FAMdust: Fast Adaptive Machine-Learning Framework for Global Martian Dust Optical Depth Retrieval: Applied to HOPE/EMIRS Thermal Infrared Observations

Monitoring dust optical depth (DOD) from space is critical to support China’s Tianwen Mars missions. Here, we present FAMdust (Fast Adaptive Machine-learning framework for dust retrieval), a regime-aware machine-learning framework developed to retrieve Martian DOD from satellite thermal infrared spectra. In the absence of operational Chinese thermal infrared sounders at Mars, FAMdust is applied to observations from the Emirates Mars Infrared Spectrometer (EMIRS) onboard the Hope spacecraft, which has provided global thermal infrared measurements of the Martian atmosphere since 2021. Using EMIRS radiance spectra, FAMdust produces global DOD estimates that show strong agreement with the Montabone gridded dust dataset and independent ground-based observations from the Mars Science Laboratory (MSL) Mastcam. On an independent test set, the retrieval achieves a coefficient of determination of R2 = 0.96, with 96.60 % of retrievals falling within the expected error envelope of ±(0.05+0.15×DOD). Global DOD maps derived from EMIRS successfully capture major dust structures and regional enhancements while maintaining consistency at key landing sites. Time series comparisons with Mastcam observations demonstrate strong temporal correlation (R = 0.86) and low mean bias error (MBE = 0.08) across Martian Years 36–37. Relative to current products, the FAMdust-derived DOD exhibits improved temporal continuity and extends global dust monitoring into MY37. These results demonstrate the capability of FAMdust to provide fast, quantitative, and globally consistent DOD retrievals and highlight its adaptability to future Mars orbiters equipped with similar infrared sensing capabilities.