结合地形因子的酉阳县针叶林实际地表生物量遥感估算

发布时间：2018-07-29 20:56

【摘要】：森林生物量是衡量生态系统生产力的重要指标,也是研究森林生态系统物质循环的重要基础。随着遥感技术的快速发展,影像数据参与森林生物量反演的方法与手段日益推广和完善。山地地区地形复杂,运用传统生态学方法进行大区域的生物量测量耗时长、受限较大,在该区运用遥感数据反演生物量具有重要的现实意义。在遥感反演过程中,地形起伏对像元生物量影响显著,而目前生物量反演研究中的地形校正主要为地表辐射校正,较少关注地形因素引起的像元实际地表面积和投影面积差异,而这种差异将直接导致植被生物量估算结果的偏差。本研究以重庆市酉阳县为例,以GF-1 WFV数据和DEM数据为基础,结合影像因子和地形因子建立酉阳县针叶林生物量反演模型。在此基础上,结合地表面积计算方法和物质守恒定律,得到地形辐射校正基础上的针叶林实际地表生物量,并且定量地分析和讨论了地形起伏对酉阳县针叶林生物量估算的影响,旨在探讨GF-1WFV数据在反演针叶林生物量信息等方面的潜力,同时也为研究区内的林业资源管理、生态系统研究工作以及林业生态工程实施提供理论依据与参考。主要研究结果包括:(1)结合GIS空间分析,得到研究区的实际地表面积。影像地形校正后,单位像元面积最高值为965平方米,校正前后像元面积变化量最高达709平方米,约为影像像元单位面积的2.7倍。高值区域主要为酉阳县两大高盖地及南部山区,低值主要集中在地形较平坦、坡度较小的东部平坝区,与研究区地形起伏特点大体一致。(2)在SPSS软件平台上分析了41种影像因子、5种地形因子与针叶林生物量的相关性。结果表明,3*3窗口下,GF-1 WFV影像红光波段生成的纹理信息与生物量信息相关性总体较低,而影像近红外波段、近红外波段生成的均值纹理、差值植被指数和修正型植被指数与生物量信息的相关性显著。(3)对遥感影像进行分类实验,包括最大似然法监督分类和BP人工神经网络分类两种。分类结果显示,与基于统计理论的传统监督分类方法相比,人工神经网络在数据信息不完备的情况下,依然能够在模式识别、知识处理等方面获得较理想的效果。加入近红外波段correlation纹理的三层BP人工神经网络总体分类精度为86.62%,比传统监督分类的精度增加了9.38%。与基于统计理论的传统监督分类方法相比,人工神经网络方法是获取GF影像地物空间分布信息的更佳途径。(4)利用多元线性逐步回归分析得到基于影像和地形因子的地面生物量反演模型,模型的相关系数为0.853。利用模型分别反演三维尺度与二维尺度上的像元生物量,研究结果表明,地形校正前后,单位像元面积上的生物量变化最大值为258.5t/hm2,研究区针叶林地表实际生物量总值比校正前增加了431423.98t/hm2,生物量变化比率为8.54%,地形对生物量遥感反演精度的影响不可忽视。
[Abstract]:The forest biomass is an important index to measure the productivity of the ecosystem, and it is also an important basis for the study of the material circulation of the forest ecosystem. With the rapid development of remote sensing technology, the methods and means to participate in the inversion of forest biomass are increasingly popularized and perfected. The terrain of the mountain area is complex and the traditional ecological methods are used in the large area. The measurement of biomass is very time-consuming and limited. It is of great practical significance to use remote sensing data to retrieve biomass in this area. In the process of remote sensing inversion, the terrain undulation has a significant influence on the biomass, while the topographic correction in the study of biomass inversion is mainly the surface radiating correction, less attention to the pixel reality caused by the terrain factors. The difference between the surface area and the projected area will lead to the deviation of the estimated results of vegetation biomass. This study takes Youyang County, Chongqing, as an example, based on GF-1 WFV data and DEM data, and combines image and topographic factors to establish a coniferous forest biomass back model in Youyang county. On the basis of this, the surface area is calculated. The law of conservation of material and the law of conservation of matter, the actual surface biomass of coniferous forests on the basis of Topographic Radiation Correction, and the quantitative analysis and discussion of the effects of topographic fluctuations on the estimation of the biomass of coniferous forests in Youyang county are analyzed and discussed. The purpose is to explore the potential of GF-1WFV data in the inversion of the biomass information of coniferous forests and to the forestry capital in the study area. Source management, ecosystem research work and forestry ecological engineering implementation provide theoretical basis and reference. The main research results include: (1) the actual surface area of the study area is obtained with the GIS spatial analysis. After the image terrain correction, the maximum area of unit pixel area is 965 square meters, and the area change of the corrected pixel area is up to 709 square meters before and after the correction. It is about 2.7 times the unit area of the image pixel. The high value area is mainly two high cover and southern mountainous area of Youyang county. The low value is mainly concentrated in the eastern flat dam area with relatively flat terrain and smaller slope. (2) 41 kinds of image factors, 5 terrain factors and coniferous forests are analyzed on the SPSS software platform. The results show that the correlation between the texture information generated by the red band of the GF-1 WFV image and the biomass information under the 3*3 window is low, but the correlation of the mean texture, the difference vegetation index and the modified vegetation index in the near infrared band of the image is significant. (3) the remote sensing image is divided into two parts. Class experiments, including two kinds of maximum likelihood supervised classification and BP artificial neural network classification, show that, compared with the traditional supervised classification based on statistical theory, artificial neural networks can still achieve better results in pattern recognition and knowledge management in the case of incomplete data information. The overall classification precision of the three layer BP artificial neural network with band correlation texture is 86.62%. Compared with the traditional supervised classification, the accuracy of the artificial neural network is increased by 9.38%. and the traditional supervised classification based on statistical theory. The artificial neural network method is a better way to obtain the spatial distribution information of the GF image objects. (4) multivariable linear stepwise regression is used. An inversion model of ground biomass based on images and terrain factors is obtained. The correlation coefficient of the model is 0.853. using the model to invert the pixel biomass on the three-dimensional scale and the two-dimensional scale. The results show that the maximum value of the biomass on the unit pixel area is 258.5t/hm2 before and after the topographic correction, and the coniferous forest land in the study area is real. The total biomass ratio increased by 431423.98t/hm2 and the biomass change ratio was 8.54% before the correction. The influence of topography on the accuracy of biomass remote sensing inversion can not be ignored.
【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S718.5;S771.8

【参考文献】