Estimating Plant Species Richness With Sentinel and Landsat Data Across Ecosystems in China.

The existing remote-sensing models for estimating plant alpha diversity typically exhibit relatively low accuracy. This study endeavors to develop a high-precision remote-sensing model for plant species richness across multiple scales and diverse ecosystems in China. Sentinel-2 and Landsat data were collected. A range of spectral vegetation indices, band indices, and their statistical metrics (mean, variance, coefficient of variation) were selected as predictor variables, with plant species richness serving as the response variable. Three modeling approaches, namely simple linear regression, multiple stepwise regression, and partial least squares regression (PLSR), were employed to evaluate the predictive capabilities of these variables across five vegetation types. The findings revealed that the PLSR model demonstrated the optimal performance, followed by multiple stepwise regression and simple linear regression. Among the spectral indices, EVI, DVI, PSRI, NDVI, PRI, GNDVI, and GMEVI were identified as powerful indicators for predicting plant species richness. Additionally, Sentinel-2 data showed higher prediction accuracy compared to Landsat data, highlighting the crucial role of spectral and spatial resolution in the estimation of plant diversity.