Improving ocean reanalyses of observationally sparse regions with transfer learning
Abstract Oceanic subsurface observations are sparse and lead to large uncertainties in any model-based estimate. We investigate the applicability of transfer learning based neural networks to reconstruct North Atlantic temperatures in times with sparse observations. Our network is trained on a time...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-01-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-86374-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832585781154152448 |
|---|---|
| author | Simon Lentz Sebastian Brune Christopher Kadow Johanna Baehr |
| author_facet | Simon Lentz Sebastian Brune Christopher Kadow Johanna Baehr |
| author_sort | Simon Lentz |
| collection | DOAJ |
| description | Abstract Oceanic subsurface observations are sparse and lead to large uncertainties in any model-based estimate. We investigate the applicability of transfer learning based neural networks to reconstruct North Atlantic temperatures in times with sparse observations. Our network is trained on a time period with abundant observations to learn realistic physical behavior. Evaluating it within a consistent data assimilation framework, this network learns and reproduces its training data’s physical patterns. Additionally, the network is able to transfer these patterns towards a historical ocean heat content estimate in times with sparse observations. Consequently, with infrequent input data, machine learning reconstructions exhibit similar physical structures, while correcting for known errors compared to state-of-the-art data assimilation products. In this manner, transfer learning can impact the initialization and evaluation of climate hindcasts. Furthermore, by exhibiting the capability to accurately transfer results from high to low-frequencies, transfer learning based neural networks showcase their relevance in mixed-frequency measurements beyond climate science. |
| format | Article |
| id | doaj-art-6ee8524444074228ab605ce58e5ae2a1 |
| institution | Kabale University |
| issn | 2045-2322 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-6ee8524444074228ab605ce58e5ae2a12025-01-26T12:29:26ZengNature PortfolioScientific Reports2045-23222025-01-011511810.1038/s41598-025-86374-4Improving ocean reanalyses of observationally sparse regions with transfer learningSimon Lentz0Sebastian Brune1Christopher Kadow2Johanna Baehr3Institute of Oceanography, Center for Earth System Sustainability, Universität HamburgInstitute of Oceanography, Center for Earth System Sustainability, Universität HamburgGerman Climate Computing Centre, DKRZInstitute of Oceanography, Center for Earth System Sustainability, Universität HamburgAbstract Oceanic subsurface observations are sparse and lead to large uncertainties in any model-based estimate. We investigate the applicability of transfer learning based neural networks to reconstruct North Atlantic temperatures in times with sparse observations. Our network is trained on a time period with abundant observations to learn realistic physical behavior. Evaluating it within a consistent data assimilation framework, this network learns and reproduces its training data’s physical patterns. Additionally, the network is able to transfer these patterns towards a historical ocean heat content estimate in times with sparse observations. Consequently, with infrequent input data, machine learning reconstructions exhibit similar physical structures, while correcting for known errors compared to state-of-the-art data assimilation products. In this manner, transfer learning can impact the initialization and evaluation of climate hindcasts. Furthermore, by exhibiting the capability to accurately transfer results from high to low-frequencies, transfer learning based neural networks showcase their relevance in mixed-frequency measurements beyond climate science.https://doi.org/10.1038/s41598-025-86374-4Transfer learningConvolutional U-netsPhysics-informedClimate hindcast initializationOcean heat content |
| spellingShingle | Simon Lentz Sebastian Brune Christopher Kadow Johanna Baehr Improving ocean reanalyses of observationally sparse regions with transfer learning Scientific Reports Transfer learning Convolutional U-nets Physics-informed Climate hindcast initialization Ocean heat content |
| title | Improving ocean reanalyses of observationally sparse regions with transfer learning |
| title_full | Improving ocean reanalyses of observationally sparse regions with transfer learning |
| title_fullStr | Improving ocean reanalyses of observationally sparse regions with transfer learning |
| title_full_unstemmed | Improving ocean reanalyses of observationally sparse regions with transfer learning |
| title_short | Improving ocean reanalyses of observationally sparse regions with transfer learning |
| title_sort | improving ocean reanalyses of observationally sparse regions with transfer learning |
| topic | Transfer learning Convolutional U-nets Physics-informed Climate hindcast initialization Ocean heat content |
| url | https://doi.org/10.1038/s41598-025-86374-4 |
| work_keys_str_mv | AT simonlentz improvingoceanreanalysesofobservationallysparseregionswithtransferlearning AT sebastianbrune improvingoceanreanalysesofobservationallysparseregionswithtransferlearning AT christopherkadow improvingoceanreanalysesofobservationallysparseregionswithtransferlearning AT johannabaehr improvingoceanreanalysesofobservationallysparseregionswithtransferlearning |