Remote sensing of forest health : the detection and mapping of Thaumastocoris peregrinus damage in plantation forests.

Abstract

Thaumastocoris peregrinus (T. peregrinus) is a sap-sucking insect that feeds on Eucalyptus leaves. It poses a major threat to the forest sector by reducing the photosynthetic ability of the tree, resulting in stunted growth and even death of severely infested trees. The foliage of the tree infested with T. peregrinus turns into a deep red-brown colour starting at the northern side of the canopy but progressively spreads to the entire canopy. The monitoring of T. peregrinus and the effect it has on plantation health is essential to ensure productivity and future sustainability of forest yields. Insitu hyperspectral remote sensing combined with greater availability and lower cost of new generation multispectral satellite data, provides opportunities to detect and map T. peregrinus damage in plantation forests. This research advocates the development of remote sensing techniques to accurately detect and map T. peregrinus damage, an assessment that is critically needed to monitor plantation health in South Africa. The study first provides an overview of how improvements in multispectral and hyperspectral technology can be used to detect and map T. peregrinus damage, based on the previous work done on the remote sensing of forest pests. Secondly, the utility of field hyperspectral remote sensing in predicting T. peregrinus damage was tested. High resolution field spectral data that was resampled to the Hyperion sensor successfully predicted T. peregrinus damage with high accuracies using narrowband normalized indices and vegetation indices. Field spectroscopy was further tested in predicting water stress induced by T. peregrinus infestation, in order to identify early physiological stages of damage. A neural network algorithm successfully predicted plant water content and equivalent water thickness in T. peregrinus infested plantations. The result is promising for forest health monitoring programmes in detecting previsual physiological stages of damage. The analysis was then upscaled from field hyperspectral sensing to spaceborne sensing using the new generation WorldView-2 multispectral sensor, which contains key vegetation wavelengths. Partial least squares regression models were developed from the WorldView-2 bands and indices and significant predictors were identified by variable importance scores. The red edge and near-infrared bands of the WorldView-2 sensor, together with pigment specific indices predicted and mapped T. peregrinus damage with high accuracies. The study further combined environmental variables and vegetation indices calculated from the WorldView-2 imagery to improve the prediction and mapping of T. peregrinus damage using a multiple stepwise regression approach. The regression model selected the near infrared band 8 of the WorldView-2 sensor and the temperature dataset to predict and map T. peregrinus damage with high accuracies on an independent test dataset. This research contributes to the field of knowledge by developing innovative remote sensing techniques that can accurately detect and map T. peregrinus damage using the new generation WorldView-2 sensor. The result is significant for forest health monitoring and highlights the importance of improved sensors which contain key vegetation wavelengths for plantation health assessments.