NSF – Frost Institute for Data Science & Computing

Predicting El Niño and La Niña

February 26, 2024February 26, 2024

For forty years, climate scientists have been trying to develop a computational model that accurately predicts the El Niño–Southern Oscillation (ENSO) phenomenon. But that work has stagnated in recent years with only minor, gradual improvements to their predictions.

“The question is why? What are we not understanding?” said Ben Kirtman, the Deputy Director of the University of Miami’s Institute for Data Science and Computing (IDSC), head of the IDSC Earth Systems program, and Director of the Rosenstiel School of Marine and Atmospheric Science’s Cooperative Institute for Marine and Atmospheric Sciences (CIMAS).

Kirtman is now leading a multi-institutional group to try to answer those questions. They submitted a four-year proposal to the National Science Foundation (NSF) that, if approved, would allow Kirtman and his team to test new algorithms designed to better understand the oceanic and atmospheric phenomena that drive El Niño, and would use that information to test new forecasting models as well. Any improvements to the current-day understanding of El Niño would have profound implications since El Niño influences rainfall and temperature patterns throughout the United States, Mexico, and other countries bordering the Pacific Ocean.

For Kirtman, the key to unlocking the secrets of El Niño starts underwater. He wants to get a better grasp on what’s happening below the surface of the western Pacific Ocean, the driving force behind El Niño. To do so, his team would employ a vast constellation of NOAA satellites that measure the height of the sea surface, buoys that record the temperature of the water below the surface, and drones that track the shifting currents. Then comes the hard part.

“The real tricky part is we know it’s not just what’s going on below the surface, but noisiness that happens in the atmosphere,” he said.

Kirtman said that “noise” has been driving the scientific community mad in recent years. No matter how confident they are in their computer models designed to predict El Niño, those atmospheric disturbances continuously muddle their results. That’s why Kirtman’s proposal will closely examine two atmospheric phenomena that he thinks are to blame.

The first is the Madden-Julian oscillation, a large-scale weather pattern that moves quickly from the Indian Ocean east over the Pacific. The second is the Indian monsoon, a period of torrential storms that regularly bombards India and its surrounding waters each summer. Combined, Kirtman said those phenomena consistently cripple their results. “It’s a shock to the system,” he said. “The predictability technology we’re developing will give us a way to remove that shock.”

To form his team, Kirtman first turned to a former student whose Ph.D. he supervised at UM: Sarah Larson, now an assistant professor and climate scientist at North Carolina State University. But he’s also looking to recruit interns from underrepresented groups. Kirtman intends to target students from Miami Dade College and the American Indian Higher Education Consortium (AIHEC), a network of 37 Tribal colleges and universities (TCU’s) in the U.S. But Kirtman doesn’t cite moral reasons for that decision—he explains it by pointing back to science.

“For me it’s a very quantitative thing. Diversity of models, diversity of ideas, improves the forecast, improves your product,” he said. “Everything I’ve done in the last 20 years tells me diversity makes things better, so I want to tap into that as best I can.”

As part of Kirtman’s proposal, he would publicly release the data on an ongoing basis throughout the four-year study. He acknowledges that he runs the risk of other scientists using his data to do a quick analysis and beating him to publication, but much like scientists have been cranking out research based on images and data released by NASA from its James Webb Space Telescope, Kirtman feels the overall benefit to science trumps being first to publish.

“It’s a balance, it’s a challenge, but I always err on the side of just sharing everything,” he said. “It grows the science of what we’re working on, it creates a community. That’s something I really like to do.”

Story by Alan R. Gomez

Next-G Wireless Networks

August 30, 2022February 27, 2024

Congratulations to IDSC core faculty member Dr. Mingzhe Chen on his award from the National Science Foundation (NSF) for his proposal ‘‘Collaborative Research: NeTS: Small: Digital Network Twins: Mapping Next Generation Wireless into Digital Reality’’ (Award Abstract #2312139). Dr. Chen is also an Assistant Professor in the Department of Electrical and Computer Engineering, and a recent hire who received his PhD from Beijing University of Posts and Telecommunications and did his Postdoctoral research at Princeton.

Abstract

Next-generation (NextG) wireless networks provide users with customized, instant services, especially for bandwidth-hungry and latency-sensitive applications. Despite the significant advantages of NextG wireless networks (e.g., 5G/6G and millimeter-wave / Tera Hertz), realizing them faces several key deployment and evaluation challenges:

1) how to speed up the deployment of novel yet complex NextG network technologies; and

2) how to provide flexible testbed facilities with high availability.

In this regard, there is an urgent need for a virtual solution that could create a digital model to replicate as accurately as possible the NextG network ecosystem and help tackle the above obstacles before the full realization of a real system. To this end, this project explores methodologies to run faithful digital network twins that replicate the physical NextG networks, and then to build and optimize the twins over the actual networks while considering communication, computing, and networking resource constraints. The built network twins provide an overarching architecture involving the whole life cycle of physical networks, serving the critical application of innovative technologies such as network planning, construction, optimization, and predictive evaluation, and improving the automation and intelligence level of the wireless networks. This transformative research provides a holistic framework for the implementation and optimization of digital network twins, thus catalyzing the deployment and operation of future network systems with major societal impact.

This proposed research lays the foundations of digital network twins by developing a novel framework that merges tools from machine learning, communication theory, and distributed optimization to advance the networking technologies in:

1) novel mapping approaches that integrate data-driven modeling, ray-tracing analysis, wireless channel derivation, and regression-based predictions to map NextG wireless networks into digital network twins and then to evolve the mapped twins adaptively;

2) new digital network twin management and optimization framework that combines graph neural networks, distributed learning, and reinforcement learning, to allow distributed devices in a physical network to first independently determine their mapping methods and resource utilization, and then collaboratively maximize the digital network twin performance over actual network environments;

3) design of the twinning platform and evaluation methodology based on simulation and experiments to demonstrate the fidelity, efficacy, and optimality of the built network twins.

The project provides a rich environment and virtualized platform that facilitate educating and training students at multiple levels.