Brief communication: Rapid machine-learning-based extraction and measurement of ice wedge polygons in high-resolution digital elevation models
We present a workflow for the rapid delineation and microtopographic characterization of ice wedge polygons within high-resolution digital elevation models. At the core of the workflow is a convolutional neural network used to detect pixels representing polygon boundaries. A watershed transformation...
| Published in: | The Cryosphere |
|---|---|
| Main Authors: | , , , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
Copernicus Publications
2019
|
| Subjects: | |
| Online Access: | https://doi.org/10.5194/tc-13-237-2019 https://www.the-cryosphere.net/13/237/2019/tc-13-237-2019.pdf https://doaj.org/article/b989ebc5f8114d9fb04ee30e276e6b59 |
| Summary: | We present a workflow for the rapid delineation and microtopographic characterization of ice wedge polygons within high-resolution digital elevation models. At the core of the workflow is a convolutional neural network used to detect pixels representing polygon boundaries. A watershed transformation is subsequently used to segment imagery into discrete polygons. Fast training times (<5 min) permit an iterative approach to improving skill as the routine is applied across broad landscapes. Results from study sites near Utqiaġvik (formerly Barrow) and Prudhoe Bay, Alaska, demonstrate robust performance in diverse tundra settings, with manual validations demonstrating 70–96 % accuracy by area at the kilometer scale. The methodology permits precise, spatially extensive measurements of polygonal microtopography and trough network geometry. |
|---|