基于HJ卫星数据的甘蔗长势监测与估产研究

发布时间：2018-06-03 11:15

本文选题：甘蔗 + 植被指数　；参考：《南京信息工程大学》2015年硕士论文

【摘要】：广西兴宾蔗区是全国最大的蔗糖生产基地,蔗糖业已成为当地经济的主导和支柱产业。及时准确地获取甘蔗种植面积、监测甘蔗长势、进而对产量进行预测,对甘蔗生产管理、稳定食糖市场、保障食糖安全及经济健康循环发展具有重要意义。目前,“3S”技术在甘蔗长势动态监测与产量预测方面的应用研究在国内方兴未艾,特别是缺乏国产卫星数据的应用示例。本文选用国产HJ-1A/B星CCD数据,以监督分类、多时相迭代法相结合的方法提取甘蔗种植面积；通过引入随机变量的标准差和离均差的概念并结合田间农学观测数据构建甘蔗长势监测模型；在此基础上利用表征甘蔗长势的植被指数和产量数据构建甘蔗单产遥感估产模型。主要研究成果包括以下三方面：(1)甘蔗种植面积是甘蔗遥感长势监测与产量估算的基础内容,基于多源数据,利用多种遥感方法相结合提取甘蔗种植面积。首先,建立研究区典型地物的遥感解译标志；其次,采用监督分类法,结合广西耕地图层、坡度等本底信息数据库获取甘蔗可能种植区；最后,根据解译标志逐时相设定甘蔗NDVI阂值,逐步向下通过掩膜迭代剔除其他地物信息,获得甘蔗种植面积信息。利用野外实测样点和农业部门提供的部分乡镇甘蔗面积验证遥感解译结果,总体精度达91.18%,Kappa系数为0.8101,与农业部门统计数据的平均相对误差为10.17%。(2)建立了基于随机变量理论的长势监测模型。针对甘蔗长势遥感监测缺乏统一分级标准的研究现状,通过分析多种遥感指标与田间农学观测数据的相关性差异,确定增强型植被指数EVI和归一化差值植被指数NDV1分别作为甘蔗茎伸长期和工艺成熟期的长势监测指标。长势监测模型假设甘蔗生长满足正态分布规律,引入离均差的概念,通过离均差与标准差的差异定量研究了兴宾区2009～2013年的甘蔗长势情况。以甘蔗产量的整体波动情况和2013年地面观测点数据推测的甘蔗长势样区以及在县(区)尺度的实际应用结果对长势遥感监测结果进行验证。对比分析结果表明,基于随机变量的标准差和离均差理论的长势评价模型可以满足县(区)尺度的甘蔗长势动态监测要求,适用于不同蔗区和多种遥感数据。(3)构建了基于植被指数的甘蔗单产估算模型。分析甘蔗植被指数与其单产的数学关系,建立了甘蔗关键生育期和全生育期的“遥感植被指数-甘蔗产量”估产模型。从模型本身和当年甘蔗实际单产的相对误差分析模型精度,其结果表明：基于全生育期的植被指数估产模型效果最好,拟合系数最高,相对误差最低；关键生育期以茎伸长中期的估产模型精度最高,其中又以线性方程模型的估产效果较佳；糖分积累期次之；工艺成熟期的估产模型拟合效果最差,相对误差最大。
[Abstract]:Xingbin Sugarcane region in Guangxi is the largest sugarcane production base in China, which has become the leading and pillar industry of local economy. It is of great significance to obtain sugarcane planting area timely and accurately, monitor sugarcane growth and forecast the yield, which is of great significance for sugarcane production management, sugar market stability, sugar safety and economic healthy cycle development. At present, the application of "3s" technology in dynamic monitoring and yield prediction of sugarcane growth is in the ascendant in China, especially in the absence of domestic satellite data. In this paper, the CCD data of domestic HJ-1A/B star were used to extract sugarcane planting area by the method of supervised classification and multi-time iterative method. By introducing the concepts of standard deviation and deviation mean deviation of random variables and combining with the field agronomic observation data, the sugarcane growth monitoring model was constructed, and the remote sensing yield estimation model of sugarcane yield was constructed by using the vegetation index and yield data to characterize sugarcane growth. The main research results include the following three aspects: (1) Sugarcane planting area is the basic content of sugarcane remote sensing growth monitoring and yield estimation. Based on multi-source data, sugarcane planting area is extracted by using multiple remote sensing methods. First, establish the remote sensing interpretation mark of typical features in the study area; secondly, use the supervised classification method, combining with the background information database of Guangxi cultivated land layer, slope and other background information to obtain the possible sugarcane planting area; finally, According to the interpretation marker, the NDVI threshold value was set up, and the information of sugarcane planting area was obtained by removing other features information through the mask iteration step by step. The results of remote sensing interpretation were verified by using field survey samples and sugarcane area of some villages and towns provided by the agricultural sector. The Kappa coefficient is 0.8101 and the average relative error with the statistical data of the agricultural sector is 10.17. The growth monitoring model based on the theory of random variables is established. In view of the lack of unified classification standard for sugarcane growth remote sensing monitoring, the correlation difference between various remote sensing indexes and field agronomic observation data was analyzed. The enhanced vegetation index (EVI) and normalized difference vegetation index (NDV1) were determined as the growth monitoring indexes of sugarcane stem extension and process maturity respectively. The growth monitoring model assumes that sugarcane growth meets the normal distribution law and introduces the concept of deviation mean deviation. The growth of sugarcane in Xingbin district from 2009 to 2013 is quantitatively studied by the difference between the average deviation and the standard deviation. The results of remote sensing monitoring of sugarcane growth were verified by the fluctuation of sugarcane yield and the actual application results of sugarcane growth sample area and county scale based on the data of ground observation points in 2013. The results of comparative analysis show that the model based on the theory of standard deviation and mean deviation of random variables can meet the requirements of dynamic monitoring of sugarcane growth at county (district) scale. The estimation model of sugarcane yield per unit yield based on vegetation index was established for different sugarcane regions and various remote sensing data. Based on the analysis of the mathematical relationship between the vegetation index of sugarcane and its yield per unit yield, a model of "remote sensing vegetation index-sugarcane yield" for the key growth period and the whole growth period of sugarcane was established. The accuracy of the model was analyzed from the model itself and the actual yield of sugarcane. The results showed that the vegetation index based on the whole growth period had the best effect, the fitting coefficient was the highest, and the relative error was the lowest. The precision of the key growth stage was the highest in the middle stage of stem elongation, among which the linear equation model was the best; the sugar accumulation stage was the second; the fitting effect of the yield estimation model at the technological maturity stage was the worst, and the relative error was the largest.
【学位授予单位】：南京信息工程大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S566.1;S127

【参考文献】