PS 33-211 - An elevator-to-extinction: why high-quality mapping of alpine vegetation matters.

Abstract: Within mountain systems, warming temperatures associated with climate change is altering plant species distributions and the upper elevational limits for many species are moving upslope, tracking favourable climate. Species at mountain summits are also expected to face population declines as favourable habitat decreases with elevation. This ‘elevator-to-extinction’ process requires a detailed understanding of species distributions and available habitat so that models of competition can be developed to and tested in key alpine environments such as within the Central Mountain Range of Taiwan. This research uses a combination of fine spatial resolution imagery (aerial photographs taken with UAV devices, Sentinal-2 MSI and PlanetScope) to identify broad habitat classes. Within class variability was assessed using Principal Components Analysis, Independent Component Analysis, Tasselled-cap Transformation and K-means classification using R packages. The clusters identified using these exploratory data analyses fed forward into a Random Forest classification to map habitat boundaries for stratification of field-based sampling of vegetation types. Training and testing data for the algorithm were separated using K folds cross validation. Field data on vegetation structure, field photos, expert knowledge and high-resolution aerial photography were used as reference data. The fine-scale habitat maps for the Hehuanshan mountain region created for each imagery type were compared based on classification accuracy. Eleven habitat classes were mapped, with each class having over 90% classification accuracy across all imagery types. Classification accuracy was greater for more abundant classes (e.g. Bamboo grassland) and less accurate for rarer habitat types (e.g. Juniper mixtures). Classification accuracy was greater using the coarser scale Sentinal-2 and PlanetScope imagery, despite the reduction in image resolution compared to the UAV imagery. The findings of the research contribute to a greater understanding of the impact of habitat squeezing in the Central Mountain Range and other similar alpine zones that face high levels of alpine plant loss under climate change.