Rockfalls are difficult to predict events, owing to the suddenness of the phenomenon and to the little information available about the mass internal structure. Back analysis of collapses in steep limestone showed that rockfalls were initiated by intact rock bridge failures and that the persistence of joints was the main factor controlling the failure probability. First, we used the seismic noise recorded by autonomous monitoring systems to evaluate the degree of coupling of a prone-to-fall column to the rock massif. Seismic data were acquired on top of a limestone cliff in Chamousset (Vercors) during 4 months until 2 week prior the collapse of a 21,000 m3 column. The column 1st natural frequency was easily identified from ambient noise processing and exhibited a significant drop prior the column failure. Thanks to the monitoring system, thousands of seismic events were detected, out of which, the most energetic ones were classified. Some of these events exhibit clear P and S waves, making possible an analysis of the rupture mechanism as well as the calculation of their magnitude and hypocentral distance. Numerical simulations of rock bridge rupture allowed favouring traction rupture to explain the observed polarizations. Other numerical simulations showed that rupture events can trigger the column resonance, explaining the recurrent presence of specific frequencies in the late coda as well as the higher seismic motions on the column. Finally, we studied the thermomechanical response of another column. The influence of meteorological parameters on the column natural frequencies was shown. Both reversible and irreversible effects on the 1st natural frequency were identified.