Exploring vegetation indices adequate in detecting twister disease of onion using Sentinel-2 imagery

Abstract

Traditional plant disease detection is time consuming and costly, thus an inexpensive and faster alternative method of detection is needed to send early warning to farmers to prevent pests and disease infestation and for proper intervention. To provide timely and accurate detection in twister disease of onion, remote sensing was exploited using Sentinel 2 imageries. Vegetation indices (VIs) derived from the VIS–NIR region of the image were evaluated for their capability to detect twister disease. VIs were subjected to regression analysis to evaluate the relationship between vegetation indices and severity index of onion twister disease. Vegetation indices with strong relationship to twister disease were selected and further used in unsupervised ISODATA classification. Overall accuracy of classification generated from vegetation indices were calculated based on confusion matrix using ground truth points collected from field work to identify the most suitable index based on highest overall accuracy. It was found out that NDVI and GNDVI has the highest coefficient of determination (R2) indicating a strong relationship to the disease severity. Results of the classification shows that GNDVI, PSSRa and NDVI obtained the highest overall accuracy of 83.33%, 80.95% and 78.57% respectively. This indicates that these 3 VIs can be used for detection of twister disease in the field since it gives better discrimination and high accuracies. Hence, VI’s generated from Sentinel 2 imagery has the potential in detection, monitoring and management of twister disease of onion in the field.