MIT technique improves climate AI predictions with less training data

MIT researchers developed a training method that helped AI climate models predict unfamiliar emissions scenarios using less data, MIT News reported. In their tests, a model trained with optimized synthetic scenarios captured the effects of reducing one local pollutant while other emissions rose; traditional models they evaluated failed that test. The algorithm repeatedly adjusts training scenarios to reduce prediction errors, producing rapid emissions variations that help the model learn the climate system's behavior. The approach could reduce the full climate simulations needed to generate training data from six or eight runs to one. Testing with a scenario generated by a full earth system model remains future work.