Accurate estimation of plant alpha diversity is crucial for understanding ecosystem dynamics and advancing biodiversity conservation. However, quantifying alpha diversity in grasslands remains challenging due to the small size of plant individuals and complex background interference. This study explores the use of UAV-acquired hyperspectral remote sensing data to estimate plant diversity in the Hunshandak Sandland, Inner Mongolia-a temperate continental monsoon grassland characterised by arid climates and desertified landscapes. Field surveys were conducted across areas with contrasting vegetation cover, collecting data on species richness, height and coverage. Four alpha diversity indices (species richness, Shannon-Wiener index, Simpson index and Pielou's evenness index) were calculated from the field data. A total of 1274 spectral vegetation indices, derived from spectral variation, principal components and texture features, were analysed. Spearman correlation analysis and random forest models were used to evaluate relationships between plant diversity indices and spectral metrics across gradients of vegetation coverage (indicated by NDVI) and species richness. The results showed that spectral indices derived from characteristic bands reflecting leaf pigment and photochemical traits-such as NDMI, NPCI and SRPI-and indices from the R package rasterdiv (e.g., NDMI-REN1, ARI-REN0, MVI-REN1) achieved high accuracy in estimating species richness. For the Shannon-Wiener index, Simpson index and Pielou's evenness index, the most effective indices were NDVI-REN1, ARImn and REP-CRE, respectively. Overall, spectral indices performed better in estimating the Shannon-Wiener (H), Simpson (D) and Pielou (P) indices than species richness. Estimation accuracy improved with increasing gradients of community coverage and complexity in grasslands. This study demonstrates the potential of UAV-based hyperspectral data for grassland diversity monitoring and highlights the importance of selecting context-appropriate vegetation indices to address challenges posed by vegetation cover and community complexity. These findings provide a foundation for developing efficient ecological monitoring models in sandy grasslands and analogous ecosystems.