Assessing tree species diversity in forests using combined remote sensing data

Abstract

Tree species diversity (TSD) is a critical biological variable for ecosystem functioning and sustainable forest management. This study aimed to model TSD in forest ecosystems using spectral and texture metrics extracted from Sentinel-2, WorldView-3, and PlanetScope satellite imagery, and structural metrics derived from LiDAR data. Tree species data collected through field surveys were used to calculate Shannon, Simpson, and species richness biodiversity indices. These biodiversity indicators were then related to explanatory variables derived from remote sensing (RS) metrics. The Boruta algorithm was applied for variable selection and random forest regression analysis was used for modeling. The results show that combining LiDAR and satellite data is a promising approach for estimating TSD. In particular, the integration of structural heterogeneity metrics from LiDAR with spectral and textural measures from Sentinel-2 significantly improved the prediction performance of the Shannon and Simpson indices. For species richness models, high-resolution imagery played a more prominent role. These findings suggest that the integration of different RS sources can serve as a powerful tool for monitoring forest biodiversity and provide a scientific basis for precision forest management.