Abstract: In order to realize adaptive modulation communication in Arctic ice-water mixed waters, this paper studies underwater acoustic (UWA) channel prediction algorithms. Using the sparseness of UWA channels, channel prediction is performed in the time domain which reduces the computational complexity and improves the prediction accuracy by removing most noise effects. This paper compares the performance of channel prediction methods based on different prediction coefficient acquisition algorithms using the data from the 9th Arctic scientific expedition of China. Processing results show that the prediction steady-state error of the proposed algorithms can reach -30dB. It is proved that the effective prediction of the UWA channel can be realized, which can provide reliable channel state information in ice-water mixed communication environment.