Debris-covered glaciers mapping based on machine learning and multi-source satellite images over Eastern Pamir
Debris-covered glaciers present significant challenges for accurately mapping and monitoring glacier dynamics, particularly in regions like the Eastern Pamir Plateau. This study shows a new hybrid ensemble classifier that uses random forest and decision tree algorithms to make mapping debris-covered...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-04-01
|
| Series: | Big Earth Data |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/20964471.2025.2463726 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Debris-covered glaciers present significant challenges for accurately mapping and monitoring glacier dynamics, particularly in regions like the Eastern Pamir Plateau. This study shows a new hybrid ensemble classifier that uses random forest and decision tree algorithms to make mapping debris-covered glaciers more accurate using data from multiple satellites. The method leverages features derived from the SDGSAT-1, Sentinel-2, ASTER GDEM, and ITS_LIVE datasets, including color, texture, topography, land surface temperature, and velocity data. Conventional glacier mapping techniques often misclassify debris-covered areas due to their spectral similarity to the surrounding terrain, making this work crucial in addressing these limitations. To improve the accuracy of recognition between debris-covered glaciers and non-glaciated areas by capitalizing on the strengths of multiple machine-learning algorithms and diverse data sources. The hybrid ensemble classifier did better than single-classifier models, with an overall accuracy of 92% and a Kappa coefficient of 0.885. It successfully delineated debris-covered glacier boundaries that closely matched established glacier inventories while offering a more detailed mapping of debris extent. Key innovations in this research include integrating SDGSAT-1 data, which opens new avenues for glacier monitoring, and the development of an advanced feature selection strategy that enhances classification accuracy. Further, the study introduces new spectral indices and temperature-based metrics specifically designed for debris-covered glacier identification. This was a significant step forward from previous work in the area. |
|---|---|
| ISSN: | 2096-4471 2574-5417 |