Using visible-near-infrared spectroscopy to classify lichens at a Neotropical Dry Forest

• Published in: Ecological indicators 2020-04, Vol.111, p.105999, Article 105999
• Main Authors: Guzmán Q., J. Antonio, Laakso, Kati, López-Rodríguez, José C., Rivard, Benoit, Sánchez-Azofeifa, G. Arturo
• Format: Article
• Language: English
• Subjects: Lichen spectroscopy; Reflectance; Species discrimination; Spectral classification; Wavelet analysis
• ISSN: 1470-160X; 1872-7034
• DOI: 10.1016/j.ecolind.2019.105999

•Corticolous lichen classification using visible-near infrared spectra.•Linear spectral mixing between lichen and bark was evaluated.•Overall lichen signatures tend to present low spectral bark mixing.•Lichen classification is more accurate at higher, rather than lower taxonomic ranks.•Linear discriminant analysis presents the highest performance classifying species. The optical properties of lichens have been traditionally explored in the context of geological mapping where the encrustation of lichens on rocks may influence the detection of minerals of interest. As of today, few studies have looked into the potential of using the optical properties of lichens to classify them; however, none has investigated the classification of tropical lichens using spectroscopy. Here we explore the use of the visible-near infrared reflectance (VNIR; 450–1000 nm) to discriminate Neotropical corticolous lichens; the most abundant lichens in tropical forests. Reflectance measurements on lichens and their bark substrate were performed on 282 lichens samples of 32 species attached to their host's bark. Using these measurements, we first explored the degree of spectral mixing of bark and lichens by linear unmixing each lichen spectrum with the corresponding average species spectrum and bark spectrum. Overall, the results reveal that the lichen signatures tend to mask the spectral contributions from bark; however, there are some specific groups of species with high bark mixing probably due to their nature and the similarities between the lichen and bark spectra. Next, we classified the lichen spectra based on growth forms and taxonomic ranks (i.e., family, genus, species) using five machine learning classifiers. This analysis was conducted on raw reflectance spectra and wavelet-transformed spectra to enhance the absorption features prior to classification. As expected, the classification of lichen spectra is less accurate at species-specific levels, rather than higher taxonomic ranks. The wavelet transformation was found to enhance the general performance of classification; however, the accuracy of the classification depends on the classifier. Of the classifiers used in this study, linear discrimination applied to reflectance spectra presents the highest performance at the species level. Our results reveal the potential of using the VNIR reflectance as a method to discriminate Neotropical lichens. The introduced methodology may be conducted in the field, thus allowing the mon