Effects Of Different Sampling Methods About Modeling Data On Predict Precision Of Individual Tree Volume For Dahurian Larch

In this study,different sampling methods(uniform,normal,right and left skewed distribution)are compared for individual tree volume equation for Larix gmelinii.Based on different sampling methods,prediction precision of volume equations is studied.Data with different distributions were sampled using simple random sampling(SRS)of proc surveyselect module and conditional statements in SAS.To reduce the uncertainty of data sampling,each distribution sampling process is repeated three times.Shapiro-Wilk method is used for normality test.Allometric model is fitted using GNLS in S-PLUS.Variance functions(exponential function,power function and constant plus power function)were incorporated into generalized allometric model to reduce heteroscedasticity.The variance function variable adds the combined variable when considering the volume observation value(V),the predicted value((?)),the DBH(D),and the binary modelD~2H.Coefficient determination(R~2),root mean square error(RMSE),mean absolute bias(MAB),and mean percentage of bias(MPB)were employed to evaluate the precision of different individual volume models.In order to reduce the uncertainty of the small sample,the stochastic method is used to generate random data of uniform,normal,right deviation and left partial distribution in the univariate model,and the comparison is made by using the confidence interval of the predicted value.The results show that the exponential function,the power function and the constant power function can eliminate the influence of the heteroskedasticity of the four kinds of standing timber product equations.The results show that the RMSE of the normal,right and left partial models decreases with respect to the uniform model,especially the left model decreases.Stochastic simulation shows that in the unary model,regardless of which distribution,the confidence interval width decreases with the increase of the number of samples.When the number of samples is less than 1000,the prediction accuracy based on the left partial volume model is better than the other volume model.When the number of samples is greater than 1000,the prediction accuracy of the four distribution models is very close.When the number of samples reaches 10000,the prediction accuracy of the four distribution models is basically the same.Under the overall test,the left model has the highest accuracy in the univariate model,and the normal model in the binary model is relatively optimal.