基于XGBoost-ANN的城市绿地净碳交换模拟与特征响应
|
摘要
为分析城市绿地净生态系统碳交换(Net ecosystem exchange,NEE)对环境因子的响应,利用涡度相关法测量了2013—2016年生长季白天的NEE数据,使用XGBoost以及ANN模型对NEE进行模拟和分析,并通过决定系数(R~2)、平均绝对误差(MAE)、均方根误差(RMSE)和一致性系数(IA) 4个指标评价模拟精度。结果表明,当输入因子为光合有效辐射(PAR)、饱和水汽压差(VPD)、空气温度(Ta)、相对湿度(RH)、土壤温度(Ts)、风速(WS)、10 cm处土壤含水率(VWC10)时,模拟效果达到最优。其训练集精度R~2为0. 712,RMSE为4. 394μmol/(m~2·s),MAE为3. 129μmol/(m~2·s),IA为0. 911;测试集精度R~2为0. 748,RMSE为4. 253μmol/(m~2·s),MAE为2. 971μmol/(m~2·s),IA为0. 920。在考虑因子间相互作用后,环境因子对NEE的重要性排序从大到小依次为PAR、VPD、Ta、RH、Ts、WS、VWC10;就单环境因子而言,对NEE的重要性由大到小依次为Ta、Ts、RH。通过计算生态系统净生产力(Net ecosystem productivity,NEP,即-NEE)对主要环境因子(PAR、VPD、Ta)的偏导数可知,生态系统光合作用表观量子效率最大值为0. 087,并且当PAR大于1 200μmol/(m~2·s)时,其不再是影响光合作用的主要因素; VPD偏导数的变化趋势表明,VPD对植物光合作用的影响以抑制性为主,当VPD过大时,偏导数趋近于0,此时植物叶片气孔闭合,抑制光合作用; Ta偏导数的变化趋势说明,随着温度的升高,光合作用速率逐渐大于呼吸作用的速率。研究表明,基于XGBoost与ANN模型能够更为精确地模拟NEE动态,在相关环境因子中,PAR、VPD、Ta是影响NEE变化的主导因子,NEE对主要影响因子的生态特征响应趋势可为理解碳循环关键过程提供参考。
Aiming to analyze the responses of urban green-land's net ecosystem exchange( NEE) to the climatic controls and provide theoretical and technical support for carbon cycle simulation between land and atmosphere. In growing season,half-hourly daytime NEE based on eddy covariance flux data collected from 2013 to 2016 were simulated by XGBoost and back propagation artificial neural network( ANN) model. Moreover,the accuracy of model was evaluated by using the coefficient of determination( R~2),root mean square error( RMSE),mean absolute error( MAE) and index of agreement( IA). The experimental results showed that ANN model presented that seven input variables( photosynthetically active radiation( PAR),vapor pressure deficit( VPD),air temperature( Ta),relative humidity( RH),soil temperature( Ts),wind speed( WS) and volumetric water content at 10 cm depth) performed best,yielding R~2 of 0. 712,RMSE of 4. 394 μmol/( m~2·s),MAE of 3. 129 μmol/( m~2·s) and IA of 0. 911 on train dataset,and R~2 of 0. 748,RMSE of 4. 253 μmol/( m~2·s),MAE of 2. 971 μmol/( m~2·s) and IA of0. 920 on test dataset. After considering the function and interaction among the factors,the importance score of each environmental factor was decreased in the following order: PAR,VPD,Ta,RH,Ts,WS and VWC10,otherwise Ts would be more important than RH. In particularly,after calculating the numerical partial derivatives of main climatic controls for each half-hourly point,the numerical partial derivatives of PAR showed the ecosystem quantum yield with the value of 0. 087,and it also indicated that PAR was no longer a main impact factor when value was greater than 1 200 μmol/( m~2·s). Besides,the numerical partial derivatives of VPD expressed that VPD could mainly inhibit the photosynthesis,and the higher VPD aggravated the inhibition of photosynthesis by affecting photosynthetic rate. Furthermore,the numerical partial derivatives of Ta demonstrated that the photosynthetic rate was increased bit by bit and made the photosynthetic rate overpass respiration rate gradually. According to the result,PAR,VPD and Ta played an important role in controlling the NEE of urban green-land ecosystem. Also,XGBoost and ANN could be capable in capturing NEE dynamics and simulating the NEE with high accuracy.Meanwhile,the present result provided instant insight in underlying ecosystem physiology.
