Abstract: Selecting ground observation data, sounding data, MICAPS data and ERA5 reanalysis data from 1991 to 2024 in northern Tibet. According to the strong cooling grade standards of Grade I, II, and III, the Mann Kendall test, Morlet wavelet analysis, correlation analysis, threshold analysis and other methods are used to analyze the spatiotemporal distribution characteristics of strong cooling in northern Tibet. A strong cooling prediction model and indicators are constructed and the predictive performance is tested. The results show that: (1) In the past 30 years, a total of 660 strong cooling events have occurred in the northern Tibetan Plateau, concentrated from October to May of the following year, with particularly high frequencies in January, February, and December. Spatially, these events are more common in the central region and less frequent in the eastern and western areas. (2) The annual average frequencies of Grade I, Grade II, and Grade III strong cooling events are 22, 10, and 4 times, respectively. All show a decreasing trend, with climate tendency rates of−1.2/10 a, −0.5/10 a, and −0.04/10 a, respectively. (3) Both Grade I and Grade III strong cooling events exhibit characteristics of climate abrupt changes. The former occurred in 2004, with a decrease in frequency after the abrupt change, while the latter occurred in 2002 and 2018, with a decrease in frequency after 2002 and an increase after 2018. All three types of strong cooling events exhibit a quasi-6-year oscillation cycle. (4) The circulation backgrounds for the development of strong cooling events in the northern Tibetan Plateau can be divided into three types: large trough eastward movement, short-wave trough fluctuation, and ridge-front type. (5) The strong cooling forecast index (Y) can be used to distinguish strong cooling weather and cooling intensity. 6 ≤ Y < 8 corresponds to Grade I strong cooling events, 8 ≤ Y < 10 corresponds to Grade II strong cooling events, and Y ≥ 10 corresponds to Grade III strong cooling events, with a forecasting accuracy of up to 88%.