Stage-dependent topographic controls on vegetation recovery and landscape-pattern evolution under contrasting restoration treatments in an abandoned mine from UAV multispectral time series

Assessing ecological restoration in mining areas requires tracking both temporal vegetation dynamics and spatial pattern evolution, yet most studies rely on single indices and rarely quantify topographic controls across restoration approaches. Using UAV multispectral imagery, we compiled an 11-epoch series (Q1, November 2024–Q11, November 2025) and derived nine vegetation indices (NDVI, GNDVI, SAVI, OSAVI, EVI2, NDRE, CL_RE, NIRv, and TDVI). Landscape metrics (PLAND, PD, LPI, and COHESION), together with global and local Moran's I (LISA), characterized clustering, and Mantel tests quantified stage-dependent constraints of slope and aspect. Vegetation improved overall but fluctuated: most indices peaked at Q9, declined at Q10, and remained relatively high at Q11; regraded backfill with topsoil and “obstacle removal + topsoil” units showed stronger and more stable recovery, whereas the flexible planting-bag area exhibited high variability and a late-stage decline, with NDVI and CL_RE most sensitive to these differences. Most indices were highly positively correlated across Q1–Q11, indicating coherent greenness–structure responses, but correlations between late-season chlorophyll signals and structure-related indices weakened, suggesting stage-wise decoupling. Spatially, patterns shifted from expansion and patch proliferation to core-patch nucleation and enhanced connectivity, and NDVI showed significant positive spatial autocorrelation throughout (Global Moran's I = 0.8865–0.9507, all p < 0.001). Topographic controls varied by stage, with stable early slope effects, strengthened slope–aspect effects during vigorous growth, weakened effects under high coverage, and rebounding late-season aspect effects. • UAV Q1–Q11 series links restoration engineering to vegetation patterns. • Nine indices increased overall: Q9 peak, Q10 dip, and high values at Q11. • Landform/topsoil treatments diverged; NDVI and CL_RE best discriminate. • Landscape metrics + Moran/LISA show clustering and stronger connectivity. • Mantel tests: slope dominates early; aspect strengthens late season.