Browsing by Subject "hyperspectral imaging"

Arctic ecosystems face drastic changes in community structure due to warming, shrubification, permafrost loss, and other environmental changes. Due to the spatial heterogeneity of these ecosystems, understanding such changes on a local scale requires high-resolution data. Earth observation using satellite imagery and aerial photography has become a staple in mapping large areas and general patterns. Advances in sensor technology, the proliferation of unmanned aerial vehicles (UAVs), and increases in processing capacity enable the use of higher spatial and spectral resolutions. As a result, more detailed ecological observations can be made using remote sensing methods. In this thesis, I assess how increased spectral resolution affects the remote-sensing based modelling of plant communities in low-growth oroarctic tundra heaths. Based on a large field observation dataset, I estimate biomass, leaf area index, species richness, Shannon's biodiversity index, and fuzzy community clusters. I then build random forest models of these with image data of varying spectral, spatial, and temporal specifications and topographical data. Finally, I create maps of the vegetation. Leaf area index and biomass are best estimated of the response variables, with R2 values of 0.64 and 0.59, respectively, with multispectral data proving the most important explanatory dataset. Biodiversity metrics are best estimated with R2 values of 0.40–0.50 with the most important explanatory variables being topographical and hyperspectral, and community cluster with R2 values of 0.27–0.53, with the importance of various explanatory variables depending on the cluster being estimated. These results can help choose a suitable high-resolution remote sensing approach for modelling plant communities in similar conditions.