As the PI of the Computational Climate and Ocean Group at UC Davis, I target grand challenges with the aim of improving ocean and climate understanding and resilience. Solutions to the challenges humanity, and the world, face are inherently interdisciplinary, and so is my research. Blending cutting-edge computational and Earth science tools and knowledge, I combine theory, observations, and numerics to pioneer methods and create insight.

While the research objectives go hand-in-hand, my work can be seen through both an 'Earth Science' and a 'Computational' lens. The deeply interdisciplinary approach I use guides us in innovating in both fields to reach the reseaarch objectives. Specifically, I focus on several research areas:

Computational: Injecting knowledge to guide innovation

Earth Science: Objective advancement of our understanding of the Earth

Open Source Software and its adoption by the oceanographic community is something I am very invested in. I highlight some of the efforts below. See maikejulie and compClimate on GitHub for more.

A selection of projects are listed below and selected student projects are on the group website.

Advances in data mining for messy geospatial data

Within the Earth sciences data is increasingly becoming unmanageably large, noisy and nonlinear. Most methods that are commonly in use employ highly restrictive assumptions regarding the underlying statistics of the data and may even offer misleading results. To enable and accelerate scientific discovery, I drew on tools from computer science, statistics and dynamical systems theory to develop the Native Emergent Manifold Interrogation (NEMI) method. Nemi is intended for wide use within the Earth sciences and applied to an oceanographic example here. Using domain specific theory, manifold representation of the data, clustering and sophisticated ensembling, NEMI is able to highlight particularly interesting areas within the data. In the paper, I stresses the underlying philosophy and appreciation of methods to facilitate understanding of data mining; a tool to gain new knowledge. Implications include:

  • NEMI scales and performs well on very complex and nonlinear data
  • A generalisation of SAGE that is less parametric and generic in application
  • Allows domain specific input `native' to the research problem

Package available: NEMI on GitHub via PyPi

  • Sonnewald, M., submitted, A hierarchical ensemble manifold methodology for new knowledge on spatial data: An application to ocean physics, JAMES

    • The Southern Ocean supergyre: a unifying dynamical framework for a region key to climate

    The Southern Ocean closes the global overturning circulation and is key to the regulation of carbon, heat, biological production, and sea level. However, the dynamics of the general circulation and upwelling pathways remain poorly understood. Here, a unifying framework is proposed invoking a semi-circumpolar `supergyre' south of the Antarctic circumpolar current: a massive series of ‘leaking’ sub-gyres spanning the Weddell and Ross seas that are connected and maintained via rough topography that acts as scaffolding. The supergyre framework challenges the conventional view of having separate circulation structures in the Weddell and Ross seas and suggests that idealized models and zonally-averaged frameworks may be of limited utility for climate applications. Machine learning was used to reveal areas of coherent driving forces within a vorticity-based analysis. Predictions from the supergyre framework are supported by available observations and could aid observational and modelling efforts to study this climatologically key region undergoing rapid change. Implications include:

    • Better understanding of circulation key to climate
    • Improvement of idealised and realistic models through understanding
    • Enhance collection of observations for monitoring

    Collaborators: Krissy Reeve and Redouane Lguensat.

  • Sonnewald, M., Reeve, K.A. and Lguensat, R., in review, The Southern Ocean supergyre: a unifying dynamical framework identified by machine learning, Communications Earth an Environment. Available here.

    • eXplainable AI and UQ for oceanography and science

    The use of deep learning is becoming widespread, but verification of the source of network skill is still often lacking. I highlight two core aspects that are needed for safe/trustworthy applications of deep learning: 1) Quantification of uncertainty (UQ), and 2) Understanding of the source of skill. In this study, we show that a Bayesian Neural Network (BNN) can be applied to an oceanographic problem, with uncertainty quantified through entropy, and also implement two eXplainable AI (XAI) techniques. The two XAI methods look both inside the BNN, at the inner workings, and outside effectively assessing the latent space. Implications include:

    Collaborators: Mariana Clare (student), Redouane Lguensat, Julie Deshayes and V. Balaji

  • Clare, M, Sonnewald, M., Lguensat, R.Julie Deshayes and V. Balaji Explainable Artificial Intelligence for Bayesian Neural Networks: Toward Trustworthy Predictions of Ocean Dynamics, 2022, JAMES

    • Unlocking the impact of mesoscale turbulence on the subpolar North Atlantic

    The subpolar North Atlantic is important for global heat transport and a place where important dynamics are difficult to observe and to model. Building on the SAGE method, using also methods from topology and unsupervised machine learning, I am uncovering regions of coherent dynamics that can be used to infer mechanisms important for understanding the circulation. Interactions of currents with canyons and regions associated with export of water from the shelf into the deep are topics I am investigating. Implications include:

