Predicting Tree Species Diversity Using Geodiversity and Sentinel-2 Multi-Seasonal Spectral Information

Measuring and monitoring tree diversity is a prerequisite for altering biodiversity loss and the sustainable management of forest ecosystems. High temporal satellite remote sensing, recording difference in species phenology, can facilitate the extraction of timely, standardized and reliable information on tree diversity, complementing or replacing traditional field measurements. In this study, we used multispectral and multi-seasonal remotely sensed data from the Sentinel-2 satellite sensor along with geodiversity data for estimating local tree diversity in a Mediterranean forest area. One hundred plots were selected for in situ inventory of tree species and measurement of tree diversity using the Simpson’s (D1) and Shannon (H′) diversity indices. Four Sentinel-2 scenes and geodiversity variables, including elevation, aspect, moisture, and basement rock type, were exploited through a random forest regression algorithm for predicting the two diversity indices. The multi-seasonal models presented the highest accuracy for both indices with an R2 up to 0.37. In regard to the single season, spectral-only models, mid-summer and mid-autumn model also demonstrated satisfactory accuracy (max R2 = 0.28). On the other hand, the accuracy of the spectral-only early-spring and early-autumn models was significant lower (max R2 = 0.16), although it was improved with the use of geodiversity information (max R2 = 0.25).