Building a Cross-Disciplinary Network to Tackle Climate Change with Machine Learning

We present the results of a cross-disciplinary study of the highest impact opportunities to tackle climate change with machine learning [0]. We have found that these opportunities cover a wide range of scientific and engineering approaches, machine learning techniques, and pathways to implementation. We believe that this is cause for optimism: climate change is a problem with many solutions.

In climate and atmospheric science in particular, recent trends have created new opportunities for ML to advance the art of climate prediction. Satellites have now generated petabytes of climate observation data, and large climate model ensembles have generated petabytes of simulation data.

These datasets, and the rise of new techniques that exploit them, offer glimpses towards a new paradigm for climate and scientific modeling, in which neural-network model components improve continuously as we gather more observations, preemptively detect their own errors and crashes, and run near-instantaneously on GPUs. Many climate scientists have now begun to explore these visions [e.g. 1,2,3] and to train and test new models [e.g. 4,5]. The most successful new models, from the point of view of technical success and likelihood of being deployed at a large enough scale to help the public adapt to climate change, are closely integrated into existing scientific models, and are developed by close-knit teams of climate and computer scientists.

Here, we discuss the most exciting new applications of machine learning to climate prediction according to their potential climate change impact. We focus on recent machine learning advances that may solve previously-intractable Earth science problems. We highlight tools that could be deployed in production on rapid time scales that enable individuals, companies, and governments to make better decisions about risk and infrastructure in the face of climate change. Finally, we present a set of new collaboration tools and opportunities to connect the climate science and machine learning community.

[0] Rolnick, Donti, Kaack, Kochanski, Lacoste et al. (2019). ‘Tackling climate change with machine learning’. arXiv:1906.05433

[1] Schneider, Lan, Stuart and Teixeira (2017). ‘Earth system modeling 2.0: a blueprint for models that learn from observations and targeted high-resolution simulations’. Geophysical Research Letters, 44 12396-12417.

[2] Monteleoni, C., G.A. Schmidt, F. Alexander, A. et al. (2013). “Climate Informatics,” in Computational Intelligent Data Analysis for Sustainable Development; Data Mining and Knowledge Discovery Series. Yu, T., Chawla, N., and Simoff, S. (Eds.), CRC Press. Ch. 4 81–126.

[3] Gil, Y, S. Pierce, Hassan Babaie, et al. (2019). "Intelligent Systems for Geosciences: An Essential Research Agenda", Communications of the ACM, 62(1) 76-84.

[4] Gentine, Pritchard, Rasp et al. (2018). ‘Could machine learning break the cloud convection parameterization deadlock?’ Geophysical Research Letters, 45 5742-5751.

[5] Anderson and Lucas (2018). ‘Machine learning predictions of a multiresolution climate model ensemble’. Geophysical Research Letters, 45 4273-4280.