Mapping Larch (Larix Olgensis) Plantation Aboveground Biomass Using UAV-LiDAR Data

As an important indicator reflecting the nature and state of forest ecosystems,forest biomass is the basic feature for evaluating ecosystem service functions,and has been widely used in greenhouse gas accounting,global climate change maintenance,and carbon balance regulation and other fields.Estimating forest biomass by remote sensing technology makes up for the shortcomings of traditional forest survey in large-scale and dynamic monitoring and has become a hotspot in the field of forest ecology and remote sensing.In recent years,with the rapid development of UAV remote sensing platform and the improvement of spatial resolution of remote sensing data,remote sensing technology provides the possibility to obtain fine-scale forest resource parameters.With the development of remote sensing estimation of forest biomass to multi-platform,multi-sensor,multi-scale estimation.Light detection and ranging(LiDAR)remote sensing has the ability to provide high precision 3D information directly through the vegetation canopy,showing great advantages in biomass remote sensing estimation.In this study,a new unmanned aerial vehicle(UAV)remote sensing platform equipped with LiDAR sensor(UAV-LiDAR)was used as the main data source to construct a multi-scale forest aboveground biomass estimation system from individual tree to plot to small area to forest farm.It provides technical support for multi-scale remote sensing estimation of forest biomass.Specifically,this study conducted in Mengjiagang Forest Farm in Jiamusi City,Heilongjiang Province,and selected 11 larix olgensis plantation areas with different forest conditions for UAV-LIDAR data acquisition.Firstly,an allometric model of biomass was established based on historical destructively sampled trees,and the uncertainty of field measured biomass was analyzed by error propagation method,which could be used as a reference for biomass estimation at individual and stand scales.For individual tree level biomass estimation,we first developed a new individual tree point clouds segmentation method called hierarchical region-merging,then we extracted UAV-LiDAR characteristics at tree-and plot-levels,and build a mixed effects tree-level DBH estimation model,and combined with allometric model for treelevel biomass estimation.For plot-level biomass estimation,a diameter distribution model based on UAV-LiDAR was firstly developed diameter distribution models using the mixed-effects seemingly unrelated regression.Secondly,an area-based regression model based on UAVLiDAR was constructed and compared with the biomass of the sample plots accumulated from individual biomass estimation and biomass distribution simulation results.For small-area biomass estimation,this paper adopts the unit level small area estimation model to estimate small area biomass of 11 UAV acquisition areas,and analyzes the estimation uncertainty and error sources.In this paper,UAV-LiDAR and Sentinel-2 optical images were used to estimate the biomass of larch plantation at the scale of forest field,and the error transfer process and the contribution of UAV-LiDAR to the up-scale estimation were analyzed.In addition,considering that the "true" value of biomass could not be obtained,this paper used the measured data combined with monte-Carlo simulation method to estimate the biomass of single tree,diameter scale and plot scale.For small area and stand scale,the standard error of the population mean was used to evaluate the estimation results.In summary,the main conclusions of the study included:(1)At individual tree scale,the region-based hierarchical combined point cloud segmentation method proposed in this paper can provide better segmentation results than the classical algorithm(marker-controlled watershed and point cloud region-growing)in middle and young stands with high stand densities.The relative RMSE of individual tree biomass indirectly obtained by DBH estimation model is about 37.24%,and the calibration accuracy of mixed effect model can effectively improve the prediction accuracy.In addition,the uncertainty of tree biomass estimation based on measured biomass is poor(the relative error is 15.08% on average),and the monte-Carlo simulation test proposed in this paper avoids the overestimation of the accuracy of remote sensing estimation results by using measured biomass estimation values.(2)At plot scale,the parameter recovery model was better than the parameter prediction method for simulating the quantity and biomass distribution of each diameter step in the plot.For biomass estimation of sample plots,the single-tree integration method benefited from the high-precision single-tree segmentation method and single-tree scale model,and achieved better prediction results(13.10% relative RMSE%)in the unlocally calibrated prediction,but better prediction results were obtained through quasi-stand mean regression of local calibration.In contrast,although the distribution accumulation method provides more detailed information at the plot level,its results are not as good as stand mean method.(3)At small-area level,using unit-level small area estimation can directly use the ABA of the plot scale developed above,and the small area biomass estimation based on UAV-LiDAR can obtain high-precision estimation results.The relative error was 4.06%,and it decreased with the increase of stand biomass.(4)At forest farm level,the estimation accuracy using all the data is slightly lower than that of the traditional "satellite-field" mode in the "satellite-airborne-field" integration estimation with UAV-LiDAR as the intermediate layer,and the error mainly comes from the first level.The UAV-LiDAR model accounts for 62.46% of the overall error.Ignoring this part of the error will result in an overestimation of the accuracy.In the process of simulating the reduction of the number of plots,the contribution of UAV-LiDAR is also reflected.When the sample size reduced to30,UAV-LiDAR brings an average of 2.6 times the efficiency improvement,which proves that it can be achieved by increasing the ratio of the area of the UAV flight to the area of the plot to improve the estimation accuracy of the "satellite-airborne-field " integration.