【作者】 吴见；

【导师】 彭道黎；

【作者基本信息】 北京林业大学 ， 森林经理学， 2012， 博士

【摘要】 沙漠化属于全球性环境恶化现象,国家需要及时准确地掌握其动态以便进行科学防治。但遥感技术在沙漠化监测与评价方面存在很多问题,例如评价指标选取不恰当、权重不客观、指标反演精度低等,至今仍缺乏一套被广泛认同的实用的定量评价指标体系。不同的地区和地物类型主导因子或许相同,但各因子对沙漠化的影响程度一定有差异。目前国内还没有以地物类型为基础的沙漠化遥感定量评价研究报道,如何从地物类型的角度对沙漠化进行遥感定量评价?解决这一问题将有助于认识土地退化过程的机制和成因等内容,有利于建立沙漠化评价指标体系。本论文以京津风沙源治理工程区为例,分别探讨了多光谱和高光谱遥感对干旱半干旱区土地沙漠化进行评价的具体方法,得到的主要结果归纳如下：1.提出将线性光谱混合分解模型、植被指数和专家知识相结合的地物信息分层次提取模型,实现了地物信息高精度分层次提取。2.筛选出区分树种信息的多光谱遥感指标,并引用改进的SVM算法提取了退耕还林地树种信息。结果表明,该方法平均精度较传统方法提高9.2%,对快速评价工程质量有重要意义。3.将纹理、空间信息融入到高光谱影像地物信息提取中。通过反射率光谱分析,结合纹理特征对地物信息进行提取,并采用基于空间信息的方法进一步对植被类型进行分类,平均分类精度较最大似然法提高17.8%。分析了高光谱遥感树种分类可行性,选取差异较大的波段及光谱特征参量,引用改进的BP神经网络模型完成林地树种信息提取。4.建立了基于地物类型的沙漠化评价遥感指标体系,明确了“基准”的确定方法,在分析大量实测数据的基础上提出了一种新的指标权重计算方法。经过验证,本文模型较传统模型的评价精度提高,评价结果更接近土地沙漠化的真实状况。5.提出了利用高分辨率卫星影像修正线性光谱混合分解模型分解的TM影像的植被分量,建立提取干旱半干旱地区植被覆盖度的模型。结果表明,该模型不仅提供了更纯的植被光谱信息,而且降低了对土壤背景的敏感度,更适合于中等分辨率卫星影像量化干旱半干旱地区植被覆盖度。6.利用高光谱遥感数据对森林蓄积量进行预测研究,确定了与蓄积量之间相关系数达到极显著水平的19个特征参数；比较了目前流行的多种高光谱植被盖度提取方法,结果表明基于一阶微分的PLSR模型效果最好。7.提出了通过高光谱影像分解剔除植被光谱干扰,从而更合理地预测土壤含水量的具体方法；分析了最小噪声变换回归模型和主成分回归模型预测土壤含沙量的能力。更多还原

【Abstract】 Desertification belongs to the deteriorative phenomenon of global environment. Its dynamic changes should be monitored timely and accurately by the county in order to control the desertification scientifically. The remote sensing technology has a lot of problems in terms of desertification evaluation, such as improper evaluation indexes selection, subjective weight, the low index inversion precision, and so on. A quantitative evaluation index system that is widely recognized and practical is still lacked. Perhaps the factors in a leading role are the same in different regions and different land use types, but the influence degree of the factors to the desertification must have differences. At present, there is still no remote sensing quantitative evaluation research report of desertification based on the land use interiorly. How to evaluate desertification quantitatively using remote sensing from the point of view of land use? To solve this problem will help to know the content of the mechanism and genesis in the process of land degradation and help to establish desertification evaluation index system. In this paper, taking Beijing-Tianjin sandstorm source control project district as an example, the specific desertification evaluation methods of multispectral and hyperspectral remote sensing based on land use in arid and semiarid regions were discussed respectively. The mainly achieved results were summarized as follows:1. The stratified remote sensing information extraction model of land use combining linear spectral mixture model, vegetation index and expert knowledge was put forward. The model achieves stratified extraction of land use information with high precision.2. The multispectral remote sensing indexes distinguishing tree species were sifted, and the improved SVM algorithm was also quoted to extract tree species information in returning farmland to forest region. The results have shown that the average accuracy was increased by 9.2%than that of the traditional method. The method in this paper has the important meaning to the fast evaluation of the project quality of returning farmland to forest.3. Texture features and spatial information were blended in land use information extraction of hyperspectral images. The land use type information of hyperspectral images was extracted by spectral analysis of the reflectance and texture features. And then the vegetation types were further classified using the method of spatial information. The results show that the average classification accuracy is increased by 17.8%than that of maximum likelihood method. The feasibility of tree species classification of hyperspectral remote sensing was analyzed, and the bands and spectral characteristic parameters that have great differences were selected. At last, the improved BP neural network model was quoted to complete the tree species information extraction of forest land.4. The remote sensing index system of desertification evaluation based on land use was set up firstly and the determination method of the "standards" was determined. A new index weight calculation method was put forward on the basis of analyzing large quantities of measured data. The evaluation accuracy of the model in this paper was higher than that of the traditional model and the evaluation results were closer to the real condition of desertification.5. A model that was fit for the extraction of vegetation coverage in arid zones was presented. In this model, TM image was decomposed by linear spectral mixture mode and then the vegetation component of TM image was amended by the high-resolution satellite image. The result shows that the model can not only provide more pure vegetation spectrum information but also reduce the sensitivity to the soil background, so it is more suitable for quantifying vegetation coverage of arid and semi-arid regions by medium-resolution satellite images.6. In this study, nineteen characteristic parameters that had significant level correlation coefficients with forest volume were selected to forecast forest volume. A variety of hyperspectral vegetation coverage extraction methods that are currently popular were compared. The conclusion is that the partial least-squares regression model based on first order differential is the best.7. The specific method that eliminated interferences of vegetation spectrum by decomposing hyperspectral imaging to predict soil water content more reasonably was put forward. The prediction ability of soil sediment concentration of the minimum noise fraction regression model and the principal component regression model were analyzed.更多还原