Abstract:

【Objective】In this study, the Camellia oleifera base in Mingyue Village, Changsha County, was used as the study area to explore the feasibility of unmanned aerial vehicle (UAV) oblique photography technology for extracting C. oleifera crown volume and estimating the tree height and yield. 【Method】Remote sensing variables such as band reflectance, vegetation indices, texture factors, height characteristics, and crown parameters were extracted from UAV orthophotos and dense matched point clouds. The Kriging, inverse distance weighting, natural neighbor, and filtered triangulation methods were used to obtain the crown volume of C. oleifera. The multiple linear regression, random forest, and K-nearest-neighbor models were established to estimate the height and yield of C. oleifera, and the accuracy of the estimation results was evaluated using the crown volume obtained from the ground 3D laser point clouds and the actual measured height and yield of the sample plots as the measured values. 【Result】The filtered triangulation method was the most effective for obtaining the crown volume with an average relative error of 31.54%, which was better than 36.73% for the inverse distance weighting method, 37.04% for the Kriging method, and 38.54% for the natural neighbor method. The accuracy of all three estimation models for the tree height and yield was improved by using the crown volume as a characteristic variable in the modeling. The relative root mean squared errors (rRMSEs) for tree height were reduced by 3.77%, 0.78% and 0.64%, respectively. In addition, the rRMSEs for the yield were reduced by 1.32%, 0.34% and 0.16%, respectively. The multiple linear regression, random forest and K-nearest-neighbor models were compared, and the coefficients of determination of the random forest model were all better than those of the multiple linear regression and K-nearest-neighbor (the R² of tree height was 0.78, 0.51 and 0.19, and the R² of yield was 0.61, 0.48 and 0.24, respectively). There is no significant difference in the accuracy of using the estimated tree height and measured tree height to participate in yield modeling. 【Conclusion】The combination of crown volume and tree height participation modeling can effectively improve the accuracy of C. oleifera yield estimation, and the results of this study can provide a reference for conducting C. oleifera height and yield surveys using UAV remote sensing technology on a regional scale.

Key words: non-wood forest, yield of Camellia oleifera, random forest(RF), unmanned aerial vehicle, crown volume, spatial interpolation