Abstract ID: 3.11578 | Accepted as Talk | Talk/Oral | TBA | TBA

Treeline dynamics have been studied using a range of remote sensing and GIS methods depending on the scale of analysis (regional, landscape, hillslope and stand scale). At the landscape scale, alpine forest can be delineated from grassland using medium resolution imagery (10-30 m spatial resolution). At the hillslope scale, high resolution imagery (5 m spatial resolution) can help detect the general pattern of treeline ecotone and any vertical or lateral shifts, while very high-resolution imagery (<1 m spatial resolution) could help detect the pattern within the ecotone by telling individual trees or clusters of trees apart from the surrounding low-stature vegetation. However, there is a dearth of such very high-resolution data for treeline ecotone sites around the world, especially those in the global south. Therefore, there is the need for a treeline pattern detection method that works efficiently using limited data and that can be applied with high level of accuracy to diverse geographical regions. We are developing a deep learning method trained on the best available data; but that also works well for lower-quality data from new alpine treeline sites. Through this process, we are studying what level of detail and spatial accuracy is needed to characterise different aspect of treeline spatial pattern to answer different ecological questions. An effective method should be able to detect treeline-ecotone patterns in a consistent and comparable manner to allow a global comparison of patterns and their relation to driving factors and processes.