By contrasting nematode infection with drought stress, the study highlights both the potential and limitations of remote sensing for separating overlapping crop stress signals, paving the way for non-invasive, large-scale monitoring and improved pest management.
Potato cyst nematodes (Globodera pallida and G. rostochiensis) are major quarantine pests that can cause serious yield losses yet often remain undetected for years because early infections show few or no visible symptoms. Current detection methods rely on soil sampling and laboratory diagnostics, which are invasive, labor-intensive, and unsuitable for large-scale early monitoring. Remote sensing provides a promising alternative by assessing plant health through spectral signals rather than visible damage. Hyperspectral imaging captures detailed information across visible, near-infrared, and short-wave infrared wavelengths, revealing subtle biochemical and physiological changes. However, distinguishing nematode stress from abiotic stresses such as drought remains a key challenge due to overlapping plant responses.
A study (DOI: 10.1016/j.plaphe.2025.100127) published in Plant Phenomics on 14 October 2025 by Uroš Žibrat's team, Agricultural Institute of Slovenia, demonstrates that hyperspectral imaging combined with explainable machine learning can detect early, otherwise invisible potato cyst nematode stress while revealing why drought signals dominate and complicate stress separation, highlighting both the promise and current limits of precision crop monitoring.
Using a controlled greenhouse experiment, the researchers inoculated potato plants with Globodera pallida or G. rostochiensis at two initial population levels under well-watered and water-deficient conditions, and integrated nematode reproduction data, plant morphology and physiology, hyperspectral imaging, and explainable machine-learning analyses to disentangle biotic and abiotic stress signals. Nematode reproduction and basic plant traits were quantified across days after inoculation, while hyperspectral data collected at multiple growth stages were explored with UMAP to assess stress separability in low-dimensional space. Twelve machine-learning classifiers were then screened using cross-validation, with performance summarized by F1 scores across a range of binary and multi-class tasks, followed by in-depth evaluation of top models. In parallel, linear mixed models assessed treatment effects on photosynthetic rate, PSII quantum yield, and stomatal conductance, and SHAP-based analyses identified the most informative spectral regions and evaluated reduced-band performance. The biological assays confirmed successful infections: both nematode species reproduced under both watering regimes, reaching 0.5 (±0.2) cysts/cm³ at low inoculum and 1.4 (±0.5) cysts/cm³ at high inoculum, with reproduction factors declining at higher inoculum and G. pallida reproducing more strongly at low Pi. Despite infection, plant morphology was largely unaffected, aside from transient height reduction at early stages under combined water deficit and low G. pallida inoculation. UMAP analysis revealed that plant growth stage dominated spectral variation, overshadowing stress effects in two dimensions. Classification results reflected this hierarchy: drought stress was most readily detected (F1 up to 0.95), while nematode infection, species, and inoculation level were identified with moderate accuracy (F1 ~0.65–0.80), and multi-stress classification performance declined markedly (F1 ~0.58). Physiological analyses showed water deficit consistently reduced photosynthesis, PSII efficiency, and stomatal conductance, whereas nematode effects were subtle and often non-significant. Spectral importance centered on water- and biochemistry-related bands in the near- and short-wave infrared, explaining why drought signals dominated and why separating early nematode stress remains challenging.
The findings demonstrate that hyperspectral imaging can reveal "hidden" nematode stress before visible symptoms appear, offering a pathway toward earlier and more spatially precise detection than current methods. Such tools could help farmers and regulators identify infestation hotspots, optimize targeted control measures, and reduce unnecessary pesticide use.
