文献类型：期刊文献

中文题名：基于机载激光雷达的崇礼冬奥核心区林分地上生物量反演

英文题名：Inversion of Aboveground Biomass in the Core Area of Chongli Winter Olympics Based on Airborne LiDAR

作者：陈星京[1,2] 冯林艳[2,3] 张宇超[4] 刘清旺[2,5] 杨朝晖[1] 符利勇[2,3] 白晋华[1]

第一作者：陈星京

机构：[1]山西农业大学林学院,晋中030801;[2]中国林业科学研究院资源信息研究所,北京100091;[3]国家林业和草原局森林经营与生长模拟重点实验室,北京100091;[4]国家林业和草原局林草调查规划院,北京100714;[5]国家林业和草原局林业遥感与信息技术重点实验室,北京100091

年份：2022

卷号：58

期号：10

起止页码：35-46

中文期刊名：林业科学

外文期刊名：Scientia Silvae Sinicae

收录：CSTPCD;;EI(收录号：20232814385059);Scopus;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：张家口市崇礼区森林防火综合体系建设无人机巡护监测系统(DA2020001);国家自然科学基金面上项目(31971653)。

语种：中文

中文关键词：机载激光雷达;林分地上生物量;混合效应模型;贝叶斯模型

外文关键词：arborne LiDAR;forest aboveground biomass;mixed effect model;Bayesian model

分类号：S771.8;S758.4

摘要：【目的】基于机载激光雷达数据建立结构稳定的林分地上生物量预测模型,考虑最小二乘、混合效应和贝叶斯等参数估计方法对最优生物量预测模型选择进行探讨,为生物量建模方法研究、生物量估测提供科学依据,为冬奥核心区实现“双碳”目标和生物量模型计算提供技术支撑。【方法】基于崇礼冬奥核心区2种森林类型(华北落叶松和白桦)62块实测样地及对应的激光雷达数据,通过变量筛选分别建立最小二乘、混合效应和贝叶斯生物量模型,应用确定系数(R^(2))、均方根误差(RMSE)、残差、总体相对误差(TRE)评价模型,采用留一交叉法验证模型精度。【结果】筛选出相关性较高的激光雷达变量共20个,最终进入模型的自变量3个。拟合效果最好的是Logistic混合效应模型(RMSE=22.99 t·hm^(-2),R^(2)=0.768,TRE=6.08%),分树种建立模型后华北落叶松模型拟合效果提升(RMSE=22.92 t·hm^(-2),R^(2)=0.795,TRE=7.45%),白桦模型预测精度提高(RMSE=23.34 t·hm^(-2),R^(2)=0.440,TRE=4.35%)。利用训练好的模型预测崇礼冬奥核心区生物量并制图。【结论】基于机载激光雷达和地面实测数据估测林分地上生物量,非线性模型优于线性模型;以龄组为随机效应的非线性混合效应模型预测精度最高;贝叶斯估计受先验条件影响较大,本研究样本量偏少,贝叶斯估计具有进一步探讨的价值。
【Object】This study was implemented to develop the stand aboveground biomass model with stable structure based on light detection and ranging(LiDAR)data,considering the ordinary least squares,mixed effects and Bayesian parameter estimation method,and then discussed the selection of the optimal biomass prediction model with the aims to provide a scientific basis for biomass modeling method and biomass estimation and to provide a technical support for achieving the“double carbon”target and biomass model calculation in the core area of the Winter Olympics.【Method】Based on the LiDAR data and field measurement of 62 sample plots distributed across the Larix principis-rupprechtii and Betula platyphylla forest stands in the core areas of the Winter Olympics,ordinary least squares(OLS),mixed effects and Bayesian biomass models were established by variable screening.Determination coefficient(R^(2)),root mean square error(RMSE),residual error and total relative error(TRE)were used to evaluate the model,and reserve-one crossover method was used to verify the accuracies of the models.【Result】A total of 20 LiDAR variables with high correlations were filtered out,and 3 independent variables were finally entered into the models.The best fitting was the Logistic mixed effect model(RMSE=22.99 t·hm^(-2),R^(2)=0.768,TRE=6.08%).After establishing the model by tree species,the accuracy of larch model was improved(RMSE=22.92 t·hm^(-2),R^(2)=0.795,TRE=7.45%),and the accuracy of birch model decreased(RMSE=23.34 t·hm^(-2),R^(2)=0.440,TRE=4.35%).Using the trained model,the biomass of Chongli Winter Olympic core area was predicted and mapped.【Conclusion】As for the estimation of the stand aboveground biomass based on the LiDAR and field measurement data,the nonlinear model was superior to the linear model.The nonlinear mixed effect model with age group as random effects might have the highest prediction accuracy of biomass.Bayesian estimation may be greatly affected by prior conditions and might have further discussion values although with a small sample size in this study.