Description: The collapse of turbulence in a pressure driven, cooled channel flow is studied by using 3-D direct numerical simulations (DNS) in combination with theoretical analysis using a local similarity model. Previous studies with DNS reported a definite collapse of turbulence in case when the normalized surface cooling h/L (with h the channel depth and L the Obukhov length) exceeded a value of 0.5. A recent study by the present authors succeeded to explain this collapse from the so-called Maximum Sustainable Heat Flux (MSHF) theory. This states that collapse may occur when the ambient momentum of the flow is too weak to transport enough heat downward to compensate for the surface cooling. The MSHF theory predicts that in pressure driven flows, acceleration of the fluid after collapse eventually will cause a regeneration of turbulence, thus in contrast with the aforementioned DNS results. Also it predicts that the flow should be able to survive ’supercritical’ cooling rates, in case when sufficient momentum is applied on the initial state. Here, both predictions are confirmed using DNS simulations. It is shown that also in DNS a recovery of turbulence will occur naturally, provided that perturbations of finite amplitude are imposed to the laminarized state and provided that sufficient time for flow acceleration is allowed. As such, it is concluded that the collapse of turbulence in this configuration is a temporary, transient phenomenon for which a universal cooling rate does not exist. Finally, in the present work a one-to-one comparison between a parameterized, local similarity model and the turbulence resolving model (DNS), is made. Although, local similarity originates from observations that represent much larger Reynolds numbers than those covered by our DNS simulations, both methods appear to predict very similar mean velocity (and temperature) profiles. This suggests that in-depth analysis with DNS can be an attractive complementary tool to study atmospheric physics in addition to tools which are able to represent high Reynolds number flows like Large Eddy Simulation.