My research uses Graph Signal Processing and graph learning to combine measurements from different sensors, resolutions, and physical contexts—recovering information that a single modality cannot reliably provide alone.
Pixels or measurements become graph nodes; edges encode spatial, terrain, spectral, temporal, or sensor-derived relationships. This representation supports reconstruction on irregular domains, the incorporation of ancillary variables, and the preservation of local structure.
My remote-sensing research combines data from NASA's Soil Moisture Active Passive (SMAP) mission, MODIS products, and Global Navigation Satellite System Reflectometry (GNSS-R) measurements from the SMAP reflectometer with terrain elevation, land-surface temperature, vegetation optical depth, and vegetation water content. The work addresses coarse resolution, missing observations, and mismatched spatial scales, with validation against in-situ measurements at calibration sites across the United States.
Established a GSP reconstruction framework in which high-resolution ancillary data define graph relationships for improving coarse remote-sensing observations.
Extended the framework with data-fitting, structural-model, and graph-smoothness terms to integrate soil moisture, temperature, vegetation, and terrain information while maintaining physical consistency.
Introduced a GCN framework that learns edge relationships from multimodal features and transfers them to unseen geographic regions when ancillary data are available.
