Spatial Pattern Analysis And Biomass Estimation Model Of Chinese Fir Plantation

Forest carbon sinks play an important role in achieving the goal of carbon neutrality.Accurate estimation of forest biomass is the basis for obtaining effective forest carbon sink data.The estimation of classical statistical methods is difficult to explain and comprehensively analyze the spatial distribution pattern of forest attribute characteristics and its influence on biomass prediction,leading to biases in model estimation.Techniques such as econometric models,decision tree methods,machine learning,and geostatistics（modern spatial statistics）provide analysis tools for forestry-related spatial data to explore the current situation of forest biomass spatial pattern,the law of spatial structure changes,and the impact of spatial effects on tree growth.In order to explore the changes in the spatial pattern of Chinese fir plantation and its impact on tree growth,this study focused on the spatial pattern and biomass of Chinese fir plantation.Using geostatistical methods such as spatial structure parameters,spatial autocorrelation,variogram,and Kriging interpolation,the changes of spatial patterns and the influence of spatial effects on the estimation of forest biomass are analyzed from the level of individual trees and stands.Combine relevant estimation models to estimate forest biomass.The conclusions reached are as follows:First,using spatial autocorrelation analysis to study the spatial pattern of Chinese fir plantations changes with age or research scale.In young forests,the trees show a positive spatial autocorrelation,and the size of the trees is not much different from the surrounding neighboring trees.As the age increases,the distribution pattern is random,and the size difference between trees gradually becomes obvious.There is a significant negative spatial autocorrelation in mature forests and over-mature forests.That is,the trees are surrounded by small trees（or small trees are surrounded by large trees）,and the size of the trees in the forest stands is highly different.The size of the trees is uneven.In terms of scale changes,in nearly mature forests,mature forests,and over mature forests,the negative spatial autocorrelation in the scale of 2 to 5 meters may be related to the secondary trees in different forest layers.Second,analyze the effects of spatial patterns on tree growth from the stand and single tree level.At the stand level,clustered patterns usually appear in young forests,while dispersed patterns are common in nearly mature forests and mature forests.The spatial pattern has an important influence on the diameter growth,but it is mainly manifested in the stage of young forest to middle-age forest,and the extent is limited.At the single tree level,from the clumped structure to the regular structure,the proportion of trees does not change with age,but the annual average growth diameter at breast height and the size of the trees in different spatial structures are significantly different,especially in middle-age forests.Third,to explore the impact of spatial effects on biomass estimation a spatial lag model,a spatial error model,and a geographically weighted regression model was constructed.The coefficient of determination for least-squares regression（OLS）that without spatial effects considered is 0.9475,while the spatial error model（SEM）and the geographically weighted regression model（GWR）with the spatial effect considered are 0.9493 and 0.9696,respectively.Considering the spatial effect can improve the prediction accuracy of the biomass model,but the improvement is small.Fourth,stepwise regression,Cubist regression model,and random forest model are used to evaluate the explanatory power of different factors on the components and total above-ground biomass of Chinese fir plantations.In terms of variable selection,Cubist regression gives the role and contribution of each variable.The random forest model can obtain the importance ranking of all explanatory variables.In terms of model estimation,the random forest model has the best fitting accuracy,but the independent estimation level is not much different from the Cubist regression model.In terms of solving spatial effects,the residuals of the stepwise regression model have spatial autocorrelation,and biomass estimation is affected by spatial dependence.The Cubist regression and random forest model residuals did not show spatial autocorrelation,indicating that the spatial effect is explained and the estimation of biomass is not affected by the spatial dependence.Fifth,the spatial variation of the above-ground biomass of Chinese fir plantations in Jiangle County is isotropic at the county scale.That is,the biomass per unit area only changes with distance,regardless of direction.The above-ground biomass structure ratio（₀/（₀+））is less than 25%,and it has a medium-strength spatial autocorrelation.The Kriging method was used to estimate above-ground biomass.The coefficient of determination is increased from 0.2 of ordinary kriging to 0.8 of co-kriging when using average tree height as a covariate.the accuracy was significantly improved,and the estimation is relatively ideal.