Laboratory vs. in-field spectral proximal sensing for early detection of Fusarium head blight infection in durum wheat

A comparison between two VIS–NIR spectral based systems performed in laboratory vs. in-field for the early detection of Fusarium head blight infection in two cultivars of durum wheat (Creso and Simeto) was carried out. The VIS–NIR spectrophotometric data were analysed with multivariate statistical tools. For both, laboratory and in-field experiments two analytical conditions were tested for two cultivars: diseased plants (artificial infection without fungicide treatment) and healthy plants (treatment with Folicur SE). Spectral measurements were performed at the different sampling times 6, 8, 12 and 15 d after artificial infection which correspond to the following Zadoksʹ scale growth stages: GS70, GS71, GS73 and GS75. The infection visible onset (VO) was then evaluated by an expert at the soft dough growth stage GS85. Since the growth stages revealed a great influence on spectral reflectance data, separated PLSDA models were adopted to differentiate diseased and healthy plants at the different sampling times. Using the Euclidean distance matrix cladogram results for the laboratory, three models were used considering spectral data from GS70, GS71 + GS73, GS75, while for in-field data from GS70 + GS71 and GS73 + GS75. In the laboratory good performance of classification (86%) was observed at GS71 + GS73 i.e., only 8–10 days after the infection. The in-field measurement showed a lower percentage of correct classification at the same growth stages. Finally the VIS–NIR spectral analysis could facilitate detection of Fusarium disease anticipating visual assessment.