多源遥感地表温度数据时空融合研究及应用

发布时间：2017-12-28 21:03

本文关键词：多源遥感地表温度数据时空融合研究及应用　出处：《武汉大学》2016年博士论文　论文类型：学位论文

【摘要】：陆地表面温度是反映地表状况的重要参数,其对水文、生态、环境和生物地球化学等研究有重要意义。卫星遥感技术的出现,使得大范围、连续性的地表温度产品获取成为可能。然而受到遥感卫星载荷量的限制及传感器制造上的技术瓶颈,现行的主流卫星传感器均难以获取到同时具有较高时空分辨率的地表温度数据,从而限制了遥感地表温度数据的推广和应用。多源卫星传感器的遥感定量信息时空融合是解决这一问题的有效途径,论文以遥感定量信息时空融合研究为主线,分别研究遥感反射率时空融合方法和地表温度降尺度方法,并基于这两种方法提出新的地表温度时空融合模型框架,最后,利用时空融合技术辅助实现遥感森林火灾的自适应监测应用。论文的研究内容主要包括以下四个方面：(1)基于卷积神经网络和稀疏表示理论的遥感反射率时空融合模型研究。传统的反射率时空融合模型基于邻域内相似地物对应像元的插值实现,当用于融合的影像内的地物出现较大的物候变化或明显的类型变更时,会出现较大融合误差。论文针对这一问题,从影像的空间特征信息出发,提出将反射率时空融合过程视为变化反射率影像的超分辨率重建过程,利用卷积神经网络和稀疏表示算法实现变化反射率影像的轮廓和特征细节信息的重建,并将不同观测时相对预测时相的重建影像经时间权重进行叠加实现预测时相的反射率融合。分别选取覆盖武汉和北京城郊的高地表异质性实验区进行实验,实验结果表明论文提出的地表反射率时空融合模型相对于已有方法生成的融合结果能够更好的预测出由物候作用和地物类型突变现象造成的反射率变化,并且由于同时引入了卷积神经网络和稀疏表示算法分步学习影像的特征信息,使得融合反射率影像的细节信息更清晰和准确。该融合模型能够针对高异质性区域的不同应用研究提供同时具有高时空分辨率的高精度反射率融合影像。(2)综合多地表参数的BP神经网络地表温度影像空间降尺度方法研究。遥感地表温度产品总是面临更高分辨率上信息不足的问题,需要引入更高空间分辨率的信息才能实现降尺度转换。基于这一理论,论文首先在低分辨率尺度下建立能够描述NDVI-LST分布特征的尺度转换模型,并将该模型按显式和隐式关系分别进行建模,从而提高尺度转换模型的泛化能力,以干燥地物的NDVI-LST分布特征作为显式关系进行模拟,并利用BP神经网络实现由可见光和近红外数据生成的多种地表参数与地表温度残差间的隐式复杂关系映射；然后将更高空间分辨率的地表参数引入尺度转换模型,最终实现地表温度的降尺度转换。该方法可有效缓解地表温度数据与地球系统过程精细化研究中所需要的空间特征尺度不匹配的现象。(3)基于空间降尺度技术的地表温度时空融合模型研究。传统地表温度时空融合方法容易受到像元辐射率的时空变异性影响,造成在高异质性区域的时空融合应用中会出现较大融合误差；针对这一问题,论文提出将地表温度时空融合视为预测时相地表温度空间尺度转换的新思路；将反射率时空融合模型生成的预测时相的高分辨率地表参数引入由预测时相低分辨率影像建立的地表温度降尺度模型中,最终实现预测时相地表温度影像的空间尺度转换；针对复杂地表区域,并考虑模型对于不同季节实验数据的适用性,利用多源遥感数据分别对论文模型和传统模型进行实验,实验结果表明论文提出的地表温度时空融合模型相较于传统方法更适用于高异质性地表区域的应用,并且在不同季节的实验中都体现出了良好的融合精度。该模型能够为地表温度数据在全球温度变化领域定量应用的拓展提供可靠的技术支持。(4)基于时空融合技术的森林火灾自适应监测模型研究。传统的森林火灾遥感监测模型易受地域和季节环境的影响,存在火情的漏检或误判问题。论文依据亮度温度的时空季节分布特征构建典型晴空地表亮度温度模型,利用时空融合技术得到的高分辨率亮度温度影像丰富模型的训练样本数据,以提高模型的参数率定精度；依据典型晴空地表亮度温度模型和地表发射辐射理论,建立函数化的时空动态阈值,并搭建遥感森林火灾自适应监测模型；以环境减灾小卫星和MODIS为实验数据,建立黑龙江省的典型晴空地表亮度温度模型,并对该省的火灾案例进行了实验分析,实验结果表明论文方法有效的提高了遥感森林火灾监测的时空自适应程度和监测精度。
[Abstract]:The surface temperature of the land is an important parameter reflecting the condition of the surface, and it is of great significance to the research of hydrology, ecology, environment and biogeochemistry. The emergence of satellite remote sensing technology makes it possible to obtain large scale and continuous surface temperature products. However, due to the limitation of remote sensing satellite load and technical bottlenecks in sensor manufacturing, the current mainstream satellite sensors are difficult to get the surface temperature data with high temporal and spatial resolution, which limits the promotion and application of remote sensing land surface temperature data. The remote sensing information space-time multi-source satellite sensor fusion is an effective way to solve this problem, based on quantitative remote sensing information temporal spatial fusion research as the main line, respectively study on remote sensing reflectance fusion method and surface temperature temporal downscaling method, and based on these two methods provided a framework, a new space-time surface temperature fusion model finally, adaptive monitoring the application of remote sensing technology using temporal spatial fusion assisted forest fire. The main contents of this paper include the following four aspects: (1) research on spatio-temporal fusion model of remote sensing reflectance based on convolution neural network and sparse representation theory. The traditional reflectivity temporal and spatial fusion model is based on the interpolation of the corresponding pixels in the neighborhood. When there are larger phenological changes or obvious types of changes in the fusion images, there will be large convergence errors. Aiming at the problem, starting from the spatial information of the image, the reflectivity of temporal spatial fusion process as changes in reflectance image super-resolution reconstruction process using convolutional neural network and sparse reconstruction algorithm in change of reflectance image contours and feature details, and different observation time relative prediction image reconstruction phase by the time the weight stack to achieve prediction phase reflectivity fusion. Experimentation area were selected for heterogeneity of higher surface coverage of Wuhan and the Beijing suburb of experiment, experimental results show that the proposed space-time surface reflectance fusion model integrates the existing methods of generating better results from the effect of phenology and object types mutation phenomenon caused by the