论文部分内容阅读
回归模型可用于预测森林生态系统地上生物量,其中最为常用的是最小二乘回归模型.在预测灌木,尤其是多茎灌木的地上生物量时,最小二乘法与贝叶斯方法的比较很少被研究.我们开发了小叶锦鸡儿(CaraganamicrophyllaLam.)生物量预测模型.小叶锦鸡儿是科尔沁沙地广泛分布的多茎灌木,对减少风蚀、固定沙丘具有重要作用.本研究建立6种表征生物量的异速增长模型,并基于统计标准选择在预测生物量方面表现最佳的1种,然后分别用最小二乘法与贝叶斯方法对模型中的参数进行估计.参数估计过程中用自助法考察样本量大小的影响,同时区分测试集与训练集.最后,我们比较了最小二乘法与贝叶斯方法在小叶锦鸡儿地上生物量预测中的表现.异速增长的6个模型均达到显著水平,其中幂指数为1的模型表现最佳.研究结果表明,采用无先验信息与有先验信息的贝叶斯方法进行估计,得到的均方误差在测试集上低于最小二乘法.另外,基径作为预测变量在最小二乘法与贝叶斯方法中均不显著,表明在生物量预测模型中应谨慎选择合适变量.本研究强调贝叶斯方法、自助法和异速增长模型相结合能够提升沙地灌木生物量预测模型的准确度.“,”Aims In forest ecosystems,different types of regression models have been frequently used for the estimation of aboveground biomass,where Ordinary Least Squares(OLS)regression models are the most common prediction models.Yet,the relative performance of Bayesian and OLS models in predicting aboveground biomass of shrubs,especially multi-stem shrubs,has relatively been less studied in forests.Methods In this study,we developed the biomass prediction models for Caragana microphylla Lam.which is a widely distributed multi-stems shrub,and contributes to the decrease of wind erosion and the fixation of sand dunes in the Horqin Sand Land,one of the largest sand lands in China.We developed six types of formulations under the framework of the regression models,and then,selected the best model based on specific criteria.Consequently,we estimated the parameters of the best model with OLS and Bayesian methods with training and test data under different sample sizes with the bootstrap method.Lastly,we compared the performance of the OLS and Bayesian models in predicting the aboveground biomass of C.microphylla.Important Findings The performance of the allometric equation(power = 1)was best among six types of equations,even though all of those models were significant.The results showed that mean squared error of test data with non-informative prior Bayesian method and the informative prior Bayesian method was lower than with the OLS method.Among the tested predictors(i.e.plant height and basal diameter),we found that basal diameter was not a significant predictor either in OLS or Bayesian methods,indicating that suitable predictors and well-fitted models should be seriously considered.This study highlights that Bayesian methods,the bootstrap method and the type of allometric equation could help to improve the model accuracy in predicting shrub biomass in sandy lands.