基于NDVI时间序列重构的经济型人工林时空分布信息提取研究

发布时间：2018-03-06 22:03

本文选题：经济型人工林　切入点：桉树　出处：《中国科学院大学(中国科学院遥感与数字地球研究所)》2017年博士论文　论文类型：学位论文

【摘要】：植被是全球陆地生态系统重要的组成部分,它不仅为陆地生态系统提供了维持其正常运转的能源动力,也为生态系统中其他物种提供了栖息场所。在全球变化和人类活动日益加剧的条件下,全球植被的类型和格局发生了巨大变化,伴随着大量的原生林砍伐,人工林种植面积急剧增加,使得某些地区原有生态系统失去平衡,严重减少了该地区的生物多样性。因此,经济型人工林时空动态信息的提取对陆地生态系统功能的保护具有重要的意义。但是,目前的对经济型人工林的识别使用的数据多为时间分辨率较高、空间分辨率较低的MODIS数据,通过传统的分析算法,如分块差值算法、标准化欧氏距离算法、Bounding Envelope算法等。但是这些算法对于中小区域尺度的经济型人工林识别中,并不适用。本文在使用Landsat 5 TM、Landsat 7 ETM和HJ 1重构了归一化植被指数(NDVI)时间序列数据的基础上,提出了一种倒三角形面积算法,并利用该算法成功提取了韶关、河源和赣州3个市桉树人工林的时空分布,另外还通过分析了生长期类型和影像获取时间之间的关系间接地推算出了相对精确的种植年份,从而得出了如下结论:1.基于Landsat 5 TM、Landsat 7 ETM和HJ 1数据重构NDVI时间序列Landsat 5 TM数据的大量缺失和质量问题使得本文采用了Landsat 7 ETM和HJ 1数据来填补空缺。不同的数据源,在光谱响应函数、卫星运行轨道、获取影像时太阳高度角、观测角度等方面可能存在差异,为了探究这种差异的大小,本文对2012年的4对LANDSAT 5 TM和HJ影像进行了分析,在分别对比了100多个同名点的NDVI值后,发现二者的NDVI相关性都达到了98%以上,证明LANDSAT 5 TM/ETM、HJ数据虽然存在些微差异,但是不足以对分类造成影响。说明使用LANDSAT 5 TM/ETM、HJ数据来重构NDVI时间序列进行地物分类具是具有可行性的。2.基于倒三角形面积算法识别小区域尺度桉树人工林针对本研究重构的NDVI时间序列存在不连续、最小时间间隔偏大的问题,在充分考虑桉树人工林像元参考时间序列规律的基础上,提出了一种新的算法——倒三角形面积算法(ITA)。相比其他三种桉树识别算法,它避免了对浮动值较大的单个时间点NDVI值进行描述,而是利用稳定性较强的倒三角形面积来描述NDVI的整体走向。在此基础上,利用超高分辨率的无人机合成相片对本研究提出的算法进行了训练和验证,结果表明,使用ITA算法识别桉树,用户精度和制图精度接近于80%,而其他三种传统的算法的分类精度只有50%左右。为了更好地验证ITA的适用性,本文还使用了GF-1数据,选取了两块较大区域对ITA算法进行验证,结果也表明,分类的用户精度和制图精度达到了76%,从而证明了使用倒三角形面积算法进行小区域尺度桉树识别的可行性。3.桉树人工林种植时间估算在分析了135个参考桉树像元快速生长期内影像最早获取时间和生长期类型(一年长和两年长)之间的关系后,本论文进行了大胆的假设:引入了影像最早获取时间与砍伐时间之间的时间间隔?t这一参数,利用其来推算桉树相对精确的种植时间。在得到初步的估算结果后,又使用了迭代的方式对?t进行修正,提高估算的准确度。结果表明,使用该方法估算的种植时间和实际种植时间之间的均方根误差为3个月左右,这样的估算精度是满足实际应用需求的。4.韶关市、河源市、赣州市桉树种植时空变化情况在绘制广东省韶关市、河源市、赣州市三市桉树种植时空变化情况时,结合了本文提出的倒三角形积算法和BE算法进行桉树像元识别,利用本文提出方法进行桉树种植时间估算。在此基础上获得广东省三市从2000年至2014年的桉树种植时空分布信息。结果显示,在这15年间,该研究区每年桉树种植面积都处于变化状态,其中,2007年和2008年到达种植最高峰,种植面积超过250km~2,其余年份种植面积控制在100km~2左右。
[Abstract]:Vegetation is an important part of the global terrestrial ecosystem, it is not only for the terrestrial ecosystem provides energy and power maintain its normal functioning, but also provide a habitat for other species in the ecosystem. In the global change and human activities in the growing conditions, great changes have taken place in the type and pattern of global vegetation, accompanied by a lot of native forest felling, plantation area increased dramatically, making the original ecological system in some areas is out of balance, reduce the biodiversity of the region. It is of great significance to extract economic dynamic artificial forest information protection of terrestrial ecosystems. However, recognition of economic plantation at present, the use of data for the higher time resolution, MODIS data of low spatial resolution, through the analysis of the traditional algorithms, such as block difference algorithm, standardization The Euclidean distance algorithm, Bounding Envelope algorithm and so on. But the economic plantations in recognition of these algorithms for small regional scale, is not applicable in this article. The use of Landsat 5 TM, Landsat 7 ETM and HJ 1 reconstruction of normalized difference vegetation index (NDVI) based on time series data, presents an inverted triangle area algorithm. Using the algorithm successfully extracted from Shaoguan, Heyuan and the temporal and spatial distribution of the 3 cities of Ganzhou eucalyptus plantation, also through the analysis of the relationship between growth period type and image acquisition time indirectly calculated the exact year of planting, and draws a conclusion as follows: 1. based on the Landsat 5 TM, a large number of missing and quality Landsat 7 ETM and HJ 1 Landsat 5 NDVI time series data reconstruction of TM data makes the Landsat ETM 7 and HJ 1 data to fill the vacancy. Different sources of data, the spectral