基于环境小卫星和GIS的灌区土壤盐渍化研究

发布时间：2018-04-02 10:11

本文选题：干旱半干旱灌区　切入点：土壤次生盐渍化　出处：《中国农业大学》2016年博士论文

【摘要】：土壤盐渍化是干旱、半干旱地区最主要和极易发生的土地退化现象,严重影响生态环境质量,制约着人类社会和经济的发展。灌区土壤次生盐渍化已成为限制我国生态和经济发展的主要因素,更对我国的粮食生产造成严重威胁。平罗县作为我国传统的农业灌溉区和产粮大县,土壤次生盐渍化现象十分严重。亟需对该区的土壤盐渍化程度和分布情况进行快速、全面和深入的了解。本文利用遥感和地理信息系统技术等手段对研究区的土壤盐渍化进行有效监测,并深入分析了影响土壤盐渍化形成与发展的自然与人为因素,在此基础上实现了该区土壤盐渍化的模拟和预测。研究成果如下：(1)研究区的土壤盐渍化有明显的表聚现象,土壤表层的含盐量有较强的空间变异。利用冗余分析得到K+、Na+、SO42-、Cl-与土壤全盐量的相关性很强,HC03-和Ca2+与pH值相关性更强,而且S042-和Na+是对盐渍化程度贡献最强的阴阳离子。各层土壤含盐量都具有中等强度的空间相关性,其半方差函数模型均可以用指数模型进行拟合。表层与深层土壤盐分的空间分布格局存在一定的差异。(2)对环境小卫星的高光谱数据进行线性光谱混合分解。利用纯净像元指数和最小噪声分离法提取了水体、盐分、植被和暗色物质等端元。对不同条件下的线性光谱混合分解方法进行对比分析,得到全约束条件下的线性光谱混合分解的效果最好且物理意义更明确。基于该方法的盐分丰度结果探讨了环境小卫星高光谱数据在土壤盐渍化等级分类与制图中的应用。(3)充分利用环境小卫星多光谱数据的优势,基于研究区内的农业种植模式和物候信息,建立了适合研究区的土地利用／覆被分类系统。构建能够反映地表植被信息变化的NDVI时间序列,提取了表示该区物候信息且对各地类有较强分异性的时间维特征参数。结合光谱特征参数构建了基于专家知识的决策树,实现了研究区高精度的土地利用/覆被分类。(4)以采样点的实际控制面积为土壤盐渍化研究尺度,利用环境小卫星遥感数据、DEM等地理数据以及土地利用数据等,提取了影响土壤盐渍化的自然因素和人为因素。构建了样方尺度中既能间接反映不同作物对盐分的响应,又能直接反映不同土地利用方式对土壤盐渍化影响的以作物面积为权重的冠层响应盐分指数这一综合指标。利用BP神经网络建立各指标因子对EC的预测模型。研究区土壤盐渍化程度受到自然因素和人为因素的共同影响,且不同因素之间存在着相互作用和不同的尺度效应,对盐渍化的预测精度有一定影响。
[Abstract]:Soil salinization is the most important and easily occurring phenomenon of land degradation in arid and semi-arid areas, which seriously affects the quality of ecological environment. The secondary salinization of soil in irrigated area has become the main factor that restricts the ecological and economic development of our country. Pingluo County, as a traditional agricultural irrigation area and a big grain-producing county in China, has a very serious secondary salinization phenomenon. It is urgent to carry out rapid soil salinization and distribution in this area. In this paper, we use remote sensing and GIS technology to monitor soil salinization effectively, and analyze the natural and human factors that affect the formation and development of soil salinization. On this basis, the simulation and prediction of soil salinization in this area have been realized. The research results are as follows: (1) the phenomenon of soil salinization in the study area is obvious. There is strong spatial variation in the salt content in the surface layer of soil. By using redundant analysis, the correlation between K ~ (2 +) Na ~ (2 +) so _ (42) ~ (-) Cl- and soil total salt content is very strong. HC03- and Ca2 are more closely correlated with pH value. Moreover, S042- and Na are the most important ions that contribute to the salinization degree. The semi-variance function model can be fitted by exponential model, and the spatial distribution pattern of salt in surface and deep soil is different. The hyperspectral data of environmental small satellite are decomposed by linear spectral mixing. The pure pixel index and the minimum noise separation method were used to extract the water body. Salt, vegetation and dark matter, etc. The linear spectral mixing decomposition methods under different conditions are compared and analyzed. The results show that the linear spectral mixing decomposition under the condition of full constraint is the best and the physical significance is more clear. Based on the salt abundance results of this method, the classification and mapping of soil salinization by environmental small satellite hyperspectral data are discussed. To make full use of the advantages of environmental small satellite multispectral data, Based on the agricultural planting pattern and phenological information in the study area, a land use / cover classification system suitable for the study area was established, and a NDVI time series which could reflect the change of surface vegetation information was constructed. The time dimension characteristic parameters which represent phenological information in this area and are different from each other are extracted, and the decision tree based on expert knowledge is constructed by combining the spectral characteristic parameters. The high precision land use / cover classification of the study area is realized. The actual control area of the sampling point is taken as the scale of soil salinization research, and the geographic data such as Dem and land use data are used. The natural and human factors affecting soil salinization were extracted. It can also directly reflect the effect of different land use patterns on soil salinization, which is a comprehensive index of canopy response salt index with crop area as weight. BP neural network is used to establish the prediction model of EC by each index factor. The degree of soil salinization is affected by both natural and human factors. The interaction between different factors and the different scale effect have certain influence on the prediction accuracy of salinization.
【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S156.41

【参考文献】