A hybrid model creates climate ensembles at full model resolution without GPUs or retraining.

Studying how climate change will affect our world requires running huge batches of simulations across many weather variables. Full-scale Earth system models show these details well, but running enough of them takes massive computing power. To fix this, researchers created GCMagicc. It is a fast tool that combines a simple physics climate model with machine learning. Instead of burning up heavy computing power, it runs without graphics processing units and needs no retraining when given new future emissions scenarios. It generates detailed patterns for 10 climate variables at full model resolution. When tested on a severe 2025 drought in Iran, GCMagicc showed that human-driven climate change played a massive role. The simulated odds of such a drought were 29% under human emissions, but zero under natural conditions alone. In the future, these drought conditions are projected to worsen materially, raising risks for farming, food security, and water conflicts.

Why it matters

Climate impact studies currently require enormous computing power to run the many scenarios needed for risk assessment. GCMagicc makes this fast and accessible, allowing researchers to explore a wide range of future pathways and attribute extreme events like droughts to climate change. This helps policymakers prepare for food security risks and water conflicts.

Questions answered