【作者】 黄昌；

【导师】 吴健平； 陈芸；

【作者基本信息】 华东师范大学 ， 地图学与地理信息系统， 2014， 博士

【摘要】 洪水,作为常见的自然现象之一,不仅影响着人们的生活和生产,也影响着自然界多种生物群体的生存与繁荣。作为洪水最重要的性质,洪水淹没一直是洪水研究重点关注的问题。对洪水淹没的研究包括对洪水淹没的探测、对淹没范围的制图、以及以淹没概率估测和淹没范围预测为代表的洪水淹没建模。随着空间技术的发展,地理信息系统(Geographic Information System, GIS)和遥感(Remote Sensing, RS)技术在其中的应用日益突出,相关研究方法日益丰富和成熟。不过,就目前看来,依然有很多问题值得进一步研究。首先是基于遥感技术探测洪水淹没的精度和稳定性问题,以及遥感影像自身空间分辨率和时间分辨率之间的相互制约对洪水淹没探测和制图的影响。其次,对一个区域来说,不仅需要探测在某次洪水中是否被淹没,而且需要分析其被洪水淹没的概率,但目前只有一些单纯基于时间序列流量数据分析洪水概率的方法,结合遥感和GIS方法的应用很少,因而就很难获取具有空间特征的洪水淹没概率。最后,对河漫滩区域来说,是否被洪水淹没与上游来水有密切关系,但同时又和地形有关。目前的一些研究基于河道观测流量和河漫滩淹没之间的相关关系建立了经验关系模型,但这些模型大多未考虑地形因素,对淹没范围及水深的分析与建模缺乏足够的理论支持。本文在分析了国内外已有研究的基础之上,提出基于RS和GIS的河漫滩洪水淹没分析与建模方法,方法涵盖了洪水淹没的探测、制图和建模等方面。论文的主要研究内容和成果如下：(1) MODIS (Moderate-resolution Imaging Spectroradiometer)数据具有覆盖范围广、重复周期短、易获取等优点,是比较理想的大范围洪水淹没探测工具。本文详细介绍了MODIS数据及其在洪水淹没探测中的应用,并概括了利用MODIS数据探测洪水淹没的方法,重点介绍了水体指数法。考虑到现有的指数方法大多需要人为地调整水体分割的阈值,导致它们不太适合对洪水淹没范围的自动提取,本文引入开放水体似然性(Open Water Likelihood, OWL)指数,它的特点是其在时间序列遥感影像上一致和稳定的表现。本文利用更高空间分辨率的同期Landsat影像,对MODIS影像上基于OWL指数的水体判读的精度和可靠性进行了分析和论证,分析结果显示利用OWL指数从MODIS影像中提取水体具有较高的精度和较好的稳定性,可以使用统一的阈值对影像上的淹没范围进行自动识别。(2)水文站观测流量数据最大的优点是其具有很长的历史记录,而遥感技术的特点是其可以快速获取洪水淹没的空间分布。本文提出了基于时间序列观测流量数据和MODIS数据的河漫滩洪水淹没时空分析方法,充分发挥两类数据各自的优势,实现了对大区域尺度洪水淹没的时间和空间特征的分析。本文以澳大利亚的Murray-Darling Basin (MDB)作为实验区,首先对该实验区基于河道网络数据和水文站位置信息进行了一个分区的工作,保证了在各个分区内观测流量和遥感探测的洪水淹没之间的密切联系。基于这个分区框架,本文利用时间序列观测流量数据和MODIS数据得到了这个大的流域盆地内的包括淹没持续时间、年际淹没模式、淹没频率和淹没概率等洪水淹没特征。(3)考虑到当前的基于河道流量数据和河漫滩淹没范围经验关系模型的洪水淹没建模方法在理论基础上的不足,本文引入地形数据,提出了河段尺度的基于观测流量数据、Landsat影像以及数字高程模型(Digital Elevation Model, DEM)数据的洪水淹没连通性及水深分析与建模方法。该方法被应用于MDB内的一段典型河段,得到了该区域内下游河漫滩淹没状况与上游水文站观测流量之间的相关关系,并以此预测不同流量条件下河漫滩区域的各处是否被淹没及淹没的水深。(4)在利用遥感影像进行洪水淹没制图的时候,混合像元问题是制约制图精度的重要因素,尤其是对于MODIS这种空间分辨率相对较低的遥感数据。本文从混合像元分解与重构的角度探讨了提高洪水淹没制图精度的可行性,提出了基于DEM改进的河漫滩洪水淹没亚像元制图算法,并利用实验数据对该算法的结果进行了分析和评价。分析结果表明,相较于传统的亚像元制图算法,改进的算法得到的淹没范围无论在形态上还是在精度上都有了明显的提高。更多还原

