Estimation Of Tree Diameter Structure And Logging Priority Level Zoning In Chinese Fir And Eucalyptus Plantations

Plantation plays a strategically significant role globally in achieving the balance of timber supply,improving forest layout,and mitigating climate change.Against the backdrop of increasing demand for timber resources in China,precise silviculture of high-quality and efficient artificial afforestation has become an inevitable path to address the demand for timber.To better utilize forest resources,it is imperative to achieve technological precision in all aspects of the highly intensive cultivation process.Accurate estimation of the diameter at breast height（DBH）structure and its dynamic changes in fast-growing tree species such as Chinese fir（Cunninghamia lanceolata）and eucalyptus（Eucalyptus robusta）,which are typical in southern China’s planted forests,is beneficial for maximizing the resource utilization efficiency and ecological functions of planted forests.It is of great significance for precise silviculture,quality improvement,and precise enhancement of forest quality.Traditional methods of obtaining forest stand structure parameters in planted forests mainly rely on ground measurements.However,this method is time-consuming and difficult to obtain structural parameters and their dynamic changes during the growth process.The application of active and passive remote sensing technologies,such as Light Detection and Range（Li DAR）technology and optical satellite imagery,enables rapid,quantitative,and accurate acquisition of breast-height diameter（DBH）structure parameters and their dynamic changes,facilitating accurate monitoring of forest silviculture and real-time diagnosis of forest health.This study focuses on typically planted forests in southern China,such as Chinese fir and eucalyptus plantations.It utilizes multi-source remote sensing data to accurately estimate the breast diameter structure and its dynamic changes in these planted forests,providing new technological and theoretical approaches,as well as accurate data support,for precision silviculture and efficient utilization of plantations.Specifically,this study used airborne Li DAR and tree height-DBH relationship curves to invert and estimate plantation DBH.Different modal distributions of plantation DBH were distinguished based on airborne Li DAR,and the predictive capability of different distribution models for multi-modal DBH distribution was explored.In addition,multi-temporal DBH distribution information was obtained and dynamic analysis was conducted based on multi-period optical imagery and Li DAR data,and various key information was integrated for predicting the probability of logging risk in harvested areas,and the logging priority level was divided based on the predicted probability.The main results of this study are as follows:（1）An improved method for estimating tree diameter at breast height（DBH）using airborne Li DAR data was proposed in this study.The method integrates the tree height-DBH relationship curve with Li DAR-derived variables.The predictive performance of this method was compared with a conventional Li DAR-based method for estimating DBH at plot level.Based on field-measured tree height and DBH data,optimal tree height-DBH relationship curve models were constructed for different tree species and forest density levels.The Li DAR-derived variables were then extracted and regressed against the model parameters（β₀ andβ₁）and mean tree height（H）of the tree height-DBH curve,resulting in inversion models for the model parameters and mean tree height of the tree height-DBH curve.Finally,the inverted model parameters and mean tree height were used as inputs in the original optimal tree height-DBH relationship curve model to estimate the plot-level DBH.The results show that the Meyer model was selected as the optimal fitting tree height-DBH relationship curve model for all plots（general model）,low forest density plots,and high forest density plots（determination coefficient（R²）=0.48-0.51,root mean square error（RMSE）=2.82-4.25,relative root mean square error（r RMSE）=20.18-26.10%）,while the Logarithm and Linear models were selected as the optimal fitting tree height-DBH relationship curve models for fir（R²=0.76,RMSE=1.71,r RMSE=12.91%）and eucalyptus（R²=0.73,RMSE=2.51,r RMSE=16.44%）,respectively.For plot-level DBH estimation,the improved plot-level method（R²=0.77-0.86,RMSE=1.29-2.03 cm,r RMSE=1.21-18.87%）outperformed the conventional ALS-based plot-level DBH estimation method（R²=0.74-0.82,RMSE=1.37-2.23cm,r RMSE=1.25-22.90%）.In the improved plot-level DBH estimation method,the inversion of plot-level tree height and tree height-DBH relationship model parameters achieved good predictive accuracy（β₀:R²=0.78-0.88 and r RMSE=12.10-17.03%;β₁:R²=0.82-0.93 and r RMSE=6.26-17.68%）.