response function The number of satellite orbit, acquiring the image when the sun angle, observation angle etc. there may be differences, in order to explore the differences in the size of the 4 in 2012 to 5 LANDSAT TM and HJ images were analyzed respectively in comparing the more than 100 points NDVI, NDVI found that the correlation between the two are reached more than 98%, 5 TM/ETM HJ LANDSAT that, although the data are slightly different, but not enough to affect the classification. It proved that the use of LANDSAT 5 TM/ETM, HJ NDVI data to reconstruct the time series classification is feasible.2. NDVI time series inverted triangle area recognition algorithm of regional scale for Eucalyptus Plantation in this study, there is a continuous reconstruction based on the minimum time interval is too large, considering the pixel of Eucalyptus Plantation reference sequence rule, this paper proposes a new algorithm- Inverted triangle area algorithm (ITA). Compared with the other three Eucalyptus species identification algorithm, it avoids the floating value of a single point in time NDVI large values are described, but the use of strong stability of inverted triangle area to describe the overall trend of NDVI. On this basis, the use of UAV photo synthesis was trained and validated on this study proposes an algorithm for ultra high resolution. The results show that the ITA algorithm is used to identify users of Eucalyptus, precision and mapping accuracy close to 80%, while only 50% of the classification accuracy of the other three traditional algorithms. In order to better verify the applicability of ITA, this paper also used the GF-1 data, the selection of the two large area to validate the ITA algorithm, the results also show that the classification accuracy and user mapping accuracy reached 76%, which proves that the use of inverted triangle area algorithm for regional scale identification of the feasibility of.3. Eucalyptus Eucalyptus The tree plantation planting time estimation in the analysis of the 135 reference pixel fast growth period of Eucalyptus is the earliest image acquisition time and growth period (a type of older and the relationship between the two years long), this paper carried out bold assumption: the time interval between the first introduced image acquisition time and cut down t? This parameter, use it to calculate time of planting Eucalyptus. Relatively accurate estimation results obtained in the preliminary, and use the iterative method of? T is modified, improve the estimation accuracy. The results show that the root mean square error between the use of the method to estimate the actual planting time and planting time is 3 this month, the estimation accuracy is to meet the needs of practical application of the.4. in Shaoguan City, Heyuan City, Ganzhou City, temporal and spatial variation of Eucalyptus Plantation in drawing Guangdong city of Shaoguan Province, Heyuan City, Ganzhou City, three city, temporal and spatial variation of Eucalyptus Plantation When combined with the proposed algorithm inverted triangle and BE algorithm using pixel identification of Eucalyptus, this paper proposes the method of Eucalyptus planting time estimation. Eucalyptus Plantation distribution information on the basis of the three cities of Guangdong province from 2000 to 2014. The results showed that, in the past 15 years, the study area every year Eucalyptus planting area a state of change, among them, in 2007 and 2008 to reach the peak of planting, planting area of more than 250km~2, the rest of the year of planting area of control in about 100km~2.

【学位授予单位】：中国科学院大学(中国科学院遥感与数字地球研究所)
【学位级别】：博士
【学位授予年份】：2017
【分类号】：S771

【相似文献】