Skip to main content

2024 Winter meeting

The COMPUTE Winter Meeting 2024 will be on Friday, March 22nd. The theme is "Computational Science and Sustainability". This is a joint event with the LU profile area Nature based future solutions. It is supported by the LU eScience Hub. We will have talks and discussions about the intersection between computational science and sustainability.

The meeting takes place in Lundmarksalen.


  • 9:30 Fika/mingle
  • 10.00-10.15 COMPUTE News and Introduction (Sonja Aits/Philipp Birken/Hakim Abdi)

  • 10.15-11.00 Maja Essebo (LUCSUS): The role of algorithmic practices in, for, and against sustainability (and why stories matter)

  • 11.00-11.30 Breakout Sessions where lead questions are discussed, see below

  • 11.30-12.00 Presentations from breakout sessions

  • 12.00-13.00 Sandwich lunch

  • 13.00-13.40 Lightning Talks by COMPUTE PhD students

  • 13.40-14.20 Ankit Kariryaa (UCPH): Sustainability from space: mapping individual trees and carbon using AI

  • 14.20-15.00 Clas Jacobson (Carrier): Can simulations help with energy efficiency of buildings?

  • 15:00-15.30 Fika

Lightning Talks

The point of a lightning talk is to briefly present the main point about a topic. The theme is sustainability in its broadest sense.

  • 1. Wan Ni Lin (INES): An Overview of the Vegetation Fire Dynamic in the Middle East

Fire plays an important role in the ecosystem, but it can also make great impacts on environment and be harmful for the human health. Meanwhile, the fire intensity and frequency have increased in recent decades. However, most research focus on area with larger fire activities and the relevant studies in America, Austria, and Africa. Relatively less attention of vegetation fire has brought to the Middle east, which is an area with increasing risk of burned areas due to the high temperature, wind speed and underlying conflict uncertainties. To fill the knowledge and research gap of vegetation fire patterns and drivers in the Middle East, this project attempt to provide an overview of the vegetation fire dynamic during 20 years of long-term observation. To achieve the goal, remote sensing data is utilized due to its time trackability and its ability to cover country level or cross-country level study area.

  • 2. Simon Liedtke (Chemistry): Solar Fuel Simulations

We develop and apply reactive Molecular Dynamics (MD) methods to simulate coarse-grained photosensitiser and quencher dynamics that serve to model the interactions with physically accurate size, force fields, solvent properties, molecular concentration, etc. The simulated rates and dynamic interactions are able to be observed on the molecular level, which can be compared with the rates of quenching which are observed in experiments. There are a number of theories which are currently used to describe this fluorescence-quenching behavior, such as Stern-Volmer and Smoluchowski, though these theories tend to break down at high concentrations, complexation, and activated quenching processes. Our MD simulations are able to provide a physically accurate picture of when and why these theories break down, and also be used to simulate more complex systems which are not described by theory. The accurate simulation of such processes, in comparison with theory and experiment, can serve to provide better understanding of the nature of these interactions. The aim of this research is motivated by the desire to understand charge separation (CS) dynamics, which can be used for energy extraction in solar cells and solar fuels. Through improved understanding of photo-excitation and charge carrier dynamics, this information can be used to fine-tune and select the most efficient molecular systems for sustainable energy conversion.

  • 3. Valentina Schüller (Mathematics): Numerical Methods for Coupled Groundwater and Surface Flows

Floods caused by changing rainfall patterns or extreme events affect the water table. Similarly, the distribution of groundwater in the soil has an influence on the severity and extent of floods. These physical processes can be modeled by coupling the Richards equation for groundwater flow to a shallow water model for rivers or lakes. This leads to a coupled system of nonlinear partial differential equations which act on vastly different time scales. Our research aims to develop efficient and accurate numerical coupling algorithms for this application, so-called waveform iterations. We analyzed an idealized, linearized formulation of this coupling problem, yielding an optimal parameter choice to guarantee fast convergence of waveform iterations. Our theoretical analysis is supported by numerical results for a fully nonlinear test case.

