Multi-scale estimation of species diversity in an urban forest wetland with fractional-order derivative of UAV hyperspectral imagery

Semi-mangrove communities in urban coastal wetlands provide critical ecological functions but face increasing pressures that threaten biodiversity. The Heritiera littoralis community in Baguang, Shenzhen, China represents one of the most intact semi-mangrove communities embedded within a rapidly urbanizing coastal landscape, offering an ideal setting for advancing assessment of community-level species diversity using remote sensing techniques. However, non-destructive estimation of species diversity remains challenging due to the scale dependence of spectral diversity across spatial resolutions. In this study, we developed a multi-scale framework that integrates UAV hyperspectral imagery and field survey data with fractional-order derivatives (0-2) and ensemble machine learning to estimate four species diversity indices (Simpson’s dominance ( D ), Shannon-Wiener diversity ( H ′), Margalef’s richness ( F ), and Pielou’s evenness ( E )) in H. littoralis wetlands at 0.2, 0.5, 1.0, and 2.0 m resolutions. Results showed that fractional-order derivatives significantly enhanced the correlations between spectral features and diversity indices, with optimal derivative orders varying among indices. Red-edge and near-infrared coefficient of variation features (1.4 order) were most informative for D and H′ , while blue-band heterogeneity (1.6 order) and near-infrared differential bands (1.8 order) dominated F and E , respectively. Model accuracy improved with coarser spatial resolution, peaking at 2.0 m (R 2 > 0.6), revealing pronounced scale effects, with 1.0 m providing the best trade-off between global spatial autocorrelation and local heterogeneity. Moreover, AdaBoost achieved the most stable and accurate performance across indices. These findings deepen the understanding of spectral-scale mechanisms underlying species diversity in semi-mangrove communities and provide an efficient approach for biodiversity conservation, monitoring, and management of H. littoralis communities and other urban wetland ecosystems.