主办单位:中国气象局沈阳大气环境研究所
国际刊号:ISSN 1673-503X
国内刊号:CN 21-1531/P

气象与环境学报 ›› 2024, Vol. 40 ›› Issue (1): 105-112.doi: 10.3969/j.issn.1673-503X.2024.01.0013

• 快报 • 上一篇    

基于遥感和机器学习的中国典型城市碳排放驱动因子及预测分析

吴宇恒1(),白景昌2   

  1. 1. 中国遥感应用协会, 北京 100094
    2. 北京航天世景信息技术有限公司, 北京 100089
  • 收稿日期:2023-08-24 出版日期:2024-02-28 发布日期:2024-03-25
  • 作者简介:吴宇恒, 女, 1988年生, 工程师, 主要从事卫星遥感应用方面研究, E-mail: 317788194@qq.com
  • 基金资助:
    中国科协咨询团队项目(20220615ZZ08010034)

Analysis and prediction of driving factors for carbon emissions in typical cities of China based on remote sensing data and machine learning method

Yuheng WU1(),Jingchang BAI2   

  1. 1. China Association of Remote Sensing Application, Beijing 100094, China
    2. Beijing Aerospace Shijing Information Technology Co., Ltd, Beijing 100089, China
  • Received:2023-08-24 Online:2024-02-28 Published:2024-03-25

摘要:

采用国家统计数据和遥感夜间灯光数据, 通过对数平均迪氏指数(LMDI)对15个典型城市的碳排放影响因素进行贡献率分析, 并构建了3组变量用于机器学习ridge和Lasso回归模型的预测分析。结果表明: 城市生产总值、能源消费量(ES)、人口(P)、房地产开发施工面积(RECA)、夜间灯光强度(NL)、货物运输量(CT)和旅客运输量(PT)等7个因子对CO2排放起促进作用, 能源消费结构(EI)和第三产业占比(TIR)对CO2排放起抑制作用。城市的成熟度越高, 产业越丰富, 则碳排放的影响因子越多样。回归预测模型ridge和Lasso在变量组合1至组合3中模拟结果与测试数据集的相关系数均为0.8以上, 其中组合1结果最好, 其次是组合2, 最后是组合3。

关键词: 遥感, 机器学习, 碳排放, 影响因子

Abstract:

Utilizing national statistical data and remote sensing nighttime light data, this study performs a contribution rate analysis of the influencing factors on carbon emissions in 15 typical cities through the Logarithmic Mean Divisia Index (LMDI), and constructs three sets of variables for predictive analysis using machine learning Ridge and Lasso regression models. The results indicate that seven factors including urban Gross Domestic Product (GDP), Energy Consumption (ES), Population (P), Real Estate Construction Area (RECA), Nighttime Light Intensity (NL), Cargo Transportation Volume (CT), and Passenger Transportation Volume (PT) play a promoting role in CO2 emissions, whereas Energy Consumption Structure (EI) and the Proportion of Tertiary Industry (TIR) have an inhibiting effect on CO2 emissions. The more mature a city, the richer the industry, the more diverse the impacting factors of carbon emissions. The correlation coefficient exceeds 0.8 between simulated results from predictive models of ridge and Lasso regression across variables set 1 to set 3 and the results from the test datasets. Among them, the result from set 1 is the best, followed by set 2, and finally, set 3.

Key words: Remote sensing, Machine learning, Carbon emissions, Influencing factors

中图分类号: