Surbhi Sharma's Successful PhD Defense: Semantically Consistent and Data-Efficient Deep Learning for Land-Cover and Land-Use Mapping

On October 23, 2026, Surbhi Sharma successfully defended her PhD thesis, “Advancing Land-Cover and Land-Use Classification with Semantically Consistent and Data-Efficient Deep Learning on Multi-Source Earth Observation Data,” in Computational Engineering at the University of Iceland. Surbhi’s work addresses key challenges in satellite-based land-use and land-cover (LULC) mapping, from semantic inconsistencies between satellite imagery and in-situ observations to the limited availability of annotated training data.

Her thesis follows four complementary research directions. She used geo-tagged field photographs to study the semantic relationship between satellite and field observations, showing that many apparent disagreements stem from ambiguities in field interpretation and land-cover taxonomies. She produced Sen4Map, a large-scale benchmark dataset combining Sentinel-2 imagery with harmonized LUCAS field observations. She developed sub-pixel LULC estimation methods, using geospatial foundation model embeddings to estimate built-up fractions from Sentinel-2 imagery. Finally, she explored few-shot learning with meta-learning to reduce the dependence on large labeled datasets. Alongside these contributions, she designed an HDF5-based data architecture for efficient processing of large multi-source Earth observation datasets on High-Performance Computing (HPC) systems.

Together with Prof. Morris Riedel, her advisor, and Dr. Claudia Paris, it was a pleasure to serve on Surbhi’s doctoral committee. Congratulations, Dr. Sharma, and all the best for what comes next.

‍

More information (PhD dissertation page, University of Iceland announcement, IHPC announcement)