干形数据采样方法对树干削度方程构建的影响

Influence of different sampling methods for stem data on the construction of taper equations

【目的】基于落叶松15个点的直径-树高干形数据,比较树干不同采样方法对曾伟生(1997)、Bi(2000)和Max&Burkhart(1976)削度模型预测精度的影响。【方法】以大兴安岭343株落叶松干形数据为研究对象,设计了30种数据采样方法,使用留一交叉验证法对基于原始数据和不同采样数据的模型进行检验,选用确定系数(R2)、平均绝对误差(MAB)、均方根误差(RMSE)和相对误差(MPB)作为检验的评价指标,并求出以上4个指标的平均相对排序值。使用Tukey多重比较法对基于不同采样数据的模型进行成对比较,分析当拟合数据改变的情况下各削度模型在预测落叶松直径时是否有显著差异。最后为了直观地表示不同采样数据对削度模型的影响,从落叶松数据中分别随机抽取一株大树和一株小树进行树干模拟。【结果】1)相对于使用原始数据时的模型,基于一部分采样方法的模型检验精度有略微提高,其中使用第27种方法时,曾伟生模型的精度最高;使用第26种方法时,Bi的模型精度最高;使用第9种方法时,Max&Burkhart的模型精度最高;2)Tukey多重比较结果表明,基于原始数据和基于3种最优采样方法的削度模型在预测落叶松直径时,模型两两之间均没有显著差异;3)通过对落叶松小树和大树的树干进行模拟发现,使用原始数据时的削度模型,与使用最优采样数据时的削度模型,对树干的模拟效果几乎相同。【结论】针对不同的削度模型,其最适用的采样方法也不同。当使用曾伟生的模型对大兴安岭落叶松树干削度预测时可以选择方法 27中7个点的直径-树高数据进行拟合,即将树干分成7个区分段进行干形测量;当使用Bi的模型时可以选择方法 26中的7个区分段进行干形测量;当使用Max&Burkhart的模型时可以选择方法 9中的10个区分段进行干形测量;当同时考虑这3个模型时,可以选择方法 20中的8个区分段进行干形测量,此时3个模型对落叶松树干不同位置处直径的预测精度均会略有提高。

【Objective】The influences of different sampling methods on the prediction accuracy of Zeng(1997),Bi(2000)and Max&Burkhart(1976)taper models were compared based on the diameter-tree height stem taper data at 15 points of larch.【Method】The stem taper profile data of 343 Larix gmelinii in Daxing’an Mountains were studied,30 kinds of data sampling methods were designed,and the models based on the original data and different sampling data were tested by using the leave-one-out cross-validation method.The determination coefficient(R2),mean absolute bias(MAB),root mean square error(RMSE)and mean percentage of bias(MPB)were selected as the evaluation indexes of the test,and the average relative ordering values of the above four indexes were calculated.The models were compared in pairs by using Tukey multiple comparison method to analyze whether there was significant difference in predicting larch diameter of each taper model under the condition of changing fitting data.Finally,a large tree and a small tree were randomly selected from larch data for trunk simulation in order to directly represent the impact of different sampling data on the taper model.【Result】1)Compared with the model using original data,the test accuracy of model based on partial sampling method was slightly improved,when the 27th method was used,Zeng(1997)had the highest accuracy,and when the 26th method was used,Bi(2000)had the highest accuracy,and when the 9th method was used,Max&Burkhart(1976)had the highest accuracy;2)Tukey multiple comparison results showed that there was no significant difference between the original data and the three optimal sampling methods in predicting larch diameter;3)Through the simulation of larch small trees and large tree trunks,it was found that the taper model with the original data was almost the same as the taper model with the optimal sampling data.【Conclusion】For different taper models,the most applicable sampling methods are also different.When Zeng(1997)model was used to predict the stem taper of larch in Daxing’an Mountains,the diameter-tree height data of 7 points in method 27 could be selected for fitting,that is,the tree trunk could be divided into 7 segments for stem taper measurement.When Bi(2000)model was used,7 segments in Method 26 can be selected for stem measurement.When Max&Burkhart(1976)model was used,10 segments in method 9 can be selected for stem measurement.When the three models are considered at the same time,8 segments in method 20 can be selected to measure the stem,and the prediction accuracy of the three models for the diameter at different positions of the larch trunk will be slightly improved.