Selecting Observationally Constrained Global Climate Model Ensembles Using Autoencoders and Transfer Learning
Climate modes of variability are recurring patterns that influence climate phenomena across spatial scales. Accurately representing these modes in Global Climate Models (GCMs) is crucial for assessing model performance and reducing uncertainty in future climate projections. In this study, we present...
Uloženo v:
| Vydáno v: | Journal of geophysical research. Machine learning and computation Ročník 2; číslo 1 |
|---|---|
| Hlavní autoři: | , , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Wiley
01.03.2025
|
| Témata: | |
| ISSN: | 2993-5210, 2993-5210 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | Climate modes of variability are recurring patterns that influence climate phenomena across spatial scales. Accurately representing these modes in Global Climate Models (GCMs) is crucial for assessing model performance and reducing uncertainty in future climate projections. In this study, we present a novel approach utilizing autoencoder neural networks (AEs) combined with transfer learning to evaluate the representation of monthly sea level pressure (SLP) modes over North America across five GCMs: the Geophysical Fluid Dynamics Laboratory Climate Model (GFDL‐CM4), Centro Euro‐Mediterraneo sui Cambiamenti Climatici Climate Model (CMCC‐CM2‐SR5), Canadian Earth System Model Version 5 (CanESM5), Institut Pierre‐Simon Laplace Climate Model (IPSL‐CM6A‐LR), and Hadley Center Global Environment Model Version 3 (HadGEM3‐GC31‐LL). We derived the reference regional SLP modes using autoencoders (AE) from the European Center for Medium‐Range Weather Forecasts Reanalysis (ERA5), capturing more physically consistent SLP patterns. Transfer learning was employed to adapt the pre‐trained AE, from ERA5 to the GCM outputs, enabling a direct and robust evaluation of each model’s ability to produce the observationally constrained SLP modes. This approach allowed us to rank the GCMs based on how well they replicated the reference SLP modes, providing an observationally constrained assessment of model performance. The congruence coefficients between the modeled and reference modes exceeded 0.91 for all GCMs, demonstrating strong performance in simulating regional SLP modes over North America. Among the models, HadGEM3‐GC31‐LL achieved the highest performance with an average congruence coefficient of 0.94. These results highlight the effectiveness of neural network techniques in evaluating and ranking GCMs for model intercomparison projects.
Plain Language Summary
Global climate models (GCMs) are instrumental in understanding climate systems and projecting future changes in the climate. Nonetheless, the accuracy of the model output is constrained by biases stemming from various sources, thus warranting a thorough evaluation of the model outputs before their applications. One challenge in evaluating GCMs is the dependency on the techniques utilized. Applying linear techniques evaluates aspects of the model concerned with linear relationships. However, the climate system is often characterized by complex feedback and nonlinear interactions. Therefore, in this study, we show that using nonlinear artificial neural network models such as autoencoders can result in unique sea level pressure (SLP) modes; and transfer learning can be further utilized to produce the reference modes in GCMs. The implication is a more thorough evaluation of the ability of the GCMs to produce autoencoder‐based SLP modes. Our results showed that the ability of the GCMs to produce the reference modes can depend on the modes considered but on average HadGEM3‐GC31‐LL outperformed the other GCMs during our analysis period. Hence, we present a novel technique that can enhance the ranking of an ensemble of climate models beyond linear techniques.
Key Points
A transfer learning approach with an autoencoder is used to evaluate sea level pressure (SLP) modes found in European Center for Medium‐Range Weather Forecasts Reanalysis (ERA5) compared to Global Climate Models (GCMs)
Linear techniques such as principal component analysis fail to produce patterns from autoencoder, with congruence matches less than 0.91
Among the analyzed GCMs, HadGEM3‐GC31‐LL performed best in producing the reference SLP modes in North America |
|---|---|
| ISSN: | 2993-5210 2993-5210 |
| DOI: | 10.1029/2024JH000528 |