Add your Insight
31 January 2021
Machine-learning Applications in the Atmospheric and Oceanic Sciences
Atmosphere-Ocean, the principal scientific journal of the Canadian Meteorological and Oceanographic Society (CMOS), is inviting contributions to a Special Issue on Machine-learning (ML) Applications in the Atmospheric and Oceanic Sciences.
Although ML has been used in the atmospheric and oceanic sciences for decades, it is undergoing a rapid paradigm shift driven by three factors. The first is bigger datasets, which present an opportunity for ML but also bring about the need for ML, as many modern environmental datasets are too large to be exploited by traditional methods. The second is an increase in computing resources, fueled not only by Moore’s Law, but also by smarter parallelization and hardware architectures, including the use of graphical processing units. The third is novel ML methods, including the rise of deep learning. Two deep-learning algorithms have been especially important: convolutional neural networks and long-short-term memory, because they can learn directly from gridded spatiotemporal data, which abound in the atmospheric and oceanic sciences. The ability to reason with gridded data has led not only to better predictions, but also to more interpretable and physically consistent ML.
Past uses of ML in the atmospheric and oceanic sciences have included post-processing of physical models, clustering, and dimensionality reduction, with recent expansion to a much wider variety of tasks. These include feature segmentation (e.g., outlining fronts in gridded data); prediction of extremes (e.g., severe thunderstorms); subseasonal, seasonal, and long-range climate prediction (e.g., MJO, ENSO, and decadal/centennial projections); parameterization of subgrid-scale processes in physical models (e.g., radiative transfer and deep convection); and even emulation of full physical models. As ML becomes more integrated with other tools and more widely adopted for decision-making, there is an emerging focus on physics-guided and -constrained ML methods. Similarly, interpretation methods – including ones developed in the ML literature, such as saliency maps and class-activation maps – are used to understand the physical relationships learned by ML systems.
Looking to Publish your Research?
We aim to make publishing with Taylor & Francis a rewarding experience for all our authors. Please visit our Author Services website for more information and guidance, and do contact us if there is anything we can help with!
This Special Issue of Atmosphere-Ocean welcomes submissions on applied ML in the atmospheric and oceanic sciences, especially novel methods/applications and ML for decision-making. Novel methods may include (but are not limited to) new algorithms and datasets, physics-guided and -constrained ML, interpretable ML, uncertainty quantification, etc. We also welcome position pieces on the current and future impacts/priorities of ML in the atmospheric and oceanic sciences. Topics may include (but are not limited to) transparency and reproducibility, how to communicate ML results with end users (research to operations), how to ensure that ML leads to better decisions and not just better predictions (research to operations to practice), how to prevent “skill atrophy” from increasing adoption of and reliance upon ML, etc.
Special Issue Editors:
Dr. Carlos Gaitain, Benchmark Labs, Inc.
Dr. Ryan Lagerquist, Colorado State University, Cooperative Institute for Research in the Atmosphere (CSU/CIRA)
Please select “Machine-learning Applications in the Atmospheric and Oceanic Sciences” when submitting to ScholarOne. Your article will be handled by Editor-in-Chief Dr. Alex. J. Cannon and the Special Issue Editors.
View the latest tweets from tandfenviro