Arno Rüegg

Ionospheric Slant TEC Modeling Based on GNSS Data with Machine Learning

Development of a data-driven model for slant total electron content (STEC) estimation directly from GNSS observations, bypassing single-layer assumptions and mapping functions. The approach leverages deep learning to model spatiotemporal ionospheric variability and provides uncertainty estimates alongside the STEC. Models are pre-trained on historical data and fine-tuned daily to achieve improved accuracy compared to standard GIM-based corrections.

Work in progress · Conference material (open access)

Links: Slides (PDF)

Fusion of GNSS and VLBI Data for Global Ionospheric Mapping

Machine learning–based data fusion framework combining dense GNSS-derived VTEC with sparse but complementary VLBI observations. Two fusion strategies are investigated: direct assimilation of station-based VLBI VTEC and end-to-end integration of VLBI differential TEC (DTEC). A weighted loss balances heterogeneous data volumes, with evaluation against independent Jason-3 altimetry measurements over a full year.

IAG – Geodesy for a Changing Environment 2025 (under review) · Poster (open access)

Links: Poster (PDF)

A Comparative Study of VTEC Estimates Derived from S/X VLBI and VGOS Observations

Comparative analysis of vertical total electron content (VTEC) estimates derived from colocated S/X VLBI and VGOS observations during simultaneous sessions. Results show good agreement with GNSS-based Global Ionosphere Maps (GIM) and indicate improved stability for VGOS under increased solar activity, highlighting its potential for more robust ionospheric monitoring during disturbed conditions.

Space Weather (manuscript) · Code + data available

Links: AGU Space Weather · Dataset · VLBIono (GitLab)

Master Thesis: A Shared Deep Feature Embedding of Sentinel-1 and Sentinel-2 for Building Detection

Joint deep learning framework for building segmentation using SAR (Sentinel-1) and optical (Sentinel-2) imagery. The work explores domain consistency through adversarial learning and input augmentation, enabling improved single-domain inference while leveraging multi-sensor information during training.

Links: Thesis (PDF)

Master Project: Vegetation Mapping with ICESat-2

Evaluation and bias correction of a global canopy height map using ICESat-2 laser altimetry data across multiple test regions. Systematic over- and underestimation effects were analyzed and mitigated using tile-wise mean corrections and spatial smoothing.

Links: Report (PDF)

Bachelor Thesis: Classification of Solid and Loose Rock on Orthophotos with Deep Learning

Deep learning–based classification of solid rock and loose material in aerial imagery for permafrost-related mapping applications. Implemented with a U-Net architecture and transfer learning, evaluated via cross-validation and data augmentation.

Links: Thesis (PDF)