【Abstract】 Flood is one of the most common natural phenomena across lots of regions in the world. It not only affects the living and production of human being, but also affects the survival and prosperity of flora and fauna communities along the rivers and around the lakes. As a most important characteristic of flood, inundation has always been a focus in flood studies. Researches on flood inundation involve sensing, mapping and modeling of flood inundation represented by inundation probability estimation and inundation extent prediction. With the development of space technologies, Geographic Information System (GIS) and Remote Sensing (RS) are playing a more and more important role in these studies, with increasing number of related researches emerging. However, there are still a lot of issues that need to be investigated. The first one is the accuracy and stability issue of inundation detection using remote sensing technology, as well as the impact from mutual restrain of spatial and temporal resolutions of remotely sensed imagery on flood inundation detection and mapping. In addition to that, for a specific region, analysing the historical inundation as well as future inundation probability is equally important. However, there are only some flood frequency analysis methods utilizing time-series observed flow data to reveal the flood probability. Few of these methods used GIS and remote sensing technologies, which hampered the derivation of flood inundation probability that has spatial characteristics. One more thing, for floodplain area, inundation has a close relationship with the water quantity upstream, as well as the terrain. Several studies have established empirical relationship models between in-channel observed flow and floodplain inundation, but these models did not take the terrain into consideration. Therefore, the analysis and modeling of floodplain inundation based on these relationship models did not have strong theoretical basis.After elaborated the progress of flood inundation studies, this thesis proposed a floodplain inundation analysis and modeling method based on RS and GIS. The method involved the sensing, mapping and modeling of flood inundation. Main contents and contributions of this thesis were summarized as follows: (1) MODIS (Moderate-resolution Imaging Spectroradiometer) data have several advantages such as broad coverage, short revisit time and high accessibility, which makes them an ideal tool for flood inundation detection over broad areas. This thesis made a detailed introduction on the MODIS data, and then described its application in flood inundation detection. The methods for inundation delineation from MODIS image were summarized, represented by the water index methods. It was found that most of these methods were not suitable for automatic delineation because they generally required human intervention. Therefore, an Open Water Likelihood (OWL) index was introduced here. An advantage of the OWL index is that it appears to be stable and consistent on a time-series of images. Using Landsat images which cover the same period but have a much higher resolution, a comprehensive evaluation was then conducted to ensure its accuracy and reliability in flood inundation detection from MODIS OWL imagery. Evaluation results demonstrated that inundation extents detected from a time-series of MODIS imagery using OWL index have high accuracy and strong stability, which means a universal threshold is applicable to automatically delineate inundation extent.(2) Observed flow data have a long record history, while remotely sensed data are able to reflect the spatial distribution of flood inundation quickly and efficiently. This thesis thus proposed a method for analyzing the spatio-temporal dynamics of floodplain inundation using a combination of these two types of data. This method made use of the advantages of both data to derive the spatial and temporal characteristics of flood inundation at large basin scale. The Murray-Darling Basin (MDB) in Australia was selected as a case study area. A zoning process was conducted using stream network data and gauge location data, in order to ensure the close relationship between the observed flow and remotely sensed inundation within each zone. Based on the zoning results, a series of flood inundation characteristic maps, including inundation duration map, annual inundation map, inundation frequency map and inundation probability map, were derived for this big river basin through the method of combining time-series observed flow data and MODIS imagery. (3) Existing flood inundation modeling methods that are based on the empirical relationship models between in-channel observed flow and floodplain inundation did not have strong theoretical basis. Therefore, through introducing terrain data, this thesis proposed an analysis and modeling method for flood inundation connectivity and depth at river reach scale using a combination of observed flow data, Landsat imagery and DEM (Digital Elevation Model) data. This method was then applied to a typical river reach in MDB. The relationship between downstream floodplain inundation and upstream observed gauge flow was established with stronger theoretical basis. Based on this relationship, inundation conditions including connectivity and depths over the floodplain area can be predicted under different flow conditions.(4) When remotely sensed data are utilized for flood inundation mapping, the existence of mixed pixel is an important but unavoidable factor that limits the mapping accuracy, especially for those data with coarse spatial resolutions such as MODIS. This thesis investigated the feasibility of using DEM to improve the resolution and accuracy of flood inundation maps through the method of pixel unmixing and reconstruction. It then proposed a DEM-based modified sub-pixel mapping algorithm for enhancing floodplain inundation mapping. Test data were employed to evaluate the performance of this algorithm. Evaluation results demonstrated that the modified algorithm is able to derive a better inundation map than the traditional algorithm, either in the form of shape or accuracy.更多还原