Mapping Capability of Hyperspectral Information on Dominant Arctic Vegetation Species Using Terrestrial Hyperspectral Imagery

Hyperspectral remote sensing serves as a valuable tool to monitor the rapid changes in the Arctic vegetation ecosystem induced by global warming. Nevertheless, studies on mapping Arctic vegetation using hyperspectral information have been limited due to data acquisition challenges in the Arctic region. To overcome this, we collected hyperspectral images measured in the Adventdalen Valley, Svalbard. We constructed labeled datasets by extracting pixel data from ground-based hyperspectral images and evaluated the classification performance of three machine learning-based classifiers, random forest (RF), support vector machine (SVM), and one-dimensional convolutional neural network (1D-CNN). The 1D-CNN exhibited the highest statistical accuracies of 0.97 and 0.93 for two independent test sets and generated the most precise vegetation map. Furthermore, we developed a spectral library using mean spectra from the labeled datasets, which serves as a fundamental reference for understanding the spectral characteristics of Arctic vegetation. Our approaches and findings have broad applicability across various remote sensing datasets, facilitating the effective monitoring of changes in Arctic vegetation as a consequence of climate change.