【作者】 李凡；

【导师】 李玉霞；

【作者基本信息】 电子科技大学 ， 控制科学与工程， 2021， 硕士

【摘要】 植被不仅是陆地生态系统的重要组成部分之一,同时在调节全球碳元素的动态平衡、维持全球气候稳定和研究全球气候变化等方面也有着举足轻重的作用。植被参数不仅能从多方面表征植被特征,同时也可以分门别类的对植被从多种角度进行细致的研究讨论。在植被的光合作用和呼吸作用中,水分也是至关重要的主体和参与者,植被含水量的改变不仅直接反映了植被自身的生物物理基础过程,同时这些生物物理基础过程也与全球范围内的水循环和碳循环有着密不可分的关系;植被净初级生产力NPP（Net Primary Productivity）在调节世界范围内的气候变化中扮演着重要的角色;植被物候的发展变化不仅会影响植物生物量大小,还会影响到全世界大气中CO₂的浓度含量,进而可能会对全球气候变化和全球碳循环产生重要影响。目前基于遥感技术对于植被参数的反演模型大致可以分为两大类:传统经验反演模型和物理反演模型:传统经验反演模型虽然结构简单但是结果精度有限;而物理反演模型虽然具有扎实的科学技术理论基础,但模型输入参数繁杂难获取、普适性不高。本文选择四川省作为研究区域、以MODIS卫星遥感影像数据和部分实测数据、气象数据作为主要数据源,在植被参数的反演过程中引入了机器学习的方法模型,不仅构建了基于机器学习的植被参数反演模型,同时还研究植被参数对气候变化的响应关系。本文的主要研究内容和结论如下:（1）植被燃料含水量的反演本文研究植被燃料含水量FMC（Fuel Moisture Content）时引入了机器学习模型,选取两类方法:基于植被水分指数的反演方法和四种机器学习方法模型,利用两类方法模型分别反演植被燃料含水量FMC,并对比四种机器学习方法的反演精度,取最佳的XGBoost模型对研究区域四川省进行植被燃料含水量FMC结果图的反演,并且对其进行时空分析研究。（2）植被净初级生产力的反演本文研究植被净初级生产力NPP选取的方法模型:光能利用率CASA模型。利用CASA模型对研究区域四川省进行NPP反演,并与遥感产品数据进行对比分析,同时对CASA模型的反演结果进行时空分析研究。（3）植被物候的反演本文研究植被物候引入了机器学习模型,选取两类方法:传统方法阈值法和GBDT（Gradient Boosting Decision Tree）机器学习方法,利用两种方法模型分别反演同一研究地区四川省的植被物候并进行结果对比,同时对GBDT机器学习方法模型的反演结果图进行时空分析研究。（4）植被参数对气候变化的响应本文对于三种植被参数与气候变化之间响应关系进行研究与分析,分别研究了四川省区域在2011年-2015年间植被燃料含水量FMC、植被净初级生产力NPP、植被物候分别对于气温和降水变化的响应。更多还原

【Abstract】 Vegetation is not only an important part of terrestrial ecosystem,but also plays an important role in regulating the dynamic balance of global carbon,maintaining global climate stability and studying global climate change.Vegetation parameters can not only represent the characteristics of vegetation from many aspects,but also can be classified to study and discuss vegetation from various angles.In the photosynthesis and respiration of vegetation,water is also a vital subject and participant.The change of vegetation water content not only directly reflects the basic biophysical processes of vegetation itself,but also has an inseparable relationship with the global water cycle and carbon cycle.NPP of vegetation regulates the global gas cycle The development and change of vegetation phenology will not only affect the size of plant biomass,but also affect the concentration of CO₂in the world atmosphere,which may have an important impact on global climate change and global carbon cycle.At present,the inversion models based on remote sensing technology can be divided into two categories:traditional empirical inversion model and physical inversion model:Although the traditional empirical inversion model is simple in structure,the result accuracy is limited;Although the physical inversion model has a solid theoretical foundation of science and technology,the input parameters of the model are complex and difficult to obtain and are not universal.In this paper,Sichuan Province is chosen as the research area,MODIS satellite remote sensing image data,some measured data and meteorological data as the main data sources.In the process of vegetation parameter inversion,the machine learning method model is introduced.Not only the model of vegetation parameter inversion based on machine learning is built,but also the response relationship between vegetation parameters to climate change is studied.The main contents and conclusions of this paper are as follows:（1）The inversion of FMC of vegetation fuel water contentIn this paper,the machine learning model is introduced in the study of FMC of vegetation fuel water content.Two methods are selected:the inversion method based on vegetation moisture index and four machine learning method models.The FMC of vegetation fuel water content is retrieved by using two kinds of methods,and the inversion accuracy of four machine learning methods is compared,and the best xbboost model is used to study the vegetation fuel content in Sichuan Province The inversion of FMC results of water volume is carried out and studied in time and space.（2）Inversion of NPP of vegetationThis paper studies the NPP selection method model of net primary productivity of vegetation:CASA model of light energy utilization rate.The NPP inversion is carried out in Sichuan Province by CASA model,and the data of remote sensing products are compared and analyzed.Meanwhile,the results of CASA model are analyzed in time and space.（3）Inversion of vegetation phenologyIn this paper,we introduce machine learning model to vegetation phenology,and select two methods:traditional threshold method and gbdt machine learning method.We use two methods to retrieve vegetation phenology of Sichuan Province in the same research area and compare the results.At the same time,the paper also makes a spatiotemporal analysis of the inversion results of gbdt machine learning method model.（4）Response of vegetation parameters to climate changeThis paper studies and analyzes the response relationship between the three vegetation parameters and climate change,and studies the response of FMC,NPP and phenology to the change of temperature and precipitation in Sichuan Province during 2011-2015.更多还原