Forest Parameters Estimateofmulti-remote Sensing Data Basedon ArboliDAR

Forests are the main body of terrestrial ecosystem and the most complete resource library of nature features.They can not only provide the materials for the human survival and development,but also play an important role in the maintenance of the land and the global ecological balance.While China is one of the countries rich in forest resources,timing and regular forest inventory,supervision of the forest growth,can promote sustainable development of forestry.At present,remote sensing technology,as a means of forest resource inventory,is widely used in forestry investigation because of its fast,accurate and rich information.As an advanced active remote sensing technology,light detection and ranging technology has a significant advantage in the detection of the spatial structure of forest vegetation and the inversion of forest parameters.Therefore,the combination of laser radar and other remote sensing technology can be used to detect and check the forest structure in order to accurately evaluate the forest growth and make a reasonable plan for forestry development.In this study,we used lidar point cloud data and aerial remote sensing data of Zhangye city in Gansu province Qilian Mountains Dayekou forest area as the data source.ArboLiDAR is used to generate the height raster and density raster through the classification of source data and the normalization of height,and then combined with the classification raster of aerial data to generate the variable raster.Based on multi source remote sensing data and using region growing algorithm based on gradient,we need to find the optimal segmentation result through the different parameter settings of segmentation parameters and merging parameters.Next,on the basis of stand segmentation image,through the calculation of different variables and combined with the measured data,the regression model between the measured stand parameters and LiDAR variables was established using sparse bayesian algorithm.And then,estimating the stand parameters and accuracy evaluation.The results are as follows:(1)Classification filtering of LiDAR point cloud data is firstly separated by high threshold method and over low search algorithm.Then,the irregular triangulation algorithm is used to classify ground points and forest points.The main filter parameters include search radius,iterative angle,iterative distance,maximum height,minimum height,etc..Classification of aerial data using supervised classification method,the Kappa coefficient of classification result is 0.8,indicating the classification results are better,we can use this classification result to complete subsequent operations.(2)The research of stand automatic segmentation based on ArboLiDAR,7 groups of segmentation parameters(band weights,priority function)and combined parameters(mean difference,standard deviation difference,band weights)are set in this study.By comparing the 7sets of segmentation results with the result of stand manual segmentation,we can see that the fifth groups have the best segmentation effect.Because its segmentation effect is obvious in non forest land and low vegetation area,it does not divide the non forest land into one subarea.And the result is similar to the result of manual segmentation.(3)The estimation results of stand parameters by using regression model showed that the estimation accuracy of average HGW of stand is the highest by using of multi-source remote sensing data,with R~2 of 0.744 and the average estimate accuracy of 98%;Followed by stand volume,the average estimate accuracy is 96%;And because of the influence of forest age structure and stand distribution,the estimation accuracy of average DBH and G(total breast height area)are low,but the average estimate accuracy are above 80%.(4)Compared with traditional forest resource survey method,the results of stand automatic segmentation are more detailed and comprehensive based on ArboLiDAR.The number of stand automatic segmentation is about 380,and the number of manual segmentation is 332.At the same time,the extraction of forest parameters by partitioning can make the extraction result more accurate,and can display the distribution of forest parameters in the study area more intuitively.