Projects per year
Project Details
Description
AI4ARCTIC will
1. Exploit deep learning techniques, which have proven their capabilities in the computer vision domain, in the field Earth observation of the cryosphere
Deep Learning (DL), in particular deep convolutional neural networks (CNNs), will be used for Earth observation (EO) applications within the cryosphere, focusing on sea ice and snow. We will train DL systems from relevant training data and demonstrate the capability of DL by applying it to large-scale inference of cryosphere related variables.
2. Build new models (machine learning architecture) and methodologies which will also take into consideration physical knowledge and scales of the phenomenon under analysis during the training phase
Several physical processes related to the observed EO signal as well as the physics and scale of the observed phenomenon are often well understood and will be used as guideline for designing efficient DL models. Such prior physical knowledge may also specify constraints that should hold over the output space in order to ensure physically acceptable predictions.
3. Contribute to the evolution of the current European infrastructure capabilities (e.g. DIAS) feeding it with requirements to exploit machine learning, during both training and production phases, for large scale EO applications
The requirements for the most important European infrastructure initiatives, such as DIAS, will be reviewed and related to the infrastructure needs of the applications developed in this project. The outcome requirements are then to be provided to relevant European infrastructure capabilities.
4. Contribute to the evolution of the European capabilities with the provisioning of open-source API to support the exploitation of deep learning techniques in particular for EO data access in the training and production phases
The dataset and example code on how to apply deep learning for extraction of snow and sea ice information in polar regions will be made available as open-source.
5. Increase exploitation capabilities to apply machine learning techniques in Earth observation data in the polar region by contributing to existing European initiatives
Few applications in polar regions have so far used DL. This project represents an initiative to change this by making DL readily available for polar region applications.
6. Demonstrate how the developed models would run on a large-scale operational use case
The models are to be demonstrated on real applications in a large geographical area of the polar region in close collaboration with the respective user communities. The sea-ice application will be demonstrated in the vast Greenland waters, while snow will be demonstrated in the Scandinavian Peninsula.
7. Engage new stakeholders (e.g. the hydro power community) offering them new exploitation capabilities for observation data
Users, preferably representing a larger community, and typically not currently using DL, are to be contacted and preferably collaborated with to demonstrate the potential of using DL in cryospheric applications.
8. Offer to the AI4EO community case studies providing open-data/open-source and evolving computer vision methodologies (architecture)
Case studies from the project consisting of data and open-source code will be made available to relevant ML communities exploiting EO applications.
1. Exploit deep learning techniques, which have proven their capabilities in the computer vision domain, in the field Earth observation of the cryosphere
Deep Learning (DL), in particular deep convolutional neural networks (CNNs), will be used for Earth observation (EO) applications within the cryosphere, focusing on sea ice and snow. We will train DL systems from relevant training data and demonstrate the capability of DL by applying it to large-scale inference of cryosphere related variables.
2. Build new models (machine learning architecture) and methodologies which will also take into consideration physical knowledge and scales of the phenomenon under analysis during the training phase
Several physical processes related to the observed EO signal as well as the physics and scale of the observed phenomenon are often well understood and will be used as guideline for designing efficient DL models. Such prior physical knowledge may also specify constraints that should hold over the output space in order to ensure physically acceptable predictions.
3. Contribute to the evolution of the current European infrastructure capabilities (e.g. DIAS) feeding it with requirements to exploit machine learning, during both training and production phases, for large scale EO applications
The requirements for the most important European infrastructure initiatives, such as DIAS, will be reviewed and related to the infrastructure needs of the applications developed in this project. The outcome requirements are then to be provided to relevant European infrastructure capabilities.
4. Contribute to the evolution of the European capabilities with the provisioning of open-source API to support the exploitation of deep learning techniques in particular for EO data access in the training and production phases
The dataset and example code on how to apply deep learning for extraction of snow and sea ice information in polar regions will be made available as open-source.
5. Increase exploitation capabilities to apply machine learning techniques in Earth observation data in the polar region by contributing to existing European initiatives
Few applications in polar regions have so far used DL. This project represents an initiative to change this by making DL readily available for polar region applications.
6. Demonstrate how the developed models would run on a large-scale operational use case
The models are to be demonstrated on real applications in a large geographical area of the polar region in close collaboration with the respective user communities. The sea-ice application will be demonstrated in the vast Greenland waters, while snow will be demonstrated in the Scandinavian Peninsula.
7. Engage new stakeholders (e.g. the hydro power community) offering them new exploitation capabilities for observation data
Users, preferably representing a larger community, and typically not currently using DL, are to be contacted and preferably collaborated with to demonstrate the potential of using DL in cryospheric applications.
8. Offer to the AI4EO community case studies providing open-data/open-source and evolving computer vision methodologies (architecture)
Case studies from the project consisting of data and open-source code will be made available to relevant ML communities exploiting EO applications.
Layman's description
The AI for the Arctic (AI4ARCTIC) project applies deep learning, in particular deep convolutional neural networks, for Earth observation applications within the cryosphere, focusing on sea ice and snow. The project trains deep-learning systems from relevant training data, and tests and demonstrates the capability of deep learning by applying it to large-scale inference of cryosphere-related variables.
The project focusses on two use cases, one on snow mapping in Scandinavia and the other on sea ice charting in the waters around Greenland.
The project focusses on two use cases, one on snow mapping in Scandinavia and the other on sea ice charting in the waters around Greenland.
Acronym | AI4ARCTIC |
---|---|
Status | Finished |
Effective start/end date | 03/02/2020 → 02/02/2021 |
Collaborative partners
- Technical University of Denmark
- Norwegian Computing Center (lead)
- Danish Meteorological Institute
- Nansen Environmental and Remote Sensing Center
- Swedish Meteorological and Hydrological Institute
Fingerprint
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.
Projects
- 1 Finished
-
ASIP: Automated Downstream Sea Ice Products for Greenland Waters
Pedersen, L. T., Saldo, R., Skriver, H., Malmgren-Hansen, D. & Nielsen, A. A.
01/11/2017 → 28/02/2021
Project: Research