[M-IS23] Mountain Science

[MIS23-13] An automated method for drawing land cover maps from time-lapse photographs of alpine landscapes

*Ryotaro Okamoto¹, Hiroyuki Oguma², Reiko Ide² (1.University of Tsukuba, 2.National Institute for Environmental Studies)

Keywords:Remote-sensing, Deep Leaning, Alpine Area

Alpine ecosystems are one of the most vulnerable ecosystems to the recent climate changes. Some studies have already reported on-going changes in physiological (e.g., Glaciers, Snow-melting) and biological (e.g., Species distributions, Phenology) environments. Over-time monitoring of alpine ecosystems is required to detect their changes against climate change to define the conservation plans. We have conducted 10-years long monitoring of Japanese alpine ecosystems using automated time-lapse cameras. This study aims to clarify the space-time distributions of such as vegetation and snow cover by developing a set of open-source software. It has two steps, 1. semantic segmentation of time-lapse photographs with semi-supervised deep learning, 2. georectification of segmentation results. The developed software will be published as open-source tools to expand alpine ecosystems' monitoring network using time-lapse cameras. In the session, we will explain the overview and future applications of it.