Turning Computational Frustration into Innovation: Combining Too Coarse Meshes with Simplified Physics Models for Efficient Modelling

High-performance computing (HPC) is a powerful tool in geoscience, yet its complexity often creates barriers. Running large-scale simulations requires significant resources and time, and results are rarely immediate—making iterative improvements (or debugging) slow and frustrating. During my research, and after long HPC runs, I discovered that my chosen mesh discretization still suffered from significant discretization errors, forcing me to restart the workflow and endure additional delays.

Rather than viewing this as wasted effort, I analyzed that data and realized that the “incorrect” data contained valuable information. This observation sparked a new idea: could we bypass costly, high-resolution simulations by combining approximate physics models with coarse-mesh simulations to reduce error without sacrificing interpretability? Over a single weekend, I developed a prototype workflow that integrates these concepts [1-2]. I have now developed an approach that avoids full 3D forward models that requires specialists’ expertise and a tremendous amount of computational resources, not readily available to everyone. It allows users to check their multidimensionality assumptions, without relying on costly simulations. Recently, I updated the approach to actually replace full 3D simulations with a combination of coarse meshes and simplified physics models. This experience highlights how unexpected challenges can lead to creative solutions.

[1] Deleersnyder, W., Dudal, D., Hermans, T. (2022). Novel Airborne EM Image Appraisal Tool for Imperfect Forward Modeling. Remote Sensing, 14 (22), Art.No. 5757. doi: 10.3390/rs14225757

[2] Deleersnyder, W., Slob, E. (2025). Fast approximate physics method for 3D time-domain EM modelling. Geophysics https://doi.org/10.1190/geo-2025-0566