change of reflectivity with respect to, and by introducing the convolutional neural network and sparse representation step by step algorithm learning image feature information, the detail information fusion reflectance image is more clear and accurate. The fusion model can provide high precision reflectivity fusion images with high spatial and temporal resolution for different applications of high heterogeneity regions. (2) research on the spatial reduction method of surface temperature image space of BP neural network with multiple surface parameters. Remote sensing surface temperature products are always faced with the problem of insufficient information in higher resolution. It is necessary to introduce the information of higher spatial resolution to realize the downscaling transformation. Based on this theory, firstly, in the low resolution scale set can describe the distribution characteristics of NDVI-LST scale model, and the model according to the explicit and implicit relations are modeled respectively, so as to improve the scaling of the generalization ability of the model, with NDVI-LST distribution features as dry explicit relation was simulated, and the use of BP neural network implicit by a variety of surface parameters and the surface temperature of the residual visible and near infrared data generated by the complex relationship between mapping; then the surface parameters of higher spatial resolution based on scale transformation model, realize the surface temperature of the downscaling. This method can effectively alleviate the mismatch between the surface temperature data and the spatial feature scale needed in the fine study of the earth system process. (3) research on spatio-temporal fusion model of surface temperature based on space reduction technology. The traditional surface temperature temporal spatial fusion method is easily affected by the temporal and spatial variability effects of pixel radiance, a larger error will cause the fusion fusion application in high spatial heterogeneity of the region; to solve this problem, this paper will surface temperature as predicted by the combination of time and space surface temperature new idea of spatial scaling; prediction temporal reflectance the fusion model is generated by high resolution surface parameters of surface temperature predicted by the phase of low resolution image to establish the downscaling model, realize the prediction of spatial scale phase surface temperature image conversion; for complex surface areas, and consider the model is suitable for different seasons of the experimental data, the use of multi-source remote sensing data respectively. The experiment of the model and the traditional model, the experimental results show that the surface temperature of the time compared with the traditional fusion model The method is more suitable for the application of the high heterogeneity surface area, and the good fusion precision is shown in the experiments of different seasons. The model can provide reliable technical support for the expansion of surface temperature data in the field of global temperature change. (4) research on adaptive monitoring model of forest fire based on spatio-temporal fusion technology. The traditional remote sensing model of forest fire remote sensing is easily affected by regional and seasonal environment, and there is a problem of missing or misjudged fire. Based on the seasonal distribution characteristics of the spatio-temporal brightness temperature of typical building sky brightness temperature model, using space-time fusion technology to obtain the high resolution image brightness temperature rich training data model, in order to improve the accuracy of parameter calibration model; based on typical clear sky surface brightness temperature model and surface radiation theory, space-time dynamic threshold establishment function the remote sensing of forest fire monitoring and build the adaptive model; to the environmental satellite and the MODIS experimental data, a typical clear sky surface brightness temperature model in Heilongjiang Province, and the province of the fire cases were analyzed. The experimental results show that this method can effectively improve the accuracy of remote sensing monitoring and space-time adaptive degree of forest fire monitoring the.
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P407

【相似文献】