Aiming to analyze the responses of urban green-land's net ecosystem exchange( NEE) to the climatic controls and provide theoretical and technical support for carbon cycle simulation between land and atmosphere. In growing season,half-hourly daytime NEE based on eddy covariance flux data collected from 2013 to 2016 were simulated by XGBoost and back propagation artificial neural network( ANN) model. Moreover,the accuracy of model was evaluated by using the coefficient of determination( R~2),root mean square error( RMSE),mean absolute error( MAE) and index of agreement( IA). The experimental results showed that ANN model presented that seven input variables( photosynthetically active radiation( PAR),vapor pressure deficit( VPD),air temperature( Ta),relative humidity( RH),soil temperature( Ts),wind speed( WS) and volumetric water content at 10 cm depth) performed best,yielding R~2 of 0. 712,RMSE of 4. 394 μmol/( m~2·s),MAE of 3. 129 μmol/( m~2·s) and IA of 0. 911 on train dataset,and R~2 of 0. 748,RMSE of 4. 253 μmol/( m~2·s),MAE of 2. 971 μmol/( m~2·s) and IA of0. 920 on test dataset. After considering the function and interaction among the factors,the importance score of each environmental factor was decreased in the following order: PAR,VPD,Ta,RH,Ts,WS and VWC10,otherwise Ts would be more important than RH. In particularly,after calculating the numerical partial derivatives of main climatic controls for each half-hourly point,the numerical partial derivatives of PAR showed the ecosystem quantum yield with the value of 0. 087,and it also indicated that PAR was no longer a main impact factor when value was greater than 1 200 μmol/( m~2·s). Besides,the numerical partial derivatives of VPD expressed that VPD could mainly inhibit the photosynthesis,and the higher VPD aggravated the inhibition of photosynthesis by affecting photosynthetic rate. Furthermore,the numerical partial derivatives of Ta demonstrated that the photosynthetic rate was increased bit by bit and made the photosynthetic rate overpass respiration rate gradually. According to the result,PAR,VPD and Ta played an important role in controlling the NEE of urban green-land ecosystem. Also,XGBoost and ANN could be capable in capturing NEE dynamics and simulating the NEE with high accuracy.Meanwhile,the present result provided instant insight in underlying ecosystem physiology.
引文
[1]北京统计局.北京统计年鉴[J].北京:中国统计出版社,2017.
[2]LONGDOZ B,YERNAUX M,AUBINET M.Soil CO2efflux measurements in a mixed forest:impact of chamber disturbances,spatial variability and seasonal evolution[J].Global Change Biology,2000,6(8):907-917.
[3]MOFFAT A M,PAPALE D,REICHSTEIN M,et al.Comprehensive comparison of gap-filling techniques for eddy covariance net carbon fluxes[J].Agricultural&Forest Meteorology,2007,147(3-4):209-232.
[4]王宏莹.基于涡度观测和遥感技术的城市碳源/汇研究[D].徐州:中国矿业大学,2016.WANG Hongying.Study of urban carbon sources/sinks based on eddy covariance technology and remote sensing[D].Xuzhou:China University of Mining and Technology,2016.(in Chinese)
[5]MENZER O,MEIRING W,KYRIAKIDIS P C,et al.Annual sums of carbon dioxide exchange over a heterogeneous urban landscape through machine learning based gap-filling[J].Atmospheric Environment,2015,101:312-327.
[6]CHEN T,GUESTRIN C.XGBoost:a scalable tree boosting system[C]∥Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery&Data Mining.San Francisco:ACM,2016:785-794.
[7]赵俊日,肖昕,吴涛,等.空气质量数值预报优化方法研究[J].中国环境科学,2018,38(6):2047-2054.ZHAO Junri,XIAO Xin,WU Tao,et al.A revised approach to air quality forecast based on Models-3/CMAQ[J].China Enviromental Science,2018,38(6):2047-2054.(in Chinese)
[8]陈文婧.城市绿地生态系统碳水通量研究[D].北京:北京林业大学,2013.CHEN Wenjing.Carbon and water fluxes of urban green-land ecosystem-case study of Beijing Olympic Forest Park[D].Beijing:Beijing Forestry University,2013.(in Chinese)
[9]郭智明.长沙市城市生态系统碳通量时间变化与影响因子研究[D].长沙:中南林业科技大学,2015.GUO Zhiming.The study on temporal variation and impact factors of carbon flux between urban ecosystem and atmosphere in Changsha[D].Changsha:Central South University of Forestry and Technology,2015.(in Chinese)
[10]陈文婧,李春义,何桂梅,等.北京奥林匹克森林公园绿地碳交换动态及其环境控制因子[J].生态学报,2013,33(20):6712-6720.CHEN Wenjing,LI Chunyi,HE Guimei,et al.Dynamics of CO2exchange and its environmental controls in an urban greenland ecosystem in Beijing Olympic Forest Park[J].Acta Ecologica Sinica,2013,33(20):6712-6720.(in Chinese)
[11]PAPALE D,REICHSTEIN M,AUBINET M,et al.Towards a standardized processing of net ecosystem exchange measured with eddy covariance technique:algorithms and uncertainty estimation[J].Biogeosciences,2006,3(4):571-583.
