AI for the environment - Climate modeling
Since the early days of climate modeling, software, hardware, and the ways that engineers and scientists collaborate have gone through incredible transformations. Better data and technologies will inform how we mitigate and adapt to global impacts, such as sea level rise, community destruction, and biodiversity loss.
What are climate models?
Planetary-scale Earth simulations known as global climate model projections are the primary sources of information on future climate change. Climate models are based on mathematical equations represented using a grid mesh that covers the globe: a finer grid mesh is more accurate but much more computationally expensive. Current global climate projections agree that a world with more greenhouse gases will be warmer everywhere, especially over land and at high latitudes. However, the current understanding of high-risk outcomes like rainfall extremes is more uncertain, and these changes have the potential to impact billions of people.
Refining climate predictions
The technology behind climate models was first created 50 years ago. Much has changed since then, and there is now an opportunity to make use of the latest advances in supercomputing, modern programming languages, and machine learning to improve climate models and enable more certain projections of local trends of average and extreme temperature and precipitation change in our rapidly warming climate. We're building modern machine learning (ML) into current climate models to improve their performance in key areas and ultimately to refine climate change predictions. Our ML is trained on ultra-realistic ‘digital twin’ simulations of the Earth’s atmosphere that exploit the world’s fastest supercomputers
Better climate models using finer grids
In the same way photos have become clearer because screens now pack in more pixels, fine grid ‘global storm resolving models’ (GSRMs) based on grids with less than 5 km (3 miles) horizontal spacing and 50 or more vertical grid levels spanning the depth of the atmosphere can now provide a detailed and actionable ‘digital twin’ of our world, enabling realistic simulation of airflow around mountain peaks and within thunderstorm systems that generate much of the world’s most intense rainfall.
Smarter simulations with machine learning
GSRMs are too costly to run for more than a few years, so they are not yet practical for climate modeling. But they can be run in a small selection of changed climates, and the simulations can be used to train a machine learning (ML) emulator that simulates similar climates and weather extremes, but 1000s of times faster, and is also accurate in intermediate climates. We partner with two leading climate modeling centers, NOAA’s Geophysical Fluid Dynamics Laboratory (GFDL) and the Department of Energy-funded Lawrence Livermore National Laboratory (LLNL), to design new GSRM simulations and use them for ML climate emulator training. Our group is a world leader in this area.
Creating open-source, collaborative solutions
We're developing open-source software so the broader climate modeling community can easily adopt our advances. Our partnerships with climate modeling centers ensure our work builds on their valuable experience and high-performance computing resources, and has the quickest impact. We also partner with NVIDIA and academic research groups to bring in the best new ML approaches and to work with top young minds in this rapidly evolving field.
Recent papers
Journal of Geophysical Research - Machine Learning 2024 - Application of the AI2 Climate Emulator to E3SMv2's global atmosphere model, with a focus on precipitation fidelity
Can the current successes of global machine learning-based weather simulators be generalized beyond 2-week forecasts to stable and accurate multiyear runs? The recently developed AI2 Climate Emulator (ACE) suggests this is feasible, based upon 10-year simulations with a network trained on output from a physics-based global atmosphere model using a grid spacing of approximately 110 km and forced by a repeating annual cycle of sea-surface temperature. Here we show that ACE, without modification, can be trained to emulate another major atmospheric model, EAMv2, run at a comparable grid spacing for at least 10 years with similarly small climate biases—a prerequisite to wider applicability. With an analysis that combines multiple temporal, spatial, and frequency domain perspectives, we show that ACE faithfully represents the spatiotemporal structure of EAMv2 precipitation and related variables. Finally, we show that a pretrained ACE network is able to adapt to a new global climate model simulation data set with 10mathematical equation fewer training steps than when starting from random initialization, all while still maintaining low levels of climate bias. Further analysis of these fine-tuning experiments reveal ACE's intriguing ability to interpolate between distinct global climate models.
Nature Climate Change 2024 - Pushing the frontiers in climate modelling and analysis with machine learning
Climate modelling and analysis are facing new demands to enhance projections and climate information. Here we argue that now is the time to push the frontiers of machine learning beyond state-of-the-art approaches, not only by developing machine-learning-based Earth system models with greater fidelity, but also by providing new capabilities through emulators for extreme event projections with large ensembles, enhanced detection and attribution methods for extreme events, and advanced climate model analysis and benchmarking. Utilizing this potential requires key machine learning challenges to be addressed, in particular generalization, uncertainty quantification, explainable artificial intelligence and causality. This interdisciplinary effort requires bringing together machine learning and climate scientists, while also leveraging the private sector, to accelerate progress towards actionable climate science.
Nature 2024 - Weather and climate predicted accurately — without using a supercomputer
A cutting-edge global model of the atmosphere combines machine learning with a numerical model based on the laws of physics. This ‘hybrid’ system accurately predicts the weather — and even shows promise for climate simulations.