B064-0004
Classification of boreal spruce tree health status using UAV multispectral and G-LiHT hyperspectral imagery
Classification of boreal spruce tree health status using UAV multispectral and G-LiHT hyperspectral imagery
Friday, 11 December 2020
Poster
Abstract:
The spruce beetle (Dendroctonus rufipennis) causes widespread spruce tree (Picea spp.) mortality in
Alaska and experiences population outbreaks that can impact millions of acres. Outbreaks typically
occur after one or more years of above average warm, dry conditions and are expected to increase in
severity with climate change. Detection of outbreaks typically occurs on a regional scale by visual aerial
survey of changes in foliage color. However, by the time visual changes occur, adult beetles may have
colonized new trees several kilometers away. Field inspection most accurately detects early stage
infestation that cannot be observed in visible wavelengths, but it is impractical for regional surveys.
Multispectral remote sensing has shown promise in this application, particularly if the spatial resolution
supports crown-scale analysis since infestation and thus health status can vary at this level. High
resolution spectral and structural data from Unmanned Aerial Vehicle (UAV) and airborne platforms
enable more efficient crown-scale monitoring of larger areas at extremely high spatial resolution. These
platforms potentially extend monitoring areas while accurately detecting health status change through
data fusion including combined spectral-structural metrics.
In this study, we use fusion of UAV structure-from-motion and multispectral imagery to classify crown
health status at sites located 250 km north of Anchorage, Alaska. Separately, we use fusion of lidar and
hyperspectral imagery from Goddard Lidar Hyperspectral (G-LiHT) to classify health status and compare
with UAV results. This comparison demonstrates scaling up to an airborne platform can provide accurate
crown-scale detection over larger areas. Three status classes were designated: alive or fully healthy,
green which appear healthy but have mechanical signs of infestation, and dead with visible foliage
change. Preliminary results indicate accurate health status predictions by both UAV and G-LiHT
platforms. The best classifiers easily distinguished both alive and green from dead but had limitations
separating alive from green. The regional scale and high-resolution of G-LiHT imagery could greatly
improve our knowledge of the spatial distribution of spruce beetle outbreaks and potentially aid in
targeted mitigation efforts.
Alaska and experiences population outbreaks that can impact millions of acres. Outbreaks typically
occur after one or more years of above average warm, dry conditions and are expected to increase in
severity with climate change. Detection of outbreaks typically occurs on a regional scale by visual aerial
survey of changes in foliage color. However, by the time visual changes occur, adult beetles may have
colonized new trees several kilometers away. Field inspection most accurately detects early stage
infestation that cannot be observed in visible wavelengths, but it is impractical for regional surveys.
Multispectral remote sensing has shown promise in this application, particularly if the spatial resolution
supports crown-scale analysis since infestation and thus health status can vary at this level. High
resolution spectral and structural data from Unmanned Aerial Vehicle (UAV) and airborne platforms
enable more efficient crown-scale monitoring of larger areas at extremely high spatial resolution. These
platforms potentially extend monitoring areas while accurately detecting health status change through
data fusion including combined spectral-structural metrics.
In this study, we use fusion of UAV structure-from-motion and multispectral imagery to classify crown
health status at sites located 250 km north of Anchorage, Alaska. Separately, we use fusion of lidar and
hyperspectral imagery from Goddard Lidar Hyperspectral (G-LiHT) to classify health status and compare
with UAV results. This comparison demonstrates scaling up to an airborne platform can provide accurate
crown-scale detection over larger areas. Three status classes were designated: alive or fully healthy,
green which appear healthy but have mechanical signs of infestation, and dead with visible foliage
change. Preliminary results indicate accurate health status predictions by both UAV and G-LiHT
platforms. The best classifiers easily distinguished both alive and green from dead but had limitations
separating alive from green. The regional scale and high-resolution of G-LiHT imagery could greatly
improve our knowledge of the spatial distribution of spruce beetle outbreaks and potentially aid in
targeted mitigation efforts.