针对不同陆地生态系统中净生态系统CO_(2)交换量(Net ecosystem exchange,NEE)数据的长期连续测量中存在的数据差异问题,以中国气象局青海高寒生态气象野外科学试验基地野牛沟试验站为研究对象,利用涡动协方差技术获取高寒湿地生态系统...针对不同陆地生态系统中净生态系统CO_(2)交换量(Net ecosystem exchange,NEE)数据的长期连续测量中存在的数据差异问题,以中国气象局青海高寒生态气象野外科学试验基地野牛沟试验站为研究对象,利用涡动协方差技术获取高寒湿地生态系统水平上的NEE数据。通过对比机器学习算法和通量数据后处理算法(Reddyproc)两种数据填充方法,提出了一种结合机器学习与时序异常检测(Time series anomaly detection,TAD)的新框架,用于NEE数据的空白填补。研究结果表明:1)Reddyproc算法在剔除异常值后,NEE插补决定系数(R^(2))达到0.67,数据离散度显著降低,数据质量提升;2)八种机器学习模型中,随机森林(Random Forest,RF)模型表现最优,其决定系数(Coefficient of determination,R^(2))为0.63,均方根误差(Root mean square error,RMSE)为2.17μmol s^(-1)m^(-2),且经过时序异常检测后,估算精度提升了17%;3)Reddyproc和RF估算的CO_(2)通量存在季节性差异,冷季(1—3月和10—12月)Reddyproc估算值低于RF,而暖季(4—9月)则高于RF,表明冬季Reddyproc低估了CO_(2)释放,夏季则低估了CO_(2)吸收。该新框架有效解决了数据采集不确定性和缺失导致的二氧化碳通量计算准确率问题,为研究高寒湿地生态系统的碳固持能力、对气候变化的响应以及极端事件的影响提供了关键数据支持。未来研究应进一步探索新方法的适用性、改进和优化方向,以实现更准确、可靠且适用于不同生态系统的填补模型,为生态系统建模和预测提供强大工具。展开更多
Alpine meadows,alpine wetlands,and alpine desert steppes are the three typical vegetation types on the Qinghai-Tibet Plateau.The complex terrain and harsh climatic conditions across this region lead to considerable di...Alpine meadows,alpine wetlands,and alpine desert steppes are the three typical vegetation types on the Qinghai-Tibet Plateau.The complex terrain and harsh climatic conditions across this region lead to considerable diversification in the vegetation growth environment,resulting in substantial spatial heterogeneity in ecosystem carbon flux and its controlling mechanisms.Using eddy covariance data collected from March to August 2019,this study examined the responses of carbon and water fluxes in different ecosystems on the Tibetan Plateau to typical hydrometeorological factors,focusing on Net Ecosystem CO□Exchange(NEE)and Evapotranspiration(ET).The results indicate that:1)The Longbao alpine wetland primarily acted as a carbon sink from May to August,while serving as a carbon source from March to April.In the Maqin alpine meadow,it functioned as a carbon sink during June and July but acted as a carbon source in March,April,May,and August.The Tuotuohe alpine desert strppe was predominantly a net carbon sink from March to August.Overall,after the entire growing season(March to August),the Longbao alpine wetlands,Maqin alpine meadow,and Tuotuohe alpine desert steppe all showed net carbon sink properties,with net CO_(2)uptakes of 236.12 g/m^(2),291.45 g/m^(2),and 290.28 g/m^(2),respectively.2)The importance of meteorological factors to NEE varies with scale and ecosystem type,with global radiation(Rg)being the most critical factor influencing NEE variation.Volumetric soil water content(Soil_VWC)and soil temperature(Soil_T)had a positive effect on NEE at Maqin alpine meadow and Tuotuohe alpine desert steppe,while higher values of these variables showed a negative contribution.Furthermore,the sensitivity of NEE to Soil_T at Longbao alpine wetland and Tuotuohe alpine desert steppe was greater than its sensitivity to air temperature(Tair).3)The effect of Gross Primary Productivity(GPP)on NEE in alpine desert steppes is significantly greater than in alpine meadows.Both Ecosystem Respiration(Reco)and NEE were substantially limited by GPP,with 84%of GPP in alpine wetlands contributing to Reco and 16%to NEE;92%of GPP in alpine meadows contributing to Reco and 8%to NEE;and 40%of GPP in high-altitude desert grasslands contributing to Reco and 60%to NEE.4)The strong correlation between NEE and evapotranspiration suggests that water availability is the primary factor controlling changes in the carbon and water budgets of alpine ecosystems.展开更多
