Repository for Publications and Research Data
News from the ETH Library
Recently Added
Enforcing tail calibration when training probabilistic forecast models
Item type: Journal Article
Wessel, Jakob Benjamin; Schillinger, Maybritt; Kwasniok, Frank; et al. (2026)
Probabilistic forecasts are typically obtained using state-of-the-art statistical and machine learning models, with model parameters estimated by optimizing a proper scoring rule over a set of training data. If the model class is not correctly specified, the learned model will not necessarily produce calibrated forecasts. Calibrated forecasts allow users to appropriately balance risks in decision-making, and it is particularly important that forecast models issue calibrated predictions for extreme events, since such outcomes often generate large socio-economic impacts. In this work, we study how the loss function used to train probabilistic forecast models can be adapted to improve the reliability of forecasts made for extreme events. We investigate loss functions based on weighted scoring rules, and additionally propose regularizing loss functions using a measure of tail miscalibration. We apply these approaches to a hierarchy of increasingly flexible forecast models for UK wind speeds, including simple parametric models, distributional regression networks, and conditional generative models. We demonstrate that state-of-the-art models do not issue calibrated forecasts for extreme wind speeds, and that the calibration of forecasts for extreme events can be improved by suitable adaptations to the loss function during model training. This introduces a trade-off between calibrated forecasts of extreme events and those of more common outcomes.
Progressive failure of a circular excavation drilled through faulted clay shale — Experimental setup, rock mass structures and damage evolution at excavation boundary
Item type: Journal Article
Ziegler, Martin; Wang, Rushan; Loew, Simon (2026)
The PF (Progressive Failure of Structurally Controlled Overbreaks) experiment at the Mont Terri Underground Rock Laboratory (MT URL) examines how structurally controlled rock mass damage evolves and spreads in faulted Opalinus Clay shale excavations. Installation of the PF experiment was completed in 2020 in proximity to the Main Fault of the MT URL. The experiment consists of a central borehole measuring 0.6 m in width, which represents a circular high-level waste repository drift at a scale of 1:6.5. In addition, the experiment features six parallel monitoring boreholes situated at varying distances above the central borehole. Changes to the unsupported excavation boundary of the experimental borehole were surveyed for a period of four years using a custom-built automated close-range photogrammetric survey system. In conjunction with structural data from the six monitoring boreholes, a detailed model of the fracture network surrounding the Main Fault was established. The Main Fault at the designated test location comprises a 0.6 m thick zone of scaly clay, sharp fault zone boundaries and a few mm-thick fault gouge layer. The hanging wall and footwall are transected by two sets of mm-thick tectonic faults.
The evolution of borehole convergence and visible damage inside the central experiment borehole was explored during ventilation, transient resaturation, and a fully resaturated phase. The formation of new fractures, overbreaks, and slippage along pre-existing faults was mapped, and the time-dependent formation of borehole wall breakouts and induced cracks was assessed. The PF experiment provides insights into the damage evolution around an unsupported tunnel in faulted clay shale that are unparalleled in their precision and detail. In this study, the data acquisition and processing methods are presented, and the progressive and saturation-dependent damage evolution at the excavation boundary is discussed.
Information content of JWST transmission spectroscopy of the exoplanet HAT-P-12b from the optical to the mid-infrared
Item type: Journal Article
Heinke, Linus; Min, Michiel; Bouwman, Jeroen; et al. (2026)
Context. The James Webb Space Telescope (JWST) provides low- to medium-resolution spectra with unprecedented precision and broad near- to mid-infrared wavelength coverage, thus enabling the detailed characterization of exoplanet atmospheres. Given the complexity of JWST data and the diversity of observing modes, it is essential to understand the information content of the resulting spectra to optimize observation strategies and assess the limits of atmospheric inference. Aims. This work presents a new JWST NIRISS SOSS transit observation of the warm sub-Saturn HAT-P-12b. Together with complementary NIRSpec G395M and MIRI LRS data, this enables a detailed assessment of the information content across JWST instruments over the full accessible wavelength range (excluding MIRI MRS). Methods. The NIRISS data were reduced, and the impact of specific reduction choices on the resulting transmission spectrum was assessed. Atmospheric retrievals were performed for all combinations of JWST data, supplemented by archival HST observations in select cases. The analysis further included evaluations of molecular detection significances and assumptions about the atmospheric structure. Results. The same four molecules previously reported are significantly detected: H2O, CO2, CO, and H2S. Except for H2O, all require NIRSpec coverage for detection, while H2S is only detected in multi-instrument retrievals. Abundance constraints obtained using HST WFC3 instead of JWST NIRISS SOSS are largely consistent, particularly when combining instruments, but NIRISS SOSS proved essential to establishing robust evidence for nongray cloud behavior. A moderately steep scattering slope (p < 4) is consistently retrieved across different HST STIS reductions. Conclusions. Single-instrument retrievals, even when yielding significant detections, tend to overestimate molecular abundances. In contrast, retrievals that combine spectra from multiple JWST instruments generally converge toward consistent abundance constraints. The derived C/O ratio remains sensitive to subtle differences between NIRSpec reductions, owing to the instrument's exclusive coverage of the carbon-bearing molecules CO2 and CO, so its interpretation requires caution. The results are broadly consistent with information content studies of the benchmark target WASP-39b, although differences, including the absence of a preference for non-isothermal T-P profiles, highlight variations in information content across exoplanet types.
