Estimation Of CBD And CBH Using UAV Remote Sensing Data

The accurate estimation of Canopy Bulk Density（CBD）and Canopy Base Height（CBH）)is an important basis for evaluating canopy fire hazard and predicting canopy fire behavior.As a cutting-edge technology of remote sensing data acquisition,Unmanned Aerial Vehicle（UAV）can provide stand information with high resolution,high timeliness and high precision.Light Detection and Ranging（Li DAR）technology is the most effective remote sensing method to obtain forest canopy structure at present,and optical remote sensing is an important supplement.In this study,using two methods based on Area-based and Individual Tree Segmentation,taking Qingbei Forest Farm in Jiaohe City,Jilin Province as the experimental area,the CBD and CBH estimation methods of UAV Li DAR and Multispectral images were established,and the estimation accuracy was evaluated.For the model based on Area-based,three kinds of remote sensing data sources（Li DAR,Multispectral and Li DAR+Multispectral）are studied.Using Logarithmic transformation and Box-Cox transformation to optimize modeling data.Furthermore,variable feature selection is carried out by Mutual Information（MI）,L1regularization（L1）,Stepwise Selection（SS）and Best Subset Selection（BSS）.Finally,three regression models of Multiple Linear Regression Model（MLRM）,General Additive Model（GAM）and Robust MM estimation（MM）are constructed,and the performance of different combinations（data source,data transformation,feature selection and estimation model）in the estimation of CBD and CBH is discussed.In addition,in order to reduce the dependence on the ground sample plot data and construct the CBD and CBH models of mixed tree species stand,Individual Tree Segmentation fusion model method for estimating CBD and CBH of mixed tree species stand using remote sensing data was constructed based on Individual Tree Segmentation and combining the estimation methods of tree species,tree height,tree base height and DBH.The results based on the Area-based estimation method show that:（1）Fusion of data sources（or Li DAR data sources）,Box-Cox transform,BBS（or L1）and MLRM are the best combination methods for estimating CBD and CBH.（2）Among the CBD and CBH models of different remote sensing data sources,the model with the highest accuracy is the fusion data source（R² of CBD is 0.672,RMSE is 0.067 kg/m³,R² of CBH is 0.739,RMSE is 1.429 m）,followed by Li DAR data model（R² of CBD is 0.615,RMSE is 0.072 kg/m³,R² of CBH is0.716,RMSE is 1.487 m）.（3）Height features(H_mean,H_max)and structural features（VCI,LAI）in Li DAR data sources are of great significance for the prediction of CBD and CBH.Structural Vegetation Index（VARI,SR）and blue are important features in Multispectral data sources.The results based on the estimation method of Individual Tree Segmentation show that:（1）Compared with other methods in this study,Dalponte Individual Tree Segmentation algorithm has higher segmentation accuracy（R is 0.79,P is 0.95,F is 0.87）.3%interval percentile（using MAE as penetration correction）is the best method for estimating tree base height.（2）Based on the Dalponte Individual Tree Segmentation calculation,the tree height estimation method of the highest point cloud（R² is 0.77）,the DBH estimation method of tree Height-DBH curve（R² is 0.32）,the Random Forest tree species classifier（OOB is 5%）,and the tree base height method of 3%interval percentile（R² is 0.64）,the final CBD model R² is0.406,RMSE is 0.097 kg/m³ and CBH model R² is 0.614,RMSE is 1.8 m.（3）By comparing the estimation methods based on Area-based and Individual Tree Segmentation,it is found that the estimation method based on Individual Tree Segmentation will underestimate the CBD at the high value.There is little difference in the accuracy of the two estimation methods for CBH,but in spatial prediction,the value of Individual Tree Segmentation method is higher than that of Area-based method in Pinus sylvestris stand.In this study,the remote sensing data estimation models of stand CBD and CBH were established by using two different methods,and the estimation ability and applicability of the two methods to CBD and CBH were analyzed and evaluated,which has reference value for the remote sensing estimation of UAV canopy fuel parameters.At the same time,this study can provide basic data for canopy fire behavior simulation and promote the expansion of canopy fire behavior simulation to UAV remote sensing.