地理空间分析中土壤景观模型的设计与创新应用
发布时间：2018-08-25 17:32
【摘要】：精细的土壤属性信息是开展环境管理、监测、建模及精细农业等工作必需的基础数据,而目前可用的土壤信息是土壤调查得到的土壤图,它们的图斑包含的土壤信息类型有限且可能融合其他类型,并且属性离散,难以满足社会对高精度土壤信息的需求。但以往的土壤制图研究中,已经收集了大量的包含土壤养分、土壤类型、成土环境的土壤采样数据,并且绘制了蕴含土壤专家丰富经验的传统土壤图,这些宝贵的历史资料可以建立土壤景观模型并在精细土壤信息获取中被充分利用,缩减研究成本的同时提高研究效率和准确性,然而土壤信息的获取往往是通过地理空间分析获取。为提高传统土壤制图成果中土壤信息的精度,本研究以地统计学理论、土壤景观建模为基础,在地理空间分析中,对土壤环境要素提取与融合、土壤采样布局设计与优化、土壤景观模型建立与完善、土壤综合信息概括与展示四个方面进行探索。以湖北省钟祥市为例,先后探索的内容为在土壤景观关系的基础上设计土壤采样方案、在研究区内路网空间分布的基础上优化土壤采样的空间布局、在地形单元划分的基础上预测土壤有机质的空间分布,得到的结论如下:(1)研究区内的土壤景观关系在没有人为干扰的情况下是固定不变的,因此通过比较不同采样方案中由对应的采样点数据求得的土壤景观相关系数矩阵的相似性,可以验证研究区内土壤景观关系存在固定关系模式可能性。以研究区域A内的8个地形因子和5个常规土壤养分为研究对象,利用空间分级统计方法将地形因子划分为5个级别,采用Pearson相关性分析各分区上的样点,并对不同采样布局下地形因子与土壤养分间的Pearson系数进行相似性分析。不同采样方案下地形因子与土壤养分间的Pearson系数相似性程度在99%以上,说明研究区内存在固定的关系模式。综合大量环境因素而确定的基于地形因子分级的采样方案A预测精度较高,以其关系模式代表固定关系模式评价其余采样方案的合理性是可行的。本节设计的基于网格的3种传统随机采样方案中,含有3661个采样点的方案C与采样方案A获得的关系模式的相似性最接近,因而比其他方案样点布局更合理。(2)城镇化的发展会使区域内的交通网络更加发达而趋于稳定,可为野外土壤采样提供便捷的交通条件。以湖北省钟祥市东部区域(研究区B)为研究对象,基于路网设置5种采样尺度进土壤采样,采样过程中,采样数据可能会因为一些因素而产生误差,采样点的空间布局也可能不是十分合理,影响最终研究结果的精度,因此需要剔除采样数据中误差较大的采样点。运用模拟退火算法对各样点的空间布局分别进行优化,以获取基于路网的土壤采样优化布局。在此基础上,对地形因子和优化后样点的有机质建立多元线性回归模型,同时建立基于神经网络的多层感知机模型,并用此模型精度与多元线性回归模型精度进行对比。结果表明:利用道路网制定土壤采样方案是可行的,优化后的采样点布局能够准确获取土壤景观知识,并且优于原始样点的精度。本研究利用道路空间分布格局、历史样点、数字高程数据等可利用资源设计采样方案,为减少采样成本、提高采样效率、展现有机质空间分布格局提供了有效手段与理论依据。(3)以钟祥市(研究区A)土壤有机质空间分布为例,克服传统地形分类方法中仅依据单一指标(如高程)的缺点,综合由30m精度数字高程模型生成的地形因子,依据其在不同地形条件下的层次组合规律构建地形分类规则,精确地划分为13种典型地形单元,并运用普通克里金法对不同地形单元内的土壤样本插值,获得相应区域的土壤有机质空间分布。通过组合各地形范围下的结果,以获取蕴含地形因素影响的有机质空间分布。研究发现,地形起伏较大的地形单元的预测精度与全局预测结果精度相似度达0.75,而地势平缓区域内的预测精度大幅度提升,比全局预测结果精度提升了16.39%,因此基于地形单元的空间预测能精确有效地获取土壤有机质空间特征。利用地形分区获取较高精度的有机质空间分布,进一步探讨了有机质地统计学研究中地形的协同影响。
[Abstract]:Fine soil attribute information is the basic data necessary for environmental management, monitoring, modeling and precision agriculture. At present, available soil information is the soil map obtained from soil survey. Their patches contain limited types of soil information and may merge with other types, and their attributes are discrete, which is difficult to meet the high precision of society. However, in the past soil mapping research, a large number of soil samples including soil nutrients, soil types and soil forming environment have been collected, and traditional soil maps containing rich experience of soil experts have been drawn. These valuable historical data can be used to establish soil landscape models and obtain fine soil information. In order to improve the precision of soil information in traditional soil mapping results, this study is based on Geostatistics theory, soil landscape modeling, geospatial analysis, and soil ring. Four aspects are explored: extraction and fusion of environmental factors, design and optimization of soil sampling layout, establishment and perfection of soil landscape model, generalization and display of soil comprehensive information. The spatial distribution of soil organic matter was predicted on the basis of topographic unit division by optimizing the spatial distribution of soil sampling. The conclusions are as follows: (1) The relationship of soil landscape in the study area is fixed and unchanged without human disturbance. The similarity of soil landscape correlation coefficient matrix can validate the possibility of the existence of a fixed relationship