Answers are grounded in this paper and cite their sources.
Digital sensing, genomic prediction and increasingly accessible computational methods are changing how crop breeders observe environments, characterise genetic variation and prioritise candidate varieties. Yet the central question is not whether more data improve predictive fit, but whether combining information sources produces reliable, timely and economically worthwhile selection decisions outside the circumstances in which models were trained. This critical narrative review examines the convergence of plant genomics, field phenomics, environmental characterisation and artificial intelligence in predictive breeding. Peer-reviewed research and methodological literature published from 2001 to 11 September 2026 were critically selected through accessible scholarly sources and citation tracing, with attention to empirical comparisons, validation design and translational relevance. Genomic prediction provides a well-established foundation, whereas phenomic and environmental covariates yield more conditional benefits. Contrasting wheat studies illustrate that spectral and thermal information can improve prediction in some settings but can also introduce noise or fail under forward validation. Environmental descriptors become more informative when aligned with developmental stage and the target population of environments; flexible machine-learning architectures do not consistently outperform appropriately specified statistical baselines. Structural variation, pangenomes, multi-omics and genomic language models broaden the biological information available to breeders, although evidence of additional realised genetic gain remains limited for several emerging approaches. Major weaknesses include information leakage, unrepresentative validation partitions, incompatible field metadata, uncertainty neglected during selection, and scarce evaluation of operational costs and varietal delivery. The review develops a decision-centred framework in which data collection, model complexity and evaluation are governed by the breeding decision and the information genuinely available at that point. The strongest near-term opportunities lie in prospective multi-environment validation, careful phenomic measurement, stage-aware environmental modelling and interoperable datasets. Predictive breeding is therefore best understood as a disciplined integration of biological understanding, field experimentation and statistical evidence, not as a replacement for them.
No takes yet. Share an insight, caveat, or question.
Sidharthan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: