As climate change intensifies interannual weather variability, plant breeders are under pressure to evaluate more genotypes across diverse environments while keeping phenotyping costs manageable. Sparse testing — assessing only a subset of genotypes per location — offers a practical solution, but it demands robust predictive models to fill missing data. A new study published in Scientific Reports demonstrates that high-throughput phenotyping (HTP) from drone imagery can serve as a powerful substitute for genomic information in such scenarios. Researchers working with 256 potato clones across three Swedish environments (two locations over 2020–2021) extracted time‑integrated vegetation indices, canopy cover, and plant height from UAV‑based RGB images. Using principal component analysis (PCA) on these image‑derived traits, they constructed an “environmental kernel” (E‑kernel) that quantifies similarity between testing sites and compared its predictive performance against the standard genomic relationship matrix (G‑kernel) under data sparsity levels ranging from 10% to 66%.
The results are striking: in two‑environment scenarios with 50% data completeness, the E‑kernel significantly outperformed both the G‑kernel and their combination across all traits, boosting prediction accuracy for total tuber yield by 60% and for the 50–60 mm size fraction by 62%. At 33% completeness, starch content was the only trait with r² > 0.5, and only the E‑kernel delivered that level. The genomic kernel proved more robust only under extreme sparsity (10–20%), while conventional meteorological covariates showed limited ability to discriminate between nearby locations. Notably, combining G and E kernels did not improve predictions, suggesting overlapping information. The authors acknowledge limitations — modest sample size and environmental leakage — but emphasize that image‑based environmental kernels are cost‑effective, scalable, and particularly valuable for slow‑cycling crops like tetraploid potato. This approach offers a pragmatic pathway to accelerate genetic gain in resource‑limited breeding programs, especially for yield‑related traits, while reducing the phenotyping burden through reusable HTP data.
