Phenotypic Traits Extraction And Genetic Characteristics Assessment Of Eucalyptus Plantation Based On UAV Multi-source Remote Sensing

Forest tree phenotype describes the physical,physiological and biochemical characteristics of forest tree that are determined or influenced by genes and environment.It refers to the expression of height,yield,color,wood property,stress resistance and other traits or characteristics in the process of tree growth and development.Forest tree phenotyping helps to understand the growth and development,the response to environmental changes and its adaptability of tree,and to reveal the genetic and performance laws of forest tree.Accurate and precise extraction of phenotype is a prerequisite for forest phenotyping.However,the traditional forest phenotype acquisition commonly relies on manual measurements which is time-consuming,labor-intensive and destructive sampling,small scale and low efficiency,rendering it difficult to meet the high-throughput needs of forest tree phenotyping and modern tree breeding analysis.UAV remote sensing technology,as an efficient and low-cost tool for forest tree high-throughput phenotyping,can obtain forest phenotype in a large-scale,real-time,non-destructive,high-precision and-throughput manner with various types of sensors.The forest tree high-throughput phenotyping based on UAV remote sensing technology opens up new ways for accurate phenotypic evaluation of large-scale breeding populations,helps to select high-yield,high-resistance and high-quality varieties,accelerate forest tree breeding,improve precision silviculture and management,promote improvement of forest quality and promote sustainable forest management.Eucalyptus is of great value for its contributions to industrial timber in countries due to its fast growth rate,barren-tolerance and strong resistance,thus it plays an important role in ensuring the national timber reserve and supply.In this study,we disposed a comprehensive UAV-based remote sensing campaign in the Eucalyptus breeding trials of Gaofeng Forest Farm in Guangxi province of southern China,using UAV-borne Li DAR data（of high-density point cloud）and RGB images（of high-resolution true-color orthophoto）collected by multi-rotor UAV system.Combined with field data,we explored the methods and potential of UAV remote sensing technology in obtaining and analyzing high-throughput phenotype of forest tree:Firstly,the accuracies of three individual tree segmentation methods（i.e.Marker-controlled Watershed Segmentation（MWS）,Point Cloud-based Cluster Segmentation（PCS）,Multiresolution Segmentation（MRS））based on Li DAR data and image were compared in three Eucalyptus Plantations with different stand densities,aiming to obtain high-throughput phenotype at the individual tree scale with high-efficiency and-precision.Then,this paper compared the differences of phenotypic traits（tree height（H）,diameter at breast height（DBH）and crown width（C））and characteristics（structural metrics（n=18）,spectrum indices（n=9）and texture features（n=24））obtained by different phenotype acquisition methods（i.e.based on manual measurement and UAV-based extraction）and different scales of UAV-based extraction（i.e.individual tree scale and plot scale）to explore the validity and reliability of UAV technology in obtaining forest tree high-throughput phenotype,and explore the ways to obtain high accuracy phenotype by comparing the accuracy of different scales of UAV-based extraction.Finally,this paper carried out phenotypic traits variation analysis and comprehensive selection,and compared differences between UAV-based extractions and manual measurements for genetic characteristics analysis to explore the potential of UAV remote sensing technology in genetic characteristics analysis of forest trees.This paper provides process technical methods support for accelerating forest tree breeding,improving precision silviculture and management,and promoting improvement of forest quality.The research indicates:（1）The sensitivity analysis results showed that MRS has higher segmentation accuracy when the segmentation scale is 30.The result of individual tree segmentation accuracy shows that:the overall segmentation accuracy（F）of MWS,PCS,and MRS were 0.80,0.80,and 0.83,respectively in the lower density site（760 n/ha）,the F of MWS(F_（2）=0.93,F_（3）=0.86)was higher than that of PCS(F_（2）=0.84,F_（3）=0.74)and MRS(F_（2）=0.83,F_（3）=0.78)in the medium（982 n/ha）and high density sites（1239 n/ha）.With the increase of stand density,the F of MRS gradually decreased which make it became lower than that of MWS in the two higher density sites,besides,the difference between the overall accuracy of MWS and PCS gradually expanded.（2）The result of comparions showed that compared with manual measurements,the values of plot-scale Li DAR extractions（H,DBH）are lower;The individual-tree-scale Li DAR extractions（H,DBH,C）performed better for its accuracy（R² is 0.91～0.99,r RMSE is 1.07%～6.18%,Pearson r is 0.94～0.99）is much higher than that of plot-scale Li DAR extractions（H,DBH）（R²is 0.27～0.90,r RMSE is 10.92%～27.14%,Pearson r is 0.51～0.95）,and its mean ranking of phenotypic traits among clones is closer to the results of manual measurements.（3）The ANOVA result showed that there were significant or extremely significant difference in manual measurements and individual-tree-scale Li DAR extractions（H,DBH,C）among clones,while there were not significant difference in plot-scale extractions（H,DBH）among clones in low-density site.The result of genetic parameters estimation showed that phenotypic coefficients of variation（PCV）,genotypic coefficients of variance（GCV）,repeatability of clones()and repeatability of individual trees(₄))of manual measurements were 0.066～0.125,0.054～0.067,0.537～0.858 and 0.281～0.669 respectively,theshowed middle to strong repeatability indicated that traits were greatly affected by heredity.Genetic parameters of individual-tree-scale and plot-scale Li DAR extractions are higher and lower respectively,and the relative error of plot-scale Li DAR extractions is larger.Most of structural metrics（29/36）and spectrum indices（17/18）were significantly or extremely significant difference among clones,the repeatability of selected characteristics（structural metrics（the high percentile）,vegetation index（EXG-R,BI,EXG）and texture feature（ME））is higher than that of traits in low-density site and lower in high-density site.（4）The result of clonal selection and genetic gain estimation showed that:The clones evaluation and selection based on manual measurements and individual-tree-scale Li DAR extractions are different but small,the matching rates of clones optimized by them were 100%,67%and 100%respectively in low-density site,and 0%,100%and 100%in high-density site with the selection rates of 10%,20%and 30%,demonstrated that the pure clones（of E.urophylla）,the hybrid clones（of E.urophylla as the female parent）and the hybrid clones（of E.wetarensis and E.grandis）perform better.There are differences but consistent trends between the genetic gains calculated by the two phenotype acquisition methods.