Sebastian Schemm

Last Name

Schemm

First Name

Sebastian

ORCID

0000-0002-1601-5683

Show all metadata

Search Results

Publications1 - 10 of 45

Graphframex: Towards systematic evaluation of explainability methods for graph neural networks
Amara, Kenza; Ying, Rex; Zhang, Zitao; et al. (2022)
Proceedings of Machine Learning Research ~ Proceedings of the First Learning on Graphs Conference
As one of the most popular machine learning models today, graph neural networks (GNNs) have attracted intense interest recently, and so does their explainability. Users are increasingly interested in a better understanding of GNN models and their outcomes. Unfortunately, today’s evaluation frameworks for GNN explainability often rely on few inadequate synthetic datasets, leading to conclusions of limited scope due to a lack of complexity in the problem instances. As GNN models are deployed to more mission-critical applications, we are in dire need for a common evaluation protocol of explainability methods of GNNs. In this paper, we propose, to our best knowledge, the first systematic evaluation framework for GNN explainability, considering explainability on three different “user needs”. We propose a unique metric that combines the fidelity measures and classifies explanations based on their quality of being sufficient or necessary. We scope ourselves to node classification tasks and compare the most representative techniques in the field of input-level explainability for GNNs. For the inadequate but widely used synthetic benchmarks, surprisingly shallow techniques such as personalized PageRank have the best performance for a minimum computation time. But when the graph structure is more complex and nodes have meaningful features, gradientbased methods are the best according to our evaluation criteria. However, none dominates the others on all evaluation dimensions and there is always a trade-off. We further apply our evaluation protocol in a case study for frauds explanation on eBay transaction graphs to reflect the production environment.
Impacts of North American forest cover changes on the North Atlantic Ocean circulation
Bauer, Victoria; Schemm, Sebastian; Portmann, Raphael; et al. (2025)
Earth System Dynamics
Planetary-scale forestation has been shown to induce global surface warming associated with a slowdown of the Atlantic Meridional Overturning Circulation (AMOC). This AMOC slowdown is accompanied by a negative North Atlantic sea surface temperature (SST) anomaly resembling the known North Atlantic warming hole found in greenhouse gas forcing simulations. Likewise, a reversed equivalent of the SST response has been found across deforestation experiments. Here, we test the hypothesis that localised forest cover changes over North America are an important driver of this response in the downstream North Atlantic Ocean. Moreover, we shine a light on the physical processes linking forest cover perturbations to ocean circulation changes. To this end, we perform simulations using the fully coupled Earth system model CESM2, where pre-industrial vegetation-sustaining areas over North America are either completely forested (“forestNA”) or turned into grasslands (“grassNA”). Our results show that North American forest cover changes have the potential to alter the AMOC and North Atlantic SSTs in a manner similar to global ones. North American forest cover changes mainly impact the ocean circulation through modulating land surface albedo and, subsequently, air temperatures. We find that comparably short-lived cold-air outbreaks (CAOs) play a crucial role in transferring the signal from the land to the ocean. Around 80 % of the ocean heat loss in the Labrador Sea occurs within CAOs during which the atmosphere is colder than the underlying ocean. A warmer atmosphere in forestNA compared to the “control” scenario results in fewer CAOs over the ocean and thereby reduced ocean heat loss and deep convection, with the opposite being true for grassNA. The induced SST responses further decrease CAO frequency in forestNA and increase it in grassNA. Lagrangian backward trajectories starting from CAOs over the Labrador Sea confirm that their source regions include (de-)forested areas. Furthermore, the subpolar gyre circulation is found to be more sensitive to ocean density changes driven by heat fluxes than to changes in wind forcing modulated by upstream land surface roughness. In forestNA, sea ice growth and the corresponding further reduction in ocean-to-atmosphere heat fluxes forms an additional positive feedback loop. Conversely, a buoyancy flux decomposition shows that freshwater forcing only plays a minor role in the ocean density response in both scenarios. Overall, this study shows that the North Atlantic Ocean circulation is particularly sensitive to upstream forest cover changes and that there is a self-enhancing feedback between CAO frequencies, deep convection, and SSTs in the North Atlantic. This motivates studying the relative importance of these high-frequency atmospheric events for ocean circulation changes in the context of anthropogenic climate change.
A climatological characterization of North Atlantic winter jet streaks and their extremes
Bukenberger, Mona; Fasnacht, Lena; Rüdisühli, Stefan; et al. (2025)
Weather and Climate Dynamics
The jet stream is a hemisphere-wide midlatitude band of westerly wind. Jet streaks, which are regions of enhanced wind speed within the jet stream, characterize it locally. Jet streaks are frequent upper-tropospheric flow features that accompany troughs and ridges and form in tandem with surface cyclones. Upper-level divergence in their equatorward entrance and poleward exit regions couples them to surface weather via vertical motion, and these are regions prone to precipitation formation, which feeds back into the strength of upper level divergence and wind speed via diabatic heat release. This reanalysis-based study presents a systematic characterization of the life cycle of jet streaks and extreme jet streaks over the North Atlantic during winter, their occurrence during three different regimes of the eddy-driven jet, and their relation to Rossby wave breaking (RWB) from a potential vorticity (PV) gradient perspective. Extreme jet streaks are most frequent when the North Atlantic jet is in a zonal regime, while they are least