[12]KUNWOR S,STARR G,LOESCHER H W,et al.Preserving the variance in imputed eddy-covariance measurements:alternative methods for defensible gap filling[J].Agricultural&Forest Meteorology,2017,232:635-649.
[13]MOFFAT A M,BECKSTEIN C,CHURKINA G,et al.Characterization of ecosystem responses to climatic controls using artificial neural networks[J].Global Change Biology,2010,16(10):2737-2749.
[14]窦贤明.机器学习方法在陆地生态系统碳水通量模拟中的应用研究[D].徐州:中国矿业大学,2018.DOU Xianming.Applications of machine learning methods in modeling carbon and water fluxes of terrestrial ecosystems[D].Xuzhou:China University of Mining and Technology,2018.(in Chinese)
[15]JING X,XIN J,HE G,et al.Environmental control over seasonal variation in carbon fluxes of an urban temperate forest ecosystem[J].Landscape&Urban Planning,2015,142:63-70.
[16]张红梅,吴炳方,闫娜娜.饱和水汽压差的卫星遥感研究综述[J].地球科学进展,2014,29(5):559-568.ZHANG Hongmei,WU Bingfang,YAN Nana.Remote sensing estimates of vapor pressure deficit:an overview[J].Advances in Earth Science,2014,29(5):559-568.(in Chinese)
[17]李清毅,周昊,林阿平,等.基于网格搜索和支持向量机的灰熔点预测[J].浙江大学学报(工学版),2011,45(12):2181-2187.LI Qingyi,ZHOU Hao,LIN Aping,et al.Prediction of ash fusion temperature based on grid search and support vector machine[J].Journal of Zhejiang University(Engineering Science),2011,45(12):2181-2187.(in Chinese)
[18]陈元鹏,罗明,彭军还,等.基于网格搜索随机森林算法的工矿复垦区土地利用分类[J].农业工程学报,2017,33(14):250-257.CHEN Yuanpeng,LUO Ming,PENG Junhai,et al.Classification of land use in industrial and mining reclamation area based grid-search and random forest classifier[J].Transactions of the CSAE,2017,33(14):250-257.(in Chinese)
[19]李翱翔,陈健.BP神经网络参数改进方法综述[J].电子科技,2007(2):79-82.LI Aoxiang,CHEN Jian.Summarize of parameter improve methods for BP neural network[J].Electronic Science and Technology,2007(2):79-82.(in Chinese)
[20]MOFFAT A M.A new methodology to interpret high resolution measurements of net carbon fluxes between terrestrial ecosystems and the atmosphere[D].Berlin:Friedrich-Schiller-Universitt,2012.
[21]张诗青,王建伟,郑文龙.中国交通运输碳排放及影响因素时空差异分析[J].环境科学学报,2017,37(12):4787-4797.ZHANG Shiqing,WANG Jianwei,ZHENG Wenlong.Spatio-temporal difference of transportation carbon emission and its influencing factors in China[J].Acta Scientiae Circumstantiae,2017,37(12):4787-4797.(in Chinese)
[22]QUINLAN J R.Induction of decision trees[J].Machine Learning,1986,1(1):81-106.
[23]WEN X,ZHAO Z,DENG X,et al.Applying an artificial neural network to simulate and predict Chinese fir(Cunninghamia lanceolata)plantation carbon flux in subtropical China[J].Ecological Modelling,2014,294:19-26.
[24]JIA X,ZHA T,GONG J,et al.Multi-scale dynamics and environmental controls on net ecosystem CO2exchange over a temperate semiarid shrubland[J].Agricultural and Forest Meteorology,2018,259:250-259.
