EGI OpenIR
基于遥感时空融合算法的哈萨克斯坦春小麦旱情及产量监测研究
Alternative TitleStudy on Spring Wheat Drought and Yield Monitoring in Kazakhstan based on Spatial and Temporal Fusion Algorithm
姚远
Subtype博士
Thesis Advisor陈曦
2019-06-30
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline理学博士
Keyword农业旱情 遥感 时空融合 哈萨克斯坦 Agriculture Drought Remote Sensing Spatiotemporal Fusion Kazakhstan
Abstract在全球气候变化的背景下, 干旱灾害已成为当前全球范围内发生频率最高、影响范围最广、持续时间最长的自然灾害类型。在不同干旱灾害类型中,农业干旱对全球生态安全和粮食安全的影响最为直接也最为明显,严重影响社会稳定和区域经济发展。 哈萨克斯坦作为我国“丝绸之路经济带”建设的核心区域, 既是世界第六大粮食出口国和全球最大的小麦出口国,也是我国在农业粮食方面最为重要的战略合作伙伴。 因而, 进一步提高我国农业旱情遥感监测技术在中亚农情、旱情等方面的监测能力和应用水平,对于促进我国的遥感监测技术走出国门,保障我国和中亚地区粮食安全有着极为重要的现实意义。本研究首先系统梳理了当前遥感数据在农业旱情监测研究中的现状,总结了当前利用遥感卫星数据开展区域农业旱情监测研究存在的问题。在此基础上, 以Landsat 8 和 MODIS 遥感影像为基础数据源, 以遥感时空融合技术为主要研究方法, 以春小麦为研究对象,以占哈萨克斯坦小麦总产量 80%以上的北部 3 州(北哈萨克斯坦州、 库斯塔奈州和阿克莫拉州)为研究区, 重点从以下几个方面开展了研究区农业旱情的监测研究工作:(1) 根据遥感反射率时空融合模型和地表温度时空融合模型对高、低空间分辨率输入影像的要求, 首先,采用卷云波段结合遥感数据的红外波段和可见光波段光谱特征的方法去除中低空其它薄云对 Landsat 8 高空间分辨率基础输入影像的影响,以受云影响的 Landsat 8 数据为示例, 从图像去云前后 1~5 波段的视觉效果以及平均值、标准差和不同剖面反射率值变化等方面进行定性和定量分析验证去云效果。 研究结果表明,该方法可以有效弱化和消除薄云对 Landsat 8 基础输入影像的影响。 其次, 在此基础上利用基于单窗算法反演的地表温度 LST(Land Surface Temperature) 与归一化植被指数 NDVI(Normalized DifferenceVegetation Index)之间的线性关系估算了薄云覆盖区域的 LST 数据,并利用同期获取的 MODIS 地表温度产品进行交叉验证分析,结果表明估算的 LST 结果精度较高且满足研究需要。 最后,对用于遥感时空融合且受云影响的低空间分辨率MODIS 输入数据采用 YCbCr 颜色空间进行云及阴影检测, 在此基础上采用像元替换补偿的方法对 MODIS 反射率数据和地表温度数据进行像素替换去云。 结果表明, 在图像进行云影处理补偿后,其像元差异较有云影时有较大幅度降低,图像的标准差变小,表明图像的原始信息得到有效保留, 可以有效实现云影区域的去除。(2) 分别从遥感反射率融合模型中选取基于时空自适应融合模型融合方法中的代表性方法 STARFM(Spatial and Temporal Adaptive Reflectance FusionModel)、 基于线性混合模型融合方法中的代表性方法 FSDAF(Enhanced FlexibleSpatiotemporal Data Fusion) 和基于学习模型融合方法中的代表性方法 SPFMOL(spatiotemporal fusion model through one image pair learning),对研究区的遥感反射率数据进行时空融合, 从中选出最佳方法用于生成哈萨克斯坦北部 3 州 2018年春小麦关键生育期(6 月上旬至 9 月上旬)旬尺度高空间分辨率反射率遥感影像。 研究结果表明, SPFMOL 模型因其充分考虑到输入影像遥感分辨率差异并采用双层融合和逐层尺度递推的方式,结合字典训练所获得的较为丰富的纹理结构特征,因而在融合效果上要优于其它两种方法,因而本研究基于 SPFMOL 模型生成研究区春小麦关键生育期旬尺度的反射率影像数据集。(3) 针对当前地表温度时空融合模型的不足,提出一种改进型灵活的时空融合算法 EFSDAF(Enhanced Flexible Spatiotemporal Data Fusion),并与已有的STITFM 模型(Spatio-temporal Integrated Temperature Fusion Model) 和 FSDAF算法的融合结果进行比对。研究结果表明, EFSDAF 算法的融合精度和处理效率最好, 为生成研究区春小麦关键生育期旬尺度地表温度影像数据集的最优方法。在此基础上, 为获取与反射率时空融合数据相对应的 30 m 空间分辨率的旬尺度LST 数据, 采用 TUS(Temperature Unmixing with Spectral)地表温度光谱分解模型进行 LST 像元分解。同时采用升降尺度的精度验证方式,将 TUS 与最邻近法以及双线性内插法的分解结果进行比较。 研究结果表明,经过 TUS 地表温度光谱分解模型降尺度后生成的地表温度数据精度较高, 空间表现效果更接近 LST真实影像。(4) 为了构建用于研究区春小麦旱情监测的旬尺度高空间分辨率植被遥感监测指数数据集,首先,基于时空融合后生成的研究区春小麦抽穗期遥感影像,利用基于知识的面向对象分类方法准确提取了 2018 年哈萨克斯坦北部 3 州及其下属各行政区的春小麦种植面积,通过与官方统计数据进行对比分析表明,分类结果能够满足实际应用需求。 其次,基于遥感反射率时空融合模型和地表温度时空融合模型构建了研究区旬尺度高空间分辨率 NDVI, EVI(Enhanced VegetationIndex), VTCI(Vegetation Temperature Condition Index), TCI(TemperatureCondition Index), VHI (Vegetation Health Index), VCI (Vegetation Condition Index)和 SAVI(Soil Adjusted Vegetation Index) 等 7 种当前常用于农业旱情监测和作物产量估测的植被遥感监测指数数据集。最后,利用 Savitzky-Golay 滤波法、非对称高斯函数 Asymmetric Gaussian 拟合法和双 Logistic 函数拟合法对本研究所提取的高时空植被遥感监测指数进行数据重构。经过比对, S-G 滤波在多种植被遥感监测指数的处理结果均要优于其它两种方法,可用于本研究所提取的高时空分辨率植被遥感监测指数的平滑去燥。