Abstract:

Based on spectral information,vegetation index and texture characteristics extracted from GF-1 remote sensing data in Guanshanhu,Guizhou Province,combined with the measured masson pine plot data,multiple stepwise regression and random forest algorithm were used to construct forest stock estimation models with different windows of remote sensing features.The results indicated that:the optimal windows of multiple stepwise regression and random forest estimation models were both 13×13.When feature variables were extracted as DI2,B3,EN2,SM2,CO3,and the accumulation estimation model was established with the optimal window,the fitting effect of random forest model was better than that of multiple stepwise regression model.According to the resolution of remote sensing image,selecting suitable window to extract characteristic variables can further improve the modeling accuracy of forest stock estimation.

Key words: GF-1, texture features, multiple stepwise regression, random forest algorithm, forest stock