    Collaborators: V. Balaji, Alistair Adcroft, Aparna Radhakrishnan and Mathieu Le Corre

    Inferring in depth dynamics from surface fields in CMIP (Climate models)

    Climate models are key to understanding and preparing for future change, but a revolution in analysis tools is necessary due to the scale and complexity of the ensembles. The `tracking global heating with ocean regimes' (THOR) method reveals changes in the in-depth drivers in climate model data. Needing only the depth, sea surface height and wind stress, THOR applies an explainable multilayer perceptron trained on data labeled by an interpretable unsupervised ML method. Applied to a climate model, THOR reveals areas for example of deepwater formation, upwelling and the approximate location key currents like the Gulf Stream. THOR could accelerate in-depth climate model analysis with concrete dynamical intercomparisons, helping lower uncertainties. Implications include:

    Collaborators: Redouane Lguensat, Aparna Radhakrishnan, and Zouberou Sayibou

  • Sonnewald, M., and Lguensat, R. Revealing the Impact of Global Heating on North Atlantic Circulation Using Transparent Machine Learning, 2021, JAMES
  • Sonnewald, M., Lguensat, R., Radhakrishnan, A., and Sayibou Z. Revealing the Impact of Climate Modes Using Transparent Machine Learning, 2021, ICML

    • Unlocking the impact of mesoscale turbulence on the subpolar North Atlantic

    Important for the global heat transport, the subpolar North Atlantic is a place where important dynamics are difficult to observe and to model. Building on the SAGE method, using also methods from topology and unsupervised machine learning, I am uncovering regions of coherent dynamics that can be used to infer mechanisms important for understanding the circulation. Interactions of currents with canyons and regions associated with export of water from the shelf into the deep are topics I am investigating. Implications include:

    Collaborators: V. Balaji, Alistair Adcroft, Aparna Radhakrishnan and Mathieu Le Corre

    Decoding ecological complexity using the SAGE method

    Global ecology effects our climate and the ocean food chain. Defining glogal regions or provinces help monitor its health, but is difficult due to the overwhelmingly complex ecosystem data. I developed the SAGE (Systematic AGgregated Eco-province) method combining statistical tools, unsupervised machine learning and graphs to determine global provinces in the Darwin ecosystem model. SAGE delivers AEPs (Aggregated Eco-Provinces) for global or regional use, and you can set the 'aggregation' level to fit your needs using the tool here. Implications include:

    Collaborators: Steph Dutkiewitz, Chris Hill, Gael Forget

  • Sonnewald, M., Dutkiewitz, S., Hill, C. and Forget, G. Elucidating Ecological Complexity: Unsupervised Learning determines global marine eco-provinces, 2020, Science Advances

    • Using machine learning to discover global dynamical regions

    Applying an unsupervised learning algorithm the the depth integrated ECCO data, the ocean partitioned itself neatly into globally coherent dynamical regimes in terms of the barotropic vorticity. This means that the way the terms of the barotropic vorticity equation are balanced fall into a subset of configurations. For the decedal ocean, this is remarkably robust. Implications include:

    Collaborators: Carl Wunsch, Patrick Heimbach

  • Sonnewald, M., Wunsch, C. and Heimbach, P. Unsupervised Learning Reveals Geography of Global Ocean Dynamical Regions, 2019, Journal of Earth and Space Science ed. ''Geoscience paper of the future''

    • Presenting a benchmark for Sea Surface Height predictability

    Sea Surface Height is a critical quantity for vast numbers of people. Living by the coast means that small changes can accumulate, and understanding how the Sea Surface Height varies is cruicial. To check if predictions are skillful, a benchmark is nessesairy to compare against. Developed using linear methods, this suggests that using more complicated methods is merited. Key points include:

    • Sea Surface Height predictability is reasonable using linear statistical models
    • Predictability is related to planetary waves close to the equator, and more to do with advective signals further polewards
    • Complicated models can likely make progress!

    Collaborators: Carl Wunsch, Patrick Heimbach

  • Sonnewald, M., Wunsch, C. and Heimbach, P., 2018. Linear predictability: A sea surface height case study, 2018, Journal of Climate

    • Barotropic Vorticity dynamics

    Vorticity is a useful tool for understanding ocean dynamics. To assess what term does what and where and how things are dissipated requires a close look at what is going on. Using the closure in the MITgcm/ECCO, we can:

    Collaborators: Yvonne Firing, Joel J.-M. Hirschi A. George Nurser

    Open Source software initiatives

    Adopting Open Source means that oceanography can lean on innovations across a wide range of fields and industry. Open development processes lead to better software in terms of readability, ability to modify and fix bugs as well as for teaching purposes.