MODIS陆表产品数据重建与时间序列分析
发布时间：2018-09-04 20:03
【摘要】：航天遥感经过40多年的发展,累积了海量的遥感数据。云覆盖严重影响了遥感对地观测的数据质量,使传感器无法获取有效的地表观测数据,导致遥感观测数据产生空间不连续,时间间隔不规律的现象,从而降低了遥感数据时序分析的应用水平,限制了对遥感数据时间维度隐藏规律的认知。如何对遥感缺失和低质量数据进行数据重建,及对重建数据进行时序分析逐渐成为遥感应用领域一个新的研究热点。本文选择MODIS陆表产品中时序变化特征有代表性的归一化植被指数(Normalized Difference Vegetation Index,NDVI)和地表温度(Land Surface Temperature,LST)作为研究对象进行数据重建及时间序列分析。根据两类数据的时空特征分别设计了分形插值算法进行NDVI数据重建,以及基于逐步回归模型的LST时序重建算法,实现提高数据时空连续性的目的。通过对重建数据进行时序分析探求它们时间维度包含的信息。具体研究内容及研究成果包含以下几方面:(1)对研究区的云覆盖时空特征进行分析,定量阐述数据重建的必要性。设计了月无云概率(cfP),月平均无云率(cfA),月内80%无云期占比,月内连续无云期众数四个指标对云覆盖时空特征进行分析。结果表明云覆盖对整个研究区影响较大,cfP最大值为57.19%,最小值为18.95%,且时空分布存在差异;可以36°N为界将研究区分为南北两部分,研究区北部较南部受云覆盖影响更小;从时序上看3、4月份各指标总体表现相对较好,该时段云对研究区影响相对较小。(2)对NDVI的空间平稳性和分形特征进行研究,确定NDVI数据重建方法。低海拔区NDVI通常具有空间平稳性特征,但在四川西部、陕西南部和西藏东北的高海拔区非平稳特征明显(15%VC.),此外城市建成区,河流等与耕地或林地混杂交错区域也表现为非平稳特征;NDVI行(列)剖面线具有明显的分形特征,可把NDVI行(列)看作分形集,对不同季节、地类的NDVI行(列)剖面线进行抽样计算发现盒维数基本处于1.30-1.60之间。(3)根据NDVI行(列)剖面线的分形特征设计了NDVI数据的分形插值重建算法。算法先以分组的方法确定初始点集,利用解析法确定纵向压缩因子(id),并设计检核点集C控制迭代函数系(IFS)生成吸引子的精度;精度分析发现分形插值的精度对NDVI空间缺失尺度的响应规律不明显,缺失尺度较小时与普通克里格法(OK)的插值精度相当,当缺失尺度较大时分形插值的精度优于OK和距离反比插值法(IDW);并且分形插值较空间插值的方法能保留更多的纹理细节特征。NDVI重建数据是LST数据重建的基础数据。(4)在LST与高程、NDVI、经度和纬度因子相关性分析的基础上,设计了LST时序重建算法。重建算法利用后向剔除法进行自变量因子筛选,并通过赤池信息量准则对全回归模型进行压缩筛选,对单因子模型进行扩张筛选来确定最优回归函数。重建数据的误差较小,白天两个时点71.8%的数据,夜晚两个时点78.2%的数据可控制在3℃以内,总体上90%以上数据误差可控制在5℃以内。利用气象站点的实测数据进行时点LST重建数据精度验证时需将站点实测数据进行尺度扩展,并提出了站点实测数据尺度扩展的方法。(5)设计了云覆盖LST修正模型。该方法利用低日照时数天数对LST影响的突变特征,以期内低日照时数天数为判别条件对LST重建地温进行修正,该方法可提高云覆盖区LST的估计精度。(6)设计了基于加窗DTW距离的贴近度模糊分类算法。算法首先利用样点数据迭代计算得到各地类标准时间序列曲线,通过加窗处理提高DTW距离的计算效率及精度,结合贴近度模糊分类的方法对NDVI重建时序数据各像元进行分类。整体分类精度较高,总体分类精度为83.8%,Kappa系数为0.77,该方法适用于无采样数据年度NDVI时序数据植被信息提取。(7)对研究区内不同高程和不同地类的LST重建数据进行时序特征分析。发现不同高程的年平均LST呈平行分布的特点,高程1-2Km的区域年均LST比高程小于1Km的区域平均低2.0℃,高程大于2Km的区域年均LST比高程1-2Km区域平均低2.6℃,研究时段内不同高程区域的年均LST均呈缓慢增长趋势;不同地类的年均地温则呈水田旱地林地的特征。通过地类间地温差异分析,认为第7-11期和第24-26期地温数据可辅助用于水田与旱地分类。另外通过LST和NDVI的季节效应分析,认为季节效应对LST的影响比NDVI的更显著。(8)构建LST和NDVI时序数据的自向量回归模型,分析二者之间的时滞变化规律。LST和NDVI时序数据均为1阶单整时序变量,对水田、旱地、林地分别构建了VAR(7)、VAR(5)和VAR(2)模型,结合Granger因果关系分析认为LST和NDVI的滞后变量对NDVI的解释能力较强;通过脉冲分析认为LST受外部条件的某一冲击后,会给NDVI带来同向的冲击,但不同地类的冲击持续时间及强度不尽相同。利用本文数据重建算法实现了研究区2005-2014年的NDVI和LST两种代表性数据的数据重建,实现了提高两类数据时空连续性的目的。并充分利用NDVI时序变化特征进行植被信息提取,整体分类精度较高;利用自向量回归的方法对不同地类LST和NDVI时序数据进行时滞关系分析,发现LST和NDVI的滞后变量对NDVI的影响显著。
[Abstract]:Over the past 40 years, space remote sensing has accumulated a great deal of remote sensing data. Cloud cover has seriously affected the quality of remote sensing earth observation data, making it impossible for sensors to obtain effective surface observation data, resulting in spatial discontinuity and irregular time intervals of remote sensing observation data, thus reducing the response of time series analysis of remote sensing data. How to reconstruct the missing and low-quality remote sensing data and how to analyze the time series of the reconstructed data are becoming a new research hotspot in the field of remote sensing application. In this paper, we choose the normalized vegetation which has typical temporal variation characteristics in MODIS land surface products. The Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) were used to reconstruct data and analyze time series. The research contents and achievements include the following aspects: (1) The spatial-temporal characteristics of cloud cover in the study area are analyzed, and the necessity of data reconstruction is explained quantitatively. Mean cloud-free rate (cfA), 80% of cloud-free period in a month and mode of continuous cloud-free period in a month were used to analyze the spatial and temporal characteristics of cloud cover. The northern part of the study area is less affected by cloud cover than the southern part; the overall performance of the indicators in March and April is relatively good, and the impact of