發(fā)布時間：2018-04-23 12:20

  本文選題：土地覆蓋/土地利用 + 上海��； 參考：《華東師范大學》2016年博士論文

【摘要】：在對地觀測衛(wèi)星技術(shù)迅猛發(fā)展、遙感應用研究成果豐碩的今天,如何準確區(qū)分城市中具有復雜混合光譜的地表覆蓋仍然是困擾國內(nèi)外遙感研究的一個核心問題。這個問題在歷史悠久的中國城市中尤顯突出,其城區(qū)形成過程經(jīng)歷過千百年的朝代更迭和文化興衰,無論是空間結(jié)構(gòu)布局還是建筑材質(zhì)用料,復雜程度遠遠超出西方發(fā)達國家的同等規(guī)模城市。這些歷史文化和空間結(jié)構(gòu)的特殊性導致了高度復雜的地物光譜特征,給中國城市遙感研究帶來了嚴重挑戰(zhàn)。而過去三十年的改革開放,大規(guī)模的城市化進一步強化了城市地表的光譜異質(zhì)性。從鄉(xiāng)村農(nóng)田到城市土地利用類型的快速轉(zhuǎn)型,自然地表環(huán)境開發(fā)為以不透水面為主體的居住、工業(yè)或商業(yè)用地,中心城區(qū)歷史建筑的遺存及土地利用的集約型高強度重組,以及對城市水體的改造利用乃至污染,形成了高度破碎化、異質(zhì)化的城市景觀,使城市下墊面各要素的光譜特征更加錯綜復雜。面對上述嚴峻的技術(shù)和應用挑戰(zhàn),本文以上海為例,針對中國城市景觀破碎度大、地表覆蓋空間變異強度高、光譜同質(zhì)性低的特點,利用多時相、多源遙感數(shù)據(jù),研發(fā)和測試一系列適合高度異質(zhì)城市下墊面地表覆蓋特征量化的方法,研明城市發(fā)展模式與軌跡,并進一步分析其對城市熱環(huán)境的影響。研究旨在準確、高效和及時評估和監(jiān)測高度異質(zhì)的城市地表覆蓋范圍、分布結(jié)構(gòu)、地物成分以及時空變化,以利于從宏觀和中觀層面上了解城市化進程,為合理利用與規(guī)劃土地覆蓋,緩解城市化所帶來的一系列問題提供科學依據(jù)。本研究主要工作包括以下五個方面：(1)在城市水體的提取研究中,設(shè)計了一種由粗到細(coarse-to-fine, CTF)的城市水體提取策略,方法結(jié)合不同時相的熱紅外和光學影像數(shù)據(jù)來應對城市水體的高度光譜復雜性和季節(jié)性變化。傳統(tǒng)的提取方法多基于像元水平上的水體探測,無法準確地識別城市內(nèi)部細小的水體,且水體的識別易受低照不透水面和陰影的干擾。而本研究中的CTF方法可結(jié)合線性光譜混合分解模型和多時相變化檢測技術(shù)來準確識別亞像元水平上的水體特征,方法尤其適用于河網(wǎng)水系復雜的城市地區(qū)。CTF方法通過對試驗區(qū)(上海)不同時相的Landsat ETM+影像的分析結(jié)果表明：方法可以有效地在高度異質(zhì)的城市環(huán)境內(nèi)識別永久性和季節(jié)性水體,并達到滿意的精度。(2)在城市不透水面的提取研究中,利用四面體模型(植被-高照不透水面-低照不透水面-土壤,V-H-L-S)解決了傳統(tǒng)的植被-不透水面-土壤(V-I-S)理論模型與實際影像所提取端元不匹配的問題。方法首先通過四面體模型來描述最小噪聲分離(minimum noise fraction, MNF)變換后的像元在三維空間的分布特征,然后通過多目標優(yōu)化的遺傳算法來確定四面體的頂點,即城市地表的四個端元分別位于各個頂點的小四面體內(nèi),進而確定三維空間內(nèi)端元位置,計算出各地物覆蓋類型的影像端元光譜。上海主城區(qū)實例數(shù)據(jù)的驗證充分表明基于四面體的V-H-L-S模型能較好地解釋城市地表分解為四組分的情況,相比傳統(tǒng)的端元獲取方法(像元純凈度指數(shù)法和二維散點圖法)自動化程度更高,且可以獲得更為理想的不透水面定量結(jié)果。(3)在城市植被的估測研究中,將增強型時空自適應反射率融合模型(enhanced temporal adaptive reflectance fusion model, ESTARFM)用于融合Landsat和MODIS數(shù)據(jù)以生成高空間分辨率和高時相分辨率的NDVI時間序列,進而利用時空混合分析技術(shù)(temporal mixture analysis, TMA)對NDVI時相曲線進行分解,以估測農(nóng)田、常綠、落葉植被覆蓋情況。研究發(fā)現(xiàn)利用TMA方法分解NDVI時間序列生成的植被覆蓋度與地面參考數(shù)據(jù)的一致性較好(R2大于0.79,RMSE小于0.11,MAE小于0.84)�；跁r空融合模型獲取高空間和高時相分辨率的遙感數(shù)據(jù)對于復雜城市環(huán)境的植被監(jiān)測是十分關(guān)鍵的,所提取的時相端元相比傳統(tǒng)的光譜端元,可以更準確地區(qū)分出更多的植被類型。(4)在城市土地覆蓋的變化檢測研究中,采用了基于地表物理組分的長時間序列變化檢測技術(shù)。方法首先利用不同年份(1990,1995,2000,2003,2007,2013)30m空間分辨率NDVI時序數(shù)據(jù),基于時空混合分析模型獲取各個年份的城市不透水面信息,進而采用Z-score分析對亞像元尺度的歷年不透水面覆蓋度進行變化監(jiān)測。方法相比傳統(tǒng)的基于像元尺度的變化檢測技術(shù),可以獲取土地覆蓋更為精細和穩(wěn)定的變化信息,包括二值變化信息、變化強度和方向信息。航拍影像的精度驗證表明方法可以提高不透水面估測精度,且能精細地檢測出復雜的城市和郊區(qū)內(nèi)土地覆蓋變化特征。(5)一種新型的非線性處理方法——極端學習機(extreme learning machine, ELM)用于探索城市地表溫度(land surface temperature, LST)和不透水面組分之間的非線性關(guān)系。通過不同年份上海夏季城市熱島數(shù)據(jù)的測試比較發(fā)現(xiàn)ELM方法所建立的模型在不同年份LST的預測上都表現(xiàn)出比線性模型更高的精度,且算法的效率遠高于傳統(tǒng)的非線性方法。同時考慮鄰近像元的不透水面信息可以進一步提高模型的精度。研究結(jié)果充分說明ELM模型可以準確地處理LST模擬過程中的非線性,并且這種非線性關(guān)系可能是周邊像元的地表環(huán)境共同作用的結(jié)果。
[Abstract]:With the rapid development of earth observation satellite technology and the fruitful research achievements of remote sensing applications, it is still a core problem that how to accurately distinguish the surface coverage of complex mixed spectra in cities. This problem is particularly prominent in Chinese cities with a long history, and the process of urban formation has experienced hundreds of thousands of years. The changes in the dynasties and the rise and fall of the culture in the years, whether the spatial structure or the material of the building material, are far more complex than the same scale cities in the western developed countries. The particularity of these historical and cultural and spatial structures has led to the highly complex spectral