面向对象的草原植被参数反演方法及应用

发布时间：2018-08-06 11:30

【摘要】：草原是构成陆地生态系统的重要成分,对能量流动和物质循环以及人类的生存和发展起着不可估量的作用。草原植被的生长状况能够直接或间接地通过植被的生物、物理与化学参数(如叶面积指数和冠层水含量)反映。因此,通过研究草原植被参数动态监测草原生态环境状况,及时为有关部门提供有效的科学决策,具有重要的科学意义和应用价值。论文以青海湖流域为研究区,基于多源遥感数据和地面实测数据,以研究区植被冠层水含量和叶面积指数为研究对象,应用面向对象的反演方法进行定量反演。为了比较验证该方法的可行性和有效性,又分别从传统的基于像元的物理模型和神经网络的方法进行了反演。同时,采用国产遥感卫星数据(如HJ-1和GF-1)进行研究区草原植被叶面积指数的反演,探讨国产数据的应用潜力和面向对象方法的可行性和有效性。论文的主要工作和成果如下:(1)应用面向对象的方法,结合查找表算法,利用Landsat-8 OLI遥感影像数据对研究区草原植被的叶面积指数和冠层水含量进行反演。该方法通过考虑邻近像元的光谱信息,反演精度在一定程度上得到较好的提高。研究过程中为了解决模型反演的病态特性以及草原植被的非均匀性,对模型输入参数分别从定量与定性的角度进行了敏感性分析,将研究区内的植被进行分区:稀疏区和密集区,对其依次构建查找表。反演结果与实测数据比较分析显示:叶面积指数与冠层水含量的反演值与实测值的决定系数R2分别为0.88和0.81,均方根误差RMSE分别为0.59和67.31g/m2,两者均表现出较高的反演精度,从而验证了该方法的高度有效性。(2)利用相同的数据源(Landat-8 OLI),分别采用基于像元的物理模型方法和神经网络方法对研究区的草原植被叶面积指数与冠层水含量进行定量反演。研究中通过将该两种方法与面向对象的方法进行对比验证,说明后者的优越性。通过将该两种方法的反演结果与地面实测值进行对比验证,结果显示:基于像元的物理模型方法反演的叶面积指数和冠层水含量与实测值的决定系数R2分别为0.87和0.78,均方根误差分别为0.62和80.11 g/m2;而神经网络方法的反演值与实测值的决定系数R2分别为0.84和0.72,均方根误差分别为0.65和99.95 g/m2。对比本文提出方法的反演结果说明其具有较好的反演精度。(3)以前面应用的三种方法为基础,采用国产遥感卫星数据(如HJ-1和GF-1)进行研究区草原植被叶面积指数的反演。一方面为了探讨国产数据与国外遥感数据的数据质量与应用潜力,另一方面也是为了进一步验证本文提出的方法即面向对象方法的可行性和有效性。结果表明:采用相同的数据源,面向对象方法的反演精度较高,基于像元的物理模型方法次之;而相同的方法,Landsat-8影像数据应用的性能较高,其次是高分一号影像数据。
[Abstract]:Grassland is an important component of terrestrial ecosystem, which plays an inestimable role in energy flow, material circulation and human survival and development. The growth of grassland vegetation can be directly or indirectly reflected by the biological, physical and chemical parameters (such as leaf area index and canopy water content). Therefore, it is of great scientific significance and application value to study the dynamic monitoring of grassland ecological environment by studying the parameters of grassland vegetation, and to provide effective scientific decision for relevant departments in time. Taking Qinghai Lake Basin as the study area, based on multi-source remote sensing data and ground measured data, and taking the water content and leaf area index of vegetation canopy in the study area as the research object, the paper uses the object-oriented inversion method to carry out quantitative inversion. In order to compare the feasibility and effectiveness of the proposed method, the traditional methods based on pixel physical model and neural network are inversed respectively. At the same time, using domestic remote sensing satellite data (such as HJ-1 and GF-1) to invert the leaf area index of grassland vegetation in the study area, the application potential of domestic data and the feasibility and effectiveness of object-oriented method are discussed. The main work and achievements are as follows: (1) the leaf area index and canopy water content of steppe vegetation in the study area are inversed by using Landsat-8 OLI remote sensing image data by using object-oriented method and lookup table algorithm. The inversion accuracy is improved to a certain extent by taking into account the spectral information of adjacent pixels. In order to solve the pathological characteristics of model inversion and the nonuniformity of grassland vegetation, the sensitivity analysis of model input parameters from quantitative and qualitative perspectives was carried out. The vegetation in the study area is divided into sparse area and dense area, and the lookup table is constructed in turn. The comparison and analysis of the inversion results and the measured data show that the coefficient R2 of the inversion value and the measured value of the leaf area index and the canopy water content are 0.88 and 0.81, respectively, and the root mean square error (RMSE) is 0.59 and 67.31 g / m2, respectively. Both of them have high inversion accuracy. This method is highly effective. (2) the same data source (Landat-8 OLI),) is used to quantitatively invert the leaf area index of grassland vegetation and the water content of canopy in the study area by using the method of physical model based on pixel and the method of neural network respectively. In the study, the superiority of the two methods is proved by comparing them with the object-oriented method. The inversion results of the two methods are compared with the measured data on the ground. The results show that the coefficient R2 of leaf area index and canopy water content and measured values are 0.87 and 0.78, respectively, and the root mean square error are 0.62 and 80.11 g / m2, respectively. The determined coefficient R2 of the measured values is 0.84 and 0.72, and the root mean square error is 0.65 and 99.95 g / m2, respectively. The inversion results show that this method has good inversion accuracy. (3) based on the three methods used in this paper, we use domestic remote sensing satellite data (such as HJ-1 and GF-1) to invert the grassland vegetation leaf area index in the study area. On the one hand, in order to discuss the data quality and application potential of domestic and foreign remote sensing data, on the other hand, it is necessary to further verify the feasibility and effectiveness of the proposed method, namely, the object-oriented method. The results show that with the same data source, the retrieval accuracy of object-oriented method is higher, the physical model method based on pixel is the second, and the same method Landsat-8 image data has higher performance, followed by high-score No. 1 image data.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S812

【参考文献】