A review of major factors influencing the accuracy of mapping green-attack stage of bark beetle infestations using satellite imagery: Prospects to avoid data redundancy

Abstract Detecting green-attacked trees on satellite imagery with high accuracy provides very important broad-scale tools required to control the spreads of infestations to healthy trees. The scale or resolution plays a critical role in such equations but very often lacks adequate attentions in remote sensing investigations of environmental issues, including mapping green-attacked trees. We reviewed and synthesized current understanding about mapping green-attacked trees, vegetation remote sensing, and spatial statistics to characterize the uncertainty associated with major factors influencing the accuracy of detecting green-attacked trees. These factors include the spatial, spectral, and temporal resolutions of satellite imagery, and a type of model used. Satellite imagery with spatial resolutions less than 4 meters (approximately 2/3 size of the crown diameter of a host tree on average) on bidaily to weekly acquisitions are required to improve the accuracy of mapping infested trees on a continuous basis. Changes in, forest structure, needle age, leaf intercellular structure of air-to-cell wall interfaces, forest floor moisture content, and tree physiology, over the growing seasons affect the near-infrared (NIR) and shortwave-infrared (SWIR) more prominently than visible and red-edge spectral reflectance in healthy coniferous forests. These variabilities in forest parameters lead to more uncertainties in models using NIR and SWIR spectral reflectance, or their dependent spectral vegetation indices (SVIs), to map green-attacked trees. The high correlation between far-red or red-edge (red-to-NIR transition) reflectance, in particular shifts in the red-edge inflection point, with changes in the leaf chlorophyll contents makes red-edge spectra or red-edge dependent indices strong candidates to map green-attacked trees. The red-edge normalized difference vegetation index (RENDVI; red-edge NDVI or NDVI705) that incorporates red and red-edge values, found to be useful to estimate chlorophyll a and b concentrations as well as minimizing the effects of background soil reflectance. Future investigations are required to improve red-edge NDVI or explore other SVIs, by using preferably visible to red-edge spectral reflectance, to increase the accuracy of mapping green-attacked trees. When the attack occurs in, or spreads from a tree to, the stand level, causes spatial autocorrelation in the spectral properties of infested trees. Ignoring the spatial autocorrelation, as it usually is, inflates type I error in the parametric models. A non-parametric model, with no need to fulfill the assumptions of data independency, is recommended to map green-attacked trees as the spatial property of underlying process of the infested trees is usually non-stationary.