盐渍化灌区多尺度土壤水力特征参数空间变异规律及转换技术研究

发布时间：2018-04-25 07:40

本文选题：空间变异 + 尺度效应　；参考：《内蒙古农业大学》2016年博士论文

【摘要】：土壤是高度不均一的历史自然体,其特性参数空间分布具有一定的非均一性,且空间变异具有尺度效应,所以描述其不同尺度空间变异特性及变化过程等问题是当前灌溉水文学研究的热点。自然土壤特性在水平和垂直方向上都存在变异性,且不同土层土壤特性的空间变异存在着一定的相互关系,探求不同土层土壤特性空间变异特性及其相互对应的关系,对于了解土壤特性在三维空间上的变化规律及模拟和预测具有一定的意义。在土壤水力参数空间变异研究中,建立水力参数与理化参数的转换函数成为获取水力参数的重要途径,但不同尺度的主导过程并不相同,如何将某一尺度建立的模型经过尺度转换应用于其它尺度,明确其精度及消除不确定性等问题值得进一步研究。为此,论文依托于国家自然科学基金项目(51069006),以内蒙古河套灌区为例,针对上述问题进行相关研究和讨论。(1)分别采用Bayesian神经网络(BNN)及BP神经网络建立土壤基本特性参数与土壤水分特征曲线模型参数、特征含水率之间的转换函数,进行模型验证及适应性比较。两种方法均能得到较好的训练及预测效果,且BP模型的训练精度优于BNN模型,但BNN模型的预测精度整体高于BP模型,且由于BNN模型的预测值为一个区间范围,可更好的体现土壤特性参数的空间随机性和结构性特征。(2)通过经典统计、地统计及多重分形进行土壤基本物理特性参数(粘粒、粉粒、砂粒、有机质)和水力参数(饱和含水率θs、van Genuchten模型参数α、n)在不同尺度(小尺度：52.40km2、1km×1km；中尺度：1.243×103km2、4km×4km；大尺度：3.708×103km2、8km×8km)不同土层(0～20cm、20～40cm、40～70cm、70～100cm)的空间变异性分析,得到结论如下：基本物理特性参数在不同尺度不同土层表现为强烈的空间自相关性,空间分布主要受母质、气候等结构性因素影响；在研究区域上均具有明显的多重分形特征,且在中尺度呈现最强的空间变异性(除了有机质在0-20cm、20-40cm土层),同时多重分形谱曲线(除了有机质在小尺度的20-40cm及40-70cm土层)表现为左钩状,即在其空间分布中,数值较大的数据占主导地位,其概率分布较大。水力参数不同尺度不同土层整体上具有强烈的空间自相关性,空间分布主要受母质、气候、土壤类型等结构因素的影响；van Genuchten模型参数α空间分布格局具有多重分形特征,但在3个尺度下多重分形谱谱宽没有一致性变化规律,vanGenuchten模型参数n和饱和含水率θs分布格局的多重分形特征不明显,其多重分形谱谱宽均较小。(3)通过联合多重分形方法研究了基本物理特性参数表层(0-20cm)与其它土层(20-40cm、40-70cm及70-100cm)空间变异性的相关程度并建立转换函数。整体上各参数0-20cm土层与20-40cm、40-70cm、70-100cm土层空间变异性的相关性依次递减,且其相关性在小尺度及大尺度高于中尺度。通过一元函数建立基本物理特性参数在3个尺度下表层(0-20cm)与其它土层(20-40cm、40-70cm及70-100cm)的转换函数,各参数3个尺度下0-20cm土层与20-40cm土层回归关系较好,决定系数在0.41～0.65之间,与40-70cm及70-100cm土层整体回归效果较差,决定系数分布在0.038～0.401之间。(4)通过多元回归、支持向量机及BP神经网络方法建立基于中尺度的水力参数与基本物理特性参数及基本物理特性参数与高光谱的转换函数,并将其尺度上推至大尺度与尺度下推至小尺度,并对其尺度转化的适用性进行评价。基于中尺度建立的高光谱与土壤颗粒组成及有机质的反演模型均可以较好的应用于其它两个尺度：多元回归方法在其它两个尺度上的相关性在0.33～0.60之间,支持向量机为0.41-0.52,BP神经网络为0.52-0.72,BP神经网络方法建立的模型在其它两个尺度上具有更好的适用性。且颗粒组成的效果整体好于有机质含量。基于中尺度建立的水力参数(饱和含水率θs、van Genuchten模型参数α)与基本物理参数(粘粒、粉粒、砂粒及有机质)的转换函数在其它两个尺度上均具有较好的适用性：多元回归方法在其它两个尺度上的相关性在0.535～0.944之间,支持向量机为0.602～0.968,支持向量机方法具有更好的适用性。van Genuchten模型参数n的建模及模型检验效果均较差。3个参数的尺度转换结果为饱和含水率θs效果最好,其次是van Genuchten模型参数α,而van Genuchten模型参数b效果最差。
[Abstract]:The soil is a highly heterogeneous historical natural body. The spatial distribution of its characteristic parameters has a certain heterogeneity, and the spatial variation has the scale effect. So it is a hot spot in the current research of irrigation hydrology to describe the spatial variability and change process of different scales. The natural soil characteristics vary in the horizontal and vertical direction. The spatial variability of soil characteristics of different soil and different soil layers has a certain relationship, exploring the spatial variation characteristics of soil characteristics and their corresponding relationship in different soil layers. It has certain meaning for understanding the change law of soil characteristics in three-dimensional space and the simulation and prediction of soil characteristics. The transformation function of the hydraulic parameters and the physical and chemical parameters becomes an important way to obtain the hydraulic parameters, but the leading processes of different scales are different. How to apply the scale transformation to the other scales through the scale transformation, and to clarify its accuracy and eliminate the uncertainty are worth further studying. The Natural Science Fund Project (51069006) takes Inner Mongolia Hetao irrigation area as an example to study and discuss the above problems. (1) Bayesian neural network (BNN) and BP neural network are used to establish the parameters of soil basic characteristic parameters and soil moisture characteristic curve model, and the conversion function between the characteristic water content and the model is verified. Adaptability comparison. The two methods can get better training and prediction effect, and the training precision of BP model is better than that of BNN model, but the prediction accuracy of BNN model is higher than that of BP model as a whole, and the spatial randomness and structural characteristics of soil characteristic parameters can be better reflected by the prediction value of BNN model. (2) pass through the channel. The basic physical properties of soil parameters (clay, powder, sand, organic matter) and hydraulic parameters (saturated water content [theta] s, van Genuchten model parameter