Optoacoustic brain imaging: A systematic review and field mapping
Item type: Journal Article
Wang, Yikai; Khansari, Amir; Ding, Qingxiang; et al. (2026)
Optoacoustic (OA) imaging has emerged as a powerful hybrid modality for investigating brain structure and function by combining optical absorption contrast with ultrasonic detection. This approach enables high-resolution imaging beyond the optical diffusion limit while preserving the strong molecular sensitivity of optical methods, thereby providing a unique platform for mapping cerebral vasculature, hemodynamics, oxygen metabolism, and disease-specific labels across spatial scales ranging from microvascular networks to whole-brain volumes. Here we present a systematic review on OA brain imaging, including both technological advances and biomedical applications while also providing a quantitative overview of the evolving research landscape. Specifically, we map the development of the field by analyzing worldwide research activity, including the geographical distribution of contributing institutions, yearly publication trends. This analysis provides a comprehensive perspective on the state of the art and emerging research directions. We then summarize recent advances in OA microscopy and tomography dedicated to brain imaging and discuss emerging multimodal platforms integrating fluorescence imaging, electrophysiology, ultrasound, and magnetic resonance imaging to enable multiparametric interrogation of brain dynamics. Applications in animal models of major neurological disorders are reviewed, highlighting functional and molecular biomarkers of disease progression and therapy. Finally, we discuss key challenges facing the field, including spectral quantification, transcranial imaging, system standardization, and clinical translation. Overall, this review serves as a comprehensive reference for capturing the multi-scale trajectory of OA neuroimaging.
Combining horizontal-to-vertical spectral ratios with empirical amplification functions to develop high resolution site amplification maps: Application to Sion, Switzerland
Item type: Journal Article
Panzera, Francesco; Bergamo, Paolo; Janusz, Paulina Agnieszka; et al. (2026)
Site amplification significantly influences ground shaking intensity during earthquakes and is a key factor in seismic hazard assessment and microzonation. Traditional methods for mapping site amplification that use earthquake recordings often face challenges related to spatial resolution and capturing local soil heterogeneities. This study takes the advantage of Horizontal-to-Vertical Spectral Ratios (HVSR) obtained from ambient noise measurements to develop high-resolution site amplification maps. The proposed methodology combines HVSR clustering, canonical correlation analysis (CC), and geostatistical interpolation to predict site amplification across the area of Sion (Switzerland). Spatial patterns are modelled using kriging, and results are compared with independent hybrid standard spectral ratio (SSRh) amplification estimates. The results show consistent identification of strong amplification in the central part of the valley across all methods, highlighting the control of basin geometry on low-frequency site response. Among the tested approaches, CC provides the best overall agreement with SSRh in terms of spatial pattern and statistical performance (lower Root Mean Squared Error), although it systematically underestimates amplification values by up to a factor of ∼2.5. The Pearson correlation-based clustering approach produces smoother amplification fields and larger discrepancies at basin edges, with underestimation of up to a factor of ∼1.5 relative to SSRh. Despite differences in amplitude, all methods reproduce the main frequency-dependent behaviour, with low frequencies controlled by deep basin resonance and higher frequencies influenced by shallow structures. Validation against independent datasets indicates good consistency in spatial trends, while also highlighting limitations of statistical models in capturing extreme amplification in deep Alpine basins. Overall, the results demonstrate that combining HVSR-based methods with empirical amplification functions improves the spatial resolution of site response estimates. CC offers the most robust framework among the tested approaches, while HVSR clustering remains useful for rapid or data-limited microzonation. The study highlights the benefit of multi-method strategies for improving seismic microzonation in complex geological environments.