Research On Forest Key Structural Parameters Estimation Based On Airborne LiDAR Data

As a new active remote sensing technology,the airborne light detection and ranging(LiDAR)could not only record the 3D coordinates of objects,but also could record the intensity data,which has been successfully used to estimate various forest structural parameters.The majority of LiDAR sensors could emit the wavelength which belonges to near infrared spectral range.And the green vegetation shows a good reflection within this spectral range.In theory,the LiDAR intensity data should play a significant potential in the future forestry research.However,the LiDAR intensity data are affected by many factors in the acquisition process,such as:distance,incident angle,target reflectance and atmospheric attenuation.Therefore,the research on exploration the potential of LiDAR intensity data on forestry application and quantification the effects of different factors intensity normalization on forest structural parameter estimation has become the hot point of current LiDAR data forestry application research.In this study,Jingyuetan national forest park,located in Changchun Jilin Province,was selected as the study site.A series of preprocesses including denoising,classification,adjacent overlap points removing and intensity normalization were done before the raw LiDAR data were used.Then the height distribution metrics and intensity metrics were extracted from the preprocessed LiDAR data and were used to classify the different stand types of study area.And the effects of LiDAR data sampling shape and scale on forest stand mean height estimation were studied.Additionally,different affecting factors were apllied to normalize the LiDAR intensity data and the effects of normalized intensity data on forest stand types classification and forest LAI estimation were quantified.Finally,the biomasses of different stand types were estimated based on the above acquired results of forest stand types classification,stand mean height and forest LAI.The main research contents and results are as follows:(1)To quantify the effect of adjacent overlap points on the forest structural parameters estimation,the adjacent overlap points were cut from the raw LiDAR data.Then the metrics on LiDAR height and intensity information,extracted from the LiDAR data with or without overlap points,were used to estimate stand mean height and forest LAI.The results showed that the R2 and RMSE of Scotch pine stand mean height estimation with overlap points were 0.873 and 0.940 m.While the R2 and RMSE of Scotch pine without overlap points were 0.892 and 0.868 m.For Larch pine stand mean height estimation,the R2 and RMSE from LiDAR data with overlap points were 0.725 and 1.196 m.And the R2 and RMSE from LiDAR data without overlap points were 0.741 and 1.161 m.Additionally,for Scotch pine forest LAI estimation,the R2 and RMSE from LiDAR data with overlap points were 0.666 and 0.220.And the R2 and RMSE from LiDAR data without overlap points were 0.794 and 0.172.For Larch pine forest LAI estimation,the R2 and RMSE from LiDAR data with overlap points were 0.654 and 0.110.And the R2 and RMSE from LiDAR data without overlap points were 0.762 and 0.091.(2)To explore the potential of LiDAR intensity data for forest stand types classification,a series of metrics on LiDAR height and intensity were extracted from preprocessed LiDAR data.Then the metrics were used to classify Scotch pine,Larch pine,Mongolia Oak,Aspen and other tree species using the random forest algorithm.The results demonstrated that the overall classification accuracy of metrics related to LiDAR height was 87.54%.The overall classification accuracy of metrics related to LiDAR intensity was 89.23%.And the result from the metrics related to both the LiDAR height and intensity was better,with an accuracry of 92.23%.(3)In this research,the LiDAR intensity data were separately normalized with range and incidence angle.Then the effects of normalized intensity data on forest stand types classification were assessed.It was found that the result from range normalized intensity data was 89.24%,which was better than the result from raw LiDAR intensity data,with 0.29%improvement in classification accuracy.The result from scan angle normalized intensity data was nearly the same as the result from raw intensity data,with just 0.01%improvement.However,the results from incidence angle,derived from DEM,DSM,DEM and DSM,normalized intensity data,were worse than the result from raw intensity data.The classification accuracies were separately 88.67%,74.50%and 64.87%.(4)To study the influences of LiDAR data from different sampling shapes and sampling scales on stand mean height estimation,the circle and square LiDAR data from 20 m sampling scale and a series of sampling scales with square sample were separately used to estimate different stand types mean height.It was shown that the square sampling was suitable for Scotch pine and the circle sampling was suitable for Larch pine.For Scotch pine,the metric from 35 m sampling scale achieved the best result(R2=0.904,RMSE=0.820).For Larch pine,the metric from 15 m sampling scale achieved the highest estimation accuracy(R2=0.720,RMSE=1.206).For other tree species,the metric from 20 m sampling scale achieved the highest estimation accuracy(R2=0.883,RMSE=1.257).(5)To explore the effects of LiDAR intensity normalization on forest LAI estimation,this paper used range,incidence angle and target reflectance to normalize the LiDAR intensity data.Then the laser penetration indexes(LPI),extracted from LiDAR intensity data,were used to estimate Scotch pine and Larch pine forest LAI.The results indicated that for the Scotch pine forest LAI estimation,the R2 and RMSE of LPI from LiDAR raw intensity data were 0.794 and 0.172.The R2 and RMSE from range normalized LiDAR intensity data were 0.795 and 0.172.The R2 and RMSE from incidence angle normalized LiDAR intensity data were 0.806 and 0.167.The R2 and RMSE from target reflectance coefficient normalized LiDAR intensity data were 0.810 and 0.166.While for Larch pine forest LAI estimation,the R2 and RMSE of LPI from LiDAR raw intensity data were 0.762 and 0.091.The R2 and RMSE from range normalized LiDAR intensity data were 0.762 and 0.091.The R2 and RMSE from incidence angle normalized LiDAR intensity data were 0.763 and 0.091.The R2 and RMSE from target reflectance coefficient normalized LiDAR intensity data were 0.762 and 0.091.(6)The biomasses of different forest stand types were estimated using single variable and multivariate regression algorithm based on the results of forest stand types classification,stand mean height and forest LAI.The result indicated that the stand mean height had a strong relationship with forest biomass.While the relationship between forest LAI and forest biomass was relatively weak.And the combination of forest stand mean height and forest LAI could improve the estimation accuracy of forest biomass.The above results indicate that the removal of adjacent overlap points can improve the accuracy of forest structure parameters estimation.The LiDAR intensity data could realize the higher accuracy of different stand types classification and the effects of intensity normalization on stand types classification are relatively small.The effects of sampling shape and scale of LiDAR points on stand mean height vary with different stand types.The intensity normalization can improve the accuracy of forest LAI estimation.However,the improvement levels are closely related to the selected parameters and forest stand types.The combination of stand mean height and forest LAI could improve the accuracy of forest different stand types biomass estimation.