Abstract: Based on 60 years of observation data from 16 urban meteorological stations in the Liaohe River Basin from 1960 to 2020, the Stefan model and statistical methods were used to analyze the response relationship between ice thickness and freezing days at different temperatures; The trend analysis, wavelet model, Mann Kendall test and other methods were used to reveal the trend, periodicity, and mutation characteristics of frozen elements; A BP neural network model was constructed to predict changes in ice thickness. The results showed that the average temperature during the 60 year frozen period showed a fluctuating upward trend, with an average temperature of -7.51 ℃ and a period of 23 years, with a sudden change point in 1978. The ice thickness shows a gradually decreasing trend, with an average ice thickness of 29.34 cm during the frozen period, with a variation period of around 27 years and a sudden change point in 1986. The overall freezing days showed a slow downward trend, with an average freezing day of 143 days and a sudden change point in 1991. The constructed BP neural network model can effectively predict changes in ice thickness, with R² values greater than 0.95 during both the testing and validation periods. The spatial prediction results show that the average maximum ice thickness during the freezing period in Panjin City is 26.36～49.95 cm, and scenario analysis shows that the ice thickness under five temperature conditions in the Liaohe River Basin is 28.34～53.38 cm.