Abstract: Protective development of uninhabited islands is an important part of the sustainable development of marine economy, however, little is known about the impact mechanism of this process on the island ecosystem. In this study, we used high-resolution remote sensing data of Sanjiao Island and constructed the ecological vulnerability assessment model with the spatial autocorrelation analysis (Moran’s I) and GeoDetector to explore the temporal and spatial evolution of ecological vulnerability during the protective development. The results showed that the ecosystem of Sanjiao Island was extremely fragile (EVI₂₀₁₁ = 0.912, WQI₂₀₁₁ = 0.034) in 2011, and the areas with high ecological vulnerability were highly concentrated (Moran’s I₂₀₁₁ = 0.649), indicating that the ecosystem of the Sanjiao Island had suffered fundamental damage due to the previous stone mining. Although the protective development since 2017 slightly increased the ecological vulnerability (EVI₂₀₁₆ = 0.782 v.s. EVI₂₀₁₈ = 0.825), the simultaneous ecological restoration developed the ecosystem landscape diversity (Moran’s I₂₀₂₂ = 0.595), and significantly reduced the ecological vulnerability (EVI₂₀₂₂ = 0.694, WQI₂₀₂₂ = 0.223). Vegetation coverage, topography and landscape pattern together droved the dynamic change of ecological vulnerability, showing a spatio-temporal evolution mechanism of “terrain optimization driving-vegetation restoration leading-landscape pattern formation”. The experience of protective development from Sanjiao Island can provide scientific guidance for the sustainable utilization of damaged uninhabited islands.