Predictive Ability of Enviromic Modeling in GxE Interactions for Upland Rice Site Recommendations
Enviromics is an omics approach that investigates a phenomenon using all available environmental information. This study explores the use of enviromic covariates in studies of genotype x environment (GxE) interactions in upland rice in Brazil, utilizing a field trial dataset from 143 locations over 27 years, covering diverse environmental conditions. The platforms WorldClim, NASA POWER, and SoilGrids were used to extract data, resulting in 383 environmental covariates. The objective of this study was to evaluate the use of enviromic kernels to integrate GIS and genetic data for predicting upland rice productivity across Brazil and to determine the optimal number of environmental covariates required to ensure model accuracy and stability. The predictive abilities of the enviromic model peaked with around 81 covariates, stabilizing when all 383 were included, suggesting the importance of a comprehensive dataset for accurate predictions. Analysis reveals that environmental dissimilarities are more critical than geographical distance for genotypic variability, reinforcing the need to consider multiple covariates in predictive models. Heritability mapping revealed spatial variations, with regions of high heritability concentrated in southern Brazil, where genetic selection may be more efficient. The clustering of mega-environments was not efficient, highlighting the complexity of GxE interactions, and confirming that pixel-by-pixel enviromic models are a safer approach for recommending breeding actions for upland rice. This study suggests strategies to improve genotype selection for specific conditions, guiding the expansion of rice cultivation into new agricultural areas in Brazil. The findings also contribute to rice-growing regions worldwide, especially in countries cultivating upland rice under diverse conditions. Structured AbstractO_ST_ABSObjectiveC_ST_ABSThis study aimed to evaluate the predictive ability of enviromic models for site-specific recommendations in upland rice, focusing on genotype x environment (GxE) interactions by integrating environmental and phenotypic data. MethodsA total of 734 field trials conducted between 1995 and 2022 across 143 Brazilian locations were analyzed. Environmental data (383 covariates) were retrieved from WorldClim, NASA POWER, and SoilGrids using GIS-based procedures. Statistical analyses included mixed linear models and Random Forest to correct for design effects, estimate heritability, and perform spatial predictions. ResultsEnvironmental dissimilarity better explained genotypic ranking than geographic distance. Predictive ability plateaued after 81 covariates, but adding more covariates reduced variance and increased model stability, supporting the use of comprehensive environmental data. ConclusionsThe study reinforces the need for detailed environmental characterization and the use of comprehensive enviromic models. Pixel-based predictions are more reliable than broad clustering approaches, supporting the use of virtual trials to optimize breeding strategies and resource allocation.