The prediction of grassland plant diversity using satellite imagery has been the subject of intensive research. However, the accuracy of functional diversity (FD) predictions remains unclear. To address this, high-spatial-resolution WorldView-3 (WV-3) multispectral data were used to predict species diversity and FD at the pixel scale (1.2 x 1.2 m) in the central Hunshandak Sandland, Inner Mongolia, northern China. Data collected from 120 field plots (6 x 6 m) were employed to train and validate several statistical learning methods, with the primary objective of establishing links between 156 satellite-derived spectral and texture indices and 6 plant diversity indices. Among the various diversity indices tested, functional trait diversity-specifically Functional Attribute Diversity (FAD1) and Modified Functional Attribute Diversity (MFAD)-were predicted most effectively (with coefficients of determination of approximately 0.29 and 0.14, respectively; n=48) using texture indices. In contrast, species diversity (richness, H, E, or D) and other FD metrics were not well predicted by WV-3 data. Overall, WV data did not significantly improve the accuracy of plant diversity predictions in sandy grasslands. Additionally, high plot-level vegetation coverage was found to enhance the performance of spectral indices in predicting H, E, D, and FD. These results underscore the importance of accounting for variability across field conditions and demonstrate the potential of high-spatial-and-spectral-resolution satellite imagery for monitoring plant functional diversity in sandy grasslands.