model in the study area. Eight topographic factors and five conventional soil nutrients in the study area A were studied. The topographic factors were classified into five levels by using spatial classification statistical method, and each score was analyzed by Pearson correlation analysis. The Pearson coefficients between topographic factors and soil nutrients under different sampling arrangements were analyzed. The similarity of Pearson coefficients between topographic factors and soil nutrients under different sampling schemes was more than 99%, indicating that there was a fixed relationship model in the study area. Among the three traditional random sampling schemes based on grids designed in this section, scheme C with 3661 sampling points has the closest similarity to the relational pattern obtained by sampling scheme A because its relational pattern represents the fixed relational pattern. (2) The development of urbanization will make the transportation network more developed and stable, and provide convenient transportation conditions for field soil sampling. Sampling data may cause errors due to some factors, and the spatial layout of sampling points may not be very reasonable, which affects the accuracy of final research results. Therefore, it is necessary to eliminate sampling points with large errors in sampling data. On this basis, a multivariate linear regression model is established for the topographic factors and the organic matter of the optimized sample points, and a multilayer perceptron model based on neural network is established. The precision of the model is compared with that of the multivariate linear regression model. The optimized layout of sampling points can accurately acquire soil landscape knowledge, and is superior to the accuracy of the original sampling points. This study uses the available resources such as road spatial distribution pattern, historical sampling points and digital elevation data to design sampling scheme, which provides an effective way to reduce sampling cost, improve sampling efficiency and exhibit the spatial distribution pattern of organic matter. (3) Taking the spatial distribution of soil organic matter in Zhongxiang (study area A) as an example, to overcome the shortcomings of traditional topographic classification methods which only depend on a single index (such as elevation), the topographic factors generated by the 30m precision digital elevation model are synthesized, and the topographic classification rules are constructed according to their hierarchical combination rules under different topographic conditions. The land is divided into 13 typical topographic units, and the spatial distribution of soil organic matter is obtained by interpolating soil samples in different topographic units with ordinary Kriging method. The accuracy similarity between the element prediction and the global prediction is 0.75, while the prediction accuracy in the gentle terrain region is greatly improved, which is 16.39% higher than that of the global prediction. Therefore, the spatial prediction based on topographic units can accurately and effectively obtain the spatial characteristics of soil organic matter. Spatial distribution further explores the synergistic effect of topography in organic texture statistics.
