Researchers have developed a new wheat powdery mildew index (WPMI) that uses unmanned aerial vehicle (UAV) hyperspectral imagery and spatial analysis to detect and track wheat powdery mildew (WPM) across multiple scales. The study, published in the Journal of Remote Sensing (DOI: 10.34133/remotesensing.0955), addresses the critical need for rapid, field-scale monitoring of this destructive fungal disease, which can cause severe yield losses in winter wheat.
Wheat powdery mildew damages leaf tissues, weakens plant growth, and often leads to significant yield reductions. Current diagnosis relies on expert visual inspection, a labor-intensive and subjective method that struggles to capture the spatially uneven spread of the disease in smallholder farms. While hyperspectral remote sensing has shown promise for crop disease detection, existing vegetation indices were designed for general stress monitoring rather than specific pathogen-host responses. Machine learning approaches also require large, high-quality training datasets, limiting their scalability.
The research team, from the Key Lab of Smart Agriculture System at China Agricultural University, the Information Technology Research Center at the Beijing Academy of Agriculture and Forestry Sciences, the Institute of Plant Protection at the Chinese Academy of Agricultural Sciences, and the College of Land Science and Technology at China Agricultural University, developed two forms of WPMI: WPMIG = (R760 − R554)/(R661 + R554) and WPMIR = (R760 − R661)/(R661 + R554). These indices utilize disease-sensitive bands in the green, red, and near-infrared regions to distinguish healthy from infected wheat and quantify disease index (DI) at leaf, ground canopy, and UAV canopy scales.
Data were collected from greenhouse and field experiments between 2022 and 2024, including 1,260 leaf spectra and 804 canopy spectra under varying infection conditions, wheat varieties, and spatial scales. At the leaf scale, WPMI achieved overall classification accuracy (OCA) of 85% for WPMIG and 86% for WPMIR in the 2022 greenhouse experiment. In field conditions, OCA ranged from 80% to 81% across 2023 and 2024. For disease severity estimation, WPMIG reached R² values of 0.55 to 0.93 at the ground scale and 0.48 to 0.90 at the UAV scale.
UAV-derived WPMIG maps were combined with Getis–Ord Gᵢ* hot-spot analysis to identify clusters of likely infection and track spatiotemporal changes across smallholder plots over three growing seasons. The researchers noted that a disease-specific spectral index can move crop disease monitoring beyond simple image comparison. By linking UAV hyperspectral imagery with spatial hot-spot analysis, the method reveals where WPM is emerging, expanding, or declining, offering a basis for earlier warning and more targeted disease management.
Leaf spectra were collected using a handheld hyperspectral camera, while canopy spectra were acquired with a ground spectrometer and a DJI M600 UAV equipped with a Pika L hyperspectral camera. Linear discriminant analysis (LDA) selected sensitive bands and evaluated classification performance. DI was measured through field surveys following national standards, and linear regression assessed the relationship between WPMI and DI.
With further validation across regions, wheat varieties, sensors, and disease conditions, WPMI-based UAV monitoring could support precision plant protection and early warning systems for wheat production. The approach may help farmers identify disease hot spots before severe outbreaks occur, reduce unnecessary pesticide use, and improve field-level decision-making. More broadly, this strategy provides a framework for developing disease-specific remote sensing indices for other crop–pathogen systems, contributing to smarter and more resilient agricultural monitoring.