cloud on the study area is relatively small. The non-stationary characteristics of the high altitude areas in southern Shaanxi and northeastern Tibet (15% VC.) are obvious, and the non-stationary characteristics of the urban built-up areas, rivers and other mixed interlaced areas with cultivated land or forest land are also obvious; the NDVI row (column) profile has obvious fractal characteristics, which can be regarded as a fractal set, and the NDVI row (column) profile of different seasons and terrain types can be regarded as a fractal set. (3) According to the fractal characteristics of NDVI row (column) profiles, a fractal interpolation reconstruction algorithm for NDVI data is designed. Firstly, the initial point set is determined by grouping method, the longitudinal compression factor (id) is determined by analytic method, and the checkpoint set C control iterative function system (IFS) is designed. Accuracy analysis shows that the fractal interpolation accuracy has no obvious response to the missing scale of NDVI space, and the interpolation accuracy of fractal interpolation is equivalent to that of ordinary Kriging method (OK) when the missing scale is small, and the accuracy of fractal interpolation is better than that of OK and inverse distance ratio interpolation (IDW) when the missing scale is large. (4) Based on the correlation analysis between LST and elevation, NDVI, longitude and latitude, a LST temporal reconstruction algorithm is designed. The reconstruction algorithm uses the backward elimination method to filter the independent variables and uses the red pool information criterion to complete the reconstruction. The regression model is compressed and screened, and the single factor model is expanded and screened to determine the optimal regression function. In order to validate the accuracy of LST reconstructed data, we need to scale up the measured data and propose a method to scale up the measured data. (5) A cloud-covered LST modified model is designed. The method can improve the accuracy of LST estimation in cloud coverage area. (6) A fuzzy classification algorithm based on windowed DTW distance is designed. Firstly, the standard time series curves of different classes are obtained by iteration of sample data, and the calculation efficiency and precision of DTW distance are improved by windowed processing. The overall classification accuracy is 83.8% and the Kappa coefficient is 0.77. This method is suitable for extracting vegetation information from NDVI time series data in the year without sampling data. (7) The temporal characteristics of LST reconstructed data from different elevations and different land types in the study area are analyzed. The annual average LST of 1-2km is 2.0 6550 Through the analysis of the difference of soil temperature between different land types, it is concluded that the 7-11 and 24-26 soil temperature data can be used to classify paddy field and dry land. The time series data of LST and NDVI are all one-order single-integer time series variables. VAR (7), VAR (5) and VAR (2) models are constructed for paddy field, dry land and forest land respectively. Combined with Granger causality analysis, the lagging variables of LST and NDVI have stronger explanatory power to NDVI. NDVI is impacted in the same direction, but the duration and intensity of impacts of different terrain types are not the same. The data reconstruction algorithm of this paper is used to reconstruct two representative data of NDVI and LST from 2005 to 2014 in the study area, and the purpose of improving the spatiotemporal continuity of the two types of data is realized. The results show that the overall classification accuracy is high, and the lag variables of LST and NDVI have a significant effect on NDVI.
