Wetlands in the Arctic constitute an effective long term carbon storage. With increasing temperatures and permafrost melting, these greenhouse gases will be released in the atmosphere.
In this project, the student will consider an existing dataset for mapping wetland ecosystems in the Arctic (Jiang et al, 2019). Based on remote sensing images and several derived indices, this work will involve training a variety of deep learning models (e.g. He et al., 2016) and test the adequacy and accuracy of image recognition methods to delineate the carbon sinks.
Recommendations on the most adequate types of data / model architecture are expected (Zhu et al., 2017).
- Explore a dataset for classification of Arctic wetlands
- Train a series of machine learning models to predict different land types
- Provide recommendations for future work
- Jiang, von Ness, Loisel, Wang (2019). ArcticNet: a deep learning solution to classify Arctic wetlands
- He, Zhang, Ren, Sun (2016), Deep residual learning for image recognition, CVPR.
- Zhu, et al. (2017), Deep learning in remote sensing: a review and future directions, IEEE GRSM.
- Completion of GRS-34806 Deep learning course or equivalent
- Programming skills in Python (or high motivation for learning it)
- Some background in statistics and/or machine learning is an asset
Theme(s): Sensing & measuring; Modelling & visualisation
Attention: this topic is to be performed at the Swiss Institute of Technology (EPFL), in Sion, Switzerland.