Arctic sea ice plays a vital role in the climate system, influencing atmospheric circulation, oceanic processes, and Arctic ecosystems. Thin sea ice is particularly important, as it governs seasonal transitions and contributes to feedback mechanisms such as Arctic amplification. Reliable estimates of sea ice thickness (SIT), especially within the thin ice regimes (less than 1 m), are essential for ...

The increasing density of space debris poses a growing threat to satellite operations and future missions. While sub-millimeter debris is studied through impact craters and objects larger than 0.5 cm are tracked by ground-based radar, debris in the 1–3.2 mm range remains largely unmonitored. This thesis investigates the feasibility of using space-based techniques to estimate the flux of millimeter-scale ...

Given substantial emissions of greenhouse gases, carbon capture and storage (CCS) plays an increasingly important role in efforts to reduce emissions. The Snøhvit CO2 injection project began in 2008 with injection into the Tubåen Formation through well F-2 H, which due to rapid pressure build-up was terminated in 2011 and moved to the shallower Stø Formation. A second injection well G-4H was drilled ...

In this thesis, we investigate the possibility of using a deep learning based pipeline for detection and classification of seals in high-resolution optical satellite images. The pipeline includes a seal detector which uses a UNet-3+ architecture and a clustering module for point level prediction, and an image level convolutional neural network (CNN) seal classifier. Areal images targeting the seal ...

Wind turbines in cold climates can occasionally experience ice throws that travel significant distances, posing a low-probability but high-consequence safety risk. Current guidelines typically rely on the empirical Seifert safety distance, defined as d = k x (H + D), where k is the throw factor – commonly set to 1.5, H is the hub height, and D is the rotor diameter. However, both observational field ...

This thesis tracks general LLM reasoning by examining how two open-weight models—GPT2-XL (1.5 B) and GPT-Neo (1.3 B)—organise meaning across their hidden layers. Four structured text suites (unrelated, related, identical, cross-lingual) and 50-word “Country-Stories” summaries feed the models. Layer-wise activations are projected with UMAP, connected via k-nearest-neighbour graphs, and summarised ...

As neural networks grow increasingly complex and powerful, understanding their internal representations becomes critical for ensuring safe and reliable AI systems. This thesis addresses a fundamental challenge in mechanistic interpretability: how architectural choices in sparse representation learning shape our ability to understand neural network internals. We present the first comprehensive ...

During the summer months at high and mid-latitudes, VHF and UHF radars can detect strong radar echoes, known as Polar Mesospheric Summer Echoes (PMSE), at altitudes between 80 and 90 km. The formation of PMSE is closely linked to the presence of ice particles, turbulence, and irregularities in the electron density. Current theories suggest that neutral turbulence causes the spatial structuring of ...

The physical environment of the northern Barents Sea is shaped by the inflow of warm, saline Atlantic Water (AW) and the seasonal cycle of sea ice. Recent years have seen a rapid shift toward more Atlantic conditions, driven by reduced sea ice import and increased influence of AW intrusions. Although seasonal variability is crucial for understanding these changes, observations from the ice-covered ...

This thesis investigates the Polaris prospect in the southwestern Barents Sea as a potential site for geological CO₂ storage. The study’s objectives were to (1) map the storage complex, (2) assess reservoir and seal suitability, (3) estimate storage capacity, and (4) evaluate leakage potential through the regional fault system. To achieve this, 3D seismic data were interpreted in Petrel alongside ...