Estimationon Stand Characteristic Parameters And Biomass Of Forest By UAV Aerial Photogrammetry And LiDAR

It is considered as a prerequisite for the evaluating forest productivity,carbon storage of forest ecosystem and forest ecological service capacity to accurately predict and invert stand characteristic parameters and biomass of forest.In recent years,the UAV(Unmanned Aerial Vehicle UAV)technology has been developing rapidly,by virtue of its high flexibility and relatively low cost data acquisition,is widely used in different fields.With the development of image three-dimensional reconstruction technology,Digital Aerial Photography(DAP)could not only acquire plane images,but also generate high-precision three-dimensional point cloud products.Then the vertical structure of objects could be obtained,which provids a good opportunity for its research and application in forest ecological remote sensing.In this paper,UAV-DAP and LiDAR technology were combined to estimate the characteristic parameters and biomass of forest at stand scale.The following aspects are carried out:1.The forest canopy height model was created by UAV photogrammetry and LiDAR and its interpolation error was analyzed.The normalization of point cloud was achieved by LiDAR-DEM and DAP point cloud which show the upper part structural information of canopy.The canopy height model generated by spatial interpolation of normalization point clouds was the foundation for stand characteristic parameters and biomass estimation.The accuracy of interpolation results was influenced by interpolation methods,forest types,image resolution,and etc.There was also a certain correlation between interpolation error and point cloud factors such as gradient,roughness and color components.It was found that the interpolation error of natural proximity method was the smallest(RMSE=0.76-2.68),followed by irregular triangular network(RMSE=0.92-3.53),Kriging(RMSE=0.83-4.29)and inverse distance weighting method(RMSE=0.83-4.76).The interpolation error was reduced by increase of point cloud density,brightness and image resolution significantly,and increased while the slope and roughness of point cloud were increasing.The boundary error of forest land was larger with dramatic changes.2.Optimized DAP point cloud variables and modeling methods were selected by comparing different modeling methods in the Coastal plantation Research area.The result of modeling was promoted from sample plot to the whole research area and the modeling differences among different tree species were compared.A complete technological process in forest ecosystemf or estimation of stand structure parameters,biomass inversion and output and Application of achievements using UAV-DAP was constructed.It was found that the highly quantile variables were the first choices in Modeling parameters and biomass,the expressivity of P10(10%height quantile),Hmax(maximum height)and HCV(height variation coefficient)was higher in each model.For the Weak canopy penetrability of optical remote sensing,the Low-digit height variables of DAP-pointcloud were equivalent to the Medium Quantile Height Variables of LiDAR,the height information of the whole forest in the study area was reflected in Hmax to a large extent,and HCV could represent the vertical structure of canopy to some extent.The choice of statistical methods had a direct impact on the accuracy of the mod el,it was found that the common stand characteristic parameters and biomass could be better inversted by these three methods(principal component analysis,multiple linear regression and random forest).Among them,the accuracy of multiple linear regression was relatively high,and the estimation accuracy of random forest method on several particular parameters was slightly higher.Trends from high to low of the precision of Parameters Estimated by three Models were roughly the same,the accuracy of domina nt DBH(R~2=0.712~0.85,cv=10.355%~14.354%),dominant height(R~2=0.699~0.783,cv=12.289%~14.483%),average DBH(R~2=0.709~0.793,cv=12.984%~15.378%),average height(R~2=0.651~0.729,cv=15.639%~17.727%)and Lorey's height(R~2=0.722~0.792,cv=13.241%~15.315%)of were significantly higher than other parameters,followed by aboveground bio mass(R~2=0.408~0.594,cv=28.289%~32.023%),volume(R~2=0.307~0.489,cv=34.544%~38.988%),density(R~2=0.399~0.784,cv=51.417%~85.824%)and the lowest was basal area(R~2=0.073~0.262,cv=27.411%~30.691%).Forest types had a significant effect on estimation accuracy,there was little difference in the estimation accuracy between the three broad-leaved forest species.but significant in spatial level.The estimated values and spatial d istribution of poplar were concentrated,whith obvious block distribution,which has sho wn the characteristics of plantation and fast-growing forest.The parameter distribution and internal differences of Metasequoia glyptostroboides were relatively stable,compare d to the other two species,estimation of Ginkgo biloba was slightly low,probably bec ause the complex canopy structure.3.The UAV aerial photogrammetry and LiDAR were combined used for estimating common stand characteristic parameters in the subtropical forest region,the estimation accuracy of some factors was improved by the color components,such as vegetation coverage and color index which was extracted from images,for the estimation model.Fristly,The canopy model was calculated by the DAP point cloud which was generated by image three-dimensional reconstruction technology and DEM which was extracted from LiDAR.Then,the multivariate regression estimation model was constructed and its accuracy was verified by combining the measured forest stand survey data and normalized point cloud variables.The correlation analysis had showed that the point cloud variables and tree height factors had high sensitivity(R~2=0.58–0.95),followed by the volume,aboveground biomass(R~2=0.29–0.7);stand density and basal area were the lowest(R~2=0.18–0.52).The independent variables of the multivariate regression model include three groups of point cloud characteristic variables,i.e.height quantile,point cloud density and height information,and three kinds of color variables,i.e.vegetation coverage,GLI and EXG index.The degree of correlation between height quantile and vertical information of stand was higher.The correlation between height quantile and stand vertical information is high.Lorey's tree height fitness was the highest(rRMSE=5.32%),followed by stock volume(rRMSE=6.35%),aboveground biomass(rRMSE=10.41%),breast height area(rRMSE=16.38%)and plant number(rRMSE=27.04%).The color variables were correlated with the spectral information in the green band,and the model accuracy of three factors of Lorey's tree height(rRMSE increased by 0.08%),voulme(rRMSE increased by 0.58%)and aboveground biomass(rRMSE increased by 3.11%)is slightly improved,after the model was built with the color variables.To some extent,it had reflected the response relationship between green band and vegetation growth.(4)A comparative study of DAP and LiDAR point cloud was carried out to explore the similarities and differences between these two technologies in the application of forest ecological remote sensing,and to compare the differences in factors selection and estimation accuracy of stand characteristic parameters estimation models in a same region.The accuracy of DAP and LiDAR point clouds was comparable in a same region.There was a certain similarity for the upper canopy indicators(P95,P75,Hmax,etc.),however,the differences between the two types of point clouds,i.e.P25,D3 and the variables expressing the vertical structure(e.g.hcv,Oligophotic,etc.)were greater(R=0-0.45).In estimating factors with high correlation between canopy vertical structure,basal area and volume of DBH,DAP point cloud variables such as height variation coefficient,CC2 and CCmean,which could describe the vertical structure of canopy,were far less important than same LiDAR variables.While the importance was similar for tree height,DBH and other highly correlated factor estimation models.The accuracy of LiDAR(R~2=0.515-0.895)was obvious higher than DAP(R~2=0.288-0.835)for the same parameters modeling,especially vertical structure,such as underground biomass and basal area.In a word,using UAV photogrammetry with LiDAR technology could realize the stand characteristic parameters estimation and biomass inversion at stand level.The models of lorey's H,DBH and other stand characteristic parameters,have the high sensitivity to tree height information,the DAP model can achieve the same accuracy as LiDAR model,and also a good prediction effect on volume and aboveground biomass.