Soil erosion is a major ecological problem in the red-soil hilly region of southern China. This study investigates Changting County, Fujian Province, utilizing multi-temporal remote sensing imagery from 2005 to 2025 and the RUSLE model to estimate soil erosion. A soil erosion intensity transition matrix was constructed, and global and local Moran's I indices were employed to reveal the spatiotemporal evolution patterns of erosion clustering. The results show that in 2005 the area of eroded decreased from 527.74 km² (17.00% of the county area) to 202.99 km² (6.54%) in 2025. The area affected by moderate or higher levels of erosion decreased from 230.08 km² to 107.17 km², while the area affected by severe or higher levels of erosion decreased from 79.86 km² to 26.65 km². The development of soil erosion showed a gradual downward trend. The spatial autocorrelation analysis also revealed a significant clustering of improved and deteriorated areas. The improved areas accounted for 88.47% of the entire county, while deterioration was mainly concentrated in construction sites, transportation corridors, and surrounding mining areas. These results provide scientific basis for the key protection planning and monitoring of the red soil hilly areas, combined with long-term remote sensing monitoring.