Discriminating the occurrence of pitch canker fungus in Pinus radiata trees using QuickBird imagery and artificial neural networks

Abstract

Pathogenic fungi, such as Fusarium circinatum, present a serious threat to Pinus radiata plantations. The effective management of infected trees is thus paramount. Coupled with advanced techniques, high spatial resolution remote sensing data provides the necessary tools to effectively identify and map infected trees. This paper explores the utility of transformed high spatial resolution QuickBird imagery and artificial neural networks for the detection and mapping of pitch canker disease. Individual tree crowns were delineated using an automated segmentation and classification approach within an object-based image analysis environment. Subsequently, several vegetation indices including the tasseled cap transformation were calculated and incorporated into a neural network model. The feed-forward neural network showed high discriminatory power with an overall accuracy of 82.15% and KHAT of 0.65. The results of this study show great potential for the future application of crown-level mapping of the pitch canker disease at a landscape scale.