【学位授予单位】：长安大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P237
本文编号：2223188
[Abstract]:Over the past 40 years, space remote sensing has accumulated a great deal of remote sensing data. Cloud cover has seriously affected the quality of remote sensing earth observation data, making it impossible for sensors to obtain effective surface observation data, resulting in spatial discontinuity and irregular time intervals of remote sensing observation data, thus reducing the response of time series analysis of remote sensing data. How to reconstruct the missing and low-quality remote sensing data and how to analyze the time series of the reconstructed data are becoming a new research hotspot in the field of remote sensing application. In this paper, we choose the normalized vegetation which has typical temporal variation characteristics in MODIS land surface products. The Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) were used to reconstruct data and analyze time series. The research contents and achievements include the following aspects: (1) The spatial-temporal characteristics of cloud cover in the study area are analyzed, and the necessity of data reconstruction is explained quantitatively. Mean cloud-free rate (cfA), 80% of cloud-free period in a month and mode of continuous cloud-free period in a month were used to analyze the spatial and temporal characteristics of cloud cover. The northern part of the study area is less affected by cloud cover than the southern part; the overall performance of the indicators in March and April is relatively good, and the impact of cloud on the study area is relatively small. The non-stationary characteristics of the high altitude areas in southern Shaanxi and northeastern Tibet (15% VC.) are obvious, and the non-stationary characteristics of the urban built-up areas, rivers and other mixed interlaced areas with cultivated land or forest land are also obvious; the NDVI row (column) profile has obvious fractal characteristics, which can be regarded as a fractal set, and the NDVI row (column) profile of different seasons and terrain types can be regarded as a fractal set. (3) According to the fractal characteristics of NDVI row (column) profiles, a fractal interpolation reconstruction algorithm for NDVI data is designed. Firstly, the initial point set is determined by grouping method, the longitudinal compression factor (id) is determined by analytic method, and the checkpoint set C control iterative function system (IFS) is designed. Accuracy analysis shows that the fractal interpolation accuracy has no obvious response to the missing scale of NDVI space, and the interpolation accuracy of fractal interpolation is equivalent to that of ordinary Kriging method (OK) when the missing scale is small, and the accuracy of fractal interpolation is better than that of OK and inverse distance ratio interpolation (IDW) when the missing scale is large. (4) Based on the correlation analysis between LST and elevation, NDVI, longitude and latitude, a LST temporal reconstruction algorithm is designed. The reconstruction algorithm uses the backward elimination method to filter the independent variables and uses the red pool information criterion to complete the reconstruction. The regression model is compressed and screened, and the single factor model is expanded and screened to determine the optimal regression function. In order to validate the accuracy of LST reconstructed data, we need to scale up the measured data and propose a method to scale up the measured data. (5) A cloud-covered LST modified model is designed. The method can improve the accuracy of LST estimation in cloud coverage area. (6) A fuzzy classification algorithm based on windowed DTW distance is designed. Firstly, the standard time series curves of different classes are obtained by iteration of sample data, and the calculation efficiency and precision of DTW distance are improved by windowed processing. The overall classification accuracy is 83.8% and the Kappa coefficient is 0.77. This method is suitable for extracting vegetation information from NDVI time series data in the year without sampling data. (7) The temporal characteristics of LST reconstructed data from different elevations and different land types in the study area are analyzed. The annual average LST of 1-2km is 2.0 6550 Through the analysis of the difference of soil temperature between different land types, it is concluded that the 7-11 and 24-26 soil temperature data can be used to classify paddy field and dry land. The time series data of LST and NDVI are all one-order single-integer time series variables. VAR (7), VAR (5) and VAR (2) models are constructed for paddy field, dry land and forest land respectively. Combined with Granger causality analysis, the lagging variables of LST and NDVI have stronger explanatory power to NDVI. NDVI is impacted in the same direction, but the duration and intensity of impacts of different terrain types are not the same. The data reconstruction algorithm of this paper is used to reconstruct two representative data of NDVI and LST from 2005 to 2014 in the study area, and the purpose of improving the spatiotemporal continuity of the two types of data is realized. The results show that the overall classification accuracy is high, and the lag variables of LST and NDVI have a significant effect on NDVI.
