¡AI Caramba! – RealClimate

Rapid Progress in Machine Learning for Weather and Climate Models

Introduction

The use of machine learning (ML) in weather and climate models has seen significant advancements in recent years. However, it is important to distinguish between real progress and mere hype. This article aims to provide an overview of the current state of ML in weather and climate models, with a focus on the Sustainable Development Goals (SDGs).

Defining Machine Learning and Artificial Intelligence

Machine Learning (ML) refers to the statistical fitting of large datasets to complex functions. It can be seen as a form of large regression. Artificial Intelligence (AI) encompasses ML, as well as expert systems and other methods. Generative AI, such as ChatGPT and DALL-E, involves training models with massive amounts of data and a high number of degrees of freedom. However, it is important to note that these systems are not “intelligent” in the traditional sense.

Recent Success in Weather Forecasting

ML applied to weather forecasting has shown remarkable progress. Systems like FourCastNet, GraphCast, and NeuralGCM have demonstrated the ability to predict weather with skill approaching or matching physics-based forecasts. However, it is important to note that claims of exceeding physics-based forecasts are not yet supported by a wide range of metrics used by ECMWF.

Recent advancements include techniques like “bred vectors” that generate ensemble spreads matching physics-based models and the development of GraphDOP, which learns forecasts directly from raw observations.

Climate is Not Weather

While ML has shown promise in weather forecasting, it is not well-suited for climate projections. Climate models require the consideration of boundary values, such as greenhouse gases and solar irradiance, which ML models do not track. Additionally, the lack of appropriate training data for climate predictions poses a significant challenge.

Alternative Approaches

Instead of relying solely on ML, there are several alternative approaches that can be explored:

1. Whole Model Emulation: Learning from existing climate model simulations to optimize parameter sets or produce results for new scenarios.
2. Process-based Learning: Learning specific processes from detailed process models and incorporating them into existing climate models.
3. Complexity-based Learning: Implementing ML parameterizations within simpler versions of climate models.
4. Error-based Learning: Using data-assimilated models to learn from historical observations and apply online corrections in future scenarios.

Each approach has its advantages and challenges, but all are showing positive results or are being actively researched.

Predictions are Hard

The field of ML in weather and climate models is rapidly evolving. While it is difficult to predict which approaches will be widely adopted, some possibilities include ML for tuning and calibration of climate models, scenario emulation, historical emulators for attribution analysis, and predicting changes in statistical properties of the climate.

However, it is important to note that ML-enhanced models will likely have similar climate sensitivity as non-ML enhanced models. The challenge of achieving stable coupled models with ML-based components is still being addressed.

Limitations of ML Models

Despite claims made in papers or press releases, ML models based on weather or reanalyses data cannot become climate models without the relevant inputs and sufficient training data. Generative AI also cannot provide accurate predictions of climate change and prevention without regurgitating existing knowledge or making unfounded claims.

It is worth considering the potential uses of AI technology, such as converting requests into specific information demands, but relying on AI to write proposals, conduct real science, and write scientific papers would be detrimental to the scientific community.

Conclusion

While progress in ML for weather and climate models is real, it is not as fast as some may suggest. ML has the potential to improve models and predictions, but it is crucial to approach it with caution and consider its limitations. Continued research and collaboration are necessary to advance the field and achieve the SDGs.

