Mapping and inventorying rock glaciers on the Tibetan Plateau from Planet Basemaps using deep learning

Abstract

Rock glaciers are geomorphologically valuable indicators of permafrost distribution and form potentially important hydrological resources in the context of future climate change. Despite the widespread distribution of permafrost on the Tibetan Plateau and its reputation as the "water tower of Asia", this region lacks a complete inventory and systematic investigation of rock glaciers. In this study, we develop a deep-learning-based approach for mapping rock glaciers on the Tibetan Plateau. A powerful deep learning network, DeepLabv3+, is trained using Planet Basemaps as training imagery and multi-source rock glacier inventories as training labels. The well-trained model is then used to map new rock glaciers. The visually consistent and cloud-free properties of Planet Basemaps are crucial for developing comprehensive maps of rock glacier distribution; and the rock glacier inventories from multiple regions can improve the volume and diversity of the training dataset. The deep learning mapped results present strong identification and acceptable boundary delineation of rock glaciers, indicating that the deep learning model could serve as a useful tool for facilitating the inventory of rock glaciers over vast regions. Based on the deep learning outputs, we compile 4233 rock glaciers on eight subregions of the Tibetan Plateau, which are widespread in the surrounding regions while being scarcely distributed in inner areas. Talus- and glacier-connected rock glaciers are two major classes, which are dominant on the southeastern and densely distributed on the northwestern Tibetan Plateau, respectively. The regions with steep slopes are favored by rock glacier clusters with high density, and glacier-abundant regions tend to breed large rock glaciers. The proposed rock glacier mapping method effectively speeds up inventorying efforts, which will be used to map and inventory rock glaciers on the entire Tibetan Plateau. The complete inventory will offer a significant contribution to the global catalog and serves as a benchmark dataset for modeling and monitoring the state of permafrost in a changing climate.