Beucler, T., Ebert‐Uphoff, I., Rasp, S., Pritchard, M. and Gentine, P., 2023. Machine learning for clouds and climate, Chapter 16 (pp.325-345), Clouds and their Climatic Impacts: Radiation, Circulation, and Precipitation, (AGU Geophysical Monograph Series), Vol. 281, John Wiley & Sons, Dec 2023. https://doi.org/10.1002/9781119700357.ch16.
Short summary: As the title suggests, this book chapter reviews the approaches and impact of machine learning for topics in clouds and climate.
Detailed summary: This invited chapter was submitted in Dec 2020, accepted for publication in March 2021, and was finally published in Dec 2023. Given the high speed of progress in the field it is an interesting experience to have a paper published 3 years after writing it. Lessons learned from that process: 1) Books that consist of many invited book chapters can be very slow to publish since it may take a long time to collect all chapters. 2) We got permission from AGU to host a preprint of the accepted chapter on an AGU preprint server (ESSOAr), which made it available to the public 2.5 years earlier than the book version. This step is highly recommended. 3) While many details are clearly outdated, e.g., in 2020 Generative Adversarial Networks were the state-of-the-art in Generative AI and AI-based global weather forecasting did not yet exist, it is encouraging that the overall statements about key concepts, opportunities, and challenges for the use of machine learning for clouds and climate still hold today. (POC: Imme Ebert-Uphoff, iebert@colostate.edu, Funding: NSF)
Citation: White, C. H., Ebert-Uphoff, I., Haynes, J. M., & Noh, Y. J. (2024). Super-Resolution of GOES-16 ABI Bands to a Common High Resolution with a Convolutional Neural Network. Artificial Intelligence for the Earth Systems. https://doi.org/10.1175/AIES-D-23-0065.1
Short Summary: This work presents an AI-based approach to artificially increase the spatial resolution of the 1- and 2-km channels on the Advanced Baseline Imager to a common high resolution of 0.5-km.
Detailed Summary: The Advanced Baseline Imager (ABI) on the GOES-R series satellites has sixteen channels with differing spatial resolutions between 0.5-km and 2-km. We create a convolutional neural network that is trained on a synthetic low-resolution dataset to super-resolve (or sharpen) the lower resolution channels on ABI to a common high resolution of 0.5-km. Evaluations performed at various scales show that our method far outperforms simple interpolation with a 3- to 18-times reduction in error depending on the channel. We also perform a comparison of the super-resolved ABI imagery with collocated and coarsened Landsat 8/9 data at 0.5-km resolution. This comparison with Landsat separately confirms that the high-frequency detail added by our CNN is realistic and adds value beyond simple interpolation. POC: Chuck White, charles.white@colostate.edu, Funding: GOES-R