The occurrence and predictability of extreme events is often modulated by regional dynamical drivers, but their identification is not straightforward. This paper by Fiona Spuler et al. presents an innovate approach to dimensionality reduction targeted on regionally occurring phenomena. Their approach is based on a machine-learning method demonstrating, in a broad sense, that these method can identify physically interpretable drivers of targeted phenomena. The authors discuss the tradeoff between regime informativeness of local precipitation extremes and predictability of the regimes at subseasonal lead times. The novel machine-learning-based approach may find useful application beyond atmospheric and climate science.