Rigetti Computing announced it has developed an effective solution to a weather modeling problem using quantum computers. Building on existing machine learning workflows, the company applied a combination of classical and quantum machine learning techniques to produce high-quality synthetic weather radar data and improve classical models for storm prediction.
The work was performed on Rigetti’s 32-qubit system, demonstrating that practical applications are within reach for near-term quantum hardware.
“These results set the stage for achieving quantum advantage on a high-impact, operationally relevant problem,” said Chad Rigetti, founder and CEO of Rigetti Computing, in a statement. “We’ve shown that quantum computers can be integrated effectively into state-of-the-art classical workflows and perform tasks with real-world significance.”
Private weather forecasting in the United States is a $7 billion industry and growing, according to a 2017 study by the National Weather Service. The study estimates that businesses could derive as much as $13 billion in economic value from tailored weather data for a variety of weather-related applications. It also suggests that continued improvements in weather forecasting could lead to significant gains in economic value across all sectors, as businesses and governments are better able to prepare for disasters, reduce risk, and drive critical decision making.
Generative machine learning models have emerged as a powerful new tool for enhancing predictive capabilities. One such model is the Offshore Precipitation Capability (OPC), a convolutional neural network developed by MIT Lincoln Laboratory. OPC integrates several inputs including satellite imagery, lightning strike data, and numerical models to generate synthetic radar-like data for regions outside traditional weather radar coverage. These models inform critical decisions in areas such as off-shore air traffic management for civilian and military aviation.
“We believe that quantum computers will be most valuable when they operate in tandem with classical computers,” said Matt Reagor, VP of Quantum Engineering at Rigetti, in a statement. “These results confirm that quantum subroutines can be inserted directly into a practical machine learning workflow. In addition, the techniques we developed are transferable to applications in other areas such as computational finance, genomics, and image processing.” The research was funded in part by the US Government.