Enhancing enviromics based predictions in common bean multi-environment trials

Abstract

Enviromic approaches enhance predictive models by incorporating environmental data into selection frameworks. By integrating factor analytic (FA) models, enviromics, and Geographic Information Systems (GIS), the GIS-FA method was proposed to improve genotype prediction in untested environments. This study aims to refine GIS-FA by implementing Random Forest Spatial Interpolation for enhanced environmental data interpolation and optimizing spatial sampling to exclude non-agricultural areas, thereby improving environmental characterization. We applied the improved GIS-FA framework to common bean trials conducted across 23 environments in São Paulo, Brazil, evaluating 59 genotypes from the “Carioca” and “Black” market classes. The enhanced method increased empirical Best Linear Unbiased Predictions (eBLUPs) accuracy from 0.46 to 0.53 in leave-one-out cross-validation, representing a 15.2% improvement and enabling more reliable genotype performance predictions. Additionally, integrating GIS-FA with Factor Analytic Selection Tools improved the interpretation of stability and adaptability metrics by allowing predictions at untested locations. This provided a comprehensive spatial view of genotype performance across the entire Target Population of Environments (TPE). High-resolution thematic maps generated through GIS-FA facilitated genotype recommendation across São Paulo. These findings demonstrate the value of incorporating machine learning-based interpolation and spatial optimization into GIS-FA, reinforcing its potential to support selection strategies and advance environment-informed prediction in modern plant breeding.