The potential of using hyperspectral data for plant and lichen biomonitoring of environmental pollution
(1) Department of Agriculture, Food and Environment, University of Pisa, Via del Borghetto 80, 56124 Pisa, Italy, (2) Department of Biology, University of Pisa, Via Ghini 13, 56126 Pisa, Italy
Hyperspectral detection has emerged as a promising tool, being a non-destructive, rapid, and relatively low-cost technique to monitor vegetation (as well as other targets). Reflection of light in the visible, near-infrared, and short-wave infrared (350-2500 nm) can provide a comprehensive assessment of shifts in macroscopic symptoms and the underlying morpho-anatomical and physio-chemical responses of plants to environmental constraints. This spectral approach is also scalable from leaf to remote sensing level, using airborne and space platforms. However, the use of this technology for the biomonitoring of environmental changes and/or environmental pollution effects – a well-known low-cost and effective method to estimate levels of air pollutants and their impact on biological receptors – has remained underdeveloped. Therefore, the present work aims to highlight the potential of hyperspectral data for the biomonitoring of environmental changes using plants and lichens. First, it briefly reports basic concepts of vegetation spectroscopy, including the most used approaches for exploiting information from hyperspectral data (e.g., spectral indexes, trait retrieval, spectral classification). Then, it reviews the few available studies on the topic. Finally, it shows that this approach could also be used for lichen biomonitoring, as using hyperspectral data we were able to discriminate with high accuracy (>95%) native specimens of the forest lichen Lobaria pulmonaria (L.) Hoffm. under different environmental conditions (i.e., acclimated/exposed or not to sudden increase of solar radiation, in relation to the effects of forest management), as well as to predict with high accuracy key lichen stress makers, i.e., maximal photochemical efficiency of photosystem II (PSII), PSII performance index and chlorophyll content (R2 for validation: 0.6-0.8). We believe that these preliminary outcomes may encourage further research to highlight the potential of the proposed approach.
Keywords: Air quality, hyperspectral imaging, machine learning, spectral classification, spectral indexes, spectroscopy, trait retrieval