（2）An airborne Li DAR-based model was developed to discriminate and predict diameter distribution patterns for monoclinic and multi-modal stands.First,the Lorenz curve’s Gini coefficient（GC）and asymmetry coefficient（LA）were obtained using Li DAR and ground-measured data,and an inversion prediction model was established to differentiate monoclinic and multi-modal diameter distribution patterns.Next,Li DAR-derived feature variables and measured data were used to establish prediction models for diameter distribution parameters of monoclinic and multi-modal stands,using the Univariate Weibull Model（UWM）,Finite Mixture Model（FMM）,and k-Nearest Neighbor Model（KNN）.Sensitivity analysis was performed on KNN model parameters including k（the number of neighbors）,response configuration,and distance metric（i.e.,imputation method）.Results showed that Li DAR-derived measurements effectively distinguished diameter distribution patterns.Specifically,the Gini coefficient（GC）and asymmetry coefficient（LA）predicted by Li DAR had good discrimination ability for monoclinic and multi-modal stands（overall accuracy OA=59.18-77.55%）.Moreover,Li DAR exhibited great potential for predicting diameter distribution patterns in southern subtropical plantations（Mean e _R=39.98-52.31,Mean e _P=0.20-0.26）.For predicting diameter distribution patterns of complex structured stands,both FMM and k NN models performed well in predicting multi-modal diameter distribution patterns（Mean e _R=39.98-52.19,Mean e _P=0.20-0.26）,outperforming the UWM model（Mean e _R=41.36-52.31,Mean e _P=0.21-0.26）.Sensitivity analysis of the k-Nearest Neighbor Model showed that the error indices were lowest when k was set to 5,and the Random Forest（RF）imputation method outperformed the Most Similar Neighbor（MSN）method in the KNN model.（3）A dynamic model of tree diameter at breast（DBH）distribution based on multi-source remote sensing data was constructed,and a probability prediction model for logging risk in planted forests was developed,achieving prioritized classification of logging in planted forests.Tree species classification was conducted using Sentinel-2A and Li DAR data from 2018 and 2021.The Gini coefficient（GC）and asymmetry coefficient（LA）were used to distinguish between unimodal and multimodal DBH distributions based on Li DAR-derived feature variables.Inversion and prediction models were established for DBH distribution parameters of unimodal and multimodal stands using two-period data,combining the univariate Weibull maximum likelihood（UWM）model and the finite mixture model（FMM）.The transferability of the prediction model from the first period（2018）was also examined.Changes in DBH distribution of remeasured plots between2018 and 2021 were investigated,and potential logging areas in forest growth regions in the future were identified using a combination of the Logistic model,forest features,and terrain factors.Logging priority was then determined.The results showed that the classification accuracy in 2021（overall accuracy OA=75.78%,Kappa coefficient=0.60）was lower than that in 2018（overall accuracy OA=84.22%,Kappa coefficient=0.75）.The Li DAR-derived structure feature,canopy height model（CHM）,and terrain factors contributed the most,followed by vegetation indices,spectral features,and texture features.The prediction accuracy of the second-period DBH distribution parameters model based on Li DAR was good（R²=0.47-0.87,r RMSE=8.10-36.40%）,but the fitting effect of the second-period（2021）DBH distribution obtained based on the unmanned aerial vehicle（UAV）Li DAR was better（Mean e _R=25.20-60.53,Mean e _P=0.13-0.30）,but relatively lower than the first period（2018）（Mean e _R=21.65-52.31,Mean e _P=0.11-0.26）.The characteristics of DBH distribution change in remeasured plots between 2018 and 2021 showed a gradual shift from left-skewed to right-skewed distribution,with smaller fluctuations in adjacent diameter classes and flatter distribution curves,and a gradual decrease in scale parameters.The independent variables of stand density（N）,Gini coefficient（GC）,variance（Var）,whether the mode is greater than expectation（Md E）,and DBH distribution shape change（△M）had significant impacts on logging occurrence（Is L）（p-value<0.05）,and the accuracy of the Logistic prediction model for logging occurrence was good,with a generalized coefficient of determination（Cox-Snell R²）reaching 0.74.