Seeing the forest for the scenarios: Using decision trees to draw conclusions from large energy scenario sets.

Energy transition studies often present hundreds of scenarios, yielding large datasets that are challenging to interpret. Alternative approaches define a few qualitative storylines, which are more accessible but disconnected from model-based evidence. Here, we show that decision trees, a popular machine learning technique, bridge this disconnect by distilling numerous scenarios into a small set of interpretable storylines centered around policy-relevant levers. We demonstrate this approach across three global and European case studies: First, we translate 1000 integrated assessment model scenarios into three storylines distinguished by their degrees of heating and industry electrification. Second, we show that 2°C warming pathways with continued fossil fuel usage only work under unusually mild climate responses. Third, we use 441 European pathways to show that high bioenergy or electrification are required for cost-effective climate neutrality. Our transferable approach helps decision-makers interpret large scenario sets by revealing how policy-relevant outputs are associated with different transition pathways.