文摘针对不同陆地生态系统中净生态系统CO_(2)交换量(Net ecosystem exchange,NEE)数据的长期连续测量中存在的数据差异问题,以中国气象局青海高寒生态气象野外科学试验基地野牛沟试验站为研究对象,利用涡动协方差技术获取高寒湿地生态系统水平上的NEE数据。通过对比机器学习算法和通量数据后处理算法(Reddyproc)两种数据填充方法,提出了一种结合机器学习与时序异常检测(Time series anomaly detection,TAD)的新框架,用于NEE数据的空白填补。研究结果表明:1)Reddyproc算法在剔除异常值后,NEE插补决定系数(R^(2))达到0.67,数据离散度显著降低,数据质量提升;2)八种机器学习模型中,随机森林(Random Forest,RF)模型表现最优,其决定系数(Coefficient of determination,R^(2))为0.63,均方根误差(Root mean square error,RMSE)为2.17μmol s^(-1)m^(-2),且经过时序异常检测后,估算精度提升了17%;3)Reddyproc和RF估算的CO_(2)通量存在季节性差异,冷季(1—3月和10—12月)Reddyproc估算值低于RF,而暖季(4—9月)则高于RF,表明冬季Reddyproc低估了CO_(2)释放,夏季则低估了CO_(2)吸收。该新框架有效解决了数据采集不确定性和缺失导致的二氧化碳通量计算准确率问题,为研究高寒湿地生态系统的碳固持能力、对气候变化的响应以及极端事件的影响提供了关键数据支持。未来研究应进一步探索新方法的适用性、改进和优化方向,以实现更准确、可靠且适用于不同生态系统的填补模型,为生态系统建模和预测提供强大工具。
基金supported in part by the Fundamental Research Project of the Science and Technology Department of the Qinghai Province(Grant No.2025-ZJ-739)the National Natural Science Foundation of China(Grant No.U21A2021)+1 种基金the Open Fund of Greenhouse Gas and Carbon Neutral Key Laboratory of Qinghai Province(Grant No.ZDXM-2023-3)the Key Projects of Qinghai Meteorological Bureau(Grant No.QXZD2024-08)。
文摘Alpine meadows,alpine wetlands,and alpine desert steppes are the three typical vegetation types on the Qinghai-Tibet Plateau.The complex terrain and harsh climatic conditions across this region lead to considerable diversification in the vegetation growth environment,resulting in substantial spatial heterogeneity in ecosystem carbon flux and its controlling mechanisms.Using eddy covariance data collected from March to August 2019,this study examined the responses of carbon and water fluxes in different ecosystems on the Tibetan Plateau to typical hydrometeorological factors,focusing on Net Ecosystem CO□Exchange(NEE)and Evapotranspiration(ET).The results indicate that:1)The Longbao alpine wetland primarily acted as a carbon sink from May to August,while serving as a carbon source from March to April.In the Maqin alpine meadow,it functioned as a carbon sink during June and July but acted as a carbon source in March,April,May,and August.The Tuotuohe alpine desert strppe was predominantly a net carbon sink from March to August.Overall,after the entire growing season(March to August),the Longbao alpine wetlands,Maqin alpine meadow,and Tuotuohe alpine desert steppe all showed net carbon sink properties,with net CO_(2)uptakes of 236.12 g/m^(2),291.45 g/m^(2),and 290.28 g/m^(2),respectively.2)The importance of meteorological factors to NEE varies with scale and ecosystem type,with global radiation(Rg)being the most critical factor influencing NEE variation.Volumetric soil water content(Soil_VWC)and soil temperature(Soil_T)had a positive effect on NEE at Maqin alpine meadow and Tuotuohe alpine desert steppe,while higher values of these variables showed a negative contribution.Furthermore,the sensitivity of NEE to Soil_T at Longbao alpine wetland and Tuotuohe alpine desert steppe was greater than its sensitivity to air temperature(Tair).3)The effect of Gross Primary Productivity(GPP)on NEE in alpine desert steppes is significantly greater than in alpine meadows.Both Ecosystem Respiration(Reco)and NEE were substantially limited by GPP,with 84%of GPP in alpine wetlands contributing to Reco and 16%to NEE;92%of GPP in alpine meadows contributing to Reco and 8%to NEE;and 40%of GPP in high-altitude desert grasslands contributing to Reco and 60%to NEE.4)The strong correlation between NEE and evapotranspiration suggests that water availability is the primary factor controlling changes in the carbon and water budgets of alpine ecosystems.