References

1. R. Lam, A. Sanchez-Gonzalez, M. Willson, P. Wirnsberger, M. Fortunato, F. Alet, S. Ravuri, T. Ewalds, Z. Eaton-Rosen, W. Hu, A. Merose, S. Hoyer, G. Holland, O. Vinyals, J. Stott, A. Pritzel, S. Mohamed, and P. Battaglia, “Learning skillful medium-range global weather forecasting”, Science, vol. 382, pp. 1416-1421, 2023. http://dx.doi.org/10.1126/science.adi2336
2. D. Kochkov, J. Yuval, I. Langmore, P. Norgaard, J. Smith, G. Mooers, M. Klöwer, J. Lottes, S. Rasp, P. Düben, S. Hatfield, P. Battaglia, A. Sanchez-Gonzalez, M. Willson, M.P. Brenner, and S. Hoyer, “Neural general circulation models for weather and climate”, Nature, vol. 632, pp. 1060-1066, 2024. http://dx.doi.org/10.1038/s41586-024-07744-y
3. I. Price, A. Sanchez-Gonzalez, F. Alet, T.R. Andersson, A. El-Kadi, D. Masters, T. Ewalds, J. Stott, S. Mohamed, P. Battaglia, R. Lam, and M. Willson, “Probabilistic weather forecasting with machine learning”, Nature, vol. 637, pp. 84-90, 2024. http://dx.doi.org/10.1038/s41586-024-08252-9
4. G. Elsaesser, M.V. Walqui, Q. Yang, M. Kelley, A.S. Ackerman, A. Fridlind, G. Cesana, G.A. Schmidt, J. Wu, A. Behrangi, S.J. Camargo, B. De, K. Inoue, N. Leitmann-Niimi, and J.D. Strong, “Using Machine Learning to Generate a GISS ModelE Calibrated Physics Ensemble (CPE)”, 2024. http://dx.doi.org/10.22541/essoar.172745119.96698579/v1
5. D. Watson‐Parris, Y. Rao, D. Olivié, . Seland, P. Nowack, G. Camps‐Valls, P. Stier, S. Bouabid, M. Dewey, E. Fons, J. Gonzalez, P. Harder, K. Jeggle, J. Lenhardt, P. Manshausen, M. Novitasari, L. Ricard, and C. Roesch, “ClimateBench v1.0: A Benchmark for Data‐Driven Climate Projections”, Journal of Advances in Modeling Earth Systems, vol. 14, 2022. http://dx.doi.org/10.1029/2021MS002954

SDGs, Targets, and Indicators

SDGs, Targets, and Indicators

Explanation

1. SDG 13: Climate Action is addressed in the article as it discusses the use of machine learning for weather and climate models, which can contribute to improving climate change mitigation, adaptation, and resilience. The article specifically mentions the progress in weather forecasting using machine learning techniques.
2. Target 13.1: Strengthen resilience and adaptive capacity to climate-related hazards and natural disasters can be identified based on the article’s content. The article mentions the use of machine learning models to forecast weather and climate patterns, which can help in improving resilience and preparedness for climate-related hazards.
3. Target 13.3: Improve education, awareness-raising, and human and institutional capacity on climate change mitigation, adaptation, impact reduction, and early warning can also be identified based on the article’s content. The article discusses the advancements in machine learning for weather and climate models, which require education, awareness, and capacity-building efforts to effectively utilize these technologies for climate change mitigation and adaptation.
4. SDG 9: Industry, Innovation, and Infrastructure is indirectly connected to the issues highlighted in the article. The article discusses the rapid progress in machine learning for weather and climate models, which can be seen as an innovation in the field of climate science and technology.
5. Target 9.5: Enhance scientific research, upgrade the technological capabilities of industrial sectors in all countries, in particular developing countries, including, by 2030, encouraging innovation and increasing the number of research and development workers per 1 million people and public and private research and development spending can be identified based on the article’s content. The advancements in machine learning for weather and climate models require scientific research and technological capabilities to develop and implement these technologies effectively.
6. Target 9.c: Significantly increase access to information and communications technology and strive to provide universal and affordable access to the Internet in least developed countries by 2020 can also be indirectly connected to the issues discussed in the article. The use of machine learning for weather and climate models requires access to information and communications technology, including the internet, to collect and analyze large datasets.

Overall, the article highlights the potential of machine learning for weather and climate models, which can contribute to achieving SDG 13: Climate Action and SDG 9: Industry, Innovation, and Infrastructure. However, the article does not mention specific indicators that can be used to measure progress towards the identified targets.

Source: realclimate.org