[Objective] This study aims to elucidate the change trajectories of soil erosion intensity in Yunnan Province from 1990 to 2022, analyze the types of changes and their driving factors, and provide a scientific foundation for effective soil erosion control strategies. [Methods] Quantitatively assess soil erosion intensity conditions in Yunnan Province based on the RUSLE model and capture dynamic characteristics by introducing interannual change rates. Use an improved Stability Mapping Method (STD), combining change rates and frequencies to identify soil erosion intensity change trajectory types. Analyze the contribution of driving factors using the Random Forest model and compare the characteristics of driving factors across different trajectory types. [Results] The interannual change trend of soil erosion in Yunnan Province was mainly stable, with significant changes observed in areas of substantial decrease and increase. Soil erosion intensity change trajectories exhibited significant spatial differentiation, with cyclical trajectories being the most prevalent (53.90%), followed by non-continuous stepwise (14.78%) and fluctuating types (14.08%). Precipitation, slope, population density, GDP, and vegetation cover were the main driving factors affecting soil erosion intensity trajectory changes, with contributions of 17.92%, 14.56%, 12.52%, 12.67%, and 9.41%, respectively. There were differences in driving factors across different trajectory types. Areas with cyclical and non-continuous stepwise trajectories had higher precipitation and slopes, while stepwise trajectory areas had higher farmland coverage and lower forest coverage. [Conclusion] The characteristics of soil erosion intensity trajectory changes in Yunnan Province are significant, with spatial heterogeneity in driving mechanisms. Therefore, soil erosion control strategies should reflect regional differentiation and specificity, and adopt location-specific measures based on regional characteristics. The STD trajectory partitioning method based on change rates effectively captures the dynamic changes in soil erosion, providing new insights for monitoring, early warning, and partitioned control of soil erosion.