Critical transitions in experimental and theoretical systems can be anticipated on the basis of specific warning signs, raising the prospect that it might also be possible to predict future real-world events on the scale of the 2007 global financial crisis and Arab spring. But what to measure? Recent work has focused on critical slowing down, in which a system’s recovery from perturbation is reduced as the transition is approached. Another possibility is flickering, in which increasing shifts between alternative stable states are seen in the run-up to the transition. This study uses long-term data from a real system, a Chinese lake, to show that flickering can be observed and that it occurs up to 20 years before a critical transition — in this case the deterioration of a lake towards a dead ‘eutrophic’ state as algal growth consumes the last available oxygen. (from the editor’s summary in Nature)