An estimated 30,000 deaths in the U.K. each year are linked to air pollution, and researchers at the University of Manchester are applying artificial intelligence to address this critical public health issue. Physicists David Topping and Hao Zhang successfully trained NVIDIA’s Earth-2 CorrDiff, a generative downscaling model, to forecast air quality on its first attempt, using data from the U.K.’s Isambard-AI supercomputer.
“The biggest challenge is the compute required to forecast air quality,” said Topping, adding that this new approach allows them to model potential future scenarios and inform policy changes. The team demonstrated the entire workflow, from training to inference, on the NVIDIA DGX Spark personal AI supercomputer.
NVIDIA Earth-2 CorrDiff Models Forecast UK Air Quality
The team generated training data from existing climate simulations, effectively teaching the AI to predict pollution patterns without the need for exhaustive, slow chemical calculations. Simon McIntosh-Smith, director of the Bristol Centre for Supercomputing at University of Bristol and cofounder of Isambard-AI, highlighted the efficiency of the approach, noting that the project required relatively low GPU hours and minimal power consumption from the supercomputer. This reduced energy footprint is particularly notable for a climate-focused initiative, demonstrating a pathway toward sustainable AI-driven environmental modeling.
The accessibility of this technology will broaden participation in air quality research, as the DGX Spark allows for model training and inference outside of large-scale computing centers. “The fact that this model trained in two days on Isambard-AI — and can now run on a DGX Spark sitting on a desk — changes who can do this science and how quickly,” explained David Topping, a professor in the University of Manchester’s department of Earth and environmental science.
Topping further envisions a future where localized, high-resolution pollution forecasts are readily available. “With better open access to air quality observations, someone could ask our pollution model running on DGX Spark: what’s the pollution going to be like in this neighborhood tomorrow?” he said.
NVIDIA’s involvement extends beyond providing the Earth-2 framework; the company’s ongoing investment in hybrid quantum-classical computing, including the NVQLink architecture and the Advanced Quantum Computing research center in Boston, suggests a long-term commitment to accelerating scientific discovery. The September 2025 announcement of the Boston research center, focused on drug discovery, and the recent July 28, 2026 report of a quantum calibration model working across six qubit modalities, demonstrate NVIDIA’s broader strategy of integrating advanced computing paradigms to tackle complex challenges.
To improve human health, it’s essential that we understand the impact of environmental stressors in the air we breathe.
David Topping, Professor in the University of Manchester’s department of Earth and environmental science
Isambard-AI Supercomputer Accelerates Pollution Model Training
Earth-2 CorrDiff demonstrated immediate success when applied to U.K. air quality forecasting after training on Isambard-AI, a feat highlighting the adaptability of NVIDIA’s generative frameworks beyond their original climate and weather applications, the company says. Physicists used a year’s worth of hourly U.K. pollution data to generate a detailed model, achieving a resolution of 2-3 square kilometers. This rapid deployment was enabled by the model’s ability to function on the first attempt, a surprising outcome given the complexity of accurately simulating atmospheric chemistry.
“Once you put chemistry into weather models, they get really, really slow. So I said, why don’t we try using the generative frameworks that NVIDIA develops for climate and weather for pollution fields?” explained a member of the research team. The training process, completed on a single, eight-GPU node of Isambard-AI, equipped with 5,448 NVIDIA GH200 Grace Hopper Superchips delivering 21 exaflops of AI performance, required only two days.
Beyond the initial success with CorrDiff, the team expanded their work to incorporate Earth-2 StormCast, a model capable of time-dependent forecasts using real-time air quality observations. Doctoral student Hao Zhang, who trained StormCast on Isambard-AI, praised the flexibility of the NVIDIA frameworks. “The ability to switch from one NVIDIA framework to another was really impressive,” he said.
Niall Robinson, developer relations manager for weather and climate at NVIDIA, believes this accessibility will broaden participation in air quality research. “We’re just at the beginning of what these open workflows can do globally.” The team is now focused on refining the model, aiming to increase resolution and incorporate additional open data to achieve street-scale understanding of pollution patterns, according to NVIDIA.
DGX Spark Enables Desktop Air Quality Prediction
This shift allows researchers, such as those at the University of Manchester, to refine models with regional data and deploy them rapidly, a process previously limited by computational expense. The accessibility extends beyond simply running existing models; the NVIDIA framework’s flexibility allows for seamless transitions between different generative AI approaches, the firm reports. The team is now focused on integrating data from edge AI devices to further enhance the responsiveness of the system, potentially enabling real-time decision-making during events like wildfires.
Topping’s office now houses a DGX Spark system dedicated to retraining models, demonstrating a practical shift towards decentralized, localized air quality analysis. “You can now invest a few thousand dollars to get started developing powerful AI models,” Topping said, emphasizing the reduced financial commitment required to participate in this field.
The team intends to release open-source training data and workflows, fostering broader participation and enabling the creation of tailored pollution models for diverse countries and regions. “We hope that every global country and every major city with a small burst of supercomputer AI time will be able to produce their own detailed pollution models with their own local data,” Topping added.
This open approach, combined with the power of the DGX Spark, envisions a future where localized air quality predictions are readily available, empowering individuals and healthcare services to proactively address environmental health risks, and a whole chain of interactions will deliver an answer grounded on the science these frameworks represent.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
