El Niño-Southern Oscillation (ENSO) shows a large diversity of events, whose modulation by climate variability and change, and their representation in climate models, limit our ability to predict their impact on ecosystems and human livelihood. Here, we introduce a new framework to analyze probabilistic changes in event-location and -intensity, which overcomes existing limitations in studying ENSO diversity. We find robust decadal variations in event intensities and locations in century-long observational datasets, which are associated with perturbations in equatorial wind-stress and thermocline depth, as well as extra-tropical anomalies in the North and South Pacific. A large fraction of CMIP5 and CMIP6 models appear capable of simulating such decadal variability in ENSO diversity, and the associated large-scale patterns. Projections of ENSO diversity in future climate change scenarios strongly depend on the magnitude of decadal variations, and the ability of climate models to reproduce them realistically over the 21st century.