An Interpretable Deep Learning ENSO Forecasting Model
The El Niño-Southern Oscillation (ENSO) dominates Earth’s year-to-year climate variability and can often cause severe environmental and socioeconomic impacts globally. However, despite continuous ENSO theory and modeling advances, the global heat signature variations preceding ENSO events have not b...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
American Association for the Advancement of Science (AAAS)
2023-01-01
|
| Series: | Ocean-Land-Atmosphere Research |
| Online Access: | https://spj.science.org/doi/10.34133/olar.0012 |
| _version_ | 1797729081878380544 |
|---|---|
| author | Haoyu Wang Shineng Hu Xiaofeng Li |
| author_facet | Haoyu Wang Shineng Hu Xiaofeng Li |
| author_sort | Haoyu Wang |
| collection | DOAJ |
| description | The El Niño-Southern Oscillation (ENSO) dominates Earth’s year-to-year climate variability and can often cause severe environmental and socioeconomic impacts globally. However, despite continuous ENSO theory and modeling advances, the global heat signature variations preceding ENSO events have not been fully understood, especially for long-lead ENSO forecasts more than 12 months in advance. Here, we develop an interpretable, deep learning (DL)-based ENSO forecast model that uses artificial intelligence to discover the long-term spatial and temporal processes of heat signatures associated with ENSO in the global ocean. More specifically, our results highlight the critical roles of ocean interbasin interactions and tropic–extratropic interactions in ENSO forecasts and are confirmed by our sensitivity forecasting experiments. The model has good forecast performance, with an effective ENSO forecast length of 22 months on the test set (1982 to 2020) and minimal influence from the spring predictability barrier (SPB). Moreover, our experimentally validated model performance does not degrade much even with using sea surface temperature (SST) alone, which has direct implications for operational forecasts since globally complete ocean subsurface measurements are not always available. |
| first_indexed | 2024-03-12T11:23:47Z |
| format | Article |
| id | doaj.art-23f0183e023d487fa4109dbe5d5405b4 |
| institution | Directory Open Access Journal |
| issn | 2771-0378 |
| language | English |
| last_indexed | 2024-03-12T11:23:47Z |
| publishDate | 2023-01-01 |
| publisher | American Association for the Advancement of Science (AAAS) |
| record_format | Article |
| series | Ocean-Land-Atmosphere Research |
| spelling | doaj.art-23f0183e023d487fa4109dbe5d5405b42023-09-01T12:37:31ZengAmerican Association for the Advancement of Science (AAAS)Ocean-Land-Atmosphere Research2771-03782023-01-01210.34133/olar.0012An Interpretable Deep Learning ENSO Forecasting ModelHaoyu Wang0Shineng Hu1Xiaofeng Li2Key Laboratory of Ocean Circulation and Waves, Institute of Oceanology, Chinese Academy of Sciences, Qingdao, China.Division of Earth and Climate Sciences, Nicholas School of the Environment, Duke University, Durham, NC, USA.Key Laboratory of Ocean Circulation and Waves, Institute of Oceanology, Chinese Academy of Sciences, Qingdao, China.The El Niño-Southern Oscillation (ENSO) dominates Earth’s year-to-year climate variability and can often cause severe environmental and socioeconomic impacts globally. However, despite continuous ENSO theory and modeling advances, the global heat signature variations preceding ENSO events have not been fully understood, especially for long-lead ENSO forecasts more than 12 months in advance. Here, we develop an interpretable, deep learning (DL)-based ENSO forecast model that uses artificial intelligence to discover the long-term spatial and temporal processes of heat signatures associated with ENSO in the global ocean. More specifically, our results highlight the critical roles of ocean interbasin interactions and tropic–extratropic interactions in ENSO forecasts and are confirmed by our sensitivity forecasting experiments. The model has good forecast performance, with an effective ENSO forecast length of 22 months on the test set (1982 to 2020) and minimal influence from the spring predictability barrier (SPB). Moreover, our experimentally validated model performance does not degrade much even with using sea surface temperature (SST) alone, which has direct implications for operational forecasts since globally complete ocean subsurface measurements are not always available.https://spj.science.org/doi/10.34133/olar.0012 |
| spellingShingle | Haoyu Wang Shineng Hu Xiaofeng Li An Interpretable Deep Learning ENSO Forecasting Model Ocean-Land-Atmosphere Research |
| title | An Interpretable Deep Learning ENSO Forecasting Model |
| title_full | An Interpretable Deep Learning ENSO Forecasting Model |
| title_fullStr | An Interpretable Deep Learning ENSO Forecasting Model |
| title_full_unstemmed | An Interpretable Deep Learning ENSO Forecasting Model |
| title_short | An Interpretable Deep Learning ENSO Forecasting Model |
| title_sort | interpretable deep learning enso forecasting model |
| url | https://spj.science.org/doi/10.34133/olar.0012 |
| work_keys_str_mv | AT haoyuwang aninterpretabledeeplearningensoforecastingmodel AT shinenghu aninterpretabledeeplearningensoforecastingmodel AT xiaofengli aninterpretabledeeplearningensoforecastingmodel AT haoyuwang interpretabledeeplearningensoforecastingmodel AT shinenghu interpretabledeeplearningensoforecastingmodel AT xiaofengli interpretabledeeplearningensoforecastingmodel |