Estimation Of Vegetation Structure Parameters Based On LiDAR Point Cloud And Hyperspectral Image

Vegetation structure parameters are important indicators for ecological restoration monitoring.Traditional field measurement methods have low efficiency,high cost,small coverage,and large damage to the vegetation canopy.Traditional remote sensing technology has a wide range,high efficiency and short cycle,but lacks three-dimensional vegetation information.In addition,for the vegetation structure parameters,the existing research uses horizontal and vertical structure to describe the appearance,ignoring the vegetation group type and the differences within the vegetation.Therefore,in this paper,combining LiDAR and hyperspectral data,based on LiDAR technology,a vegetation structure parameter estimation model is constructed from two aspects of vegetation morphology and constituent structure,and it is applied to the Heidaigou East Dump for demonstration,and the following are obtained results and conclusions:(1)It revealed the internal connection between the intensity,height,and return characteristics of LiDAR point cloud data and the spectrum,vegetation index,and texture characteristics of hyperspectral image data.Among the point cloud features,the correlation between the intensity feature factors is the lowest.The height feature has a high correlation with the spectral feature,and the return feature has the lowest correlation with other features.Among image features,the information redundancy of the spectral feature factors is high,and spectral features have the strongest correlation with texture features.In general,the information redundancy of feature factors extracted from LiDAR data is relatively low,and the information richness is high.This shows that LiDAR data reflects different information from hyperspectral data.In order to more accurately and comprehensively perform the estimation of vegetation structure parameters,data fusion is a very important technical means.But there is redundancy information between features,the prerequisite for application is to filter the characteristic factors.(2)The features extraction for vegetation structure parameters estimation and index screening methods for structure optimization have been established.Research shows that combining correlation analysis and random forest importance ranking method can optimize the characteristics and indicators of the parameter estimation and optimization of vegetation structure.The results show that the feature with the highest contribution rate to the recognition of the vegetation communities type is the vegetation index,followed by the height feature,and the pure spectral feature has a small contribution rate to the recognition of the vegetation communities.Correlation analysis results of vegetation structure parameters show that the six structural parameters of leaf area index,foliage height diversity at 3m vertical interval,number of vegetation communities,morphological evenness,physiological richness and physiological evenness are significant indicators for evaluating vegetation structure.Among them,the correlation between morphological evenness and other structural parameters is the least significant,and foliage height diversity at 3m vertical interval has the most significant correlation with other indicators.(3)The fusion of LiDAR and hyperspectral data can accurately and reliably perform the estimation of vegetation structure parameters.In the estimation models of vegetation morphological-structure,return and height characteristics of point cloud are mainly used.The accuracy verification results show that the estimation accuracy of gap fraction is the highest,R~2=0.986,RMSE=0.023,followed by canopy coverage,R~2=0.982,RMSE=0.024,the accuracy of canopy height parameter is better than that of leaf area index,R~2=0.922,RMSE=0.098,relatively low accuracy of leaf area index,R~2=0.898,RMSE=0.480.This shows that the estimation results of vegetation morphological-structure are highly accurate and the estimation models are reliable.Comparing the results of single tree segmentation and field investigation,the results are basically the same.In the process of extracting the number of vegetation communities,the hierarchical classification method,combined with the decision tree and random forest algorithm,realizes the fine classification of the vegetation communities.The overall classification accuracy reaches 87.45%.Compared with the Iso Data unsupervised classification method,the accuracy is increased by nearly 43%,and the accuracy of other supervised classification methods is increased by 10.7%?22.7%.The estimation results of vegetation functional diversity show that the functional diversity of the mixed vegetation configuration is better than that of the single vegetation community configuration.Comparing the estimation results of the structural parameters of different vegetation communities,it is found that the constituent-structure of the reconstructed vegetation in the mining area is more regular and the difference is relatively small,and the difference in the morphological-structure is greater than the constituent-structure.(4)The thesis developed methods for evaluation and optimization of vegetation structure.Taking the selected vegetation structure parameters as indicators,combining the coefficient of variation method and the comprehensive evaluation method,the vegetation structure evaluation method is established.The results show that the overall structure of the reconstructed vegetation area is better than that of the original land.The areas with better structure account for 71.2%of the total area in the study area.The mixed vegetation structure of arbor-shrub and arbor-shrub-grass is better than a single vegetation cluster configuration,and this result is consistent with the result of functional diversity.The paper has 47 pictures,21 tables,and 118 references.