文献类型：期刊文献

中文题名：基于遥感的北京市森林地上碳储量监测

英文题名：Remote Sensing Based Monitoring of Forest Aboveground Carbon Storage in Beijing

作者：贺晨瑞[1,2] 庞丽峰[1,2] 谭炳香[1,2] 黄逸飞[1,2] 孙学霞[1,2]

第一作者：贺晨瑞

机构：[1]中国林业科学研究院资源信息研究所,北京100091;[2]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：2024

卷号：39

期号：3

起止页码：162-170

中文期刊名：西北林学院学报

外文期刊名：Journal of Northwest Forestry University

收录：CSTPCD;;北大核心:【北大核心2023】；CSCD:【CSCD2023_2024】；

基金：“十四五”重点研发计划项目(2022YFD2200505-03)。

语种：中文

中文关键词：城市森林;碳储量;XGBoost模型;Boruta算法;北京市

外文关键词：urban forest;carbon storage;XGBoost model;Boruta algorithm;Beijing

分类号：S718.56

摘要：城市是CO_(2)排放的主要区域,推动城市碳减排与低碳发展对于早日实现“双碳”战略具有重要帮助。城市森林碳储量是反映城市CO_(2)吸收能力和评估生态系统质量的重要指标。以北京市森林为对象,以Landsat8OLI遥感影像、数字高程和森林资源二类调查数据为数据源,采用逐步回归分析、递归消除算法和Boruta算法进行特征选择,然后采用多元线性回归模型、BP神经网络、随机森林算法以及极端梯度提升算法模型(XGBoost)进行北京市森林AGC模型构建,最后选择效果最好的模型对北京市整体森林AGC进行反演估测。结果表明:1)基于Boruta算法选择特征集进行4种AGC模型构建时,其R 2是最好的,优于SRA与RFE选择方法;2)XGBoost算法构建的森林AGC模型的精度最高,其根据Boruta算法选择特征集得到的训练集、测试集R 2、RMSE、RRMSE分别为0.95、0.69、3.16、5.18、17.70%、21.49%;3)2014年北京市总体森林AGC为8931820.34 t,与实际值差距较小;在空间分布上均呈西北部高、中部及东南部低的现象;密云区、怀柔区及延庆区森林AGC较多,而朝阳区、丰台区及石景山区较少。总体上说,基于Boruta的特征选择与现代集成的XGBoost森林AGC模型有着较好的估测效果。该研究为超大城市森林AGC精准监测提供了技术支撑。
Cities are the main regions for CO_(2) emissions,and promoting urban carbon reduction and low-carbon development is of great help in achieving"the dual carbon strategy"(carbon peaking and carbon neutrality)as soon as possible.Urban forest carbon storage is an important indicator reflecting urban CO_(2) absorption capacity and evaluating ecosystem quality.This study focused on the forests in Beijing,using Landsat8OLI remote sensing images,digital elevation,and secondary resource survey data as data sources.Stepwise regression analysis,recursive elimination algorithm,and Boruta algorithm were used for feature selection.Then,multiple linear regression models,BP neural network,random forest algorithm,and extreme gradient boosting algorithm models were used to construct the Beijing Forest AGC(above ground carbon)model.Finally,the most effective model was selected to invert and estimate the overall forest AGC in Beijing.The results showed that 1)when selecting feature sets based on the Boruta algorithm for constructing four AGC models,its R 2 was the best,superior to the two feature selection methods of SRA and RFE.2)The forest AGC model constructed by XGBoost algorithm had the highest accuracy.The values of R2,RMSE,and RRMSE of training and testing sets obtained by selecting feature sets based on Boruta algorithm were 0.95 and 0.69,3.16 and 5.18,17.70%and 21.49%,respectively.3)In 2014,the total forest AGC in Beijing was 8931820.34 tons,which is relatively small compared to the actual value.The spatial distribution showed a phenomenon of high in the northwest,low in the middle and southeast.Miyun District,Huairou District,and Yanqing District had more forest AGC,while Chaoyang District,Fengtai District,and Shijingshan District had fewer.Overall,the feature selection based on Boruta and the modern integrated XGBoost forest AGC model has good estimation performance.This study provides technical support for precise monitoring of forest AGC in mega cities.