AI Model Promises Precision in Weather Forecasting

NASA selects Planette, based in San Francisco, to develop QubitCast, a quantum-inspired AI system designed to predict extreme weather months in advance. This partnership aims to bridge the gap in the scientific community's current abilities to forecast far into the future.

Currently, meteorologists struggle with accurate projections beyond a 10-day period.

While entities such as the UK's Met Office offer predictions spanning two weeks, accuracy diminishes beyond this duration. Today's weather modelling requires substantial energy and computing resources, yet often misses complex patterns pointing to severe events like hurricanes and heatwaves.

This could help insurance underwriters to set more accurate parameters for insurance claims. 

QubitCast is expected to be a system that surpasses current capabilities in both accuracy and duration, while being energy-efficient. It anticipates providing forecasts by employing algorithms rooted in quantum physics concepts.

Quantum theory allows examination of multiple possibilities simultaneously, intending to handle vast atmospheric, oceanic, and land data more efficiently than traditional means.

Advanced, Energy-Saving Computing

Existing simulation models require extensive energy and computing power, while current AI models often struggle under complex data intricacies. QubitCast aims to promise more accurate predictions while minimising computing resources and energy consumption compared to current systems.

It simplifies complex information while retaining essential details for accurate forecasting.

This improvement in efficiency can make long-term forecasting universally accessible and cheaper for insurance companies needing extensive prediction horizons.

Is Quantum Theory the Answer for Long-Term Accuracy?

NASA and Planette plan for QubitCast to explore many possibilities concurrently, a quantum theory principle. This approach allows the system to handle massive data efficiently.

"You can think of it like reading the entire history of Earth's systems all at once," says Dr Kalai Ramea, Co-Founder and CTO of Planette.

"Instead of slowly scanning year by year and missing critical details, our approach allows us to spot anomalies, those needles in the haystack that signal extreme weather events, much faster and more accurately than traditional AI models ever could, while using far less energy."

Furthermore, the tech is designed to run on conventional computers instead of costly quantum hardware, which remains years from becoming widespread technology. This makes it feasible for NASA to launch QubitCast promptly.

The Technology's Commercial Impact

QubitCast targets subseasonal-to-seasonal (S2S) forecasting–predictions spanning two weeks to two years. "Too many critical decisions are made in the dark because reliable long-range forecasts simply haven't been available," explains Dr Hansi Singh, Co-Founder and CEO of Planette.

"QubitCast changes that equation by making S2S forecasting not just more accurate but practical to deploy at scale when making highly consequential decisions."

Hansi envisions applicability across diverse sectors such as agricultural planning, emergency management, and energy stability. The system could help farmers optimise crop cycles, prepare emergency responses for hurricane seasons, and allow energy companies to manage electricity flow efficiently under demand pressures.

This collaboration contributes to NASA's Small Business Innovation Research grant that Planette received, aiding in further development of its weather intelligence technologies.

Earlier this year, a Phase I SBIR grant from the National Science Foundation supported NIVA, Planette's foundational AI model for Earth's systems analysis, and the company made long-term weather forecasts publicly available via Eddy.

By bridging short-term weather and long-term climate predictions, Planette's vision is to generate insights from one week up to six months, with intentions for even longer forecasts. With NASA's collaboration, the possibilities appear limitless for QubitCast.