MIST-Net: A Heterogeneous Spatiotemporal Network for Regional Ozone Forecasting Robust to Sparse Data.

MIST-Net reduces 1-hour ozone forecasting RMSE by over 25% compared to SARIMA and DCRNN, demonstrating high accuracy and robustness under sparse data conditions.

• Why it matters: Accurate ozone prediction is hindered by complex pollutant interactions and sparse monitoring data, limiting effective air quality management and public health responses.
• What they did: The model employs a dual-graph architecture treating environmental factors as independent nodes, with mechanisms to learn from incomplete data, evaluated on datasets from Los Angeles and Madrid.
• The result: MIST-Net outperforms existing models in accuracy, maintains limited error increases under 40% missing data, and provides insights into variable interactions, enabling resilient air quality monitoring in sensor-failure scenarios.