(5) 旱情是哈萨克斯坦北部 3 州春小麦产量高低变化最重要的影响因素,为了在春小麦不同生育时期以及相应时间段内提前估测获取最终精确的春小麦单产变化信息, 从而为哈萨克斯坦的农业旱情监测提供有效的技术支持。 首先,基于重构后生成的旬尺度高空间分辨率多种植被遥感监测指数分别建立了研究区旬尺度和生育期尺度的单因子、 全时段和多因子春小麦单产估算模型。 结果表明, 基于拔节期和抽穗期 VCI 指数建立的生育期尺度多因子春小麦单产估算模型的估产效果最好, 为研究区最优春小麦单产估算模型, 该模型可以有效解决当前最为常用的 MODIS 数据和 MODIS 植被指数产品 16 天合成数据应用于农作物估产和旱情监测研究时存在的空间分辨率不足和时相信息不准确等问题。其次,利用全球降雨观测计划的 GPM/IMERG 终极多星融合反演降水产品(Final Run)的逐月降水资料结合干旱监测 Z 指数方法,计算了研究区春小麦生育期不同等级的干旱空间分布特征。 结果表明, 研究区最优春小麦单产估算模型的估产结果与春小麦生育期的干旱空间分布特征具有一致性,利用该模型可以准确反映研究区春小麦在不同空间区域的受旱情况。最后, 本研究将基于拔节期和抽穗期 VCI指数所构建的遥感单产修正模型与基于研究区春小麦历史单产统计数据所构建的趋势单产模型相结合建立了综合估产模型。结果表明,综合估产模型可以提高研究区整体及其州级、区级尺度的春小麦单产估算精度, 有效实现研究区春小麦单产估算从像元尺度到研究区整体尺度的扩展,从而为当地政府农业管理和决策部门有效开展春小麦旱情和产量监测提供技术保障和数据支撑。
Other AbstractIn the context of global climate change, drought disasters have become the mostfrequent, influential and longest-lasting type of natural disasters on a global scale.Among different types of drought disasters, agricultural drought affects globalecological security and food security, and seriously affects social stability and regionaleconomic development. As the core area of China's “Silk Road Economic Belt”,Kazakhstan is not only the world's sixth largest grain exporter country and the world'slargest wheat exporter country, but also an important strategic partner of China inagriculture. Therefore, to further improve the application ability of China'sagricultural drought remote sensing monitoring technology in Central Asia, andcontinuously improve the application level of this technology in other country in theCentral Asia, it has extremely significance for ensuring food security in China andCentral Asia.This study summarizes the application of current remote sensing data inagricultural drought monitoring research, and also summarizes the problems existingin the current regional agricultural drought monitoring research. Based on thissummary, Northern Kazakhstan state, Kustanay state and Akmola state, which accountfor more than 80% of the total spring wheat production in Kazakhstan, are selected asthe study areas. We selected spatiotemporal fusion models as the main research methodto combine the high temporal resolution of MODIS data and high spatial resolution ofLandsat 8, in order to quantitative carrying out remote sensing monitoring of springwheat drought in study area. The research content and main results were listed asbelow:(1) In order to meet the requirements of input data for s spatiotemporal fusionmodels. On the one hand, using cirrus and quality assurance bands of Landsat 8 toremoval of thin clouds, and estimation of LST (Land Surface Temperature) in thinclouds region by NDVI (Normalized Difference Vegetation Index). The result showsthat our method can effectively remove thin cloud in band 1~5, using NDVI toestimate the LST in thin area has certain feasibility. On the other hand, usingappropriate pixels within the remaining regions of the target MODIS data to fill thecloud masked pixel of cloud and shadow based on the automatic cloud and shadowdetection. The results show that this method is highly valid in MODIS data cloud removing and information restore.