alpha, n) at different scales (small scale: 52.40km2,1km * 1km; mesoscale: 1.243 * 103km2,4km x 4km; large scale: 3.708 x 103km2,8km x 8km) in different soil layers are carried out by the canonical statistics and multifractal. The spatial variability of (0 to 20cm, 20 to 40cm, 40 to 70cm, 70 to 100cm) is analyzed. The results are as follows: the basic physical parameters are strongly influenced by the spatial autocorrelation in different soil layers, and the spatial distribution is mainly influenced by the structural factors such as the parent material and the climate. The mesoscale has the strongest spatial variability (except organic matter in 0-20cm, 20-40cm soil layer), and the multifractal spectrum curve (except for the organic matter in the small scale 20-40cm and the 40-70cm soil layer) is left hook like, that is, in its spatial distribution, the large number of data is dominant and its probability distribution is larger. The hydraulic parameters are different scales and different soil. There is a strong spatial autocorrelation on the whole layer, and the spatial distribution is mainly influenced by the structural factors such as the parent material, climate, soil type and so on. The van Genuchten model parameter alpha spatial distribution pattern has multi fractal characteristics, but the multifractal spectrum width is not consistent at the 3 scales, vanGenuchten model parameter n and saturated water cut. The multifractal characteristics of the rate theta s distribution pattern are not obvious, and their multifractal spectrum spectrum is smaller. (3) the correlation degree of the spatial variability of the basic physical characteristic parameters (0-20cm) and other soil layers (20-40cm, 40-70cm and 70-100cm) is studied by the combined multifractal method. The whole parameter 0-20cm soil layer and 20-40cm are established. 40-70cm, the correlation of spatial variability of 70-100cm soil layer decreases in turn, and its correlation is higher than the mesoscale in small scale and large scale. Through one element function, the transformation function of the basic physical characteristic parameters at 3 scales (0-20cm) and other soil layers (20-40cm, 40-70cm and 70-100cm), and the 0-20cm soil layer and 20-40cm under the 3 scales of each parameter The regression relation of soil layer is good, the coefficient of decision is between 0.41 and 0.65, and the overall regression effect of 40-70cm and 70-100cm soil layer is poor, the coefficient of decision is distributed between 0.038 and 0.401. (4) through multiple regression, support vector machine and BP neural network method to establish the hydraulic parameters and basic physical properties parameters and basic physical properties of the mesoscale parameters and the basic physical properties. The number and hyperspectral conversion function are pushed to the large scale and the scale to the small scale, and the applicability of the scale transformation is evaluated. The hyperspectral and soil particle composition and the organic matter inversion model based on Mesoscale can be better applied to its two scales: multiple regression method in the other two The correlation on the scale is between 0.33 and 0.60, the support vector machine is 0.41-0.52, the BP neural network is 0.52-0.72, the model established by the BP neural network has better applicability on the other two scales. And the effect of the particle composition is better than the organic matter content. The hydraulic parameters based on the medium scale (saturated water content [theta] s, van Gen The conversion functions of uchten model parameters and basic physical parameters (clay, particle, sand and organic matter) have good applicability on the other two scales: the correlation of multiple regression methods on other two scales is between 0.535 and 0.944, support vector machine is 0.602 to 0.968, support vector machine method has better applicability. .van Genuchten model parameter n modeling and model test results are poor, the result of poor.3 parameter scale conversion result is the best effect of saturated water content theta s, followed by van Genuchten model parameter alpha, and van Genuchten model parameter B effect is the worst.

【学位授予单位】：内蒙古农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S152.7

【参考文献】