Inversion of Picea schrenkiana var. tianshanica Growing Stock Based on WorldView-2 Image and Random Forest Algorithm

【Objective】 Taking the Picea schrenkiana var. tianshanica of Qiaxi National Forest Park in Gongliu County, western Tianshan, Xinjiang as the research object, World View-2 images and the sample plot scale per tree as the data source, the volume of Picea schrenkiana var. tianshanica was retrieved by looking for the relationship between remote sensing factors and volume. The purpose of this project is to provide a reference basis for Picea schrenkiana var. tianshanica ecological restoration and scientific management after the implementation of the natural forest protection project. 【Method】 The spectral information, texture factor and vegetation index of the sample plot were extracted by eCognition Developer, and a model was established to retrieve the volume of Picea schrenkiana var. tianshanica forest by random forest algorithm. 【Result】 The random forest algorithm was used to screen 24 kinds of remote sensing factors, and the five characteristic variables which had the greatest influence on the stock were selected. The five characteristic variables with the largest impact on the accumulation were selected, respectively, NDVI1, NDVI2, RVI2, homogeneity（Homogeneity） and correlation（Correlation）, thus establishing a random forest regression model. Its interpretation was up to 81.27%, the determination coefficient R² was = 0.8648（P <0.05） and the accuracy of estimating sample plot volume was 86.38%. 【Conclusion】 The stochastic forest regression model established by random forest algorithm and WorldView-2 image can effectively retrieve the volume of Picea schrenkiana var. tianshanica.