Prediction of late blight severity in a large panel of potato genotypes using low-altitude aerial images and machine learning methods

Loayza, H.; Palacios, S.; Ninanya, J.; Silva, L.; Pujaico, F.; Rinza, J.; Gastelo, M.; Aponte, M.; Kreuze, J.F.; Lindqvist-Kreuze, H.; Kante, M.; Ramírez, D.A. 2025. Prediction of late blight severity in a large panel of potato genotypes using low-altitude aerial images and machine learning methods (Working Paper). International Potato Center (CIP). https://doi.org/10.4160/cip.2025.12.038

Potato (Solanum tuberosum L.) is a staple crop crucial to global food security, yet its production is severely threatened by late blight (LB) disease, one of the most destructive plant diseases worldwide. Breeding programs for LB resistance have traditionally relied on labor-intensive, subjective visual assessments, which limit scalability and precision, particularly in early-generation trials. Unmanned aerial vehicle (UAV)-based remote sensing combined with machine learning (ML) offers a promising alternative for objective, high-throughput disease phenotyping. This study evaluated the potential of UAV-derived multispectral imagery and ML techniques to quantify LB severity in a large panel of early-generation potato breeding populations. Specifically, we assessed the capacity of aerial phenotyping to capture genetic variation in LB resistance, compared image-derived indicators with conventional field assessments, and evaluated the predictive performance of linear vegetation index–based models versus nonlinear ML frameworks. UAV-based multispectral data were analyzed using vegetation indices and a machine learning framework combining K-means clustering with Kernel Ridge Regression (KRR). NDVI consistently showed the strongest correlation with visually assessed LB severity across trials, particularly at advanced disease stages. However, the KRR-based approach using raw multispectral reflectance outperformed linear NDVI-based models by capturing nonlinear spectral–disease relationships and improving prediction accuracy. The ML framework effectively differentiated susceptible from resistant check cultivars and captured biologically meaningful genotypic variation in disease progression. The proposed KRR-based framework represents a robust, transferable, and cost-effective solution for high-throughput monitoring of LB severity under field conditions. Its integration with conventional phenotyping and genomic-assisted selection strategies has significant potential to accelerate genetic gain for LB resistance in potato breeding programs.