Looking sideways helps drones read wheat canopies
image: Location of the study area: (a). Plots were in the breeding experimental field on the North Campus of Northwest A&F University (b, c). Sampling locations (i.e., plot boundaries); are marked in yellow and blue. Below each plot boundary, a pseudo-color composite of hyperspectral ortho-intersections (near-infrared, red, and green bands) is displayed.
Credit: Plant Phenomics
Accurately reading crop canopy structure can help breeders see how plants capture light, compete for space and convert energy into yield. A new study shows that changing the viewing angle of an unmanned aerial vehicle (UAV), rather than relying only on straight-down images, can reveal structural information that is otherwise partly hidden. The researchers developed a multi-angle hyperspectral sensing framework to estimate leaf area index (LAI), leaf inclination distribution function parameter a (LIDFa) and fractional vegetation cover (FCover) in diverse wheat breeding materials. By combining directional reflectance modeling with machine learning, the approach improved retrieval of the more angle-sensitive traits and points toward faster, non-destructive phenotyping for crop improvement.
Unmanned aerial vehicle (UAV) remote sensing is increasingly used to measure crop traits quickly and without destructive sampling, but most field phenotyping still depends on nadir, or vertically downward, observations. That view can miss reflectance anisotropy—the way a canopy appears differently as sunlight and sensor geometry change. Multi-angle measurements can capture this directional information, yet collecting every useful viewing direction in a breeding field is impractical, while physical radiative-transfer models can be computationally demanding. It has also remained unclear which canopy traits truly benefit from oblique views and which viewing angles are most informative. Given these challenges, in-depth research is needed to determine how multi-angle UAV observations can improve canopy-structure retrieval across diverse wheat breeding materials.
Researchers from Northwest A&F University and Southeast University reported (DOI: 10.1016/j.plaphe.2026.100183) the study in Plant Phenomics, with the article published online on February 12, 2026. The team combined UAV-based oblique hyperspectral imaging with a semi-empirical bidirectional reflectance distribution function (BRDF) model and predictive learning to estimate wheat canopy structure across breeding plots. Their central question was practical: whether looking at wheat from several directions can outperform the conventional straight-down view, and whether the gain depends on the trait being measured. The study also tested which observation geometries and modeling strategies produced the most reliable estimates.
The experiments covered two growing seasons and hundreds of genetically diverse wheat materials, including 342 entries in 2023 and 560 in 2024. A hyperspectral UAV collected canopy reflectance from nadir and oblique directions, while ground measurements provided reference values for LAI, LIDFa and FCover. The team converted the airborne observations into bidirectional reflectance factor (BRF) data and fitted a kernel-driven BRDF model; the LiDense–RossHotspot combination best represented the dense wheat canopy and achieved coefficients of determination (R²) around 0.88 against airborne BRF measurements. The researchers then compared transfer-learning and conventional machine-learning approaches under a strict five-fold cross-validation design. Multi-angle sensing produced its clearest advantage for LAI, where the best model reached R² = 0.59 versus about 0.38 for the best nadir-only model. LIDFa also improved, reaching R² = 0.46 versus 0.37. FCover was already well constrained from above, with both configurations reaching R² ≥ 0.73 and the best multi-angle model reaching R² = 0.75. A convolutional neural network (CNN) transfer-learning model performed best for LAI and FCover, while Random Forest using raw multi-angle spectra performed best for LIDFa. Optimal views were trait-dependent, commonly falling between 15° and 45° from nadir.
The authors said the main message is that viewing geometry should be treated as part of measurement design, not as an afterthought. They said oblique observations add the most value when a trait changes directional scattering, as with leaf area and leaf orientation. FCover, by contrast, is largely a vertical-projection property, so the nadir view already captures much of its useful signal. The authors said this trait-specific response explains why a single “best” flight setup should not be assumed for every phenotype. Matching viewing angle to the biological target can therefore improve accuracy without adding unnecessary complexity to UAV phenotyping.
The framework could help breeding programs screen large field populations more efficiently by turning UAV hyperspectral data into trait-specific measurements of canopy architecture. Better estimates of LAI and leaf orientation may support selection for light interception and yield-related plant structure, while strong nadir performance for FCover suggests simpler flights may be sufficient when vegetation cover is the main target. The study also offers a practical deployment route: select a BRDF model suited to the local canopy, identify informative viewing angles, and standardize those angles for large-scale phenotyping. Because the tests were conducted at one wheat site and mainly during heading and grain filling, local validation will be important before applying the same configuration to other crops, regions or growth stages.
###
References
DOI
Original Source URL
https://doi.org/10.1016/j.plaphe.2026.100183
Funding information
This work was supported by the Youth Project of Shaanxi Provincial Natural Science Foundation (No. 2025JC-YBQN-280) and the National Key R&D Program of China (No. 2022YFE0116200).
About Plant Phenomics
Plant Phenomics publishes breakthrough research that integrates genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics and computer science to advance plant phenotyping—from the cellular level to entire populations. The journal tackles key scientific challenges in phenomics by showcasing cutting edge technologies for data acquisition, management, interpretation and modelling. Its scope spans practical applications across all plant sciences, aiming to translate novel phenotyping tools into real world solutions for breeding, crop management and fundamental plant biology. By fostering interdisciplinary innovation, Plant Phenomics serves as a hub for researchers developing high throughput, accurate and scalable methods to decipher complex plant traits, ultimately accelerating progress in agriculture and plant science.