Mapping tree species diversity in temperate montane forests using Sentinel-1 and Sentinel-2 imagery and topography data.

Detailed information on spatial patterns of tree species diversity (TSD) is essential for biodiversity assessment, forest disturbance monitoring, and the management and conservation of forest resources. TSD mapping approaches based on the Spectral Variability Hypothesis (SVH) could provide a reliable alternative to image classification methods. However, such methods have not been tested in large-scale TSD mapping using Sentinel-1 and Sentinel-2 images. In this study, we developed a new workflow for large-scale TSD mapping in an approximately 4000 km2 temperate montane forest using Sentinel-1 and Sentinel-2 imagery-based heterogeneity metrics and topographic data. Through a systematic comparison of model performance in 24 prediction scenarios with different combinations of input variables, and a correlation analysis between six image heterogeneity metrics and two in-situ TSD indicators (species richness S and Shannon-Wiener diversity H′), we assessed the effects of vegetation phenology, image heterogeneity metrics, and sensor type on the accuracy of TSD predictions. Our results show that (1) the combination of Sentinel-1 and Sentinel-2 imagery produced higher accuracy of TSD predictions compared to the Sentinel-2 data alone, and that the further inclusion of topographic data yielded the highest accuracy (S : R 2 = 0.562 , RMSE = 1.502; H′ : R 2 = 0.628, RMSE = 0.231); (2) both Multi-Temporal and Spectral-Temporal-Metric data capture phenology-related information of tree species and significantly improved the accuracy of TSD predictions; (3) texture metrics outperformed other image heterogeneity metrics (i.e., Coefficient of Variation, Rao's Q, Convex Hull Volume, Spectral Angle Mapper, and the Convex Hull Area), and the enhanced vegetation index (EVI) derived image heterogeneity metrics were most effective in predicting TSD; and (4) the spatial distribution of TSD showed a clear decrease trend along the altitudinal gradient (r = −0.61 for S and − 0.45 for H′) and varied significantly among forest types. Our results suggest a good potential of the SVH-based approaches combined with Sentinel-1 and Sentinel-2 imagery and topographic data for large-scale TSD mapping in temperate montane forests. The TSD maps generated in our study will be valuable for forest biodiversity assessments and for developing management and conservation measures. [Display omitted] • The Spectral Variation Hypothesis was tested in a temperate montane forest using Sentinel-2 images. • Tree species diversity mapping was affected by sensor type, vegetation phenology, and image heterogeneity metric. • Texture metrics outperformed other heterogeneity metrics in predicting tree species diversity. • The EVI-derived spectral heterogeneity metrics correlated most with tree species diversity. • Topographic variables improved the mapping accuracy of tree species diversity. [ABSTRACT FROM AUTHOR]