characteristics of the ground objects, which have brought serious challenges to the research of urban remote sensing in China. In the past three Ten years of reform and opening up, large-scale urbanization further strengthened the spectral heterogeneity of the urban surface. From rural farmland to urban land use type, the natural surface environment is developed into inhabitation, industrial or commercial land, the remains of historical buildings in the central city and intensive high-strength of land use. Degree reorganization, as well as the transformation and utilization of urban water and even pollution, form a highly fragmented and heterogeneous urban landscape, and make the spectral characteristics of the urban underlying elements more complex. In the face of the severe technical and application challenges mentioned above, this paper takes Shanghai as an example, and has a large fragmentation of the Chinese urban landscape and the strong surface coverage of the space variation. With the characteristics of high degree and low spectral homogeneity, using multi time phase and multi source remote sensing data, we developed and tested a series of methods suitable for the quantification of surface cover characteristics of the undercover of highly heterogeneous cities, studied the urban development model and trajectory, and further analyzed its impact on the urban thermal environment. The research aims to accurately, efficiently and timely assess and monitor the height. Heterogeneous urban surface coverage, distribution structure, composition of ground objects and temporal and spatial changes in order to understand the process of urbanization from the macroscopic and meso level, and provide scientific basis for the rational utilization and planning of land cover and alleviating a series of problems brought by urbanization. The main work of this study includes the following five aspects: (1) in urban water In the study of body extraction, a city water extraction strategy from coarse-to-fine (CTF) is designed. The method is combined with the thermal infrared and optical image data of different phases to cope with the high spectral complexity and seasonal variation of urban water. The traditional extraction method is based on the water body detection on the pixel level and can not be accurately identified. In this study, the CTF method can be used to identify the water characteristics at the sub pixel level accurately, which is especially suitable for the.CTF square in the complex urban area of the river network. The method of analysis of Landsat ETM+ images of different phase in the test area (Shanghai) shows that the method can effectively identify permanent and seasonal water bodies in highly heterogeneous urban environment and achieve satisfactory accuracy. (2) in the study of the extraction of urban impermeable water, the tetrahedral model (vegetation - high exposure to water and low illumination) Water surface soil, V-H-L-S) solves the problem that the Traditional Vegetation - impermeable surface - soil (V-I-S) theory model does not match the end element of the actual image. Method first, a tetrahedral model is used to describe the distribution characteristics of the image element in the three-dimensional space after the minimum noise separation (minimum noise fraction, MNF) transformation, and then through multi-objective optimization. The genetic algorithm is used to determine the vertex of tetrahedron, that is, the four endpoints of the urban surface are located in the small tetrahedron of each vertex, and