丘陵区土壤样点优化布局研究

发布时间：2018-05-29 10:39

本文选题：土壤养分 + 模拟退火　；参考：《西南大学》2017年硕士论文

【摘要】：土壤养分的空间分布信息是对土壤资源进行研究的基础信息,把握研究区内土壤养分在地形上的变异特征对农业生产和环境模拟具有至关重要的作用。在成土母质同源,气候条件、耕作方式、管理措施相同的条件下,地形是影响土壤养分空间分异的主要因素,对于地形复杂的丘陵区域,土壤样本的获取需要耗费大量成本,合理布设采样点以确保土壤养分空间信息完整表达显得尤为重要。有效合理的土壤采样点布局不仅能够充分反映土壤养分的空间信息,还能大大降低成本。本文以重庆市江津区永兴镇典型丘陵区(2km2)为研究区域,运用空间分析理论方法研究探讨土壤pH值、有机质、碱解氮、有效磷和速效钾含量的空间变异规律及其与地形的关系,利用模拟退火方法确定最优采样密度和最佳取样单元,并在GIS技术的支撑下,采用神经网络方法结合地形因子进行预测性土壤养分制图。主要研究结果为:(1)土壤养分之间关系密切。土壤pH与有机质、碱解氮、有效磷之间呈显著负相关,土壤有机质与碱解氮、速效钾之间呈显著正相关,土壤有效磷与速效钾之间呈显著正相关。土壤养分存在空间自相关性。土壤碱解氮和有机质含量具有强烈的空间自相关性;土壤pH值、土壤有效磷和速效钾含量具有中等程度的自相关性。土壤养分与地形因子存在显著相关性。土壤pH值与地形湿度指数(TWI)呈显著负相关,与水平曲率(HORIZC)和坡度(Slope)呈显著正相关,即随着土壤中水分含量的增加,土壤酸碱度降低;土壤有机质、碱解氮含量与高程(Elevation)、水平曲率(HORIZC)、坡度(Slope)和相对位置指数(RPI)呈显著负相关,与地形湿度指数、坡长(SlpLen)和比汇水面积(SCA)呈显著正相关;土壤速效钾含量与地形湿度指数和坡长呈显著正相关,与其他地形因子的相关性较弱;即地形越缓,土壤中有机质、碱解氮和速效钾的含量越高;土壤有效磷含量与地形因子的相关性较弱。(2)利用模拟退火算法结合神经网络模型对训练集中原始200个土壤样点的空间分布进行了系统优化,对五项土壤养分指标都给出了最佳空间布局组合;同时,针对每一个样点组合给出了与其对应的预测误差(均方误差)。误差结果表明:土壤pH、有机质、碱解氮、有效磷、速效钾分别最少可用5、6、7、6、5个优化后的样点代替原始样点,且误差不高于原始200个样点的均方误差。优化后的土壤pH、有机质、碱解氮、有效磷以及速效钾的最优样点个数分别为68、118、87、86、60。(3)以地形因子辅助变量,分别构建原始样点和最佳布局条件下的土壤养分BP神经网络预测模型,结果表明:样点减少后的预测模型对土壤养分的解释能力上升,预测精度增加,模型复杂度降低。土壤pH、有机质、碱解氮、有效磷和速效钾的模型拟合度分别提高了16.88%、14.85%、5.29%、104.26%、59.94%;RMSE分别降低了36.20%、41.69%、19.54%、2.67%、0.1%;MAE分别降低了39.49%、53.49%、0.26%、3.52%、10.41%。(4)为了改善传统采样方案土壤样点设置不合理的情况,对研究区分坡位确定取样单元。根据研究区地形特征,将地形部位划分为上坡位、下坡位、沟谷,并根据优化后的样点分布情况确定各个土壤养分在不同坡位上的取样单元,结果表明:各个土壤养分在坡地(上坡位、下坡位)地形上的取样单元较为接近,沟谷处的取样单元较大;坡地(上坡位、下坡位)地形上的取样单元小于沟谷处的取样单元。综合各个土壤养分在不同坡位上的取样单元和野外采样的实际情况,可将各个坡位上的土壤养分取样单元平均值作为本研究区的实际取样单元:上坡位的取样单元为2.06 hm2,下坡位的取样单元为1.81 hm2,沟谷的取样单元为4.91hm2。(5)利用模拟退火算法优化过后的土壤样点结合BP神经网络模型对五项土壤养分指标的空间分布情况进行数字化制图,其精度较为可靠,空间分布特征符合实际情况,为丘陵区内土壤养分科学管理、数字农业、精准农业的实施提供重要的理论支撑。
[Abstract]:The spatial distribution information of soil nutrients is the basic information for the study of soil resources. It is very important to grasp the variation characteristics of soil nutrients in the study area for agricultural production and environmental simulation. Under the same condition of homologous parent material, climate conditions, farming methods and management measures, the topography is the influence of soil cultivation. The main factor of spatial differentiation is that the acquisition of soil samples takes a lot of cost for the terrain complex hilly areas. It is very important to arrange sampling points to ensure the complete expression of soil nutrient spatial information. The effective and reasonable layout of soil sampling points not only can reflect the spatial information of soil nutrients, but also greatly reduce the soil nutrients. Low cost. This paper takes the typical hilly area (2km2) of Yongxing Town, Jiangjin District, Chongqing as the research area, and studies the spatial variation law of soil pH value, organic matter, alkali hydrolysable nitrogen, effective phosphorus and available potassium content and its relation with the terrain, and uses the simulated annealing method to determine the optimal sampling density and the optimum sampling unit. Under the support of GIS technology, the neural network method combined with topographic factors was used to make predictive soil nutrient mapping. The main results were as follows: (1) the relationship between soil nutrients is close. Soil pH has a significant negative correlation with organic matter, alkali hydrolysable nitrogen and available phosphorus. Soil organic matter has a significant positive correlation with alkali hydrolysable nitrogen and available potassium, soil has a significant positive correlation. There is a significant positive correlation between effective phosphorus and available potassium. Soil nutrients have spatial autocorrelation. Soil alkali hydrolysable nitrogen and organic matter content have strong spatial autocorrelation; soil pH value, soil available phosphorus