By Amara Graps
The Forecasts
Hurricane Helene was forecast to be one of the largest hurricanes in years to hit the region; only three since 1988 were larger. The US Air Force Reserve aircrews flew nine weather reconnaissance missions into the hurricane, starting September 23, 2024, to collect data for the National Hurricane Center forecasts. The US Federal Government prepared for a multi-state emergency scenario. On 25 September 2024, the National Oceanic and Atmospheric Association (NOAA) released a rare news bulletin, urging major news media to help communities prepare for catastrophic, life-threatening, inland flooding. Evacuations proceeded.
Meteorologist Tyler Hamilton pointed out to his viewers that forecasting the storm’s path is so much better now. “In twenty years, the uncertainty cone for these hurricanes has shrunk by half,” he said. “Now it is just a 400 km window of uncertainty”.
Hurricane Helene made landfall on the Florida coast night 26-September, as a Category 4 hurricane. The hurricane weakened to a tropical storm over Georgia. As of this writing, at least 25 people have died, due to the hurricane’s effects.
Some public figures are stating the obvious. Stephen King, author on social media:
These once-in-a generation, historical storms are happening once or twice every month. Climate change.
With climate change proceeding with no abatement of extreme weather conditions, how can humans be better prepared? Excellent weather forecasts are the holy grail.
Weather Forecasts using Quantum for AI are still years away
One way forward is via quantum machine learning. Of the two-sided question:
- What can AI do for quantum technologies and
- What can quantum technologies do for AI?
We are addressing the second question, utilizing hybrid quantum-classical computing. We discuss the pervasiveness of QML here, and position QML in hybrid computing in part 3 of QCR’s The Many Faces of Hybrid Quantum-Classical Computing.
Quantum for AI in GQI’s Strategy Playbook doesn’t start growing noticeably until about 2026. See the dark orange and dark red items in the next Figure.
Weather Forecasts using Quantum for AI: A Pilot Study
In 2021, researchers at Rigetti, Enos et al., 2021, implemented and published Synthetic weather radar using hybrid quantum-classical machine learning, which was a pilot project to utilize QML to generate synthetic weather radar images for forecasting and decision-making in regions without traditional radar coverage.
To improve traditional convolutional neural networks for generative tasks in global synthetic weather radar, Enos et al. (2021) added quantum-assisted models to the network. In particular, the Offshore Precipitation Capability (OPC) was enhanced with a quantum convolutional layer.
The OPC expands on the Global Synthetic Weather Radar (GSWR), a cutting-edge class of methods for producing synthetic weather radar pictures by assimilating diverse meteorological data types. Convolutional neural networks (CNNs), namely the OPC-CNN, are used in a machine-learning component to integrate several kinds of high-dimensional meteorological data at different temporal and spatial resolutions. The OPC-CNN is accepted for use in real-world scenarios. In two case studies, the Enos et al., 2021 research showed how to hybridize the GSWR system using QML approaches.
Creating a generative quantum model, to imitate one of the two primary data sources in the event of data scarcity or unreliability, was the aim of the first case study. To train the entire OPC-CNN, enough samples were collected from a quantum variational autoencoder (VAE) to substitute the corresponding actual data source. Quantum-convolutional (quanvolutional) neural networks were used in the second case study. For the lightning data, the quanvolutional layer’s training and sampling produced QNN models that performed better than the CNN model.
Weather Forecasts using Quantum for AI: A Conceptual Roadmap
Since 2021, machine learning points the way for improving weather forecasting, for example: with 3D neural networks or with DeepMapper. However, I’ve not yet found any large, implemented, quantum-enhanced AI weather forecasting projects. Some researchers are decidedly pessimistic on the idea.
A 2023 Roadmap, however, emerged from the Vellore Institute of Technology, in a CS student’s thought experiment in Suhas and Divya, 2023: conference paper: Quantum-Improved Weather Forecasting: Integrating Quantum Machine Learning for Precise Prediction and Disaster Mitigation. The authors’ big picture view (see next Figure) breaks down the components used in Weather Forecasting, and where, and which, quantum algorithms, might be implemented.
Such an optimistic, big picture view is useful, and with their suggested algorithms, it demonstrates practical directions for progress in the weather forecasting field. The year-old paper by Dalzell et al., 2023 has already 60+ citations. And the number of algorithms in the Quantum Algorithm Zoo are approaching 500.
To answer the question: “Quantum for AI: Weather Forecasting. Are we There Yet?”
No, it’s still 5-10 years away, but we know what to focus on, today.
(*) This Strategy Playbook is available in the Member’s area for GQI customers. If you are interested, please don’t hesitate to contact [email protected].
September 28, 2024
