Analysis of high resolution multispectral MEIS imagery for spruce budworm damage assessment on a single tree basis

Abstract Forty centimeter resolution MEIS (Multispectral Electro-optical Imaging Scanner) data in five visible and near-infrared bands were acquired August 1984 on Cape Breton Island, Nova Scotia, Canada. Spectral signatures of individual balsam fir [Abies balsamea (L.) Mill.] trees with varying levels of cumulative defoliation caused by the spruce budworm (Choristoneura fumiferana [Clem.]) were analyzed to determine their relationships with defoliation level. The relationship between visually estimated tree defoliation and spectral features was linear for trees with >20% defoliation and curvilinear if the entire defoliation range from healthy to dead trees was considered. There was lower discrimination capability at low defoliation levels. Single variable regression models were generally sufficient for predicting defoliation. Of the five spectral bands, the 445 nm band was the best for defoliation discrimination. Transformed spectral features were better than the original bands, especially ratios or normalized differences of the 845 nm and 665 nm bands. Signatures from the whole tree and sunlit side provided the best measures for predicting individual tree defoliation, while the signature of the shaded side gave the poorest results. Single pixels representing the top of the trees were also effective for defoliation assessment and are relatively simple, fast, and efficient to specify on the imagery. The best empirical model for predicting defoliation derived from this study had a prediction interval of ± 17% defoliation, with an interval of ± 16% for trees with defoliation of 26–100 and ±19% for trees with 0–25% defoliation (α = 0.05). The highest accuracy of the models for predicting defoliation class when tested on an independent sample was 51% for eight classes, 53% for seven classes, 60% for six classes, 80% for five classes, and 92% for four classes. Results demonstrated that high resolution MEIS data has good potential for single tree defoliation assessment suitable for sample plot or site-specific surveys.