CleanAir uses a 3D U-Net to emulate CMAQ and cut a yearlong run to 10 seconds
bravo_abad · x · 2026-07-21
Researchers built CleanAir, a residual symmetric 3D U-Net surrogate for the CMAQ chemical transport model that predicts daily gridded PM2.5 changes and component-level outputs under emission-reduction scenarios.
- Trained on 2,416 CMAQ-simulated emission-reduction scenarios, about 74,000 daily samples, sampled across 15 emission dimensions at national, provincial, and grid levels.
- The model uses an adaptive multitask loss to balance multiple chemical components.
- On held-out CMAQ runs, monthly PM2.5 change reaches R = 0.999.
- It generalizes to unseen meteorology and emission inventories, tracking observed 2017–2020 trends within CMAQ’s own error.
- Runtime drops to about 10 seconds per year on one GPU, roughly 3–4 orders of magnitude faster than CMAQ.
- The paper argues this turns long-horizon scenario search from a rationed task into something that can be run thousands of times, while still inheriting the simulator’s biases and requiring validation of the underlying model.
Paper: Liu et al., Science Advances (2026).
More from Infra
- A shopping app demo ties OpenTelemetry, Dynatrace and Port into agentic ops — Pavan_Belagatti · 2026-07-21
- Nvidia Rubin is coming, pointing to the next AI compute platform — ezyang · 2026-07-21
- Tesla’s FSD v14 Lite is reportedly headed to 4 million older HW3 cars — MatthewBerman · 2026-07-21
- TSMC’s 3nm utilization reportedly tops 120% as AI demand drives a $190B capex cycle — tengyanAI · 2026-07-21
- Nativ brings local AI model running to Mac with a desktop app and localhost API — Simon Willison · 2026-07-21
- Octen says agent search now runs at 62ms P50 with only a 6ms P90 gap — aakashgupta · 2026-07-21