Inversion method for the crown density of Mongolian scotch pine from point cloud data of small-footprint LiDAR

In order to improve the accuracy of measuring coniferous forest crown density by small-footprint LiDAR, linear regression analysis was used to establish multi-variable inversion models. Three number ratio variables and three energy ratio variables were extracted by processing the point cloud data of small- footprint LiDAR, and then a series of single-variable inversion models of crown density were set up. Afterwards the multi-variable inversion models were built with multiple linear regression analysis on the basis of single variables. Finally the remaining data were used to evaluate the accuracy of inversion models. The results revealed that I2 inversion model was the best one among all single-variable crown density inversion models with fitting correlations R2 = 0.818, AdjR2 = 0.810, RMSE = 0.016 and the accuracy P =0.978. The combination model of LPI' and I'3 was the best among multi-variable inversion models with fitting correlationsR2 =0.898, AdjR2 =0.889, RMSE =0.012 and the accuracy P =0.972. The final model showed that the energy ratio variable model was better and more stable than the number ratio variable model, and the multi-variable inversion model was better than the single-variable model with higher fitting correlations and accuracy. In the future we should extract more efficient variables and further explore the potential of energy ratio variables, because the extracted parameters are relatively less and have limitations on the extraction of energy ratio variables.