We show that explainable neural networks can identify regions of oceanic variability that contribute predictability on decadal timescales in a fully coupled Earth‐system model. The neural networks learn to use sea‐surface temperature anomalies to predict future continental surface temperature anomalies. We then use a neural‐network explainability method called layerwise relevance propagation to infer which oceanic patterns lead to accurate predictions made by the neural networks. In particular, regions within the North Atlantic Ocean and North Pacific Ocean lend the most predictability for surface temperature across continental North America. We apply the proposed methodology to decadal variability, although the concept is generalizable to other timescales of predictability. Furthermore, while our approach focuses on predictable patterns of internal variability within climate models, it should be generalizable to observational data as well. Our study contributes to the growing evidence that explainable neural networks are important tools for advancing geoscientific knowledge.
Predictable internal climate variability on decadal timescales (2–10 years) is associated with large‐scale oceanic processes, however these predictable signals may be masked by the noisy climate system. One approach to overcoming this problem is investigating state‐dependent predictability—how differences in prediction skill depend on the initial state of the system. We present a machine learning approach to identify state‐dependent predictability on decadal timescales in the Community Earth System Model version 2 pre‐industrial control simulation by incorporating uncertainty estimates into a regression neural network. We leverage the network's prediction of uncertainty to examine state dependent predictability in sea surface temperatures by focusing on predictions with the lowest uncertainty outputs. In particular, we study two regions of the global ocean—the North Atlantic and North Pacific—and find that skillful initial states identified by the neural network correspond to particular phases of Atlantic multi‐decadal variability and the interdecadal Pacific oscillation.more » « less
- Award ID(s):
- NSF-PAR ID:
- Publisher / Repository:
- DOI PREFIX: 10.1029
- Date Published:
- Journal Name:
- Geophysical Research Letters
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
An open question in the study of climate prediction is whether internal variability will continue to contribute to prediction skill in the coming decades, or whether predictable signals will be overwhelmed by rising temperatures driven by anthropogenic forcing. We design a neural network that is interpretable such that its predictions can be decomposed to examine the relative contributions of external forcing and internal variability to future regional sea surface temperature (SST) trend predictions in the near-term climate (2020–2050). We show that there is additional prediction skill to be garnered from internal variability in the Community Earth System Model version 2 Large Ensemble, even in a relatively high forcing future scenario. This predictability is especially apparent in the North Atlantic, North Pacific and Tropical Pacific Oceans as well as in the Southern Ocean. We further investigate how prediction skill covaries across the ocean and find three regions with distinct coherent prediction skill driven by internal variability. SST trend predictability is found to be associated with consistent patterns of decadal variability for the grid points within each region.
The North Atlantic Oscillation (NAO) is predictable in climate models at near-decadal timescales. Predictive skill derives from ocean initialization, which can capture variability internal to the climate system, and from external radiative forcing. Herein, we show that predictive skill for the NAO in a very large uninitialized multi-model ensemble is commensurate with previously reported skill from a state-of-the-art initialized prediction system. The uninitialized ensemble and initialized prediction system produce similar levels of skill for northern European precipitation and North Atlantic SSTs. Identifying these predictable components becomes possible in a very large ensemble, confirming the erroneously low signal-to-noise ratio previously identified in both initialized and uninitialized climate models. Though the results here imply that external radiative forcing is a major source of predictive skill for the NAO, they also indicate that ocean initialization may be important for particular NAO events (the mid-1990s strong positive NAO), and, as previously suggested, in certain ocean regions such as the subpolar North Atlantic ocean. Overall, we suggest that improving climate models’ response to external radiative forcing may help resolve the known signal-to-noise error in climate models.
By synthesizing recent studies employing a wide range of approaches (modern observations, paleo reconstructions, and climate model simulations), this paper provides a comprehensive review of the linkage between multidecadal Atlantic Meridional Overturning Circulation (AMOC) variability and Atlantic Multidecadal Variability (AMV) and associated climate impacts. There is strong observational and modeling evidence that multidecadal AMOC variability is a crucial driver of the observed AMV and associated climate impacts and an important source of enhanced decadal predictability and prediction skill. The AMOC‐AMV linkage is consistent with observed key elements of AMV. Furthermore, this synthesis also points to a leading role of the AMOC in a range of AMV‐related climate phenomena having enormous societal and economic implications, for example, Intertropical Convergence Zone shifts; Sahel and Indian monsoons; Atlantic hurricanes; El Niño–Southern Oscillation; Pacific Decadal Variability; North Atlantic Oscillation; climate over Europe, North America, and Asia; Arctic sea ice and surface air temperature; and hemispheric‐scale surface temperature. Paleoclimate evidence indicates that a similar linkage between multidecadal AMOC variability and AMV and many associated climate impacts may also have existed in the preindustrial era, that AMV has enhanced multidecadal power significantly above a red noise background, and that AMV is not primarily driven by external forcing. The role of the AMOC in AMV and associated climate impacts has been underestimated in most state‐of‐the‐art climate models, posing significant challenges but also great opportunities for substantial future improvements in understanding and predicting AMV and associated climate impacts.
Extending climate prediction from seasonal to decadal timescales requires realistic initialization of not only upper ocean heat content but also the Atlantic meridional overturning circulation (AMOC). However, it remains a major challenge to realistically initialize AMOC in a coupled system while also maintaining a balanced atmosphere‐ocean initial state. This study demonstrates the feasibility of generating fully coupled historical states with realistic AMOC variability. Employing a forced ocean—sea‐ice (FOSI) model simulation as the “truth,” we show reproducibility of key features of historical AMOC decadal variability in a fully coupled model by restoring sea‐surface salinity, in addition to sea‐surface temperature restoring widely used in seasonal prediction. The atmospheric state of the restored coupled model solution is much closer to that of the free coupled simulation than to the observations used in FOSI, pointing to potential advantages of using this approach for initializing decadal predictions with reduced ocean initialization shock.