【学位授予单位】：长安大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P237
【相似文献】
相关期刊论文 前10条
1 孟小红;刘国峰;周建军;;大间距地震数据重建方法研究[J];地球物理学进展;2006年03期
2 张华;陈小宏;吴信民;;基于压缩感知理论与傅立叶变换的地震数据重建(英文)[J];Applied Geophysics;2013年02期
3 李海山;吴国忱;印兴耀;;形态分量分析在地震数据重建中的应用[J];石油地球物理勘探;2012年02期
4 孟小红;郭良辉;张致付;李淑玲;周建军;;基于非均匀快速傅里叶变换的最小二乘反演地震数据重建[J];地球物理学报;2008年01期
5 李信富;李小凡;;地震数据重建方法原理及运用[J];物探化探计算技术;2008年05期
6 刘喜武,刘洪,年静波;非均匀地震数据重建方法及其应用[J];石油物探;2004年05期
7 高建军;陈小宏;李景叶;刘国昌;马剑;;基于POCS方法指数阈值模型的不规则地震数据重建(英文)[J];Applied Geophysics;2010年03期
8 孔丽云;于四伟;程琳;杨慧珠;;压缩感知技术在地震数据重建中的应用[J];地震学报;2012年05期
9 耿丽英;马明国;;长时间序列NDVI数据重建方法比较研究进展[J];遥感技术与应用;2014年02期
10 高建军;陈小宏;李景叶;刘志鹏;张南南;;基于非均匀Fourier变换的地震数据重建方法研究[J];地球物理学进展;2009年05期
相关会议论文 前4条
1 王一多;薛威;宋建民;席平;;对瑕疵图像基于模型的部分数据重建[A];全国第一届信号处理学术会议暨中国高科技产业化研究会信号处理分会筹备工作委员会第三次工作会议专刊[C];2007年
2 林巨;金子新;;利用浅海声层析数据重建广岛湾三维潮流场[A];中国声学学会2006年全国声学学术会议论文集[C];2006年
3 白兰淑;刘伊克;卢回忆;王一博;常旭;;Curvelet域联合迭代地震数据重建[A];中国地球物理2013——第十八专题论文集[C];2013年
4 黄小刚;;F-K域加权地震数据重建[A];中国地球物理2013——第二十二专题论文集[C];2013年
相关博士学位论文 前1条
1 韩晓勇;MODIS陆表产品数据重建与时间序列分析[D];长安大学;2016年
相关硕士学位论文 前6条
1 刘丽娜;基于低秩约束的随机缺失地震数据重建[D];哈尔滨工业大学;2014年
2 杨恒;基于数据重建的江西省生态环境变化遥感分析[D];南京信息工程大学;2014年
3 王敏莹;基于压缩感知理论的地震波场数据重建[D];东北石油大学;2015年
4 陈佳铭;二维部分K空间数据重建磁共振图像[D];上海交通大学;2010年
5 郭明焕;由相叠低分辨数据重建高分辨图像算法研究[D];西北大学;2002年
6 萧牧天;IM协议分析和数据重建技术的研究与应用[D];北京邮电大学;2011年
，本文编号：2223188
本文链接：https://www.wllwen.com/shoufeilunwen/jckxbs/2223188.html
