Stock Volume Estimation Research Based on Random Forest Algorithm in Liangshui Nature Reserve

This research taking Liangshui Nature Reserve in Heilongjiang as study area, using GF1 satellite image as data source, textural features under different window sizes and spectral feature were extracted from GF1 image. Forest stock volume inversion model were constructed using random forest algorithm and Liangshui forest volume data obtained by field investigation. Experimental results showed the coefficient of determination R2 was 059 for the spectrally based volume estimation model, and the coefficient of determination R2 was 065 for the spectral and textural feature combining model. When window size is set to 3 × 3, the forest stock volume inversion model achieves the best result. The result indicate that forest stock volume estimation based on spectral and textural feature from satellite image using random forests has potential application in forest inventory.