Winter wheat lodging mapping by combining UAV hyperspectral data and machine learning techniques.

Extreme weather events such as strong winds, rain, and hailstorms impact crop production, particularly cereal crops like wheat, where they can cause complete lodging. Agricultural remote sensing of such damages using UAVs equipped with multispectral and hyperspectral sensors has gained traction in the recent past, despite the complexity and cost of these systems. Coupled with UAV remote sensing, machine learning algorithms (MLAs) have become increasingly popular for analyzing remotely sensed data. In this study we employed a supervised machine learning approach to classify lodging caused by simulated wind damage into four categories: "Fresh Lodged," "Old Lodged," "Not Lodged," and "Soil." A two-year field trial was conducted, during which winter wheat was subjected to mechanical lodging at flowering and milk maturation stages. UAV-based monitoring was carried out using a Matrice 600 Pro drone equipped with a nanoHyperspec hyperspectral camera. Clear and cloud-free satellite images corresponding to UAV flights were analyzed, and classification was performed at resolutions of 0.15m, 3m, and 10m using Support Vector Machines. The classification achieved an overall accuracy of 85.69% and a Kappa coefficient of 0.55. The "Not Lodged" and "Soil" classes were most accurately identified, while the algorithm struggled to distinguish between "Fresh Lodged" and "Old Lodged" categories, misclassifying about 20% of lodged areas as non-lodged. The results highlight that high-resolution hyperspectral imagery may be prone to noise, whereas coarser resolutions fail to capture heterogeneous crop damage patterns like lodging. This suggests that an optimal approach might be somewhere between spectral richness of hyperspectral sensors and existing satellite imagery at coarser resolutions.