Dale Durran is a professor of atmospheric and climate science at the University of Washington. As the Chief Scientist at the world’s largest climate forecasting and modeling company, he has led extraordinary breakthroughs. He collaborated with former graduate student Jonathan Weyn and current graduate student Nathaniel Cresswell-Clay. Together, they created a new, simpler kind of AI model better suited for climate called the Deep Learning Earth System Model, or DLESyM. This new model enables us to simulate 1,000 years of today’s climate within a single day. It’s the most amazing high-tech breakthrough in meteorological history!
As soon as his work—which started over five years ago—began when he first integrated artificial intelligence into weather forecasting. In collaboration with Microsoft Research, he sought to improve the speed and precision of climate simulations by an order of magnitude. The newest iteration of his AI-powered method uses a pair of neural networks. One of the models simulates the atmosphere, and the other the ocean—together allowing for complete climate system simulations.
Groundbreaking Developments in Climate Modeling
DLESyM’s modular design allows it to simulate long climatic epochs. This design is what makes it possible for it to learn from massive amounts of data. Durran emphasized the importance of data in training AI models, stating, “To train an AI model, you have to give it tons of data.” What he did see as a challenge was the lack of historical data, especially when it’s looked at by season.
Even with this limitation, DLESyM is able to accurately reproduce month-to-month and interannual variability in weather patterns. Durran’s model, the first of its kind, has performed just as well as established climate models such as CMIP6. These conventional models are celebrated for their precision, but they require significantly higher computational capacity. The depth of DLESyM’s efficiency is even more poignant. Simulation with the model takes place on a single processor, but forecasts in DLESyM are generated in 12 hours. By comparison, traditional kinetic simulations would require about 90 days of calculation on a cutting-edge supercomputer.
Durran expressed confidence in the capabilities of DLESyM, stating, “We were the first to apply this framework to AI and we found out that it worked really well.” This new predictive model speeds up the forecasting process and it cuts down on the considerable carbon footprint that traditional forecasting methods have.
Addressing Climate Variability
The forward-thinking, interdisciplinary design of DLESyM creates exciting new potential for exploring climate variability. Durran noted that the model is looking at existing climate variability to answer important questions about natural weather events. “We are developing a tool that examines the variability in our current climate to help answer this lingering question: Is a given event the kind of thing that happens naturally, or not?” This knowledge is key to consistently predicting more accurately and discerning likely consequences of climate change.
Nathaniel Cresswell-Clay underscored the model’s ability to counteract any preconceived notions about AI’s role in climate science. He remarked, “We’re presenting this as a model that defies a lot of the present assumptions surrounding AI in climate science.” This viewpoint amplifies why DLESyM is so impactful in changing the way scientists model the climate and prepare for the future.
The model’s accessibility further enhances its impact. Durran noted that “not only does the model have a much lower carbon footprint, but anyone can download it from our website and run complex experiments, even if they don’t have supercomputer access.” This democratization of advanced modeling tools allows researchers worldwide to engage with cutting-edge climate science without the need for extensive resources.
Future Implications for Climate Science
Looking ahead, Durran sees the potential to expand DLESyM’s capabilities by bringing other components of the Earth system into the fold. He stated, “Our design opens the door for adding other components of the Earth system in the future.” This adaptability means that DLESyM can and should continue to develop in parallel with better climate science and more robust technology.
The need for better forecasting methods is urgent, as international climate trends grow more unstable by the day. Big, existing models, she explained, can’t always accurately capture weather phenomena on a smaller scale. This shows the urgent need for innovative solutions such as DLESyM. “A lot of the existing climate models actually don’t do a very good job capturing this pattern,” he explained.
Durran and his team are still in the process of finetuning their model. We’re inspired by their commitment to harnessing the cutting edge of AI as they conduct climate forecasts at scale. Their focus extends beyond the ivory tower. They want to better equip policymakers and researchers with powerful tools to address pressing environmental problems.