Starling murmurations under predation

European starlings perform a great diversity of patterns of collective behaviour when hunted by aerial predators; their large flocks are changing shape, size, and internal structure continuously and rapidly, but how these patterns emerge by self-organization is still unknown. Here, we disentangle the emergence of several interconnected patterns of collective escape in starlings. We combine video footage of starling flocks pursued by a robotic predator, the RobotFalcon, with simulations of large flocks in a new data-driven 3-dimentional agent-based model. Our empirical data show that flock members often differ in their evasive manoeuvres and that several patterns of collective escape arise simultaneously at different parts in the flock. In our computational model, we identify what rules of motion, coordination and escape at the individual level lead to the emergence of group density, internal dynamics, and patterns of collective escape similar to real starling flocks. Overall, our results suggest that the emergence and dynamics of simultaneous patterns of collective escape depend on: the speed with which the escape information propagates from one or few initiators, the positions of the escaping flock members in relation to the predator, and the previous state of the flock (hysteresis).