Assessing Hemlock Decline Using Visible and Near-Infrared Spectroscopy: Indices Comparison and Algorithm Development

• Published in: Applied spectroscopy 2005-06, Vol.59 (6), p.836-843
• Main Authors: Pontius, Jennifer, Hallett, Richard, Martin, Mary
• Format: Article
• Language: English
• Subjects: Algorithms; Animals; Aphids - pathogenicity; Forestry - methods; Least-Squares Analysis; Plant Diseases - classification; Spectrophotometry, Infrared - methods; Tsuga - chemistry; Tsuga - classification; Tsuga - parasitology
• ISSN: 0003-7028; 1943-3530
• DOI: 10.1366/0003702054280595

Near-infrared reflectance spectroscopy was evaluated for its effectiveness at predicting pre-visual decline in eastern hemlock trees. An ASD FieldSpec Pro FR field spectroradiometer measuring 2100 contiguous 1-nm-wide channels from 350 nm to 2500 nm was used to collect spectra from fresh hemlock foliage. Full spectrum partial least squares (PLS) regression equations and reduced stepwise linear regression equations were compared. The best decline predictive model was a 6-term linear regression equation (R2 = 0.71, RMSE = 0.591) based on: Carter Miller Stress Index (R694/R760), Derivative Chlorophyll Index (FD705/FD723), Normalized Difference Vegetation Index ((R800 – R680)/(R800 + R680)), R950, R1922, and FD1388. Accuracy assessment showed that this equation predicted an 11-class decline rating with a 1-class tolerance accuracy of 96% and differentiated healthy trees from those in very early decline with 72% accuracy. These results indicate that narrow-band sensors could be developed to detect very early stages of hemlock decline, before visual symptoms are apparent. This capability would enable land managers to identify early hemlock woolly adelgid infestations and monitor forest health over large areas of the landscape.