Air-quality monitors do not fail at random: they often go offline for hours or days, leaving long contiguous blocks of missing readings. In those stretches, the usual trick of borrowing information from nearby sensors becomes less dependable, because the whole local picture may be distorted. A new preprint proposes a diffusion-based imputer called GAUDI that is designed specifically for these block-shaped gaps.
GAUDI is geometry-aware: it conditions on the spatial arrangement of the monitoring network, not just on the time series itself. The authors describe it as a conditional diffusion imputer that is block-specific and aligned with the GAUDI framework, meaning it is tailored to the structure of the missing block and the surrounding data. The goal is not only to fill in plausible values but to do so with calibrated uncertainty, so downstream users know how much to trust the reconstruction.
Because the paper is a preprint, it has not yet been peer-reviewed, and the abstract does not include quantitative results. Still, the problem it targets is practical: air-quality data gaps are common, and naive imputation can bias exposure estimates. A geometry-aware, block-specific approach is a sensible direction, though its real-world gains will depend on evaluation against other imputation methods.