different traits and environments. Among the vegetation indices, NGRDI, VARI, and RCC were particularly important for predicting and capturing plant responses to drought. UAS phenotyping emerged as a viable and reliable alternative for integrating abiotic stress assessments in biotechnology institutions and in low-resource or small-scale programs due to its simplicity and straightforward implementation
Thesagro:	Milho Planta Transgênica Seca
NAL Thesaurus:	Plant stress
Tipo do material:	Resumo em anais e proceedings
Acesso:	openAccess
Aparece nas coleções:	Resumo em anais de congresso (CNPMA)