Our Work

In collaboration with the Climate Modeling Alliance (CliMA)–a partnership between MIT, Caltech, and the NASA Jet Propulsion Laboratory (JPL)–we aim to improve the accuracy of climate projections. CliMA is developing a climate model that leverages artificial intelligence algorithms to improve model accuracy by learning missing physics from the vast amount of data routinely collected by satellites and ocean floats. Over the past year, the CliMA group at MIT has developed the fastest ocean model to date. BC3 scientists have contributed by developing the skeleton of the thermodynamic sea ice component of the climate model and have implemented a preliminary carbon cycle component. Together with the dynamical core, these components constitute the ocean component of the full CliMA climate model.  The CliMA groups at Caltech and JPL have, likewise, made progress toward completion of the atmosphere and land components of the full climate model. Each group is now testing these individual components and are preparing to calibrate those components against observational data. By the end of the year, we plan to couple all components together and run our first global climate simulations.

We also aim to empower a broad range of stakeholders to use climate data. We have developed different emulator architectures to predict a targeted subset of climate variables for specific applications at far greater speed than full climate models.  The development of emulators is largely independent of the actual climate models they are trained on; as such, this work can proceed independently while the full CliMA model is being completed.  Specifically, we have developed emulators that generate spatially resolved maps of temperature and its variability (i.e., the local weather) as a function of carbon dioxide emissions. These emulators have been extended to predict other variables as well, like precipitation, daily temperature maxima, and wet bulb temperature (a measure of the maximum temperature that can be tolerated by humans). We have integrated one of our emulators into Climate Interactive’s En-ROADS simulator, a web app that predicts global emissions as a function of user-input energy and policy choices. This integration provides a simple interactive tool for stakeholders who are interested in exploring how different emissions scenarios may result in different temperature trends at regional scales. Moreover, our participating faculty have developed a new MIT class on the Julia programming language to train the next generation of climate model developers.