OS012
Interpretable Machine Learning for Marine Sciences: From Data Mapping to Gaining Trustworthy New Knowledge
Session ID#: 160173
Session Description:
Machine Learning (ML) has become a key instrument in handling and analyzing the ever growing body of Marine Data, particularly geospatial data. Vast amounts of physical, chemical and biological data from numerical models, in-situ and remote observations present a treasure trove of potential insights. The ML algorithms are often used as black boxes without insight into their inner workings. Thus, the results are not always directly interpretable, verifiable, or falsifiable from a Marine Science point of view. Can we find quantitative ways to objectively compare existing results from marine research with the outcomes of Marine Data Science methods?
We invite contributions that go a step further from “simply” applying ML. We invite contributions that explore potential avenues rendering findings more interpretable, explainable, and trustworthy. This includes methods ranging from sparse or heterogeneous data mapping to advanced ML. We aim for a discussion about interpretability, its limitations and possible advances. This can include methods from the field of Trustworthy Artificial Intelligence, especially Explainable Artificial Intelligence.
Index Terms:
0555 Neural networks, fuzzy logic, machine learning [COMPUTATIONAL GEOPHYSICS]
1980 Spatial analysis and representation [INFORMATICS]
4255 Numerical modeling [OCEANOGRAPHY: GENERAL]
4262 Ocean observing systems [OCEANOGRAPHY: GENERAL]
Primary Convener: Yvonne Jenniges, Alfred Wegener Institute Helmholtz-Center for Polar and Marine Research Bremerhaven, Bremerhaven, Germany; Helmholtz School for Marine Data Science (MarDATA), Kiel, Germany
Conveners: Maike Sonnewald, NOAA/Geophysical Fluid Dynamics Laboratory, Princeton, United States; Princeton University, Princeton, NJ, United States, Carola Trahms, GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany; Helmholtz School for Marine Data Science (MarDATA), Kiel, Germany and Dani Jones, NERC British Antarctic Survey, Cambridge, United Kingdom
Student/Early Career Convener: Yvonne Jenniges, Alfred Wegener Institute Helmholtz-Center for Polar and Marine Research Bremerhaven, Bremerhaven, Germany
Enhancing the quality of in situ and satellite ocean observations through machine learning (1047019)
James R Reagan1, Korak Saha1,2, Paige D Lavin2, Tim Boyer1 and Patrick Hogan3, (1)NOAA/NESDIS/NCEI-MD, Silver Spring, United States, (2)Cooperative Institute for Satellite Earth System Studies - University of Maryland, Earth System Science Interdisciplinary Center, College Park, United States, (3)NOAA/NESDIS/NCEI-MS, Stennis Space Center, United States
Benchmarking of machine learning ocean subgrid parameterizations in an idealized model (Invited) (1069914)
Andrew Ross1, Pavel Perezhogin2, Ziwei Li3,4, Carlos Fernandez-Granda2 and Laure Zanna4, (1)NYU, New York City, United States, (2)New York University, New York, United States, (3)Peking University, Beijing, China, (4)New York University, Courant Institute of Mathematical Sciences, New York, United States
Geospatial Machine Learning to Predict Global Abundance of Seafloor Clay Mineral Species (1113764)
Taylor Vander1, Taylor Runyan Lee2, Trilby A Hill1 and Jeffrey Obelcz2, (1)Science and Engineering Apprenticeship Program (SEAP), Ocean Sciences Division, US Naval Research Laboratory, Stennis Space Center, United States, (2)US Naval Research Laboratory, Ocean Sciences Division, Stennis Space Center, United States
Prediction of coastal upwelling and estimation of factor effects using Gaussian Process Regression (1136459)
Deoksu Kim, Korea Institute of Ocean Science & Technology, Seoul, Korea, Republic of (South), Jae-IL Kwon, KIOST Korea Institute of Ocean Science and Technology, Marine Disaster Research Center, Busan, South Korea and Seongho Song, University of Cincinnati Main Campus, Department of Mathematical Sciences, Division of Statistics and Data Science, Cincinnati, United States
Particle Simulation of Floating Marine Debris with Statistical Debris Estimation in Coastal Areas of Japan (1136697)
Kenichi Sasaki, University of Colorado Boulder, Boulder, CO, United States, Tatsuyuki Sekine, Amanogi, Tokyo, Japan and William J Emery, Univ Colorado-CCAR, Boulder, CO, United States
Spatio-Temporal Prediction of Ocean Surface Currents Using Efficient Neural Networks (1140554)
Shadan Nasseri Doust, The university of Tehran, Geophysics institute, Tehran, Iran, Salman Hassanpour, Sharif University of Technology, Department of Electrical Engineering, Tehran, Iran and Mohammad Soltanian, Kharazmi University, Department of Computer Science, Tehran, Iran
Compute Importance of Observations for AMOC Variability Prediction with Machine Learning and Explainable AI (1150199)
Yannick Woelker1,2, Willi Rath1, Arne Biastoch1 and Matthias Renz3, (1)GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany, (2)Helmholtz School for Marine Data Science (MarDATA), Kiel, Germany, (3)University of Kiel, Kiel, Germany
Reconstructing Upper Ocean Carbon with Optimal Interpolation, CMIP6 Models, and Synthetic Argo Observations (1167053)
Katherine Elise Turner, University of Liverpool, Liverpool, L69, United Kingdom, Ric Williams, Liverpool University, School of Environmental Sciences, Liverpool, United Kingdom, Anna Katavouta, University of Liverpool, Liverpool, United Kingdom and Doug M Smith, Met Office Hadley Centre, Exeter, United Kingdom
Automated Analysis of Data Streams with Physics-Informed Neural Networks (1184416)
Joshua Osborne, University of Washington, Cooperative Institute for Climate, Ocean and Ecosystem Studies, Seattle, United States, Kevin O'Brien, NOAA/PMEL UW Joint Institute for the Study of the Atmosphere and Ocean, Seattle, WA, United States and Eugene F Burger, NOAA, Boulder, United States
Integrating Machine Learning and Multi-stream Observations for Sea Surface pCO2 Assessment in the Western Arctic Ocean under a Changing Climate (1186706)
Tianyu Zhou1, Yun Li2, Zhangxian Ouyang1, Zelun Wu2 and Wei-Jun Cai1, (1)University of Delaware, School of Marine Science and Policy, Newark, DE, United States, (2)University of Delaware, School of Marine Science and Policy, Newark, United States
Layer-wise Relevance Propagation for Echo State Networks applied to Earth System Variability (1041080)
Marco Landt-Hayen1, Peer Kröger2, Martin Claus3 and Willi Rath3, (1)Christian-Albrechts-Universität zu Kiel / GEOMAR Helmholtz Centre for Ocean Research, Kiel, Germany, (2)Christian-Albrechts-Universität zu Kiel, Kiel, Germany, (3)GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany
