Spectral diversity mapping is increasingly used to estimate biodiversity from satellite imagery, but the effect of input feature type remains poorly tested. We evaluated how different Sentinel-2 input feature types - surface reflectance, Principal Component Analysis (PCA) transformed surface reflectance, and spectral indices - affect biodivMapR-based tree diversity mapping across three hemi-boreal forest landscapes in Estonia. Outputs were validated against field-measured tree species Shannon diversity using Spearman correlation and bootstrap resampling. PCA transformed surface reflectance produced the strongest and most consistent correlations across sites, spectral indices performed slightly better than surface reflectance in two sites, and interquartile range-based filtering had little effect. The results identify input feature type selection as a key parameter in Sentinel-2 spectral diversity workflows.