Under the background of global warming, forest fires occur frequently. It is conducive to forest fire prevention and control. The frequency of forest fires, burned area and extreme weather events from 1980 to 2020 in Yunnan Province were selected to analyze the occurrence law of forest fires, the differences between extreme weather years and between groups of forest fires and ordinary years. The forest fire situation was classified by cluster analysis, and the transfer law of each grade and predict the probability of different grades in 2023- 2027 was used by Markov chain. The results show that the correlation coefficient between fire number and fire area is 0.931. The breakpoints of the time series of fire number are 1989 and 2012. There was a significant difference in fire number between El Nino and La Nina years from 1980 to 1989 (Sig. <0.05), but there was no significant difference in fires area (Sig. >0.05). There was a significant difference in fire number and fires area between 1990 and 2012 (Sig. <0.05). There was no significant difference in other cases (Sig. >0.05). The forest fire situation was divided into five levels as level I (mild), level II (general), level III (moderate), level IV (height) and level V (extreme height) by K-cluster. The probability of level I~II mutual transfer accounts about 78% of all transfer events, and I~III mutual transfer is higher than 90%. From 2023 to 2027, the probability of forest fire situation grade I~V in Yunnan Province is between 0.683~0.702, 0.251~0.253, 0.036~0.048, 0.003~0.006 and 0.007~0.012 respectively. The forest fire situation in Yunnan Province will be generally stable in the next five years. The probability of grade I~II is higher than 93%, and grade IV~V is lower than 2% .This conclusion can provide theoretical support for forest fire management and prevention under climate change.

forest fire, extreme climate, evolution of forest fire, difference between groups, fire situation, transfer law