Bridging AI and Climate Modelling across disciplines
AI4PEX is short for Artificial Intelligence for Process Enhancement in Earth System Models, a collabortive effort involving 18 partner institutions across nine countries. The goal is to develop hybrid modelling frameworks that combine machine learning with physically based process models to better simulate how our planet’s land, atmosphere, and oceans behave, especially under the influence of extreme events.
The central theme of the Lund meeting was knowledge exchange between disciplines bringing together ecologists, atmospheric scientists, oceanographers, and AI experts. Discussions ranged from simulating extreme weather impacts on forests, to modelling ocean heatwaves and carbon exchange, to improving cloud representation in climate models.
Development of Earth system modelling in an era of rapid change
While much of the project is still in early development, the consortium is now entering its second year, and several prototype tools and hybrid model components are beginning to emerge. As AI4PEX progresses, more public tools, datasets, and scientific outputs are expected to be released, supporting a new generation of models for understanding the Earth system in an era of rapid change.
Lund University plays a key role in the land system part of the project. The LPJ-GUESS dynamic vegetation model, developed at the Department of Physical Geography and Ecosystem Science, is being further refined to better represent how forests grow, die, and respond to stress under a changing climate.
Are AI-enhanced models genuinely improving predictions?
In parallel, Lund researchers are helping to develop new evaluation datasets and benchmarking tools for assessing whether AI-enhanced models are genuinely improving predictions. These tools are critical for validating results across domains and ensuring reproducibility.
The work also connects closely with MERGE, Lund University’s strategic research environment for modelling the regional and global Earth system.