Enviromic prediction is useful to define the limits of climate adaptation: A case study of common bean in Brazil

Abstract

Ongoing changes in the global environmental conditions foster plant breeding research to develop climate-smart cultivars as fast as possible. Data analytics are essential for achieving this goal, especially the so-called science of enviromics (large-scale environmental characterization of crop growing conditions) that could be used to pinpoint the relevant environment impacts driving the adaptation of a certain specie in a breeding framework. Here we quantified the effects of diverse climate factors on the current adaptation of elite common bean germplasm in Brazil. To capture the non-linearity of those impacts across a wide range of environments, we developed an “enviromic prediction” approach by combining Generalized Additive Models (GAM), environmental covariates (EC), and grain yield (GY) from 18 years of historical breeding trials. Then, we predicted the optimum limits for ECs at each production scenario (four regions, three seasons, and two grain types) and its respective predictions of GY adaptation. Our results indicate that the nonlinear influence of air temperature, solar radiation, and rainfall led to a huge interaction of the impacts among the development stages, seasons, and regions. This revealed that seasonality differently affected the vegetative and reproductive stages, which its impact drastically vary according to the region and season, which makes unfeasible the development of a breeding strategy for selecting for broad adaptation. Conversely, with our approach it was possible to pinpoint the effects of the region- or season-specific impacts, which helped identify the “climate limits” and critical development phases for each possible production scenario. This could allow breeders to design crop ideotypes while directing efforts to develop climate-smart varieties. Furthermore, enviromics prediction is a cost-effective way to use EC as a data analytics tool to support the visualization of regional breeding gaps for specific growing conditions. • We verified the ability of GAM-based models to determine climate drivers that explain grain yield variation. • Climatic limitations for cropping common bean in Brazil were identified across diverse seasons and regions. • “Optimum” predicted values for climate variables in different common bean production regions were obtained.