03308naa a2200301 a 450000100080000000500110000800800410001902200140006002400510007410000210012524501620014626000090030852023590031765000120267665000160268865000260270465000370273065000190276765000190278665000100280565000260281565000350284170000190287670000200289570000210291570000170293677300530295321648972024-06-18       2024    bl uuuu     u00u1 u    #d  a0378-42907 ahttps://doi.org/10.1016/j.fcr.2024.1094522DOI1 aHEINEMANN, A. B.  aHarnessing crop models and machine learning for a spatial-temporal characterization of irrigated rice breeding environments in Brazil.h[electronic resource]  c2024  aProblem: Tropical ecosystems are essential for irrigated rice production and are vital for ensuring food security in countries like Brazil. However, the uncertainty stemming from numerous environmental constraints limits current productivity levels and presents challenges for plant breeding research in developing productive and stable cultivars. Objective: To overcome these challenges, environmental characterization is essential to support breeding decision-making and help design and develop superior cultivars capable of capitalizing on genotype x environment interactions (GxE). Methods: In this study, we characterize the tropical rice zone in Brazil using crop modeling outcomes, machine learning tools (unsupervised clustering and supervised classification), envirotyping, and long-term observed yield trials from the irrigated rice-breeding program. Results: Three target population of environments (TPE) were detected across diverse geographic regions and planting dates, named after LFE (Least Favorable Environments, 44 % of occurrence), FE (Favorable Environments, 33 %), and HFE (Highly Favorable Environments, 23 %). For each TPE, we identified the key climate drivers for achieving higher yields, which involved a combination of lower temperatures (average maximum and minimum temperature: 30 ◦C and 19 ◦C), while higher radiation in vegetative and reproductive phases (692 and 723 MJ m 2) and precipitation (663 mm). The inclusion of TPE information to support the GXE analysis spanning 18 years of advanced yield trials for cultivar targeting increased the model’s heritability, even reducing the number of trials by 40 %. Conclusions: Environmental characterization pipelines in tropical rice could be leveraged by combining crop models and machine learning. Virtual simulations of historical climate trends across locations and planting dates could extend the spectrum of growing conditions screened to compose target population of environments for breeding. This is vital to support the design of new breeding programs, exploring alternative planting dates, or when breeders want to characterize a new growing region without any prior field trial data. For an extensive region, this helps breeders either improve cultivar testing analysis or identify the most prone locations to optimize seed production at commercial phases.  aClimate  aCrop models  aEnvironmental factors  aGenotype-environment interaction  aPlant breeding  aArroz Irrigado  aClima  aInteração Genética  aMelhoramento Genético Vegetal1 aCOSTA-NETO, G.1 aMATTA, D. H. da1 aFERNANDES, I. K.1 aSTONE, L. F.  tField Crops Researchgv. 315, 109452, July 2024.