Most weather-driven impacts such as crop failure, forest mortality and devastating wildfires are associated with compound events. To improve climate risk assessment and management, knowing which combinations of weather and climate drivers lead to impacts is crucial. However, it is often very challenging to identify such drivers from data. We use tools from statistics and modern machine learning to identify multiple compounding drivers that lead to large impacts. To test new methods, we use simulations from process-based impact models such as hydrological models, forest models, crop models and fire models, and compare our findings with expert knowledge encoded in those models. We further conduct factorial model simulations with such impact models to isolate drivers.