Under-ice ambient noise in the Arctic Ocean is studied using the data recorded by autonomous hydrophones at the long-term ice station during the 9th Chinese National Arctic Research Expedition. Time-frequency analysis of two 7-s-long ice-induced noise samples shows that both ice collision and ice breaking noise have many outliers in the time-domain (impulsive characteristic) and abundant frequency components in the frequency-domain. Ice collision noise lasts for several seconds while the duration of ice breaking noise is much shorter (i.e., less than tens of milliseconds). Gaussian distribution and symmetric alpha stable (sαs) distribution are used in this paper to fit the impulsive under-ice noise. The sαs distribution can achieve better performance as it can track the heavy tails of impulsive noise while Gaussian distribution fails. This paper also analyzes the meteorological variables during the under-ice noise observation experiment and deduces that the impulsive ambient noise was caused by the combined force of high wind speed and increasing atmosphere temperature on the ice canopy. The Pearson correlation coefficients between long-term power spectral density variations of under-ice ambient noise and meteorological variables are also studied in this paper.