基于BME和NNE法的农田土壤水分和养分空间插值

发布时间：2019-06-24 23:39

【摘要】：精准农业强调通过对农田资源的分布式调控,尽可能均衡、合理地利用其生产潜力,获取尽可能高的经济产量或显著的社会效益。掌握农田土壤水分和养分的空间分布规律是分布式调控的前提条件,而基于有限的采样数据进行空间插值则是掌握农田土壤水分和养分变异规律的有效途径,对于实现农田的精确灌溉和变量施肥具有重要意义。本文以江苏省扬州市北部某农田上161个采样点的含水量、全氮量、有机质、碱解氮、速效钾、速效磷含量为例,针对研究区土壤变量的空间变异特征,结合所采用的空间插值方法的特点,开展了如下研究：①利用研究区不同田块之间土壤特性差异显著、同一田块无显著差异的特点,用同一田块内土壤变量的分布特征表达待估点估值的不确定性(即构造软数据),并将此软数据和实测硬数据结合,利用现代地质统计学方法——贝叶斯最大熵法(BME)模拟土壤变量的空间分布(该方法以下简称MVBME法)；②利用神经网络方法良好的非线性表达能力,借助集成神经网络系统估计待估点的不确定性(即构造软数据),并将其结果融入BME法中,用融入该软数据的BME法(以下简称NNEBME法)模拟土壤变量的空间分布。③采取多种随机抽样方案(建模样本和验证样本),将以上空间模拟结果分别与径向基函数神经网络法(RBF)、集成神经网络法(NNE)、普通克立格法(OK)、残差克立格法(RK)的估值结果进行比较。得出的主要研究结果如下：1)研究区内土壤含水量呈弱变异性,养分均属中等程度变异。且研究区内整个试验田的变异性比单块要大很多,不同田块之间土壤特性差异达到极显著,同一田块多属无显著性差异。2)特异值处理前,含水量、全氮量、有机质、碱解氮都近似服从正态分布,速效钾与速效磷不服从正态分布,处理特异值后土壤水分及养分的空间分布均近似服从正态分布。3)MVBME法对研究区内土壤水分与养分进行空间插值,并将结果与RBF法、RK法和OK法的插值结果进行比较得出：MVBME法的平均误差(ME)为四种插值方法中最小,估计方差(MSE)相较RBF法能降低23.77%-69.14%：相比较RE法,能降低0.41%-56.17%：与OK法相比,MVBME法能使有机质、碱解氮、速效钾、速效磷的MSE降低6.24%～52.37%,水分与全氮量在大部分情况下能降低10.25%-38.18%。四种插值方法里.MVBME法最接近于无偏估计,且对变量空间波动程度反映最精确。4)NNEBME法对研究区土壤含水量与养分进行空间差值,其结果与NNE法、RK法和OK法进行比较得出：对于不同的土壤变量,NNEBME法估值的ME最接近于零,近似无偏估计；与NNE法比较,MSE缩小1.64%～45.20%,与OK法、RK法比较,除土壤水分外,NNEBME法使MSE缩小0-40.05%：并且随着已知点(即建模数据样本容量)个数的减少,NNEBME法的插值优势更为突出；MSE的组成分析表明,NNEBME法对变量均值的估计与波动程度的描述更为精确。本文利用农田土壤变量的空间变异特征和估值方法的特点,构建了能为BME法有效利用的软数据,不仅拓展了现代地质统计学的BME方法在农业水土科学领域的应用范围,而且为农田土壤水分和养分空间分布模拟精度的改善提供了新思路。
[Abstract]:The precision agriculture emphasizes the distribution and regulation of the farmland resources, as well as possible balance and reasonable utilization of the production potential of the farmland resources, so as to obtain the highest possible economic output or remarkable social benefit. The spatial distribution of soil moisture and nutrients in farmland is a prerequisite for distributed regulation, and spatial interpolation based on limited sampling data is an effective way to master the law of soil moisture and nutrient variation of the farmland. It is of great significance to realize the precise irrigation and variable fertilization of the farmland. In this paper, the water content, total nitrogen content, organic matter, alkaline solution nitrogen, quick-acting potassium and quick-acting phosphorus content of 161 sampling points on a farmland in the northern part of Yangzhou city of Jiangsu Province are taken as an example. in that invention, the soil characteristic difference between the different field blocks in the study area is obviously different, the characteristic of no significant difference of the same field block is used, the uncertainty of the estimation of the point to be evaluated (i. e., the soft data is construct) is expressed by the distribution characteristic of the soil variable in the same field block, and the soft data and the measured hard data are combined, The spatial distribution of the soil variable is simulated by the modern geostatistical method _ Bayesian Maximum Entropy Method (BME) (this method is referred to as the MVBME method), and the nonlinear expression ability of the soil variable is good by using the neural network method. The uncertainty of the point to be evaluated (i.e., the soft data) is estimated by the integrated neural network system, and the result is integrated into the BME method, and the spatial distribution of the soil variable is simulated by the BME method (hereinafter referred to as the NNEBME method) incorporated into the soft data. The results of the above-mentioned space simulation are compared with the estimation results of the radial basis function neural network method (RBF), the integrated neural network method (NNE), the ordinary Kriging (OK) and the residual Kriging (RK). The main results of the study are as follows:1) The soil water content in the study area is weak and the nutrient is of medium degree variation. and the variation ratio of the whole test field in the study area is much larger than that of the single block, the difference of the soil characteristics between the different field blocks is very significant, and there is no significant difference in the same field block.2) Before the specific value is treated, the water content, the total nitrogen amount, the organic matter and the alkaline solution nitrogen are all approximately subject to normal distribution, The spatial distribution of the soil water and nutrients in the study area was estimated by MVBME method, and the results were compared with the results of the RBF method, the RK method and the OK method. The mean error (ME) of the MVBME method is the least in the four interpolation methods, and the estimated variance (MSE) can be reduced by 23.77%-69.14% compared with the RBF method. Compared with the OK method, the MVBME method can reduce the MSE of the organic matter, the alkaline solution nitrogen, the quick-acting potassium and the quick-acting phosphorus by 6.24% to 52.37%, Water and total nitrogen can be reduced by 10.25% to 38.18% in most cases. In four interpolation methods. The MVBME method is closest to the unbiased estimation, and the spatial difference of soil water content and nutrients in the study area is reflected by the NNEBME method. The results are compared with the NNE method, the RK method and the OK method. Compared with the NNE method, the MSE is reduced by 1.64% to 45.20%, compared with the OK method and the RK method, the MSE is reduced by 0-40.05% in addition to the soil moisture, and the interpolation advantage of the NNEBME method is more prominent with the reduction of the number of known points (i.e., the sample capacity of the modeling data); the analysis of the composition of the MSE shows that, The NNNME method is more accurate to estimate the mean value of the variable and the degree of fluctuation. By using the characteristics of the spatial variation and the estimation method of the farmland soil variable, the soft data which can be effectively utilized by the BME method is constructed, and the application range of the BME method of the modern geostatistics in the field of agricultural water and soil science is not only expanded, But also provides a new thought for improving the soil moisture and the distribution of the nutrient spatial distribution of the farmland.
【学位授予单位】：扬州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S152.7;S158

【相似文献】