Evaluating the suitability of hyper- and multispectral imaging to detect endogenic diseases in grapevine

Abstract

Endogenic diseases often arise from pathogens that exist within the plant tissue, including fungi, bacteria, and viruses, which can remain latent and then emerge under stress conditions or favorable environmental conditions, causing symptoms that weaken vines or can lead to plant death. Early detection enables a proactive approach to disease management, promoting both yield quality and environmental sustainability. Spectral sensors enable objective, non-invasive detection of plant reflectance across the visible light spectrum (400–700 nm) as well as in the near-infrared (700–1000 nm) and short-wave infrared regions (1000–2500 nm).

The suitability of ground-based hyperspectral and multispetrctral analyses for the detection of grapevine phytoplasma diseases Palatinate grapevine yellows (PGY), Bois noir (BN) and Flavescence doree (FD), the virus infection grapevine leafroll disease (GLD), and the fungal Esca disease has been evaluated. Distinction between diseases and nutrient deficiencies was tested for magnesium (Mg) and iron (Fe) deficiency. Sensor data has been acquired over recent years from 2016 – 2018 and 2022 -2023 in the greenhouse as well as in  field locations in Germany and Italy.

 Disease detection models on hyperspectral data for BN and PGY could be developed under controlled conditions, models were able to classify infected plants correctly with an accuracy of up to 96%. Infected shoots collected in the field showing grapevine yellows disease were analyzed, gaining high classification accuracies of up to 100%, leading to the assumption that both diseases might also be detectable directly in the field, which was tested using a multispectral sensor in 2022-2023. GLD detection was initially tested on various greenhouse plants, achieving identification rates of 83–100% for symptomatic vines and 85–100% for infected vines without symptoms. Field plant acquisition showed comparable classification accuracy. The results also demonstrated the potential of hyperspectral analysis for detecting asymptomatic infections directly in the field, though accuracy varied significantly between experimental years. For Esca disease, field detection was conducted over three consecutive years, with hyperspectral detection models successfully developed based on original field data. The overall classification rate for distigushing the classes FD, BN, Fe, Mg and viroses using a hyperspectral sensor was 90%.