Blog

CliMA Researchers Upgrade their Uncertainty Pipeline

CliMA provides state-of-the-art data assimilation and machine learning tools that enable users to calibrate their models using large amounts of data. CliMA’s research scientists and engineers continually upgrade these tools and made several breakthroughs in the past months. CliMA’s data assimilation and machine learning (DA/ML) team, led by Oliver Dunbar, improved the scalability of one of their machine learning emulators. Think of emulators as accelerated data-driven models, trained to replicate the evolution of a complex system (in our case, a climate model).  Once trained, emulators can be run millions of times using few computational resources. These large samples are essential…
Read More

A Discussion with CliMA’s Lead Software Developer Simon Byrne on his Team’s Parallel Input/Output Work

CliMA’s software team, led by Simon Byrne, added a software interface to ClimaCore.jl for saving and loading data from distributed simulations. We caught up with Dr. Byrne near the Gong inside the CliMA conference room; an edited version of our interview is reproduced below. Leilani Rivera-Dotson: Why did your team implement this interface? Dr. Byrne: There are two main reasons we need to be able to save and load data: saving the quantities of interest, such as temperature and precipitation, for analysis and other post-processing tasks; capturing the state of the model so that we can reproduce or resume the…
Read More

A New Numerical Scheme for Nonlinear Ocean Dynamics

Simone Silvestri and Greg Wagner have made substantial progress in improving the numerical accuracy of the ocean component of the CliMA model, Clima-Ocean. Typical finite-volume ocean models for climate prediction use “second-order” schemes to discretize terms in the ocean’s momentum balance. But second-order momentum advection schemes are noisy, and therefore must be paired with artificial viscous terms to suppress spurious oscillations and prevent numerical instability. Artificial viscosity has the unfortunate side effect of artificially suppressing ocean mesoscale turbulence, which plays a fundamental role in Earth’s climate system by transporting heat poleward and determining the ocean’s density stratification. Improving the representation…
Read More

Learning about climate model parameterizations as an inverse problem

[latexpage] By Ignacio Lopez-Gomez Over the past few years, machine learning has become a fundamental component of some newly developed Earth system parameterizations. These parameterizations offer the potential to improve the representation of biological, chemical, and physical processes by learning better approximations derived from data. Parameters within data-driven representations are learned using some kind of algorithm, in a process referred to as model training or calibration. Gradient-based supervised learning is the dominant training method for such parameterizations, mainly due to the availability of efficient and easy-to-use open source implementations with extensive user bases (e.g., scikit-learn, TensorFlow, PyTorch). However, these learning…
Read More