Comparison of Hyperspectral Data for the Previsual Detection of Balsam Woolly Adelgid and Hemlock Woolly Adelgid Infestations

Our studies have used hyperspectral data collected at the branch level. Spectral data were collected using a Geophysical Environmental Research Corp. (GER) 2600 handheld spectroradiometer with a spectral resolution of 1.5 nm from 350 nm to 1050 nm and a resolution of 11.5 nm from 1050 nm to 2500 nm. In the case of balsam woolly adelgid, we have concentrated on subalpine fir (Abies lasiocarpa), the primary host of this insect in Idaho. Our studies of hemlock woolly adelgid have concentrated on its primary host in western North Carolina, Tsuga canadensis. For both insect-tree pairs, spectral data were collected from trees in various stages of infestation. Five branches were cut from each tree that was examined. Branches were cut from various heights and orientations throughout the canopy of the trees. The branches and foliage were placed on a flat black surface with negligible amounts of measurable radiation, and five measurements per tree were obtained in an iterative manner, with the foliage being rearranged between each measurement. The radiometer was placed at a height of approximately 50 cm above the branch samples, and measurements were made when the sun angle was within 10^o of solar noon. The spectra for these five replicates of branch measurements were averaged to obtain a measure of each tree’s reflectance properties. The data for each tree were smoothed using a weighted moving filter, and comparisons were made of the spectral response among infestation classes.

Fig. 1 - Spectral measurements of
subalpine firs

In Idaho, subalpine firs in three infestation categories were sampled. The categories included trees that had no current infestation with balsam woolly adelgid (BWA), trees that were infested with balsam woolly adelgid but had no apparent crown fading, and trees that were infested with balsam woolly adelgid and had visible signs of this infestation. Using Analysis of Variance procedures and the SAS statistical analysis package, significant differences were found among the three infestation categories for some wavelength regions. Our results demonstrated a consistent response in the normalized spectral reflectance curve of subalpine fir, stressed by infestation of BWA, across the reflectance spectrum shown in Figure 1. More specifically, there is an increased reflectance in the visible region of the reflectance curve (< 700 nm), decreased reflectance in the NIR plateau (centered around 1000 nm), and increased reflectance in the shortwave infrared region (beginning around 1450 nm) as visual decline becomes apparent. The overall changes in spectra are similar to those reported for other stresses in balsam fir (Luther and Carroll 1999). Multispectral aerial imagery (Landsat and SPOT data) was also collected for areas with active BWA infestations. Because of the relatively narrow canopy architecture of subalpine fir and the patchy distribution of the species in the areas of data collection, no conclusive results were obtained.

Fig. 2 - Spectral measurements of
eastern white pine

In North Carolina, hemlock trees that were recently infested (within the past year) or that had been infested for multiple years were sampled as was eastern white pine (the only other conifer present within the stands that we sampled) during June of 2005. No uninfested stands of hemlock were found within the study areas. There were visible differences in the overall spectral measurements between the hemlocks that were recently infested with hemlock woolly adelgid and those that had been infested for a longer period of time (Figure 2). The spectral signature of eastern white pine, the only other conifer present within the stands that could be confused with the hemlocks, differed significantly from both categories of infested hemlock within the stands (Figure 2). The pattern of decreased spectral values with increasing stress is similar to the decreases measured in subalpine fir infested with balsam woolly adelgid in the NIR plateau and increases again in the shortwave IR region (Figure 1). The ability to distinguish declining hemlock at the branch level also supports the prior landscape-level investigations of Pontius and others (2005), but larger data sets from a variety of geographic locations are still needed.