[25]唐祥,陈文婧,李春义,等.北京八达岭林场人工林净碳交换及其环境影响因子[J].应用生态学报,2013,24(11):3057-3064.TANG Xiang,CHEN Wenjing,LI Chunyi,et al.Net carbon exchange and its environmental affecting factors in a forest plantation in Badaling,Beijing of China[J].Chinese Journal of Applied Ecology,2013,24(11):3057-3064.(in Chinese)
[26]ICHII K,UEYAMA M,KONDO M,et al.New data-driven estimation of terrestrial CO2fluxes in asia using a standardized database of eddy covariance measurements,remote sensing data,and support vector regression[J].Journal of Geophysical Research-Biogeosciences,2017,122(4):767-795.
[27]姜海梅,张德广,王若静,等.不同生态系统呼吸模型在半干旱草原生长季碳循环研究中的比较及应用[J].北京大学学报(自然科学版),2018,54(3):593-604.JIANG Haimei,ZHANG Deguang,WANG Ruojing,et al.Comparison of different ecosystem respiration models and its application in carbon cycle research over semi-arid grassland during growing season[J].Acta Scientiarum Naturalium Universitatis Pekinensis,2018,54(3):593-604.(in Chinese)
[2]LONGDOZ B,YERNAUX M,AUBINET M.Soil CO2efflux measurements in a mixed forest:impact of chamber disturbances,spatial variability and seasonal evolution[J].Global Change Biology,2000,6(8):907-917.
[3]MOFFAT A M,PAPALE D,REICHSTEIN M,et al.Comprehensive comparison of gap-filling techniques for eddy covariance net carbon fluxes[J].Agricultural&Forest Meteorology,2007,147(3-4):209-232.
[4]王宏莹.基于涡度观测和遥感技术的城市碳源/汇研究[D].徐州:中国矿业大学,2016.WANG Hongying.Study of urban carbon sources/sinks based on eddy covariance technology and remote sensing[D].Xuzhou:China University of Mining and Technology,2016.(in Chinese)
[5]MENZER O,MEIRING W,KYRIAKIDIS P C,et al.Annual sums of carbon dioxide exchange over a heterogeneous urban landscape through machine learning based gap-filling[J].Atmospheric Environment,2015,101:312-327.
[6]CHEN T,GUESTRIN C.XGBoost:a scalable tree boosting system[C]∥Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery&Data Mining.San Francisco:ACM,2016:785-794.
[7]赵俊日,肖昕,吴涛,等.空气质量数值预报优化方法研究[J].中国环境科学,2018,38(6):2047-2054.ZHAO Junri,XIAO Xin,WU Tao,et al.A revised approach to air quality forecast based on Models-3/CMAQ[J].China Enviromental Science,2018,38(6):2047-2054.(in Chinese)
[8]陈文婧.城市绿地生态系统碳水通量研究[D].北京:北京林业大学,2013.CHEN Wenjing.Carbon and water fluxes of urban green-land ecosystem-case study of Beijing Olympic Forest Park[D].Beijing:Beijing Forestry University,2013.(in Chinese)
[9]郭智明.长沙市城市生态系统碳通量时间变化与影响因子研究[D].长沙:中南林业科技大学,2015.GUO Zhiming.The study on temporal variation and impact factors of carbon flux between urban ecosystem and atmosphere in Changsha[D].Changsha:Central South University of Forestry and Technology,2015.(in Chinese)
[10]陈文婧,李春义,何桂梅,等.北京奥林匹克森林公园绿地碳交换动态及其环境控制因子[J].生态学报,2013,33(20):6712-6720.CHEN Wenjing,LI Chunyi,HE Guimei,et al.Dynamics of CO2exchange and its environmental controls in an urban greenland ecosystem in Beijing Olympic Forest Park[J].Acta Ecologica Sinica,2013,33(20):6712-6720.(in Chinese)
[11]PAPALE D,REICHSTEIN M,AUBINET M,et al.Towards a standardized processing of net ecosystem exchange measured with eddy covariance technique:algorithms and uncertainty estimation[J].Biogeosciences,2006,3(4):571-583.
[12]KUNWOR S,STARR G,LOESCHER H W,et al.Preserving the variance in imputed eddy-covariance measurements:alternative methods for defensible gap filling[J].Agricultural&Forest Meteorology,2017,232:635-649.
[13]MOFFAT A M,BECKSTEIN C,CHURKINA G,et al.Characterization of ecosystem responses to climatic controls using artificial neural networks[J].Global Change Biology,2010,16(10):2737-2749.
