摘要
草地生态系统作为自然生态系统的重要组成部分,为畜牧经济发展提供了重要的牧草资源,对调节气候变化和维持生态系统平衡等起着非常重要的作用。草地地上生物量(aboveground biomass,AGB)是草地植被生理状态的重要指标,它的大小体现着草地初级生产力水平,是衡量草地生态系统中能量循环和物质流动的重要指标,在陆地生态系统的碳循环中起着重要的作用。近几十年来,伴随畜牧业经济快速发展和全球气候变暖,草地生态系统的稳定性降低,生态环境发生退化,草地地上生物量和固碳能力势必受到影响。大尺度、动态化、高精度的草地地上生物量监测对草地碳储量核算和畜牧业可持续发展具有重要意义,而遥感技术凭借高时空探测能力恰好为其提供了解决思路。机器学习算法凭借其优越性、高效性、稳健和精确性已被广泛应用于各个研究领域,使用机器学习算法快速、准确、大范围监测草地地上生物量是目前的研究热点。因此,构建准确的草地地上生物量估算模型,精确估算草地地上生物量及分析其空间分布特征能够有效地衡量草地生态系统的稳定性和维持草地生态资源的可持续发展利用,为该区域草地资源的可持续利用和科学管理提供依据,对该地区的生态安全保护和畜牧业可持续发展具有重要意义。本研究以青海省兴海县草地为研究区,基于野外实测的草地地上生物量数据,结合高空间分辨率的遥感数据、气候数据、地形数据和土壤数据等,利用随机森林(random forest,RF)和极端梯度提升决策树(extreme gradient boosting,XGBoost)方法构建兴海县草地地上生物量估算模型,采用决定系数R^(2)和均方根误差(Root Mean Square Error,RMSE)两个精度验证指标评价两种草地地上生物量估算模型的精度,实现草地地上生物量高精度模拟和制图,并分析其空间分布格局特征。结果表明:基于XGBoost模型的草地地上生物量估算精度(R^(2)=0.75,RMSE=44.64)高于RF的模拟精度(R^(2)=0.72,RMSE=46.36),并且XGBoost模型估算的草地地上生物量与实测的草地地上生物量值更接近。基于两种机器学习模型估算的草地地上生物量数据制作空间分布图,其空间特征与实测草地地上生物量的空间分布相似,草地地上生物量高值区位于研究区的东部,西部地区草地地上生物量值最低,但是模型模拟能更好地揭示草地地上生物量分布的空间异质性。在空间分布特征上,XGBoost模型估算的草地地上生物量空间变异细节更加详细,尤其在研究区东部。本研究基于两种机器学习算法实现草地地上生物量的高精度(30 m空间分辨率)估算和数字制图,并分析其空间分布格局,可为草地生态环境监测和草地资源可持续利用提供科学依据,对于维持生态系统平衡和预测未来气候变化对草地生态系统的影响具有十分重要的理论和实践意义。
As an important part of the natural ecosystem,grassland ecosystem provides important grazing resources for the development of livestock economy and plays a very important role in regulating climate change and maintains the balance of the ecosystem.Aboveground biomass(AGB)is an important indicator of the physiological state of grassland vegetation.Its size reflects the level of primary productivity of grassland,and is an important indicator of the energy cycle and material flow in grassland ecosystem,and plays an important role in the carbon cycle of terrestrial ecosystem.In recent decades,along with the rapid development of the livestock economy and global warming,the stability of grassland ecosystem has been reduced,the ecological environment has been degraded,and the aboveground biomass and carbon sequestration capacity of grassland are bound to be affected.Large-scale,dynamic and high-precision monitoring of aboveground biomass in grassland is of great importance to the accounting of grassland carbon stocks and the sustainable development of animal husbandry,and remote sensing technology,with its high spatial and temporal detection capability,provides the solution.Machine learning algorithms have been widely used in various research fields due to their superiority,efficiency,robustness and accuracy,and the use of machine learning algorithms for rapid,accurate and large-scale monitoring of grassland aboveground biomass is currently a hot research topic.As a result,the construction of an accurate aboveground biomass estimation model,the accurate estimation of aboveground biomass and the analysis of its spatial distribution characteristics can effectively measure the stability of grassland ecosystem and maintain the sustainable development and use of grassland ecological resources,providing a basis for the sustainable use and scientific management of grassland resources in the region,which is of great significance to the ecological security protection and sustainable development of animal husbandry in the region.In this study,the above ground biomass data of grassland in Xinghai County,Qinghai Province was used as the study area,and the random forest(RF)and extreme gradient boosting(XGBoost)methods were used to combine remote sensing data,climate data,topographic data and soil data with high spatial resolution.The aboveground biomass estimation models of grassland in Xinghai County were constructed,and the accuracy of the two models was evaluated using two accuracy verification indicators,namely R^(2) and root mean square error(RMSE),to achieve high accuracy simulation and mapping of aboveground biomass of grassland and analyze its spatial distribution pattern.The results show that:(1)the accuracy of aboveground biomass estimation in grassland based on the XGBoost model(R^(2)=0.75,RMSE=44.64)was higher than the simulation accuracy of RF(R^(2)=0.72,RMSE=46.36),and the aboveground biomass estimated by the XGBoost model was closer to the measured above-ground biomass values in grassland;(2)the spatial distribution of grassland aboveground biomass data estimated by the two machine learning models was similar to that of the measured grassland aboveground biomass,with high values of grassland aboveground biomass in the eastern part of the study area and the lowest values of grassland aboveground biomass in the western part,but the model simulations could better reveal the spatial heterogeneity of grassland aboveground biomass distribution;(3)in terms of spatial distribution characteristics,the XGBoost model estimated more detailed spatial variation in aboveground biomass of grassland,especially in the eastern part of the study area.Based on two machine learning algorithms,this study achieved high-precision(30 m spatial resolution)estimation and digital mapping of aboveground biomass in grasslands and analysed their spatial distribution patterns,which can provide scientific basis for monitoring grassland ecosystem and sustainable use of grassland resources,and is of great theoretical and practical significance for maintaining ecosystem balance and predicting the impact of future climate change on grassland ecosystem.
作者
王婷
周伟
肖洁芸
谢利娟
WANG Ting;ZHOU Wei;XIAO Jieyun;XIE Lijuan(Chongqing Jinfo Mountain Karst Ecosystem National Observation and Research Station,School of Geographical Sciences,Southwest University,Chongqing 400715,China;Chongqing Engineering Research Center for Remote Sensing Big Data Application,School of Geographical Sciences,Southwest University,Chongqing 400715,China;Beijing Piesat Information Technology Co.,Beijing 100195,China)
出处
《冰川冻土》
CSCD
北大核心
2023年第2期753-762,共10页
Journal of Glaciology and Geocryology
基金
重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0384)
中央高校基本科研业务费专项资金项目(SWU020015)
国家自然科学基金项目(41501575)资助。
关键词
草地生态系统
地上生物量
随机森林
极端梯度提升
空间分布
grassland ecosystem
aboveground biomass
random forest
extreme gradient boosting
spatial distribution
作者简介
王婷,硕士研究生,主要从事遥感反演研究.E-mail:swuwt230@email.swu.edu.cn;通信作者:周伟,副教授,主要从事生态环境遥感监测研究.E-mail:zw20201109@swu.edu.cn。
