Abstract:

Aiming at the problem that the traditional second-class survey method of forest resources is time-consuming and laborious,it is difficult to meet the needs of forest resource dynamic monitoring under the new situation.Three forest farms in Liuhe District,Nanjing city were selected as the study area in this paper.Sparse airborne laser radar was used to extract feature parameters,combined with the data of Forest Management Inventory.Two algorithms,stepwise regression and Boruta,were used for factor screening.Three modeling methods including Stepwise Multiple Linear Regression (SMLR),Support Vector Machine (SVM) and Random Forest (RF) were compared to estimate the forest stock.The results showed that:1) Height factor was the main characteristic parameter affecting forest stock volume;2) SVM and RF algorithms performed better in model fitting and verification accuracy,SVM algorithms performed slightly inferior to RF algorithms in mixed forests,and stepwise regression methods performed poorly.In general,the modeling results of lidar extraction factor and forest stock were good,and Sparse airborne laser radar had good applicability to forest resource survey,which provided a new idea for forest resource survey in the future.

Key words: LiDAR, sparse data, forest stock estimates, forest resource survey