Abstract:

Forest reserve is an important index to show the quantity of forest resources.In this study,the boruta feature selection method and the extremely randomized trees(Extra-trees) method are used to estimate the forest resources per mu in some areas of Longquan city.The study takes small classes as the research unit to provide new methods for the estimation of forest reserves at the county level.Based on the secondary survey data of forest resources,GF-2 remote sensing image data and digital elevation model data,multiple features are collected to form the original feature set.Through the boruta selection method,the original feature set is screened,the forest volume estimation model is established by extra-trees method,and the ten fold cross validation method is used to test the model,which is compared with the random forest(RF) method and gradient boosting method.The results show that:(1) the features found by the boruta feature selection method are soil thickness,age,canopy density,altitude and slope;(2) the optimal parameter combination obtained by the grid search and parameter adjustment of the Extra-trees method is:the number of tree is 250,and the maximum depth of the tree is 14;(3) the testing accuracy of the forest volume estimation model based on the boruta and extra-trees method is 84.14%,R ² is 0.92,RMSE is 19.65m³/hm²,and MAE is 13.95m³/hm².The model is superior to the random forest method and the gradient lifting method.It shows that the boruta feature selection method combined with the Extra-trees method can achieve better results in estimating the forest reserve.

Key words: boruta feature selection, extremely randomized trees, random forest, forest reserves, machine learning