| dc.description.abstract | Sustainable intensification of agriculture requires breeding systems that
simultaneously improve productivity, resilience, and ecosystem services. However, most
breeding programs remain optimized for monoculture yield, despite growing interest in
intercrops and cover crops. Breeding directly for these systems is constrained by the
combinatorial burden of testing genotype combinations across species, species-specific
phenotyping in mixed canopies, limited quantitative genetic characterization of tropical
intercrops, and the difficulty of measuring root traits. This dissertation evaluated
genomic prediction (GP), unmanned aerial vehicle (UAV) phenomics, root phenotyping,
and quantitative genetic approaches across crimson clover–oat mixtures, dual-purpose
white lupin, and cassava–cowpea intercropping systems.
In crimson clover–oat mixtures, UAV-derived vegetation indices predicted
species-specific biomass in mixed canopies with accuracies of r = 0.70–0.87, and
calibration analyses showed that harvesting only 25–50% of plots preserved over 90%
of expected response to selection, enabling a 50–75% reduction in destructive
sampling. In cassava–cowpea intercropping, 120 cassava clones showed moderate-to
high intercrop heritability (H² = 0.50–0.75), and genetic correlations between
monoculture and intercrop performance were near-unity (rg = 0.92–0.97), yet realized
selection efficiency was only 26–44%, showing that high genetic correlation does not
ensure effective indirect selection. General mixing ability dominated specific mixing
ability, which was negligible, with producer effects explaining 20–47% of intercrop
variance. In white lupin, root traits linked to phosphorus mobilization revealed a trade-off
with grain yield, motivating the Alabama Ecosystem Service Index for dual-purpose
2
selection, which shows potential for GP (r = 0.18). Integrating genomic and phenomic
kernels improved prediction for high-heritability traits such as alkaloid status (r = 0.55
0.59, exceeding genomic-only prediction by 0.16–0.17), whereas structural root traits
remained poorly predicted across all data sources.
Collectively, these studies demonstrate that GP, UAV phenomics, quantitative
genetic analysis, and direct root phenotyping can improve phenotyping efficiency and
selection accuracy in breeding programs targeting multispecies cropping systems and
ecosystem services. Belowground architectural traits remain difficult to predict indirectly,
supporting continued integration of direct root phenotyping with UAV measurements and
the use of selection indices when ecosystem service and yield objectives are negatively
correlated | en_US |
