Deep Learning-Based Inventory of Debris-Covered Glaciers in the Tianshan Mountains for 2020 and 2025
收藏资源简介：
# Deep Learning-Based Inventory of Debris-Covered Glaciers in the Tianshan Mountains for 2020 and 2025 ## Overview This repository contains the deep learning code and a dual-epoch (2020 and 2025) inventory of debris-covered glaciers (DCGs) across the Tianshan Mountains. Multi-source remote sensing data—including Sentinel-2 optical imagery, Sentinel-1 SAR, Landsat-derived land surface temperature (LST), and DEM-based topographic features were extracted and composited on the Google Earth Engine (GEE) platform. These features were used to train a UNet++ deep learning semantic segmentation framework, which generated 10 m resolution spatial distribution products for bare ice and supraglacial debris. ## Repository Contents Tianshan_DCG_Mapping_2020&2025.zip ├── GEE-Feature_Extraction.js # GEE feature extraction script ├── train_unetpp.py # UNet++ model training ├── inference_postprocess.py # Inference and post-processing ├── Tianshan_Glacier_Debris_2020_10m.tif # Glacier inventory 2020 └── Tianshan_Glacier_Debris_2025_10m.tif # Glacier inventory 2025 ## Inventory Products Description - Format: GeoTIFF (Raster) - Spatial Resolution: 10 m - Coordinate System: EPSG:4326 (WGS 84) - Pixel Values: 0: Background (No glacier) 1: Bare Ice 2: Supraglacial Debris ## Software Requirements - GEE Code: Run directly in [Google Earth Engine](https://code.earthengine.google.com/). - Deep Learning Code: Python 3.8+, PyTorch, Rasterio, Geopandas, SciPy. (Optional: CuPy & CuCIM for GPU-accelerated post-processing). ## Authors - **Qi Wang** — China Institute of International Rivers and Eco-Security, Yunnan University Email: wqi876@stu.ynu.edu.cn - **Kunpeng Wu** (Corresponding author) — China Institute of International Rivers and Eco-Security, Yunnan University Email: wukunpeng@ynu.edu.cn ## Acknowledgements We thank ESA for Sentinel-1/2 data, NASA/USGS for Landsat data, the Copernicus Climate Change Service for ERA5-Land reanalysis data, and GLIMS for the RGI 7.0 dataset.