[14]窦贤明.机器学习方法在陆地生态系统碳水通量模拟中的应用研究[D].徐州:中国矿业大学,2018.DOU Xianming.Applications of machine learning methods in modeling carbon and water fluxes of terrestrial ecosystems[D].Xuzhou:China University of Mining and Technology,2018.(in Chinese)
[15]JING X,XIN J,HE G,et al.Environmental control over seasonal variation in carbon fluxes of an urban temperate forest ecosystem[J].Landscape&Urban Planning,2015,142:63-70.
[16]张红梅,吴炳方,闫娜娜.饱和水汽压差的卫星遥感研究综述[J].地球科学进展,2014,29(5):559-568.ZHANG Hongmei,WU Bingfang,YAN Nana.Remote sensing estimates of vapor pressure deficit:an overview[J].Advances in Earth Science,2014,29(5):559-568.(in Chinese)
[17]李清毅,周昊,林阿平,等.基于网格搜索和支持向量机的灰熔点预测[J].浙江大学学报(工学版),2011,45(12):2181-2187.LI Qingyi,ZHOU Hao,LIN Aping,et al.Prediction of ash fusion temperature based on grid search and support vector machine[J].Journal of Zhejiang University(Engineering Science),2011,45(12):2181-2187.(in Chinese)
[18]陈元鹏,罗明,彭军还,等.基于网格搜索随机森林算法的工矿复垦区土地利用分类[J].农业工程学报,2017,33(14):250-257.CHEN Yuanpeng,LUO Ming,PENG Junhai,et al.Classification of land use in industrial and mining reclamation area based grid-search and random forest classifier[J].Transactions of the CSAE,2017,33(14):250-257.(in Chinese)
[19]李翱翔,陈健.BP神经网络参数改进方法综述[J].电子科技,2007(2):79-82.LI Aoxiang,CHEN Jian.Summarize of parameter improve methods for BP neural network[J].Electronic Science and Technology,2007(2):79-82.(in Chinese)
[20]MOFFAT A M.A new methodology to interpret high resolution measurements of net carbon fluxes between terrestrial ecosystems and the atmosphere[D].Berlin:Friedrich-Schiller-Universitt,2012.
[21]张诗青,王建伟,郑文龙.中国交通运输碳排放及影响因素时空差异分析[J].环境科学学报,2017,37(12):4787-4797.ZHANG Shiqing,WANG Jianwei,ZHENG Wenlong.Spatio-temporal difference of transportation carbon emission and its influencing factors in China[J].Acta Scientiae Circumstantiae,2017,37(12):4787-4797.(in Chinese)
[22]QUINLAN J R.Induction of decision trees[J].Machine Learning,1986,1(1):81-106.
[23]WEN X,ZHAO Z,DENG X,et al.Applying an artificial neural network to simulate and predict Chinese fir(Cunninghamia lanceolata)plantation carbon flux in subtropical China[J].Ecological Modelling,2014,294:19-26.
[24]JIA X,ZHA T,GONG J,et al.Multi-scale dynamics and environmental controls on net ecosystem CO2exchange over a temperate semiarid shrubland[J].Agricultural and Forest Meteorology,2018,259:250-259.
[25]唐祥,陈文婧,李春义,等.北京八达岭林场人工林净碳交换及其环境影响因子[J].应用生态学报,2013,24(11):3057-3064.TANG Xiang,CHEN Wenjing,LI Chunyi,et al.Net carbon exchange and its environmental affecting factors in a forest plantation in Badaling,Beijing of China[J].Chinese Journal of Applied Ecology,2013,24(11):3057-3064.(in Chinese)
[26]ICHII K,UEYAMA M,KONDO M,et al.New data-driven estimation of terrestrial CO2fluxes in asia using a standardized database of eddy covariance measurements,remote sensing data,and support vector regression[J].Journal of Geophysical Research-Biogeosciences,2017,122(4):767-795.
[27]姜海梅,张德广,王若静,等.不同生态系统呼吸模型在半干旱草原生长季碳循环研究中的比较及应用[J].北京大学学报(自然科学版),2018,54(3):593-604.JIANG Haimei,ZHANG Deguang,WANG Ruojing,et al.Comparison of different ecosystem respiration models and its application in carbon cycle research over semi-arid grassland during growing season[J].Acta Scientiarum Naturalium Universitatis Pekinensis,2018,54(3):593-604.(in Chinese)
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |