植被冠层反射率模型弱敏感参数遥感反演方法

发布时间：2018-03-02 17:03

本文选题：植被冠层反射率模型　切入点：模型弱敏感参数　出处：《电子科技大学》2017年博士论文　论文类型：学位论文

【摘要】：20世纪70年代以来,植被冠层反射率模型研究得以蓬勃发展,涌现出SAIL、GEOSAIL等众多经典的冠层反射率模型。基于这些经典模型的植被关键参数反演研究也在近20年中如火如荼地开展着,并广泛应用于农作物估产、生态环境监测与保护、自然灾害(干旱、火灾等)评估及预警、水文水资源管理等关乎国家安全、全球气候变化的重大需求中。但是,近年来植被冠层反射率建模研究进展缓慢;植被关键参数反演方面也一直主要集中在几个模型敏感参数(如叶面积指数、冠层含水量等)的反演,而重要但模型弱敏感的参数研究却鲜有踏足,导致理论研究与实际应用之间的落差越来越大。鉴于此,本论文站在前人对模型敏感参数成功反演的基础上,对模型弱敏感参数进行遥感定量反演研究,重点研究了模型弱敏感参数干物质重量(DMC)及其衍生的草地地上生物量(AGB)和冠层可燃物含水率(FMC)的反演理论与方法。AGB是区域碳循环研究中的重要一环,同时也是农作物估产的研究重点;FMC是描述植被点燃概率及火灾蔓延速率的关键指标因子,因此是众多火灾模型的关键输入参数。论文通过对这二者参数反演方法及应用研究,旨在构建一套适用于模型弱敏感参数反演的方法体系,同时拓展植被冠层反射率模型的应用范围,服务于农作物估产、区域生态安全、野火风险评估及预警、全球气候变化等领域。论文主要研究工作和成果如下:(1)分析植被冠层反射率模型及参数敏感性的基础上,提出了模型弱敏感参数遥感反演策略。论文指出模型弱敏感参数成功反演的关键在于增强弱敏感参数的敏感性,并提出五个弱敏感参数反演策略:基于高光谱数据的模型弱敏感参数反演;参数化模型敏感参数,提高模型弱敏感参数的敏感性;基于面向对象的模型弱敏感参数反演;基于多时相遥感数据的模型弱敏感参数反演;基于真实实测数据的模型弱敏感参数反演。(2)研究了缓解植被冠层反射率模型参数反演的病态问题,提出了基于贝叶斯网络算法的反演方法。病态反演问题会极大地降低模型敏感参数的反演精度,对模型弱敏感参数正确反演的影响更为强烈。为缓解这一问题,本论文以贝叶斯网络算法为基础,通过考虑模型参数之间固有的关联特性,构建了更符合自然界实际情况的自由参数先验联合概率分布,降低了错误自由参数组合出现的概率,缓解了病态反演问题,提高了叶面积指数(LAI)及植被冠层含水量的反演精度,为后续基于植被冠层反射率模型的草地AGB及植被冠层FMC反演奠定基础。(3)提出了一套新的反演草地AGB的方法,突破了传统基于实测统计拟合估算草地AGB的思路。该方法假设草地AGB可通过LAI与叶片干物质重量(DMC)的乘积近似表示。首先,基于PROSAIL模型对实验区草地LAI及DMC进行反演;同时通过考虑模型参数间的相关特性及融入MODIS LAI产品等方法提高草地AGB的敏感性,反演获得草地AGB。实验中以我国青海湖流域草地AGB反演为例,论证该方法的有效性;同时,实验中使用了植被指数法、偏最小二乘法(PLSR)、人工神经网络法(ANN)等三种传统经验估算草地AGB的方法进行对比分析。实验结果显示,使用本论文方法反演的草地AGB精度高于植被指数法及ANN,与PLSR精度不相上下,但本论文方法不依赖地面实测数据,因此比PLSR更具普适性,是具有前景的草地AGB反演方法。(4)构建了植被冠层FMC反演方法体系,包括基于区间估计LAI的草地冠层FMC遥感反演方法和基于耦合辐射传输模型的森林冠层FMC遥感反演方法。FMC是一个与LAI无关的量,但在基于物理模型的FMC反演中,LAI的不确定性对FMC的正确反演具有较强的干扰作用。为缓解这一问题,提高FMC的敏感性,实验结合MODIS LAI产品,利用降尺度及区间估计的LAI参数化模型LAI输入,以此降低LAI的不确定性,从而提高草地冠层FMC的反演精度。在森林冠层FMC的反演中,针对本研究区森林植被具上层乔木下层草本植被的特点,即双层冠层结构,实验通过耦合描述草地冠层反射率特征的SAIL模型及描述森林冠层反射率特征的GEOSAIL模型,近似模拟该双层冠层结构的森林植被反射率特征,降低模型系统误差对弱敏感参数FMC的影响,从而正确反演了森林冠层FMC。(5)构建了基于遥感技术的区域大尺度野火风险评估及预警雏形。论文以澳大利亚和四川西昌泸山为例,首先应用上述植被冠层FMC反演方法,进行植被冠层FMC定量反演,生产了 2001-2015年的澳大利亚15年的植被FMC产品。其次,基于该套产品,结合MODIS历史火灾产品(MCD64),应用Logistic模型计算得出澳大利亚2001-2015年燃烧指数(FI)产品,以此对野火风险进行评估及预警。通过澳大利亚历史上三次重大火灾(2003年堪培拉火灾、2009年维多利亚州火灾及2013年新南威尔士州火灾)爆发前植被冠层FMC及FI演变情况显示,该两套产品对野火风险具有较好指示作用,该工作对于今后全球植被冠层FMC产品化及野火风险评估应用推广具有重要的示范意义。
[Abstract]:Since 1970s, the research of canopy reflectance model of vegetation flourish, the emergence of SAIL, GEOSAIL and many other classical models of canopy reflectance. Study on the key parameters of the classical model based on inversion of vegetation also in the past 20 years to carry out like a raging fire, and widely used in crop yield estimation, monitoring and protection of the ecological environment, natural disasters (drought. Fire) assessment and early warning, hydrology and water resources management related to national security, great demand of global climate change. However, recent progress in modeling the vegetation canopy reflectance is slow; the key parameters inversion of vegetation has been mainly focused on the sensitive parameters of several models (such as leaf area index and canopy water content inversion, etc.) the important parameters of the model but rarely set foot weakly sensitive, leading to more and the gap between the theoretical research and practical application of the larger. In view of this, the The station based on the predecessor model sensitive parameters inversion of success, the quantitative remote sensing model of weak inversion of sensitive parameters, focusing on the model of weak sensitive parameters of dry weight (DMC) and its derivative on the grassland biomass (AGB) and canopy fuel moisture content (FMC).AGB inversion theory and method the study area is an important part of the carbon cycle at the same time, and also a research focus in crop yield estimation; FMC is the key factor of vegetation index and the rate of fire spread fire probability description, so it is a key input parameter of many fire model. Based on the parameter inversion method and application of the two, aims to establish a suitable model weak sensitive parameter inversion method system, and expand the scope of application of the vegetation canopy reflectance model, in crop yield estimation, regional ecological security, wildfire risk assessment and early warning, global climate Changes in other fields. The main research work and results are as follows: (1) analyzing the sensitivity of the model and the parameters of canopy reflectance, proposes a model of weak sensitive parameters of remote sensing inversion strategy. The paper pointed out that the key to success of the model of weak sensitive parameters inversion is to enhance the sensitivity of weak sensitive parameters, and puts forward five strategies of weak inversion sensitive parameters hyperspectral data model: weak sensitive parameter inversion based on parametric model; sensitive parameters, improve the sensitivity of model weak sensitive parameters; object oriented model of weak sensitive parameter inversion based on multitemporal remote sensing data model; weak sensitive parameter inversion based on real measured data model; weak inversion based on sensitive parameters (2) were studied. To alleviate the ill posed problem of vegetation canopy reflectance model inversion, the inversion algorithm based on Bayesian networks. The ill posed inversion problem will be greatly reduced The precision of inversion parameters of the low sensitive model, weak influence on the model of sensitive parameters inversion is more intense. In order to alleviate this problem, this thesis is based on the Bayesian network algorithm, by considering the inherent characteristics of correlation between model parameters, build more in line with the actual situation of the free parameters a priori nature of joint probability distribution, reducing the probability of error free parameter combinations, alleviate the ill posed inversion problem, improve the leaf area index (LAI) and vegetation canopy water content inversion precision, for subsequent AGB and grassland vegetation canopy FMC inversion of vegetation canopy reflectance model based on the foundation. (3) proposed a new method of the inversion of the grassland AGB, breakthrough the traditional statistical fitting estimation of grassland AGB based on the idea. The method assumes that the grassland AGB by LAI and leaf dry weight (DMC) of the product approximation. First, based on the PROSAIL Inversion of experimentation area grassland LAI and DMC model; at the same time by considering the correlation between model parameters and integration of MODIS LAI products and other methods to improve the sensitivity of grassland AGB, derived from the AGB. experiment in Qinghai Lake Valley meadow grassland in China AGB inversion as an example to demonstrate the effectiveness of the method; at the same time, the vegetation index method using experiment in the partial least squares (PLSR), artificial neural network (ANN) analyzed the grassland AGB estimation method of three kinds of traditional experience. The experimental results show that using the method of inversion accuracy in AGB grassland vegetation index method and the accuracy of PLSR and ANN, be roughly the same, but the method does not rely on the ground the data, which is more universal than PLSR, is the grassland AGB inversion method with great prospect. (4) constructed the system of vegetation canopy FMC inversion method, including interval estimation of LAI grass canopy FMC based on Remote Sensing Modeling method and.FMC forest canopy FMC remote sensing inversion method based on coupled radiative transfer model is an independent of LAI, but in FMC based on physical model inversion, inversion LAI uncertainty of FMC has the strong jamming. To alleviate this problem, improve the sensitivity of FMC, combined with MODIS LAI products, with model LAI input and the interval estimation of scale parameter of LAI, in order to reduce the uncertainty of LAI, so as to improve the inversion precision of grassland canopy FMC. In the inversion of the forest canopy FMC, according to the characteristics of forest vegetation in the study area with lower tree layers of herbaceous vegetation, namely double canopy structure, GEOSAIL model by describing the coupled reflectance characteristics of SAIL model and describe the forest canopy characteristics of canopy reflectance of grassland experiment, approximate simulation of forest vegetation reflectance characteristics of the double canopy structure, reduce the system model The error of weak sensitive parameters affecting FMC, and thus the correct inversion of the forest canopy FMC. (5) constructed the remote sensing technology of large scale regional wildfire risk assessment and early warning based on prototype. In order to Australia and Sichuan Xichang Lushan as an example, the first application of the FMC inversion of vegetation canopy, vegetation canopy FMC quantitative inversion, production 2001-2015 years of Australia 15 years of vegetation FMC. Secondly, the set of products based on the combination of MODIS (MCD64), history of fire products using Logistic model to calculate the Australian 2001-2015 year combustion index (FI) products, in order to carry out assessment and early warning of wildfire risk. Through the history of Australia three major fire (in Canberra in 2003 fire, fire in Vitoria in 2009 and 2013 the state of New South Wales) before the outbreak of the vegetation canopy fire display FMC and FI evolution, the two sets of products with wildfire risk This work has an important demonstration significance for the application and popularization of FMC production and wildfire risk assessment in the world.

【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP79

【参考文献】