Abstract. The velocity of unstable large ice masses from hanging glaciers increases as a power-law function of time prior to failure. This characteristic acceleration presents a finite-time singularity at the theoretical time of failure and can be used to forecast the time of glacier collapse. However, the non-linearity of the power-law function makes the prediction difficult. The effects of the non-linearity on the predictability of a failure are analyzed using a non-linear regression method. Predictability strongly depends on the time window when the measurements are performed.Log-periodic oscillations have been observed to be superimposed on the motion of large unstable ice masses. The value of their amplitude, frequency and phase are observed to be spatially homogeneous over the whole unstable ice mass. Inclusion of a respective term in the function describing the acceleration of unstable ice masses greatly increases the accuracy of the prediction.