Rock glaciers automatic mapping using optical imagery and convolutional neural networks
Abstract Despite their relevance in alpine environments, rock glaciers remain uncharted in several regions of the globe due to the considerable efforts required in the mapping process. To develop a support tool for rock glacier mapping, this study investigates the feasibility using artificial intell...
| Published in: | Permafrost and Periglacial Processes |
|---|---|
| Main Author: | |
| Other Authors: | |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
Wiley
2020
|
| Subjects: | |
| Online Access: | http://dx.doi.org/10.1002/ppp.2076 https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fppp.2076 https://onlinelibrary.wiley.com/doi/pdf/10.1002/ppp.2076 https://onlinelibrary.wiley.com/doi/full-xml/10.1002/ppp.2076 |
| Summary: | Abstract Despite their relevance in alpine environments, rock glaciers remain uncharted in several regions of the globe due to the considerable efforts required in the mapping process. To develop a support tool for rock glacier mapping, this study investigates the feasibility using artificial intelligence to recognize these landforms on satellite optical imagery. The results of this exploratory analysis indicate that convolutional neural networks can accurately classify unshadowed and snow‐free rock glaciers images (88% true positives and 79% true negatives on the test set). A simple detection algorithm is also tested, achieving inferior, but still promising, performances. The algorithm works on large orthoimages and is capable of spotting large rock glaciers characterized by a ridge–furrow structure, although the detection results in several false positives. Overall, the results suggest that further developments of this methodology can produce useful tools to facilitate efforts in mapping rock glaciers for large uncharted regions. |
|---|