【学位授予单位】：华中农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S159
本文编号：2203601
[Abstract]:Fine soil attribute information is the basic data necessary for environmental management, monitoring, modeling and precision agriculture. At present, available soil information is the soil map obtained from soil survey. Their patches contain limited types of soil information and may merge with other types, and their attributes are discrete, which is difficult to meet the high precision of society. However, in the past soil mapping research, a large number of soil samples including soil nutrients, soil types and soil forming environment have been collected, and traditional soil maps containing rich experience of soil experts have been drawn. These valuable historical data can be used to establish soil landscape models and obtain fine soil information. In order to improve the precision of soil information in traditional soil mapping results, this study is based on Geostatistics theory, soil landscape modeling, geospatial analysis, and soil ring. Four aspects are explored: extraction and fusion of environmental factors, design and optimization of soil sampling layout, establishment and perfection of soil landscape model, generalization and display of soil comprehensive information. The spatial distribution of soil organic matter was predicted on the basis of topographic unit division by optimizing the spatial distribution of soil sampling. The conclusions are as follows: (1) The relationship of soil landscape in the study area is fixed and unchanged without human disturbance. The similarity of soil landscape correlation coefficient matrix can validate the possibility of the existence of a fixed relationship model in the study area. Eight topographic factors and five conventional soil nutrients in the study area A were studied. The topographic factors were classified into five levels by using spatial classification statistical method, and each score was analyzed by Pearson correlation analysis. The Pearson coefficients between topographic factors and soil nutrients under different sampling arrangements were analyzed. The similarity of Pearson coefficients between topographic factors and soil nutrients under different sampling schemes was more than 99%, indicating that there was a fixed relationship model in the study area. Among the three traditional random sampling schemes based on grids designed in this section, scheme C with 3661 sampling points has the closest similarity to the relational pattern obtained by sampling scheme A because its relational pattern represents the fixed relational pattern. (2) The development of urbanization will make the transportation network more developed and stable, and provide convenient transportation conditions for field soil sampling. Sampling data may cause errors due to some factors, and the spatial layout of sampling points may not be very reasonable, which affects the accuracy of final research results. Therefore, it is necessary to eliminate sampling points with large errors in sampling data. On this basis, a multivariate linear regression model is established for the topographic factors and the organic matter of the optimized sample points, and a multilayer perceptron model based on neural network is established. The precision of the model is compared with that of the multivariate linear regression model. The optimized layout of sampling points can accurately acquire soil landscape knowledge, and is superior to the accuracy of the original sampling points. This study uses the available resources such as road spatial distribution pattern, historical sampling points and digital elevation data to design sampling scheme, which provides an effective way to reduce sampling cost, improve sampling efficiency and exhibit the spatial distribution pattern of organic matter. (3) Taking the spatial distribution of soil organic matter in Zhongxiang (study area A) as an example, to overcome the shortcomings of traditional topographic classification methods which only depend on a single index (such as elevation), the topographic factors generated by the 30m precision digital elevation model are synthesized, and the topographic classification rules are constructed according to their hierarchical combination rules under different topographic conditions. The land is divided into 13 typical topographic units, and the spatial distribution of soil organic matter is obtained by interpolating soil samples in different topographic units with ordinary Kriging method. The accuracy similarity between the element prediction and the global prediction is 0.75, while the prediction accuracy in the gentle terrain region is greatly improved, which is 16.39% higher than that of the global prediction. Therefore, the spatial prediction based on topographic units can accurately and effectively obtain the spatial characteristics of soil organic matter. Spatial distribution further explores the synergistic effect of topography in organic texture statistics.
【学位授予单位】：华中农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S159
【参考文献】
相关期刊论文 前1条
1 赵明松;张甘霖;王德彩;李德成;潘贤章;赵玉国;;徐淮黄泛平原土壤有机质空间变异特征及主控因素分析[J];土壤学报;2013年01期
，本文编号：2203601
本文链接：https://www.wllwen.com/kejilunwen/nykj/2203601.html
