Connecting River Flows and El Niño with AI
Deep learning increases the accuracy of river flow predictions
What do tropical water temperatures have to do with winter conditions in Washington State? Both are affected by the climate phenomenon known as the El Niño-Southern Oscillation (ENSO). In fact, this climate event can influence weather across the globe, from flood timings in Africa to river flows in India. Predicting ENSO’s consequences—or even its causes—has been a challenge for scientists. Researchers from Pacific Northwest National Laboratory (PNNL), Northeastern University, the National Aeronautics and Space Administration (NASA) Ames Research Center, and the Bay Area Environmental Research Institute made these predictions more accurate with the help of artificial intelligence (AI). Their results were published in Nature Communications.
The ENSO phenomenon itself is not fully understood. Over the course of years, the temperatures across the surface of the Pacific Ocean fluctuate depending on wind conditions. Trade winds blow warm water from the region near South America westward toward Asia. This allows for an upwelling of cold water from the depths of the ocean to replace the warm water, beginning the “La Niña” phase of ENSO. As the trade winds die down, the warm water travels back to South America, leading to the “El Niño” phase.
Effects of ENSO can easily be observed in North America—flooding in the southern United States and drier conditions in the Pacific Northwest correlate with the “El Niño” phase. Conversely, the “La Niña” phase results in drought conditions for the southern United States and flooding in the Pacific Northwest.
ENSO also affects river flows across the globe—though understanding and predicting these effects remain a challenge. A team of scientists led by Northeastern University College of Engineering distinguished professor and PNNL joint appointee Auroop Ganguly investigated this connection using deep learning.
Ganguly and his team leveraged powerful computing techniques to combine climate data—including observed and predicted weather patterns and ocean temperatures—to more accurately predict ENSO’s effect on river flows in the Congo, Amazon, and Ganges rivers. Their work—which was supported by the National Science Foundation and the NASA Ames Research Center—highlights the value of incorporating explainable, physics-informed machine learning techniques in climate studies.
As a member of the core faculty leading the climate-AI area at the Institute for Experiential AI at Northeastern University, Ganguly is well aware of this value.
“The power of these approaches is in the ability to extract this information from the vast ocean—pun intended!—of data, rather than through oversimplified indices of this complex phenomenon,” said Ganguly.
At PNNL, researchers are incorporating these types of approaches into Earth System Models (ESMs), such as the Department of Energy’s Energy Exascale Earth System Model (E3SM), and other applications through projects such as Physics-Informed Machine Learning for Energy and Environment and Scalable, Efficient and Accelerated Causal Reasoning Operators, Graphs and Spikes for Earth and Embedded Systems. Other PNNL researchers, like those involved in the Mathematics for Artificial Reasoning in Science initiative, are working to improve the hardware, algorithms, and mathematics that underpin these approaches.
Besides improving climate-informed water resource projections, future extensions of these lines of work can help advance the state of ESM couplers, which are modules that connect ESM components such as ocean, atmosphere, and land. Machine learning and AI models are sometimes viewed as black boxes by scientists and engineers, as well as the stakeholder and policy communities, thus further research in domain-specific explainability can help enhance the trustworthiness of AI-informed results.
Published: April 3, 2023