common when the jet is in a poleward-shifted regime. Maximum wind speed on average occurs on the 330 K isentrope, and the peak intensity of jet streaks, defined as the maximum wind speed throughout their evolution, scales with the strength of the PV gradient, with mean values of 1.7 PVU (100 km)−1 for wind speeds exceeding 100 m s−1. The peak intensity of jet streaks also increases with their lifetime, and extreme jet streaks exhibit a prolonged intensification period as well as increased acceleration rates. A positive trend in jet streak intensity seems to have been emerging since 1979, but decadal variability still dominates the 43-year time series. Clustering jet streak events identifies typical Rossby wave patterns in which jet streaks reach peak intensity and their preferred location and orientation within the large-scale environment. In case of anticyclonic RWB, the jet streak sits upstream of the ridge axis, while in case of no RWB, the jet streak is zonally oriented and is located slightly downstream of the ridge axis. In some cases, the jet streak is found farther downstream of the ridge axis, but no case of well-marked cyclonic RWB is found at maximum jet streak intensity. As expected, the presence of an extreme jet streak is associated with a meridionally aligned pair of surface cyclones and anticyclones. More specifically, a cyclone is located poleward of an anticyclone and, in some cases, a mesoscale cyclone upstream of both, which is associated with intense precipitation. This motivates a detailed follow-up study on the relative roles of diabatic and adiabatic processes in the formation of extreme jet streaks.
The life cycle of upper-level troughs and ridges: a novel detection method, climatologies and Lagrangian characteristics
Schemm, Sebastian; Rüdisühli, Stefan; Sprenger, Michael (2020)
Weather and Climate Dynamics
Eddy Lifetime, Number, and Diffusivity and the Suppression of Eddy Kinetic Energy in Midwinter
Schemm, Sebastian; Schneider, Tapio (2018)
Journal of Climate
Wave activity flux applied to baroclinic development
Schemm, Sebastian (2009)
Tuning Atmospheric Turbulence Parameters with Machine Learning Surrogates
Grund, Dana; Fuhrer, Oliver; Mishra, Siddhartha; et al. (2024)
Parameterizations of subgrid-scale (SGS) processes, like cloud microphysics, radiation, or turbulence, cause considerable uncertainty in numerical climate and weather models at various spatiotemporal scales. Tuning the involved model parameters is challenging, given the immense computational cost of model evaluations, and the reliance on empirical judgement. The transition of numerical weather prediction to convective scales (spatial resolutions of hundreds of meters) is accompanied by new data assimilation methods including parameter estimation. However, their performance is limited by either simplified model representations or repeated model evaluations. For more objective calibration, using iterative Bayesian methods (MCMC algorithms), fast and accurate model surrogates are needed. The recent advances of data-driven full-model emulators, that avoid explicit SGS modeling, motivates the extension of such models to capture the effects of SGS parameters. Here, we focus on turbulence parameterizations in large-eddy simulations (LES) with resolutions of tens of meters. In order to accurately represent turbulence, emulators of LES simulations have to capture both the variability of the resolved turbulent motion (probabilistic/ensemble forecast) and its mean state. To this end, we compare extensions of deterministic forward emulators, such as neural operators, for probabilistic forecasting of idealized atmospheric test cases, in order to assist model calibration.
Jet stream dynamics from a potential vorticity gradient perspective: The method and its application to a kilometre-scale simulation
Bukenberger, Mona; Rüdisühli, Stefan; Schemm, Sebastian (2023)
Quarterly Journal of the Royal Meteorological Society
The influence of adiabatic and diabatic processes on the midlatitude circulation is a formidable research question, especially considering their projected changes under global warming. This study presents the prospects, merits, and caveats of a potential vorticity (PV) gradient perspective as a means to disentangle the contributions of adiabatic and diabatic processes affecting the midlatitude circulation. Theoretical considerations reassess the link between the PV gradient and the jet stream. They reveal that the maximum isentropic PV gradient is consistently located on the stratospheric side of the jet, whereas the gradient of ln(PV) is shifted to the tropospheric side but, in general, is better aligned with the jet axis. The stratospheric shift of the PV gradient results from variations in stability across the tropopause, whereas the tropospheric shift of the ln(PV) gradient results from variations in vorticity. Regions of high PV gradient may serve as a proxy for the curvature of the wind field in the case of sufficiently small variations in stability. Otherwise, they depict variations in both wind and thermal stratification along tropopause-intersecting isentropic surfaces. Lagrangian “PV gradient thinking” is demonstrated in two case studies of jet streak evolution in a simulation with 1.1 km grid spacing performed with the graphics-processing-unit-enabled numerical weather prediction model Consortium for Small-Scale Modelling featuring on-line air parcel trajectories. Dry deformation drives the Lagrangian evolution of the PV gradient in the first case, whereas there is a pronounced influence of diabatic modification in the second case. The Lagrangian PV gradient perspective presented offers fresh insight into adiabatic and diabatic processes underlying the midlatitude circulation variability and change.
Global Climatologies of Eulerian and Lagrangian Flow Features based on ERA-Interim
Sprenger, Michael; Fragkoulidis, Georgios; Binder, Hanin; et al. (2017)
Bulletin of the American Meteorological Society
Assigning precipitation to mid-latitudes fronts on sub-daily scales in the North Atlantic and European sector: Climatology and trends
Hénin, Riccardo; Ramos, Alexandre M.; Schemm, Sebastian; et al. (2019)
International Journal of Climatology

Publications1 - 10 of 45

Sebastian Schemm

Last Name

First Name

ORCID

Organisational unit

Filters

Search Results