(2) Compare three spatiotemporal fusion models (STARFM, FSDAF, SPFMOL)under a one Landsat 8-MODIS pair estimation mode using remote sensing data fromthe study area, and quantitatively assess estimation accuracy of different fusionmodels, in order to selecting the best model to generate high spatiotemporal resolutionreflectance images of the spring wheat key growth period in the study area. Theexperimental results demonstrated that SPFMOL model performed better than otherspatiotemporal fusion algorithms.(3) In this study, we propose a novel LST spatiotemporal fusion model called theEFSDAF (Enhanced Flexible Spatiotemporal Data Fusion) to generate LST image onestimation time. This method was tested with an actual LST data, and also comparedwith other well-know LST spatiotemporal fusion model (STITFM and FSDAF). Theresult show that the proposed method is more suitable for the study area and exhibitsgood fusion LST precision. In order to increase the spatial resolution of predict LSTimage based on the EFSDAF method form 100 m to 30 m, we used TemperatureUnmixing with Spectral Model to make the pixel decomposition of predict LST imagewith high spatial of 30 m. We also used upscale-downscaled verification method toprove the feasibility of our method. Firstly, we scale the LST with the spatialresolution of 100 m up to 300 m, then downscaled to 100 m spatial resolution withTUS method, and comparing this result with nearest neighbor method and bilinearinterpolation method. The result shows that TUS method after pixel decompositionwith the highest accuracy for decomposition of LST image than other methods.(4) In order to construct a high spatiotemporal resolution vegetation index dataset for spring wheat drought monitoring in the study area. Firstly, we estimated springwheat planting area in each administrative district that belongs to NorthernKazakhstan state, Kustanay state and Akmola state using predicted image of springwheat heading stage based on SPFMOL model. The knowledge-based object orientedclassification method was used to extract spring wheat planting area. Accuracyassessment proves that the classification results can meet the requirement ofapplication. Secondly, retrieval of NDVI, EVI (Enhanced Vegetation Index), VTCI(Vegetation Temperature Condition Index), TCI (Temperature Condition Index), VHI(Vegetation Health Index), VCI (Vegetation Condition Index) and SAVI (SoilAdjusted Vegetation Index) with high spatiotemporal resolution predicted data basedon spatiotemporal fusion model. Finally, we compared the performance of three time-series reconstruction methods, including Savitzky-Golay filtering, AsymmetricGaussian fitting and Double Logistic function fitting, on high spatiotemporalresolution vegetation index data sets. The result showed that Savitzky-Golay filteringare more suitable for reconstruction of vegetation index, the reconstruction curveshave high consistency in expression vegetation phenology.