Evaluation of red-edge features for identifying subtropical tree species based on Sentinel-2 and Gaofen-6 time series

Fast and accurate tree species mapping using remote sensing technology is invaluable to forest sustainable management. However, due to the limitation of ecosystem complexity and the frequent cloudy weather, it remains challenging to obtain detailed knowledge about the varieties in subtropical areas. In this study, 18 scenes of Gaofen-6 (GF6) and 13 scenes of Sentinel-2 (S2) images were harmonized using linear regression models, and the identification abilities of 23 red-edge vegetation index (REVI) time-series and the normalized difference vegetation index (NDVI) time-series were evaluated using the random forest classifier (RF). The feature reduction methods were employed to boost the effectiveness of the input features, and the optimal red-edge combination consisting of spectral and textural features was built to map the dominant tree species precisely in Southeast China. Our results showed as follows: (1) GF6 and S2 could be harmonized with a mean coefficient of determination (R²) greater than 0.94 and the root means squared error (RMSE) less than 0.02, which was helpful to improve the consistency between multi-source remote sensing data and subsequently derived products. (2) Most of the REVI time-series performed better than NDVI time-series, and the transformed chlorophyll absorption ratio indexes (TCARI₂) got the highest overall accuracy of 86.28%, which could be further improved to 88.37% by adding 26 textural features. (3) The phenological characteristics from different seasons were conducive to tree species classification compared to the effect of different sensors, and the contribution of spring images played the most important role in RF, followed by autumn and winter, while the mid-summer images seemed to be negligible in the classifier. Overall, this study proposed an accurate and cost-efficient approach for tree species identification, which demonstrated good potential in providing a profound reference for subtropical forest landscape studying.