We use spatio-temporal cross-correlations of slopes from five Shack-Hartmann wavefront sensors to analyse the temporal evolution of the atmospheric turbulence layers at different altitudes. The focus is on the verification of the frozen flow assumption. The data is coming from the Gemini South Multi-Conjugate Adaptive Optics System (GeMS). First, the Cn2 and wind profiling technique is presented. This method provides useful information for the AO system operation such as the number of existing turbulence layers, their associated velocities, altitudes and strengths and also a mechanism to estimate the dome seeing contribution to the total turbulence. Next, by identifying the turbulence layers we show that it is possible to estimate the rate of decay in time of the correlation among turbulence measurements. We reduce on-sky data obtained during 2011, 2012 and 2013 campaigns and the first results suggest that the rate of temporal de-correlation can be expressed in terms of a single parameter that is independent of the layer altitude and turbulence strength. Finally, we show that the decay rate of the frozen-flow contribution increases linearly with the layer speed. The observed evolution of the decay rate confirms the potential interest of the predictive control for wide-field AO systems.