Long-term envirotyping links persistent site climate to genotype × environment patterns in a subtropical soybean testing network

Francisco, Stephany Malfatti, Alfonso, Carlos Walter, Tartara, Sabrina Costa, Rizzo, Federico Ariel, Olivoto, Tiago, Figueiredo, Leonardo Domingues de, Giasson, Nizio Fernando, Miranda, Glauco Vieira, Benin, Giovani
BMC Plant Biology · Springer Nature · 2026-11-18 · DOI: 10.1186/s12870-026-10381-4

Abstract

Background Strata within a target population of environments help cultivar testing only if they reflect persistent site attributes and correspond to genotype responses. We characterised the long-term climate of a soybean testing network in southern Brazil and tested, in one season, whether climatic similarity corresponded to genotype-response similarity. Daily NASA POWER records for 2013/14–2022/23 were summarised into 11 covariates over the crop cycle and five developmental windows at 13 sites, and grain yield of 25 genotypes was evaluated in 14 environments in 2023/24. Results Climate varied along continuous gradients (first two principal components: 62.9% and 21.7% of the variance). Three descriptive strata summarised them: warm–humid, high evaporative demand (climatic water balance − 1.14 mm d⁻¹) and cool–humid. The three-cluster solution had the highest mean silhouette width (0.45) and was the most stable under season resampling (mean adjusted Rand index 0.88). Site accounted for 68–75% of the variance among site-season means in mean and minimum temperature, but only 8–12% in precipitation, water balance and relative humidity. In 2023/24, precipitation exceeded the historical mean at 10 of 13 sites, and no site fell in the hot, dry class (12% of historical site-seasons). Genotype, environment and G×E variances were significant ( p < 0.001). Genotype responses were more similar within than between strata for rank correlations (0.267 vs. 0.166; exact one-sided permutation p = 0.047) and centred genotype effects (0.300 vs. 0.139; p = 0.006), but not for crossover frequency ( p = 0.057) or under the best-fitting genetic covariance models. Genotype-response dissimilarity increased with long-term climatic distance (Mantel r = 0.43, p < 0.006) and with latitudinal distance ( r = 0.40, p < 0.007), but not with the climatic distance realised in 2023/24 ( r = − 0.04, p = 0.60). Conclusions Long-term envirotyping identified a persistent, mainly thermal, structure in the network climate, whereas rainfall varied mainly among seasons. In 2023/24, genotype response similarity was associated with this long-term structure, and comparably with latitude, but not with the weather of that season. The correspondence is established for one season, the strata provide a basis for multi-season trials that can test whether they are repeatable genotype-response domains.

Read at publisher →

Similar articles