(5) Drought is the most important factor affecting the change of spring wheatyield in the Northern Kazakhstan state, Kustanay state and Akmola state of northernKazakhstan. In order to obtain the final accurate information on spring wheat yieldchange in the different growth stages of spring wheat and corresponding period, andprovides effective technical support for agricultural drought monitoring in Kazakhstan.Firstly, this study established a spring wheat yield estimation model based on the highspatiotemporal resolution vegetation index in the study area. The results show that thespring wheat yield estimation model based on VCI in elongation stage and headingstage of spring wheat critical growth period has the best yield estimation effect, whichis the optimal spring wheat yield estimation model in the study area. This model caneffectively solve the problems of insufficient spatial resolution and inaccurate timeinformation of MODIS data when it applied to crop yield estimation and droughtmonitoring research. Secondly, this study constructed Z index using monthlyprecipitation data of GPM/IMERG (Final Run) data to study the regional drought. Theresults show that the spring wheat yield estimation results based on optimal springwheat yield estimation model in the study area are consistent with the spatialdistribution characteristics of spring wheat growth period. The model can accuratelyreflect the drought condition of spring wheat in different spatial regions in the studyarea. Finally, in order to further give full play to the advantages and applications ofthe optimal spring wheat yield estimation model in this study, we proposed a novelcomprehensive spring wheat yield estimation model which combines trend yieldmodel based on a long time series data of historical yield and then corrected byARIMA model, remote sensing data correction yield model based on VCI inelongation stage and heading stage of spring wheat critical growth period, and an erroritem. The results show that the accuracy of spring wheat yield by using thecomprehensive spring wheat yield estimation model is good at whole study area level,state level, country level. The model can effectively realize the expansion of springwheat yield estimation in the study area from the pixel scale to the overall scale of thestudy area, it can provide technical support and data support for the local government agricultural management and decision-making departments to effectively carry outspring wheat drought monitoring.
Subject Area地图学与地理信息系统
Language中文
Document Type学位论文
Identifierhttp://ir.xjlas.org/handle/365004/15273
Collection中国科学院新疆生态与地理研究所
研究系统
Affiliation中国科学院新疆生态与地理研究所
First Author Affilication中国科学院新疆生态与地理研究所
Recommended Citation
GB/T 7714
姚远. 基于遥感时空融合算法的哈萨克斯坦春小麦旱情及产量监测研究[D]. 北京. 中国科学院大学,2019.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[姚远]'s Articles
Baidu academic
Similar articles in Baidu academic
[姚远]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[姚远]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.