Abstract: [Objective] In this paper,we explored the possibility of defining relationships between remote sensing features that come from GF-2 and species diversity indices for subtropical forest in Gutian Mountain of Zhejiang Province,China.[Method] We extracted the reflectance values,spectral indices,texture from GF-2 data.Random forest models were used to select variables and estimate species diversity indices.We compared the texture values that come from different window size to find the best window size for species diversity estimation.[Result] Based on the random forest (RF),recursive feature elimination (RFE) was used to find small subsets of features with high discrimination levels on data sets,which provide good performance for species diversity modeling.For the multispectral (MSS) data,the best window size is 3×3,and for the panchromatic (Pan) data,the best window size is 7×7.Both texture features and spectral indices were selected for species diversity modeling and the Carotenoid reflectance index provided the best performance.The determinate coefficient and RMSE for three species diversity are 0.47 and 0.300(Shannon-Wiener diversity index),0.53 and 0.042(Simpson diversity index),0.61 and 0.051(Pielou evenness index),respectively.[Conclusion] The result demonstrated that the GF-2 data can be used to model tree species diversity effectively.Predictive map derived from the presented method ology can help evaluate spatial aspects and monitor tree species diversity of the studied forest and facilitate the evaluation of forest management and conservation strategies.

Key words: tree species diversity, GF-2 remote sensing data, texture features, spectral indices