Rapid and accurate surveys of land are essential for protecting ecosystems. This study aimed to survey the spatial–temporal land degradation patterns and explore its driving mechanism based on a geographical detector in Hainan Island from 2000 to 2020. First, the areas of forest, cultivated land, and construction land were analysed, and their mutation year was determined through the Mann-Kendall abrupt test. Then, change vector analysis was used to create a double-evaluation system by combining the normalized difference vegetation index and net primary productivity. Finally, the geographical detector was used to identify the driving mechanism of different land-use types. The results show that 2013 was the mutation year of the three land-use types. From 2000 to 2013 (Stage 1), the land degradation areas of forest, cultivated land, construction land, grassland, and unutilised land reached 1977.13, 613.25, 165.81, 261.06, and 4.25 km2, respectively. From 2013 to 2020 (Stage 2), the degraded land area accounted for only 0.91 % of the island's total area. From the single factor perspective, the elevation (DEM), least distance to residential area (LDP), and variety of mean annual temperature (VTem) were found to be the dominant factors of land degradation of cultivated land and grassland, forest, construction land in Stage 1. On the other hand, in Stage 2, the main factor of land degradation was the variety of the density of GDP (VGDP) in cultivated land, grassland, and construction land, while it was the variety of mean annual precipitation (VPre) in the forest. Moreover, all pair-factors provided a higher determinant power for degraded land areas than a single factor. From the interaction effect perspective in Stage 1, VPOP ∩ LDR (i.e. the variety of the density of population density and least distance to roads or railways), LDP ∩ LDR, and VTem ∩ LDR were the most significant pair-factors of the interaction effect on land degradation of cultivated land, forest and grassland, and construction land. In Stage 2, the most significant pair-factors of the interaction effect in different land-use types were DEM ∩ VPOP and VPre ∩ POP in degraded cultivated land, VPre ∩ DEM and VPre ∩ LDR in degraded forest land, DEM ∩ VTem in degraded grassland, and VGDP ∩ LDR in degraded construction land. Given the time-lag effects of factors and the attribution analysis of land degradation, we suggest limiting the construction of roads and residential areas, increasing biodiversity, and sheltering forests to protect terrestrial ecosystems in the future.