  • 4. Michail Boulasikis (Computer Science): Beyond AI: Using AI Hardware for COMPUTE Applications

Modern AI hardware accelerators boast impressive parallel processing capabilities and energy efficiency, outperforming conventional CPUs in tasks like neural network training and inference. Despite their current focus on AI, we foresee a broader integration of these accelerators into other applications, similar to the way GPUs evolved into GPGPUs more than a decade ago. We propose that this evolution presents a promising opportunity for scientific computing, which can benefit from the computational power and efficiency of AI accelerators to improve simulations and numerical analyses while using less energy. As an example, we will show that it's possible to express Finite Difference Methods (FDMs) for Partial Differential Equations (PDEs) in a way that conforms to the computational patterns AI hardware is built for. Namely, we show that PDEs discretized with FDMs can be modelled as convolution layers, described in TensorFlow, and successfully solved using Google TPUs. Additionally, we will present more opportunities for usage of AI hardware, such as tensor networks in quantum systems. Finally, we'll delve into the adaptation challenges, including considerations like numerical precision, mismatches in computational patterns, and the ease of programming.

  • 5. Isabel K. Erb (Sweden Water Research & Chemistry): Data-driven early warning system of E. coli levels using machine learning on raw, low resolution, flowcytometry 2D histogram pixels for monitoring of urban bathing waters.

Traditional methods for testing bathing water quality take up to two days to produce results putting bathers at risk during this period. Alternative methods that produce results more efficiently have not been well established. In this study, flow cytometric and E. coli concentration measurements were taken from 15 bathing locations in Southern Sweden. Correlations between flow cytometric measurements and E. coli concentration were explored. Applying machine learning algorithms confirmed correlations and identified patterns in flow cytometric fingerprints associated with E. coli. Random Forest algorithm was found to be the best in predicting safe and unsafe E. coli thresholds in comparison to Logistic Regression and support vector machines, improving prediction accuracy to 80% from a baseline approach of 70%. Furthermore, a two-threshold model was introduced, which only made safe predictions further improving accuracy to 87% by utilizing the prediction probability information in Random Forest. However, this approach came at the cost of only predicting 65% of the samples. A feature importance ranking was conducted using Random Forest to identify the most important part of the flow cytometry two-dimensional histogram for classification. This study suggests that flow cytometry (FCM) combined with machine learning could be used as a complementary tool to assess bathing water quality more efficiently.

  • 6. Leonard Nzabonantuma (Building and Environmental Technology): Reduction of flood risk and sediment load in and along Nyabarongo River, enhancing ecosystem services, by employing Nature-Based Solutions approach.

We investigate the use of Nature-Based Solutions (NBS) to tackle flood risk and sedimentation issues in the Nyabarongo River and its surroundings, aiming to improve ecosystem services. Environmental degradation and urban expansion exacerbate flood risks, threatening communities and ecosystems. Focused on Rwanda's Nyabarongo catchment, the study employs advanced Flood Risk Management (FRM) techniques, including machine learning and simulation models, to assess NBS effectiveness. It aims to inform sustainable policy-making by advocating for basin-based flood risk governance and promoting NBS interventions. The research adopts a comprehensive approach, integrating green infrastructure, restoring riparian zones, and engaging local communities. Through field surveys, hydrological modeling, and stakeholder consultations, the study evaluates NBS feasibility and effectiveness. It emphasizes NBS as a cost-effective and sustainable alternative to traditional engineering methods, offering a holistic strategy for reducing flood risks in the Nyabarongo catchment. Grounded in scientific inquiry, policy advocacy, and community engagement, the research presents a thorough framework for addressing the complex challenges of flood risk and sedimentation.

Breakout Sessions

Rebound Effect

Read the intro paragraph and "Suggested Solutions" in

Think about using optimization to improve efficiency of a product, e.g. reducing the use of material, the fuel usage, or energy use in production. Can the Rebound effect lead even to bad outcomes? What could one do?

Energy use of Computations

Skim the article

What is our role as researchers in this development? What is our role as teachers in this development?

Planetary boundaries (several groups)

Look at
Pick a planetary boundary, we suggest "Climate Change", "Change in Biosphere Integrity", "Land system Change" , or "Freshwater Change".

Identify a major challenge in creating a society where this planetary boundary is stable. What role can computational science play in helping achieve this society?


Please register until 18th of March 2024 using the registration form.


If you have comments or questions please contact COMPUTE via

Page Manager: | 2024-03-22