基于GARBF神经网络的耕地土壤有效磷空间变异分析

为了调整耕地管理措施、合理施用磷肥、减少磷素流失、降低水体非点源污染,该研究以高州市为例,在全市各区镇共采集了664个耕作层(0～20cm)土样,利用遗传算法优化的径向基函数(radial basis function network optimized by geneti calgorithm,GARBF)神经网络和普通克里金法(Ordinary Kriging)等方法,分析了县域耕地土壤有效磷在不同采样尺度下的空间变异特征及其空间分布格局与成因。结果表明,高州市耕地表层土壤有效磷存在半方差结构,半方差函数曲线与指数和球状模型曲线拟合较好;5种采样尺度下(训练样点数分别为100、200、300、400和500)耕地表层土壤有效磷均表现出弱的结构空间相关,在较大范围内空间自相关性较差。GARBF神经网络空间插值能力在整体上要有优于基于邻近点RBF神经网络和普通克里金法。300样本下GARBF神经网络空间插值结果表明,高州市耕地表层土壤有效磷的盈余现象比较严重,并且盈余有效磷的流失对该地区水环境会产生严重的威胁。该研究结果可以为土壤属性空间估测、合理施肥以及降低水体非点源污染提供理论依据和技术支持。

Spatial variability of available phosphorus for cultivated soil based on GARBF neural network

In order to adjust land management measures, use phosphorus fertilizer properly, minimize phosphorus loss and mitigate non-point source pollution of water, 664 soil samples in cultivate horizon were collected in Gaozhou city, Guangdong Province in this study. The radial basis function network optimized by genetic algorithm (GARBF) and Ordinary Kriging methods were applied to reveal the characteristics of spatial variability of cultivated soil variability phosphorus (AP) and its spatial distribution pattern. The results suggested that spatial variability of surface soil AP of cultivated land in Gaozhou city exhibited semi-variance structure, and its semi-variance function fitted exponential and spherical models well. The analysis showed that the spatial correlation in surface cultivated soil AP was weak in the five sampling scales (training sample points were 100, 200, 300, 400 and 500), while unobvious in wide range. Predictions of soil AP in simulation using GARBF neural network was better than that using radial basis function(RBF) neural network (Near-RBF) prediction model based on several closest neighbors and Ordinary Kriging method. In practical application, the spatial interpolation map by GARBF neural network method with 300 soil samples showed a serious trend of surplus phosphate in cropping in Gaozhou city. Diffusive surplus phosphorus made a serious threat to the water environment in this region. The results provide a theoretical basis and technical support for predicting soil property spatial distribution accurately, using fertilizer properly and mitigating non-point source pollution of water.