2026-10-13T09:27:29Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33868318 2026-09-16T20:33:31Z category_23932 item_type_3 month_year_09_2026

Dataset for Fast-tracking Miscanthus Breeding: High-Spatial Resolution Aerial Imagery as a Cost-Effective Tool for Early Yield Prediction Diego Jarquin (23136982) Crop and pasture improvement (incl. selection and breeding) Biofuel Crops Miscanthus Breeding Multi-Omics Integration Perennial Crops Phenomics Secondary Traits Unmanned Aerial Vehicles Miscanthus is a promising perennial grass as biofuel feedstock and it that typically requires three years of evaluation to confidently select superior genotypes. This study explores the use of high-throughput phenotyping with unmanned aerial vehicles (UAVs) as a potential alternative to assist the early selection in Miscanthus breeding. The effectiveness of spectral and structural information (s) extracted from multispectral imagery collected with UAVs was assessed for predicting dry biomass yield (DBY) in Miscanthus sacchariflorus (Msa 45 families) and Miscanthus sinensis (Msi.C 103, and Msi.S 123) breeding trials. Covariance matrices describing genomic (g) and phenomic similarities (s) were used for predicting the joint performance of advanced Miscanthus cycles considering different cross-validations schemes (within years predictive ability, PA; forward prediction or early predictive ability, EPA). For PA the comparison was conducted across three phenotyping levels: 70%, and 30%; while for EPA also the 100% was considered. For the PA scheme, across models and phenotyping levels (70%, and 30%) correlations ranged 0.22-0.67 (Msa), 0.52-0.92 (Msi.C), and -0.14-0.93 (0.93) with the models including s term always performing the best in contrast to g models. Under the EPA scheme, across populations and for the different phenotyping levels, the best results were obtained with those models that included the s component and these varied between 0.61 and 0.81 (30%), 0.72 and 0.84 (70%), and 0.7 and 0.81 (100%) while for those models only including genomic information ranged between -0.01 and 0.32 (30%), 0.05 and 0.44 (70%), and 0.67 to 0.83 (100%). By phenotyping only 30% of the families, the s model was able to capture 68% (Msa), 71%, (Msi.C) and 82% (Msi.S) of the genetic gain achieved through the joint performance of all years. The integration of high-throughput phenotyping using UAVs significantly improved the accuracy of early-stage yield predictions in Miscanthus. Models including the s component as main and interaction effects consistently outperformed strictly genomic models, particularly in scenarios with limited phenotypic data. 2026-09-16T20:33:31Z Dataset Dataset 10.6084/m9.figshare.33868318.v1 https://figshare.com/articles/dataset/Dataset_for_b_Fast-tracking_Miscanthus_Breeding_High-Spatial_Resolution_Aerial_Imagery_as_a_Cost-Effective_Tool_for_Early_Yield_Prediction_b_/33868318 CC BY 4.0