| Phenotype refers to the set of observable characteristics exhibited by of an individual or a population of organisms under the interaction between the control of potential genetic genes and the environment.Obtaining multi-scale and high-throughput tree phenotype data is a key bridge to study the mechanism of"gene-phenotype-environment",which is of great significance for the research of directed cultivation and improved species selection of trees.Traditional manual phenotype information collection methods are often destructive,inefficient,labor-intensive,and hard to achieve continuous dynamic monitoring.For trees with tall individuals,long growth spans,complex genome sequences and habitats,large-scale,high-precision and automatic phenotypic trait measurement is always difficult and expensive.In recent years,the development of remote sensing technology and image processing algorithm provides a new solution.The UAV photogrammetry technology can obtain high-resolution,multi-spectral images,and generate high-density three-dimensional point clouds with the help of"structure from motion"algorithm to achieve high-precision tree phenotype extraction.In addition,the ability of earth observation satellite platforms to routinely collect high-resolution imagery covering large area can further reduce survey costs,while also having the potential to be applied to the extraction of phenotypic traits at plot scale.Eucalyptus spp.,due to its fast-growing,high-yielding,has been planted in hundreds of countries and plays an indispensable role in timber production,increasing carbon sinks and alleviating global climate change.Currently,the long-term use of few clones in afforestation and the low quality of clones are among the important reasons for the decline in eucalypt stand quality.Accordingly,this study used the eucalypt clonal experimental plantations in the State-owned Dongmen Forest Farm of Guangxi Zhuang Autonomous Region as research object,in order to extract phenotype and thoroughly evaluate eucalypt clones based on unmanned aircraft and satellite multispectral images,respectively.First,to extract growth traits,point cloud metrics,vegetation indices and texture metrics of individual trees with the help of high-density digital aerial photogrammetry(DAP)point clouds and high-resolution multispectral images from UAV,to verify the accuracy of drone data in extracting phenotypic information at individual tree level,and to explore whether satellite images can identify intra-species and inter-clones differences at plot level based on sub-metre satellite images to extract vegetation indices and texture metrics;Then,the vegetation indices were used as indirect phenotype to build a genetic variation analysis system at different scales(individual tree and plot scale)based on different remote sensing data sources.Finally,the selection index method and calculated genetic parameters were used to evaluate the eucalypt clones in the trial for a multi-trait comprehensive evaluation,to select superior clones and predict the genetic gain.The main findings of this paper are as follows:(1)Compared the marker-controlled watershed algorithm(MWA)based on the canopy height model(CHM)with the multi-resolution segmentation(MRS)combining both CHM and multispectral imagery for individual tree canopy extraction.The results showed that the overall accuracy F of canopy identification using MWA reached 0.85 and 0.90 respectively in the 2 sites,both significantly higher than MRS(F values of 0.78 and 0.82,respectively);Either individual tree segmentation algorithms performed better in the sites with younger stands and more regular stand rows.(2)The use of unmanned aircraft platforms allows for high accuracy collection of individual tree phenotype and metrics.Tree height was extracted from the DAP point clouds and validated to accuracy of R~2=0.92,RMSE=0.56 m for site 1,and R~2=0.90,RMSE=0.46 m for site 2;the H-DBH model used also provided a high estimate of DBH(site 1:R~2=0.71,RMSE=0.75 cm;site2:R~2=0.69,RMSE=0.60 cm).(3)The automatic threshold segmentation using the Otsu method resulted in a precise extraction of vegetation area.The Kappa coefficients for image segmentation were 0.86 and 0.88 based on centi-metre UAV imagery;the algorithm also showed high accuracy in extracting vegetation from sub-metre multispectral imagery,with an overall accuracy of 0.86 and a Kappa coefficient of 0.83.(4)The ANOVA results showed that the metrics of different scales(individual tree scale and plot scale)obtained by different remote sensing data(UAV and satellite data)showed highly significant differences(p<0.01)among different clones,verifying the ability of UAV data to identify intra-species and inter-clones differences of Eucalyptus at individual tree level,and also indicated that sub-metre satellite data could detect intra-species and inter-clones differences at plot scale.(5)The vegetation indices were used as new indirect phenotypic traits,the genetic parameters of all phenotypic traits were counted for genetic variation analysis.The results showed that where the stand age is higher,the genetic control is higher.In site 1,except for canopy width,the clonal repeatability were all above 0.9 with individual repeatability were all exceeding 0.6;The clonal repeatability of tree height and DBH in site 2 were both 0.97,and the repeatability of vegetation indices were relatively lower(the clonal repeatability ranged from 0.65 to 0.90,and the individual repeatability ranged from 0.2 to 0.6).Among all chosen vegetation indices,the red-edge vegetation indices RECI,NDRE and m NDI showed superior genetic stability at different scales and in different sites.The near-infrared vegetation indices DVI,EVI,SAVI and MSAVI showed high genetic correlation coefficients with tree growth traits,and high genetic correlation between vegetation indices calculated by same bands.(6)A multi-trait quantitative evaluation framework was constructed to select clones and estimate their genetic gain.At 10%inclusion rate,the selected superior eucalypt clones for site 1were EA14-15,EA14-09,EC184 and EC183 in order based on UAV data of individual tree level,with predicted gain of more than 5%in tree height and DBH;Based on plot-scale satellite data,the clones EA14-09,EC221,EC151 and JJ1409 were preferably selected of site 1 by 10%inclusion rate,the mainly vegetation indices for predicted gain are ARI,NRI,RECI and RVI.The 3 superior eucalypt clones,288-4,DH26 and DH28,were selected on the basis of individual tree growth traits and vegetation indices in site 2,and estimated tree height and DBH can be increasing more than 10%. |
