By combining nine unsupervised machine learning algorithms with a custom-designed program, the ART framework detects and quantifies dense root clusters from digital images. When applied to wheat varieties with different drought tolerances, ART-based models achieved a striking 96.3% classification accuracy, outperforming traditional visual trait methods.
Understanding plant root systems is vital to improving crop productivity and climate adaptation, yet their complex underground structures remain difficult to characterize. Conventional imaging methods rely on human-defined geometric traits—like root length or diameter—limiting their ability to capture subtle, spatially complex patterns linked to stress tolerance. As drought increasingly threatens food security, identifying hidden root features that confer resilience is essential. Machine learning offers a new lens to extract “latent traits” directly from digital images, free from human bias. Because traditional root traits overlook the intricate patterns that determine drought adaptation, there is a pressing need to develop algorithmic methods that can reveal and quantify these hidden features.
A study (DOI: 10.1016/j.plaphe.2025.100088) published in Plant Phenomics on 9 July 2025 by Mirza Shoaib & Surya Kant’s team, La Trobe University, offers an objective, scalable, and high-throughput way to analyze root systems, paving the way for breeding climate-resilient crops and redefining how scientists extract biological meaning from image data.
The study first applied a multi-stage computational and physiological pipeline to quantify drought tolerance in wheat. Physiologically, genotypes were ranked using multiple drought-response metrics (RANK_1, RANK_2, RANK_3) based on traits such as stomatal conductance, relative water content, and tiller number under stress. These measurements were statistically tested (ANOVA/Kruskal–Wallis, p < 0.0001) and then used for unsupervised clustering, which grouped genotypes into tolerant and susceptible classes and showed clearer separation under drought than under control conditions, indicating biologically meaningful divergence in stress response. In parallel, the study extracted two classes of imaging-based root traits. Traditional Root Traits (TRTs) were derived from established morphology descriptors, while the new Algorithmic Root Traits (ARTs) were generated by an ensemble of eight unsupervised machine learning algorithms plus a custom algorithm. For each root image, these algorithms identified the densest root cluster and quantified its size and spatial position, producing 27 ART features. These traits, together with TRTs, were then fed into supervised classification models (e.g. Random Forest, CatBoost) to predict which genotypes were drought tolerant. The results showed that ARTs captured more complex and information-rich root architecture than TRTs, as evidenced by higher internal variability, distinct multivariate structure, and strong correlations with biologically relevant depth and biomass allocation patterns. Models trained on ARTs alone reached 96.3% accuracy (ROC AUC 0.997), outperforming TRT-only models (85.6% accuracy; ROC AUC 0.927), and combining both trait types produced the best overall performance (97.4% accuracy; ROC AUC 0.998). This combined model remained robust when validated on an independent dataset (accuracy 0.91; ROC AUC 0.96), demonstrating that algorithmically derived spatial root traits not only reflect meaningful physiological strategies—such as deeper rooting and strategic biomass concentration for water capture—but also provide a scalable, reliable basis for drought tolerance screening.
ART’s success in drought-tolerance classification demonstrates its potential as a scalable, customizable tool for high-throughput plant phenotyping. By transforming raw sensor data into interpretable biological insights, ART could greatly accelerate the screening of genotypes with superior root systems and inform breeding strategies for climate-resilient crops. Its modular framework can be integrated with genomic and metabolomic datasets to identify genetic markers linked to adaptive traits. Beyond roots, the approach can be extended to detect complex image patterns in leaves, stems, or even disease symptoms.
###
References
DOI
10.1016/j.plaphe.2025.100088
Original URL
https://doi.org/10.1016/j.plaphe.2025.100088
Funding information
This study was funded by Agriculture Victoria Research, Victoria state government, Australia.
About Plant Phenomics
Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.