Meta-learning to address diverse Earth observation problems across resolutions
Abstract Earth scientists study a variety of problems with remote sensing data, but they most often consider them in isolation from each other, which limits information flows across disciplines. In this work, we present METEOR, a meta-learning methodology for Earth observation problems across differ...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-01-01
|
| Series: | Communications Earth & Environment |
| Online Access: | https://doi.org/10.1038/s43247-023-01146-0 |
| _version_ | 1827381931558305792 |
|---|---|
| author | Marc Rußwurm Sherrie Wang Benjamin Kellenberger Ribana Roscher Devis Tuia |
| author_facet | Marc Rußwurm Sherrie Wang Benjamin Kellenberger Ribana Roscher Devis Tuia |
| author_sort | Marc Rußwurm |
| collection | DOAJ |
| description | Abstract Earth scientists study a variety of problems with remote sensing data, but they most often consider them in isolation from each other, which limits information flows across disciplines. In this work, we present METEOR, a meta-learning methodology for Earth observation problems across different resolutions. METEOR is an adaptive deep meta-learning model with several modifications that allow it to ingest images with a variable number of spectral channels and to predict a varying number of classes per downstream task. It uses knowledge mined from land cover information worldwide to adapt to new unseen target problems with few training examples. METEOR outperforms competing self-supervised approaches on five downstream tasks, showing its relevance to addressing novel and impactful geospatial problems with only a handful of labels. |
| first_indexed | 2024-03-08T14:12:07Z |
| format | Article |
| id | doaj.art-5bd59714a1e0442fa90e073dcd11187f |
| institution | Directory Open Access Journal |
| issn | 2662-4435 |
| language | English |
| last_indexed | 2024-03-08T14:12:07Z |
| publishDate | 2024-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Communications Earth & Environment |
| spelling | doaj.art-5bd59714a1e0442fa90e073dcd11187f2024-01-14T12:37:50ZengNature PortfolioCommunications Earth & Environment2662-44352024-01-015111410.1038/s43247-023-01146-0Meta-learning to address diverse Earth observation problems across resolutionsMarc Rußwurm0Sherrie Wang1Benjamin Kellenberger2Ribana Roscher3Devis Tuia4Environmental Computational Science and Earth Observation Laboratory (ECEO), École Polytechnique Fédérale de Lausanne (EPFL)Goldman School of Public Policy, University of CaliforniaEnvironmental Computational Science and Earth Observation Laboratory (ECEO), École Polytechnique Fédérale de Lausanne (EPFL)Institute of Bio- and Geosciences, Forschungszentrum Jülich GmbHEnvironmental Computational Science and Earth Observation Laboratory (ECEO), École Polytechnique Fédérale de Lausanne (EPFL)Abstract Earth scientists study a variety of problems with remote sensing data, but they most often consider them in isolation from each other, which limits information flows across disciplines. In this work, we present METEOR, a meta-learning methodology for Earth observation problems across different resolutions. METEOR is an adaptive deep meta-learning model with several modifications that allow it to ingest images with a variable number of spectral channels and to predict a varying number of classes per downstream task. It uses knowledge mined from land cover information worldwide to adapt to new unseen target problems with few training examples. METEOR outperforms competing self-supervised approaches on five downstream tasks, showing its relevance to addressing novel and impactful geospatial problems with only a handful of labels.https://doi.org/10.1038/s43247-023-01146-0 |
| spellingShingle | Marc Rußwurm Sherrie Wang Benjamin Kellenberger Ribana Roscher Devis Tuia Meta-learning to address diverse Earth observation problems across resolutions Communications Earth & Environment |
| title | Meta-learning to address diverse Earth observation problems across resolutions |
| title_full | Meta-learning to address diverse Earth observation problems across resolutions |
| title_fullStr | Meta-learning to address diverse Earth observation problems across resolutions |
| title_full_unstemmed | Meta-learning to address diverse Earth observation problems across resolutions |
| title_short | Meta-learning to address diverse Earth observation problems across resolutions |
| title_sort | meta learning to address diverse earth observation problems across resolutions |
| url | https://doi.org/10.1038/s43247-023-01146-0 |
| work_keys_str_mv | AT marcrußwurm metalearningtoaddressdiverseearthobservationproblemsacrossresolutions AT sherriewang metalearningtoaddressdiverseearthobservationproblemsacrossresolutions AT benjaminkellenberger metalearningtoaddressdiverseearthobservationproblemsacrossresolutions AT ribanaroscher metalearningtoaddressdiverseearthobservationproblemsacrossresolutions AT devistuia metalearningtoaddressdiverseearthobservationproblemsacrossresolutions |