Enviromics e modelagem preditiva aplicadas ao melhoramento de milho (Zea mays L.) no semiárido brasileiro

Abstract

Maize (Zea mays L.) is a strategic crop for food security, forage production, and income generation in the Brazilian semiarid region. However, high climatic variability and edaphic constraints result in low average productivity and elevated risk of crop failure, while also hindering the recommendation of adapted cultivars. In this dissertation, we propose an integrated framework that combines envirotyping, environmental stratification, geospatial delineation of the target population of environments (TPE), and the multi-trait mean performance and stability index (MTMPS) to support maize breeding and cultivar recommendation in the semiarid region. In the first study, the envirotyping approach was applied across selected states of Northeastern Brazil, delineating four mega-environments (MEs) based on 19 environmental covariates, ten experimental sites, and a 23-year climatological series. Across the ten environments, 20 maize hybrids were evaluated over two growing seasons, and the MTMPS was employed to select superior genotypes, simultaneously assessing grain yield (GY), plant height (PH), ear height (EH), plant density (PD), and number of ears (NE). This approach enabled the identification of productive hybrids combining high mean performance and greater consistency across the two growing seasons per ME, with G03 and G04 standing out as selected across three of the four MEs. The integration of envirotyping and MTMPS enabled the simultaneous assessment of mean performance and stability across multiple traits and may support the future rationalization of multi-location trial networks. In the second study, environmental covariables were used – through the EnvRtype package in R – to segment the Alto Sertão Sergipano into macroenvironments (MEs) based on similarities among environmental and ecophysiologically relevant covariates. Cluster analysis revealed four major environmental groups, showing that neighboring municipalities may belong to distinct climatic zones, while geographically distant areas may share similar environmental profiles. This stratification highlighted the possibility of reducing the number of testing sites by concentrating trials in more representative environments, thereby providing a basis for future environmentally informed cultivar evaluations for the region. In the third study, we developed a georeferenced TPE for maize in the Alto Sertão Sergipano using geographic information system (GIS) tools in R, including the TPEmap function. Based on the coordinates of 21 georeferenced locations, concave polygons were built, rasterized, and integrated with climatic and soil covariables (NASA POWER and SoilGrids v2.0). Cluster analysis, the environmental similarity matrix, and principal component analysis revealed that the TPE is structurally heterogeneous and composed of six edaphoclimatic sub-environments, reflecting gradients of aridity, temperature, and water availability. Using grain yield data from 352 genotypes evaluated at CEUFS (2019– 2024), mixed models (REML) were fitted to estimate BLUPs and rank genotypes for productivity and stability. Cultivars AGR Vereda, NS80VIP3, GNZ59, GNZ15, and AG8780 comprised the elite group, with high predicted performance and temporal consistency at CEUFS, which belongs to the central environmental cluster of the TPE. Taken together, the results indicate that integrating envirotyping, environmental stratification, geospatial TPE mapping, and MTMPS provides an environmentally informed basis for redesigning experimental networks, guiding breeding programs, and the future evaluation and recommendation of maize cultivars for the Brazilian semiarid region, with the potential to contribute to productivity, resilience, and food security following multi-location validation in the Alto Sertão Sergipano and similar regions.