then the end element position in the three-dimensional space is determined and the image end spectrum of each cover type is calculated. The verification of the example data of the Shanghai main city area fully shows that the V-H-L-S model based on the tetrahedron can be better. To explain the situation of urban surface decomposition to four components, it is more automated than the traditional method of endpoint acquisition (pixel pure index method and two-dimensional scatter plot method), and can obtain more ideal quantitative results of impermeable surface. (3) an enhanced spatio-temporal adaptive reflectivity fusion model (enhance) is used in the study of urban vegetation. D temporal adaptive reflectance fusion model, ESTARFM) is used to fuse Landsat and MODIS data to generate high spatial resolution and high phase resolution of NDVI time series, and then decomposes the phase curve with a spatio-temporal hybrid analysis technique (temporal mixture) to estimate farmland, evergreen, and deciduous vegetation. The study found that the vegetation coverage generated by the TMA method to decompose the NDVI time series is in good agreement with the ground reference data (R2 is greater than 0.79, RMSE is less than 0.11, MAE is less than 0.84). The remote sensing data of high space and high phase resolution based on the spatio-temporal fusion model is critical to the vegetation monitoring in complex urban environment. The time phase end element can be more accurately divided into more vegetation types compared with the traditional spectral endpoints. (4) in the study of the change detection of urban land cover, the long time series change detection technology based on the surface physical components is adopted. First, the spatial resolution NDVI of different years (199019952000200320072013) 30m is used. The time series data, based on the spatio-temporal hybrid analysis model, is used to obtain the urban water surface information of each year, and then the Z-score analysis is used to monitor the variation of the surface coverage of the sub pixel scales. The method can obtain more precise and stable change letters of land coverage compared with the traditional change detection technology based on pixel scale. Interest, including two value change information, change intensity and direction information. The accuracy verification of aerial image shows that the method can improve the accuracy of impermeable surface estimation, and can accurately detect the complex characteristics of land cover change in the complex city and suburb. (5) a new nonlinear processing square method (extreme learning machine, ELM) It is used to explore the nonlinear relationship between the urban surface temperature (land surface temperature, LST) and the impermeable water components. Through the comparison of the summer urban heat island data in different years in Shanghai, it is found that the model established by the ELM method shows a higher accuracy than the linear model in the prediction of the LST in different years, and the efficiency of the algorithm is far away. It is higher than the traditional nonlinear method. Considering the impermeable surface information of adjacent pixels can further improve the accuracy of the model. The results of the study fully demonstrate that the ELM model can accurately deal with the nonlinearity in the LST simulation process, and this nonlinear relationship may be the result of the joint action of the surrounding surface environment.

【學位授予單位】：華東師范大學
【學位級別】：博士
【學位授予年份】：2016
【分類號】：P237;X16;X87
，

本文編號：1791957