UMIT seminar - "eSSENCE projects 2025-2026"

14:10-14:45:
A novel method to transform the modeling of global lake ecosystem dynamics using Model Order Reduction, Cristian Gudasz.

14:45-15:20:
Unveiling space weather hazards – 3D spatiotemporal reconstructions of ionospheric currents, Maria Hamrin.

15:20-15:55:
Finite Element Methods for Modern Computing Architectures and Machine Learning Applications, Mats G Larson och Karl Larsson.

Location: UMIT Research Lab (map)
Zoom link: https://umu.zoom.us/my/mit.a.240

Talk #1 (14:10-14:45)

Speaker: Cristian Gudasz

Title: A novel method to transform the modeling of global lake ecosystem dynamics using Model Order Reduction

Abstract: As climate change and nutrient pollution intensify, linking local lake processes to global biogeochemical patterns is both urgent and complex. Process-based, climate-coupled models are the most robust means of integrating physical and ecosystem dynamics; yet applying them to 5.7 million lakes over decadal time scales is computationally prohibitive. We address this bottleneck with model order reduction (MOR), constructing reduced-order models (ROMs) from high-fidelity lake models based on partial differential equations (PDEs) to simulate physical dynamics with minimal loss of accuracy. ROMs of one-dimensional hydrodynamic models reproduce essential processes, vertical temperature structure and turbulence-driven mixing, at a fraction of the computational cost. This capability makes decadal, planet-scale simulations of lake dynamics feasible, thereby improving lake representation in Earth system models and transforming a long-standing computational barrier into a shared platform for synthesis, standardized evaluation, and theory building.

Talk #2 (14:45-15:20)

Speaker: Maria Hamrin

Title: Unveiling space weather hazards – 3D spatiotemporal reconstructions of ionospheric currents

Abstract: Space weather can seriously impact modern infrastructure in many ways. For example, rapid variations in ionospheric E region currents (90–150 km altitude) can generate Geomagnetically Induced Currents (GICs) in the ground, which may overload power grids and trigger blackouts. For instance, a 2003 geomagnetic storm-induced GIC caused a prolonged outage in Skåne, and a Carrington-level event today could cause unprecedented damage to our electrified society.

However, there are large gaps in the understanding of the spatiotemporal behaviour of the ionospheric currents due to a lack of in-situ observations since satellites rarely orbit the E region. Hence, studies must rely on remote ground or space based observations, far away in altitude. Fortunately, the new EISCAT 3D (E3D) radars will provide unique data from the relevant altitudes. For example, it will be the first radar facility to measure the 3D ion velocities needed for determining the currents.

To make the most of the E3D data, a method for reconstructing additional ionospheric data sets has been suggested (Stamm et al., 2023). It solves an inverse problem based on assumptions of the governing equations. However, this method has limitations: it assumes quasi‑neutrality and a static magnetic field (often invalid during geomagnetic substorms and storms), it cannot reconstruct the current density, and it suffers from boundary errors due to its finite‑difference approach. Our project will address these issues by incorporating the ideal Ohm’s law to allow quasi neutrality deviations, applying the finite element method to avoid boundary problems, and using an iterative technique and complementary ground and/or spacecraft data to reconstruct the magnetic field.

In the future, we will use our method to analyse the detailed spatiotemporal behaviour of the ionospheric E region currents, and we will use our new understanding in state-of-the art simulations of GICs in Sweden.

Talk #3 (15:20-15:55)

Speakers: Mats G Larson & Karl Larsson

Title: Finite Element Methods for Modern Computing Architectures and Machine Learning Applications

Abstract: In this talk, we outline how to implement a finite element solver that interoperates with modern machine-learning and differentiable-computing frameworks such as JAX. We demonstrate the approach on an inverse problem, where we incorporate observations of typical solution behavior to regularize the formulation, making an otherwise ill-posed problem solvable.