Feasibility study on Huanglongbing (citrus greening) detection based on WorldView-2 satellite imagery

Abstract

Huanglongbing (HLB) is a devastating citrus disease worldwide, without known cure. Since this disease shows visible symptoms on newly developed canopies, remote sensing methods based on spectral principles are considered as an effective tool. To explore a fast way to monitor HLB in large citrus groves, a satellite multispectral (MS) image with a 2-m resolution acquired by WorldView-2 was studied to evaluate its capability on HLB detection. Ground truth was conducted and two spectral libraries were constructed. Library 1 was based on RTK GPS locations, and Library 2 was combined with prior knowledge from the ground spectral features. To compare with the classification accuracy of previous work using airborne spectral images, the same classification methods were carried out. In the satellite MS image, Library 2 had higher average overall accuracy, higher Kappa coefficient, and higher producer's and user's accuracies than Library 1, indicating Library 2 which combined with prior knowledge was more effective. Mahalanobis distance had the highest overall accuracy of 81% and a Kappa coefficient of 0.464 with Library 2. The accuracy comparison between the satellite MS image and the airborne spectral images showed that the satellite MS image had the highest average producer's accuracy with Library 2, followed by the airborne MS image and the satellite MS image with Library 1, indicating that construction of an effective library was the most important step, and that the satellite MS image with a proper spatial resolution showed good potential to be a better choice than the airborne spectral image.