Research On Modeling Of Forest Stock Volume And Model's Universal Applicability

Forest volume is one of the important indicators of forest resource monitoring,and its quantitative estimation is of great significance.Therefore,in this paper we combined the Sentinel-1 radar and Sentinel-2 optical remote sensing methods with Digital Elevation Model(DEM)and forest inventory data to efficiently predict the forest unit volume using small class as the research unit,in order to explore new methods for the estimation of forest unit volume at a lower and more efficient cost.Doing so,the main conclusions are as follows:(1)The Lasso feature selection method effectively reduces the number of features,it also speed up model training and improve model generalization ability,and makes the model better interpretive.(2)The Cat Boost model has the best overall performance among the four models,it performs well in the prediction of stock volume in Chun'an and Linhai,which proves that the model has good applicability.(3)The addition of category feature variables greatly improved the model's indicators significantly in single remote sensing source and multi-remote sensing source experiments.(4)Compared with Sentinel-2 remote sensing only,the combination of Sentinel-2 remote sensing and Sentinel-1 remote sensing improves the estimation accuracy of stock volume by 3.07%(24.70%-21.63%)for Linhai city and by 4.10%(36.70%-32.60%)for Chunan city based on no category features,when adding the category features,the estimation accuracy improving by 4.83%(21.03%-16.20%)for Linhai city and by 2.20%(22.57%-20.37%)for Chunan city.