All Climate Events

Ocean exploration with machine learning: An Antidote to Chaos?

Download as iCal file

Monday, 20 May 2019, 10:30

Monday, May 20, 2019. 10:30AM. Ocean exploration with machine learning: An Antidote to Chaos?  Maike Sonnewald, MIT. Sponsored by NOAA GFDL. More information here.


Machine learning has the potential to widely influence oceanography, if applied with care. Three case studies highlight the potential for greatly accelerating the efficiency of ocean exploration using supervised (neural networks) and unsupervised (clustering) machine learning. First, two decades of data from the realistic ECCO state estimate 3D physical fields are used to objectively determine global physical regimes using k-means clustering. The identified regions correspond closely to those predicted by canonical theory from physical oceanography and the method can be scaled to analyze vast amounts of data from e.g. CMIP. Second, the high-dimensional dataset from the biogoechemical DARWIN model reveals the existence of ecological niches using t-SNE and DBSCAN clustering. Constraining ocean biomes, individual niches can be collated into larger socioeconomically relevant regions, and examined to understand how sensitive they are to climate forcing which is crucial to protecting the base of the ocean food chain. Finally, a multilayer perceptron (MLP) is trained to predict which global physical regime is present on the basis of local sea surface height. Using the results from the k-means clustering as labels, we achieve a recognition rate of >80%, with good performance across the physical regions. These case studies demonstrate that algorithms can be developed to explore the ocean that have vast potential for understanding complex problems.

Location  NOAA GFDL, Smagorinsky Seminar Room, Princeton University, Princeton, NJ