and available potassium content have moderate degree of autocorrelation. Soil nutrients and topographic factors have significant correlation. Soil pH value and Topographic Wetness The degree index (TWI) showed a significant negative correlation with the horizontal curvature (HORIZC) and the gradient (Slope), that is, soil acidity and alkalinity decreased with the increase of soil moisture content; soil organic matter, alkali hydrolysable nitrogen content was negatively correlated with Gao Cheng (Elevation), horizontal curvature (HORIZC), gradient (Slope) and relative position index (RPI), and topographic humidity. The index, the slope length (SlpLen) and the catchment area (SCA) showed significant positive correlation. The content of soil available K was significantly positively correlated with the terrain humidity index and slope length, and the correlation with other topographic factors was weak. That is, the slower the terrain, the higher the content of soil organic matter, alkaline nitrogen and available potassium, and the weak correlation between soil effective phosphorus content and topographic factors. (2) the spatial distribution of the original 200 soil sample points was optimized by simulated annealing algorithm and neural network model, and the optimum spatial distribution combination was given for the five soil nutrient indexes. At the same time, the prediction error (mean square error) was given for each sample point combination. The error results showed that the soil was soil. Soil pH, organic matter, alkali hydrolysable nitrogen, available P, and available potassium can replace the original sample at least 5,6,7,6,5, and the error is not higher than the mean square error of the original 200 samples. The optimum sample number of soil pH, organic matter, alkali solution nitrogen, available phosphorus and available potassium is 68118,87,86,60. (3) with topographic factors, respectively. The BP neural network prediction model of Soil Nutrient under the original sample and optimal layout was constructed. The results showed that the prediction model of soil nutrients increased, the prediction accuracy increased and the model complexity decreased. The model fitting degree of soil pH, organic matter, alkali hydrolysable nitrogen, available phosphorus and available potassium increased respectively. 16.88%, 14.85%, 5.29%, 104.26%, 59.94%; RMSE decreased 36.20%, 41.69%, 19.54%, 2.67%, 0.1%, respectively, and decreased by 39.49%, 53.49%, 0.26%, 3.52%, and 10.41%., respectively, to improve the layout of the soil samples in the traditional sampling scheme. The sampling units of soil nutrients at different slope positions are determined for upper slope position, downhill position and valley, and the sampling unit on the topographic topography of each soil nutrient on sloping land (upslope position, downslope position) is relatively close, and the sampling unit at the valley is larger, and the terrain on slope (upper and lower slope) is taken. The sample unit of the soil nutrient at different slope position and the actual situation of field sampling can be used as the actual sampling unit of the study area, the sampling unit of the upper slope position is 2.06 Hm2, and the sampling unit of the downslope position is 1.81 hm2. The sampling unit of the valley is 4.91hm2. (5) using the simulated annealing algorithm optimized soil sample point and BP neural network model to digitize the spatial distribution of the soil nutrient index. The precision is more reliable, the spatial distribution features conform to the actual situation, and the scientific management of soil nutrients in the hilly area, the digital agriculture and the essence. The implementation of quasi agriculture provides important theoretical support.
【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S158

【参考文献】